From 0ab2d7532b05a3e7c06fd3b0c8bd6b46c1dfb508 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Fri, 25 Mar 2022 14:55:30 -0400
Subject: drm/amdkfd: prepare per-process debug enable and disable

The ROCm debugger will attach to a process to debug by PTRACE and will
expect the KFD to prepare a process for the target PID, whether the
target PID has opened the KFD device or not.

This patch is to explicity handle this requirement.  Further HW mode
setting and runtime coordination requirements will be handled in
following patches.

In the case where the target process has not opened the KFD device,
a new KFD process must be created for the target PID.
The debugger as well as the target process for this case will have not
acquired any VMs so handle process restoration to correctly account for
this.

To coordinate with HSA runtime, the debugger must be aware of the target
process' runtime enablement status and will copy the runtime status
information into the debugged KFD process for later query.

On enablement, the debugger will subscribe to a set of exceptions where
each exception events will notify the debugger through a pollable FIFO
file descriptor that the debugger provides to the KFD to manage.

Finally on process termination of either the debugger or the target,
debugging must be disabled if it has not been done so.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h | 32 ++++++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)
 create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_debug.h

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
new file mode 100644
index 000000000000..a8abfe2a0a14
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -0,0 +1,32 @@
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef KFD_DEBUG_EVENTS_H_INCLUDED
+#define KFD_DEBUG_EVENTS_H_INCLUDED
+
+#include "kfd_priv.h"
+
+int kfd_dbg_trap_disable(struct kfd_process *target);
+int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
+			void __user *runtime_info,
+			uint32_t *runtime_info_size);
+#endif
-- 
cgit 


From 0de4ec9a03537bd2b189b5afbf83acd6b72b0258 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Mon, 4 Apr 2022 13:38:11 -0400
Subject: drm/amdgpu: prepare map process for multi-process debug devices

Unlike single process debug devices, multi-process debug devices allow
debug mode setting per-VMID (non-device-global).

Because the HWS manages PASID-VMID mapping, the new MAP_PROCESS API allows
the KFD to forward the required SPI debug register write requests.

To request a new debug mode setting change, the KFD must be able to
preempt all queues then remap all queues with these new setting
requests for MAP_PROCESS to take effect.

Note that by default, trap enablement in non-debug mode must be disabled
for performance reasons for multi-process debug devices due to setup
overhead in FW.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h             |  5 +++
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c  | 51 ++++++++++++++++++++++
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.h  |  3 ++
 drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c | 14 ++++++
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h              |  9 ++++
 drivers/gpu/drm/amd/amdkfd/kfd_process.c           |  5 +++
 6 files changed, 87 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index a8abfe2a0a14..db6d72e7930f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -29,4 +29,9 @@ int kfd_dbg_trap_disable(struct kfd_process *target);
 int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 			void __user *runtime_info,
 			uint32_t *runtime_info_size);
+static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
+{
+	return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2);
+}
+
 #endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index cfe5bd59070e..495c9238254e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -36,6 +36,7 @@
 #include "kfd_kernel_queue.h"
 #include "amdgpu_amdkfd.h"
 #include "mes_api_def.h"
+#include "kfd_debug.h"
 
 /* Size of the per-pipe EOP queue */
 #define CIK_HPD_EOP_BYTES_LOG2 11
@@ -2594,6 +2595,56 @@ out_unlock:
 	return r;
 }
 
+int debug_lock_and_unmap(struct device_queue_manager *dqm)
+{
+	int r;
+
+	if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
+		pr_err("Unsupported on sched_policy: %i\n", dqm->sched_policy);
+		return -EINVAL;
+	}
+
+	if (!kfd_dbg_is_per_vmid_supported(dqm->dev))
+		return 0;
+
+	dqm_lock(dqm);
+
+	r = unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0, 0, false);
+	if (r)
+		dqm_unlock(dqm);
+
+	return r;
+}
+
+int debug_map_and_unlock(struct device_queue_manager *dqm)
+{
+	int r;
+
+	if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
+		pr_err("Unsupported on sched_policy: %i\n", dqm->sched_policy);
+		return -EINVAL;
+	}
+
+	if (!kfd_dbg_is_per_vmid_supported(dqm->dev))
+		return 0;
+
+	r = map_queues_cpsch(dqm);
+
+	dqm_unlock(dqm);
+
+	return r;
+}
+
+int debug_refresh_runlist(struct device_queue_manager *dqm)
+{
+	int r = debug_lock_and_unmap(dqm);
+
+	if (r)
+		return r;
+
+	return debug_map_and_unlock(dqm);
+}
+
 #if defined(CONFIG_DEBUG_FS)
 
 static void seq_reg_dump(struct seq_file *m,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index bf7aa3f84182..bb75d93712eb 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -290,6 +290,9 @@ int reserve_debug_trap_vmid(struct device_queue_manager *dqm,
 			struct qcm_process_device *qpd);
 int release_debug_trap_vmid(struct device_queue_manager *dqm,
 			struct qcm_process_device *qpd);
+int debug_lock_and_unmap(struct device_queue_manager *dqm);
+int debug_map_and_unlock(struct device_queue_manager *dqm);
+int debug_refresh_runlist(struct device_queue_manager *dqm);
 
 static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
 {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
index 0fe73dbd28af..29a2d0499b67 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
@@ -88,6 +88,10 @@ static int pm_map_process_aldebaran(struct packet_manager *pm,
 {
 	struct pm4_mes_map_process_aldebaran *packet;
 	uint64_t vm_page_table_base_addr = qpd->page_table_base;
+	struct kfd_dev *kfd = pm->dqm->dev->kfd;
+	struct kfd_process_device *pdd =
+			container_of(qpd, struct kfd_process_device, qpd);
+	int i;
 
 	packet = (struct pm4_mes_map_process_aldebaran *)buffer;
 	memset(buffer, 0, sizeof(struct pm4_mes_map_process_aldebaran));
@@ -102,6 +106,16 @@ static int pm_map_process_aldebaran(struct packet_manager *pm,
 	packet->bitfields14.num_oac = qpd->num_oac;
 	packet->bitfields14.sdma_enable = 1;
 	packet->bitfields14.num_queues = (qpd->is_debug) ? 0 : qpd->queue_count;
+	packet->spi_gdbg_per_vmid_cntl = pdd->spi_dbg_override |
+						pdd->spi_dbg_launch_mode;
+
+	if (pdd->process->debug_trap_enabled) {
+		for (i = 0; i < kfd->device_info.num_of_watch_points; i++)
+			packet->tcp_watch_cntl[i] = pdd->watch_points[i];
+
+		packet->bitfields2.single_memops =
+				!!(pdd->process->dbg_flags & KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP);
+	}
 
 	packet->sh_mem_config = qpd->sh_mem_config;
 	packet->sh_mem_bases = qpd->sh_mem_bases;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 4c912b7735b5..8fca7175daab 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -816,6 +816,12 @@ struct kfd_process_device {
 	uint64_t faults;
 	uint64_t page_in;
 	uint64_t page_out;
+
+	/* Tracks debug per-vmid request settings */
+	uint32_t spi_dbg_override;
+	uint32_t spi_dbg_launch_mode;
+	uint32_t watch_points[4];
+
 	/*
 	 * If this process has been checkpointed before, then the user
 	 * application will use the original gpu_id on the
@@ -952,6 +958,9 @@ struct kfd_process {
 
 	bool xnack_enabled;
 
+	/* Tracks debug per-vmid request for debug flags */
+	bool dbg_flags;
+
 	atomic_t poison;
 	/* Queues are in paused stated because we are in the process of doing a CRIU checkpoint */
 	bool queues_paused;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index d75dac92775c..725d936b2cc7 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -1612,6 +1612,11 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev,
 	}
 
 	p->pdds[p->n_pdds++] = pdd;
+	if (kfd_dbg_is_per_vmid_supported(pdd->dev))
+		pdd->spi_dbg_override = pdd->dev->kfd2kgd->disable_debug_trap(
+							pdd->dev->adev,
+							false,
+							0);
 
 	/* Init idr used for memory handle translation */
 	idr_init(&pdd->alloc_idr);
-- 
cgit 


From 218895820e6fccade42a7c3ab9c0a44dec0a1ebc Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Tue, 5 Apr 2022 12:34:55 -0400
Subject: drm/amdkfd: add per process hw trap enable and disable functions

To enable HW debug mode per process, all devices must be debug enabled
successfully.  If a failure occures, rewind the enablement of debug mode
on the enabled devices.

A power management scenario that needs to be considered is HW
debug mode setting during GFXOFF.  During GFXOFF, these registers
will be unreachable so we have to transiently disable GFXOFF when
setting.  Also, some devices don't support the RLC save restore
function for these debug registers so we have to disable GFXOFF
completely during a debug session.

Cooperative launch also has debugging restriction based on HW/FW bugs.
If such bugs exists, the debugger cannot attach to a process that uses GWS
resources nor can GWS resources be requested if a process is being
debugged.

Multi-process debug devices can only enable trap temporaries based
on certain runtime scenerios, which will be explained when the
runtime enable functions are implemented in a follow up patch.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c |   5 ++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 148 ++++++++++++++++++++++++++++++-
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |  29 ++++++
 drivers/gpu/drm/amd/amdkfd/kfd_process.c |  10 +++
 4 files changed, 190 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index ee086a0a46df..826a99acb6fb 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -1488,6 +1488,11 @@ static int kfd_ioctl_alloc_queue_gws(struct file *filep,
 		goto out_unlock;
 	}
 
+	if (!kfd_dbg_has_gws_support(dev) && p->debug_trap_enabled) {
+		retval = -EBUSY;
+		goto out_unlock;
+	}
+
 	retval = pqm_set_gws(&p->pqm, args->queue_id, args->num_gws ? dev->gws : NULL);
 	mutex_unlock(&p->mutex);
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 898cc1fe3d13..73b07b5f17f1 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -21,13 +21,78 @@
  */
 
 #include "kfd_debug.h"
+#include "kfd_device_queue_manager.h"
 #include <linux/file.h>
 
+static int kfd_dbg_set_mes_debug_mode(struct kfd_process_device *pdd)
+{
+	uint32_t spi_dbg_cntl = pdd->spi_dbg_override | pdd->spi_dbg_launch_mode;
+	uint32_t flags = pdd->process->dbg_flags;
+
+	if (!kfd_dbg_is_per_vmid_supported(pdd->dev))
+		return 0;
+
+	return amdgpu_mes_set_shader_debugger(pdd->dev->adev, pdd->proc_ctx_gpu_addr, spi_dbg_cntl,
+						pdd->watch_points, flags);
+}
+
+/* kfd_dbg_trap_deactivate:
+ *	target: target process
+ *	unwind: If this is unwinding a failed kfd_dbg_trap_enable()
+ *	unwind_count:
+ *		If unwind == true, how far down the pdd list we need
+ *				to unwind
+ *		else: ignored
+ */
+static void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind_count)
+{
+	int i;
+
+	for (i = 0; i < target->n_pdds; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+
+		/* If this is an unwind, and we have unwound the required
+		 * enable calls on the pdd list, we need to stop now
+		 * otherwise we may mess up another debugger session.
+		 */
+		if (unwind && i == unwind_count)
+			break;
+
+		/* GFX off is already disabled by debug activate if not RLC restore supported. */
+		if (kfd_dbg_is_rlc_restore_supported(pdd->dev))
+			amdgpu_gfx_off_ctrl(pdd->dev->adev, false);
+		pdd->spi_dbg_override =
+				pdd->dev->kfd2kgd->disable_debug_trap(
+				pdd->dev->adev,
+				target->runtime_info.ttmp_setup,
+				pdd->dev->vm_info.last_vmid_kfd);
+		amdgpu_gfx_off_ctrl(pdd->dev->adev, true);
+
+		if (!kfd_dbg_is_per_vmid_supported(pdd->dev) &&
+				release_debug_trap_vmid(pdd->dev->dqm, &pdd->qpd))
+			pr_err("Failed to release debug vmid on [%i]\n", pdd->dev->id);
+
+		if (!pdd->dev->kfd->shared_resources.enable_mes)
+			debug_refresh_runlist(pdd->dev->dqm);
+		else
+			kfd_dbg_set_mes_debug_mode(pdd);
+	}
+}
+
 int kfd_dbg_trap_disable(struct kfd_process *target)
 {
 	if (!target->debug_trap_enabled)
 		return 0;
 
+	/*
+	 * Defer deactivation to runtime if runtime not enabled otherwise reset
+	 * attached running target runtime state to enable for re-attach.
+	 */
+	if (target->runtime_info.runtime_state == DEBUG_RUNTIME_STATE_ENABLED)
+		kfd_dbg_trap_deactivate(target, false, 0);
+	else if (target->runtime_info.runtime_state != DEBUG_RUNTIME_STATE_DISABLED)
+		target->runtime_info.runtime_state = DEBUG_RUNTIME_STATE_ENABLED;
+
 	fput(target->dbg_ev_file);
 	target->dbg_ev_file = NULL;
 
@@ -42,16 +107,89 @@ int kfd_dbg_trap_disable(struct kfd_process *target)
 	return 0;
 }
 
+static int kfd_dbg_trap_activate(struct kfd_process *target)
+{
+	int i, r = 0;
+
+	for (i = 0; i < target->n_pdds; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+
+		if (!kfd_dbg_is_per_vmid_supported(pdd->dev)) {
+			r = reserve_debug_trap_vmid(pdd->dev->dqm, &pdd->qpd);
+
+			if (r) {
+				target->runtime_info.runtime_state = (r == -EBUSY) ?
+							DEBUG_RUNTIME_STATE_ENABLED_BUSY :
+							DEBUG_RUNTIME_STATE_ENABLED_ERROR;
+
+				goto unwind_err;
+			}
+		}
+
+		/* Disable GFX OFF to prevent garbage read/writes to debug registers.
+		 * If RLC restore of debug registers is not supported and runtime enable
+		 * hasn't done so already on ttmp setup request, restore the trap config registers.
+		 *
+		 * If RLC restore of debug registers is not supported, keep gfx off disabled for
+		 * the debug session.
+		 */
+		amdgpu_gfx_off_ctrl(pdd->dev->adev, false);
+		if (!(kfd_dbg_is_rlc_restore_supported(pdd->dev) ||
+						target->runtime_info.ttmp_setup))
+			pdd->dev->kfd2kgd->enable_debug_trap(pdd->dev->adev, true,
+								pdd->dev->vm_info.last_vmid_kfd);
+
+		pdd->spi_dbg_override = pdd->dev->kfd2kgd->enable_debug_trap(
+					pdd->dev->adev,
+					false,
+					pdd->dev->vm_info.last_vmid_kfd);
+
+		if (kfd_dbg_is_rlc_restore_supported(pdd->dev))
+			amdgpu_gfx_off_ctrl(pdd->dev->adev, true);
+
+		if (!pdd->dev->kfd->shared_resources.enable_mes)
+			r = debug_refresh_runlist(pdd->dev->dqm);
+		else
+			r = kfd_dbg_set_mes_debug_mode(pdd);
+
+		if (r) {
+			target->runtime_info.runtime_state =
+					DEBUG_RUNTIME_STATE_ENABLED_ERROR;
+			goto unwind_err;
+		}
+	}
+
+	return 0;
+
+unwind_err:
+	/* Enabling debug failed, we need to disable on
+	 * all GPUs so the enable is all or nothing.
+	 */
+	kfd_dbg_trap_deactivate(target, true, i);
+	return r;
+}
+
 int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 			void __user *runtime_info, uint32_t *runtime_size)
 {
 	struct file *f;
 	uint32_t copy_size;
-	int r = 0;
+	int i, r = 0;
 
 	if (target->debug_trap_enabled)
 		return -EALREADY;
 
+	/* Enable pre-checks */
+	for (i = 0; i < target->n_pdds; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+
+		if (!KFD_IS_SOC15(pdd->dev))
+			return -ENODEV;
+
+		if (!kfd_dbg_has_gws_support(pdd->dev) && pdd->qpd.num_gws)
+			return -EBUSY;
+	}
+
 	copy_size = min((size_t)(*runtime_size), sizeof(target->runtime_info));
 
 	f = fget(fd);
@@ -62,6 +200,10 @@ int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 
 	target->dbg_ev_file = f;
 
+	/* defer activation to runtime if not runtime enabled */
+	if (target->runtime_info.runtime_state == DEBUG_RUNTIME_STATE_ENABLED)
+		kfd_dbg_trap_activate(target);
+
 	/* We already hold the process reference but hold another one for the
 	 * debug session.
 	 */
@@ -71,8 +213,10 @@ int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 	if (target->debugger_process)
 		atomic_inc(&target->debugger_process->debugged_process_count);
 
-	if (copy_to_user(runtime_info, (void *)&target->runtime_info, copy_size))
+	if (copy_to_user(runtime_info, (void *)&target->runtime_info, copy_size)) {
+		kfd_dbg_trap_deactivate(target, false, 0);
 		r = -EFAULT;
+	}
 
 	*runtime_size = sizeof(target->runtime_info);
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index db6d72e7930f..17481f824647 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -34,4 +34,33 @@ static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
 	return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2);
 }
 
+/*
+ * If GFX off is enabled, chips that do not support RLC restore for the debug
+ * registers will disable GFX off temporarily for the entire debug session.
+ * See disable_on_trap_action_entry and enable_on_trap_action_exit for details.
+ */
+static inline bool kfd_dbg_is_rlc_restore_supported(struct kfd_node *dev)
+{
+	return !(KFD_GC_VERSION(dev) == IP_VERSION(10, 1, 10) ||
+		 KFD_GC_VERSION(dev) == IP_VERSION(10, 1, 1));
+}
+
+static inline bool kfd_dbg_has_gws_support(struct kfd_node *dev)
+{
+	if ((KFD_GC_VERSION(dev) == IP_VERSION(9, 0, 1)
+			&& dev->kfd->mec2_fw_version < 0x81b6) ||
+		(KFD_GC_VERSION(dev) >= IP_VERSION(9, 1, 0)
+			&& KFD_GC_VERSION(dev) <= IP_VERSION(9, 2, 2)
+			&& dev->kfd->mec2_fw_version < 0x1b6) ||
+		(KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 0)
+			&& dev->kfd->mec2_fw_version < 0x1b6) ||
+		(KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 1)
+			&& dev->kfd->mec2_fw_version < 0x30) ||
+		(KFD_GC_VERSION(dev) >= IP_VERSION(11, 0, 0) &&
+			KFD_GC_VERSION(dev) < IP_VERSION(12, 0, 0)))
+		return false;
+
+	/* Assume debugging and cooperative launch supported otherwise. */
+	return true;
+}
 #endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 725d936b2cc7..e77cadadb09b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -1165,9 +1165,19 @@ static void kfd_process_free_notifier(struct mmu_notifier *mn)
 
 static void kfd_process_notifier_release_internal(struct kfd_process *p)
 {
+	int i;
+
 	cancel_delayed_work_sync(&p->eviction_work);
 	cancel_delayed_work_sync(&p->restore_work);
 
+	for (i = 0; i < p->n_pdds; i++) {
+		struct kfd_process_device *pdd = p->pdds[i];
+
+		/* re-enable GFX OFF since runtime enable with ttmp setup disabled it. */
+		if (!kfd_dbg_is_rlc_restore_supported(pdd->dev) && p->runtime_info.ttmp_setup)
+			amdgpu_gfx_off_ctrl(pdd->dev->adev, true);
+	}
+
 	/* Indicate to other users that MM is no longer valid */
 	p->mm = NULL;
 	kfd_dbg_trap_disable(p);
-- 
cgit 


From 69a8c3ae2dea84a6d571e4c1aad306f630f3ccfd Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Thu, 1 Sep 2022 11:27:15 -0400
Subject: drm/amdkfd: apply trap workaround for gfx11

Due to a HW bug, waves in only half the shader arrays can enter trap.

When starting a debug session, relocate all waves to the first shader
array of each shader engine and mask off the 2nd shader array as
unavailable.

When ending a debug session, re-enable the 2nd shader array per
shader engine.

User CU masking per queue cannot be guaranteed to remain functional
if requested during debugging (e.g. user cu mask requests only 2nd shader
array as an available resource leading to zero HW resources available)
nor can runtime be alerted of any of these changes during execution.

Make user CU masking and debugging mutual exclusive with respect to
availability.

If the debugger tries to attach to a process with a user cu masked
queue, return the runtime status as enabled but busy.

If the debugger tries to attach and fails to reallocate queue waves to
the first shader array of each shader engine, return the runtime status
as enabled but with an error.

In addition, like any other mutli-process debug supported devices,
disable trap temporary setup per-process to avoid performance impact from
setup overhead.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h            |  2 +
 drivers/gpu/drm/amd/amdgpu/mes_v11_0.c             |  7 +--
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c           |  2 -
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c             | 57 ++++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h             |  3 +-
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c  |  7 +++
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c   |  3 +-
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c   |  3 +-
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c   | 42 +++++++++++-----
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c    |  3 +-
 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c    |  3 +-
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h              |  5 +-
 .../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c |  9 +++-
 drivers/gpu/drm/amd/amdkfd/kfd_topology.c          |  7 ++-
 14 files changed, 122 insertions(+), 31 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
index d20df0cf0d88..b5f5eed2b5ef 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h
@@ -219,6 +219,8 @@ struct mes_add_queue_input {
 	uint32_t        gws_size;
 	uint64_t	tba_addr;
 	uint64_t	tma_addr;
+	uint32_t	trap_en;
+	uint32_t	skip_process_ctx_clear;
 	uint32_t	is_kfd_process;
 	uint32_t	is_aql_queue;
 	uint32_t	queue_size;
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
index 861910a6662d..c4e3cb8d44de 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
@@ -202,17 +202,14 @@ static int mes_v11_0_add_hw_queue(struct amdgpu_mes *mes,
 	mes_add_queue_pkt.gws_size = input->gws_size;
 	mes_add_queue_pkt.trap_handler_addr = input->tba_addr;
 	mes_add_queue_pkt.tma_addr = input->tma_addr;
+	mes_add_queue_pkt.trap_en = input->trap_en;
+	mes_add_queue_pkt.skip_process_ctx_clear = input->skip_process_ctx_clear;
 	mes_add_queue_pkt.is_kfd_process = input->is_kfd_process;
 
 	/* For KFD, gds_size is re-used for queue size (needed in MES for AQL queues) */
 	mes_add_queue_pkt.is_aql_queue = input->is_aql_queue;
 	mes_add_queue_pkt.gds_size = input->queue_size;
 
-	if (!(((adev->mes.sched_version & AMDGPU_MES_VERSION_MASK) >= 4) &&
-		  (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(11, 0, 0)) &&
-		  (adev->ip_versions[GC_HWIP][0] <= IP_VERSION(11, 0, 3))))
-		mes_add_queue_pkt.trap_en = 1;
-
 	/* For KFD, gds_size is re-used for queue size (needed in MES for AQL queues) */
 	mes_add_queue_pkt.is_aql_queue = input->is_aql_queue;
 	mes_add_queue_pkt.gds_size = input->queue_size;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 826a99acb6fb..d4df424e4514 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -537,8 +537,6 @@ static int kfd_ioctl_set_cu_mask(struct file *filp, struct kfd_process *p,
 		goto out;
 	}
 
-	minfo.update_flag = UPDATE_FLAG_CU_MASK;
-
 	mutex_lock(&p->mutex);
 
 	retval = pqm_update_mqd(&p->pqm, args->queue_id, &minfo);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 73b07b5f17f1..5e2ee2d1acc4 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -24,6 +24,57 @@
 #include "kfd_device_queue_manager.h"
 #include <linux/file.h>
 
+static int kfd_dbg_set_queue_workaround(struct queue *q, bool enable)
+{
+	struct mqd_update_info minfo = {0};
+	int err;
+
+	if (!q)
+		return 0;
+
+	if (KFD_GC_VERSION(q->device) < IP_VERSION(11, 0, 0) ||
+	    KFD_GC_VERSION(q->device) >= IP_VERSION(12, 0, 0))
+		return 0;
+
+	if (enable && q->properties.is_user_cu_masked)
+		return -EBUSY;
+
+	minfo.update_flag = enable ? UPDATE_FLAG_DBG_WA_ENABLE : UPDATE_FLAG_DBG_WA_DISABLE;
+
+	q->properties.is_dbg_wa = enable;
+	err = q->device->dqm->ops.update_queue(q->device->dqm, q, &minfo);
+	if (err)
+		q->properties.is_dbg_wa = false;
+
+	return err;
+}
+
+static int kfd_dbg_set_workaround(struct kfd_process *target, bool enable)
+{
+	struct process_queue_manager *pqm = &target->pqm;
+	struct process_queue_node *pqn;
+	int r = 0;
+
+	list_for_each_entry(pqn, &pqm->queues, process_queue_list) {
+		r = kfd_dbg_set_queue_workaround(pqn->q, enable);
+		if (enable && r)
+			goto unwind;
+	}
+
+	return 0;
+
+unwind:
+	list_for_each_entry(pqn, &pqm->queues, process_queue_list)
+		kfd_dbg_set_queue_workaround(pqn->q, false);
+
+	if (enable)
+		target->runtime_info.runtime_state = r == -EBUSY ?
+				DEBUG_RUNTIME_STATE_ENABLED_BUSY :
+				DEBUG_RUNTIME_STATE_ENABLED_ERROR;
+
+	return r;
+}
+
 static int kfd_dbg_set_mes_debug_mode(struct kfd_process_device *pdd)
 {
 	uint32_t spi_dbg_cntl = pdd->spi_dbg_override | pdd->spi_dbg_launch_mode;
@@ -77,6 +128,8 @@ static void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int
 		else
 			kfd_dbg_set_mes_debug_mode(pdd);
 	}
+
+	kfd_dbg_set_workaround(target, false);
 }
 
 int kfd_dbg_trap_disable(struct kfd_process *target)
@@ -111,6 +164,10 @@ static int kfd_dbg_trap_activate(struct kfd_process *target)
 {
 	int i, r = 0;
 
+	r = kfd_dbg_set_workaround(target, true);
+	if (r)
+		return r;
+
 	for (i = 0; i < target->n_pdds; i++) {
 		struct kfd_process_device *pdd = target->pdds[i];
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 17481f824647..3e56225f6ef6 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -31,7 +31,8 @@ int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 			uint32_t *runtime_info_size);
 static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
 {
-	return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2);
+	return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2) ||
+	       KFD_GC_VERSION(dev) >= IP_VERSION(11, 0, 0);
 }
 
 /*
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index 495c9238254e..44d87943e40a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -226,6 +226,10 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,
 	queue_input.paging = false;
 	queue_input.tba_addr = qpd->tba_addr;
 	queue_input.tma_addr = qpd->tma_addr;
+	queue_input.trap_en = KFD_GC_VERSION(q->device) < IP_VERSION(11, 0, 0) ||
+			      KFD_GC_VERSION(q->device) >= IP_VERSION(12, 0, 0) ||
+			      q->properties.is_dbg_wa;
+	queue_input.skip_process_ctx_clear = qpd->pqm->process->debug_trap_enabled;
 
 	queue_type = convert_to_mes_queue_type(q->properties.type);
 	if (queue_type < 0) {
@@ -1716,6 +1720,9 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
 	 * updates the is_evicted flag but is a no-op otherwise.
 	 */
 	q->properties.is_evicted = !!qpd->evicted;
+	q->properties.is_dbg_wa = qpd->pqm->process->debug_trap_enabled &&
+			KFD_GC_VERSION(q->device) >= IP_VERSION(11, 0, 0) &&
+			KFD_GC_VERSION(q->device) < IP_VERSION(12, 0, 0);
 
 	if (qd)
 		mqd_mgr->restore_mqd(mqd_mgr, &q->mqd, q->mqd_mem_obj, &q->gart_mqd_addr,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
index eb11940bec34..65c9f01a1f86 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
@@ -48,8 +48,7 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd,
 	struct cik_mqd *m;
 	uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
 
-	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
-	    !minfo->cu_mask.ptr)
+	if (!minfo || !minfo->cu_mask.ptr)
 		return;
 
 	mqd_symmetrically_map_cu_mask(mm,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
index eaaa4f4ddaaa..a0ac4f2fe6b5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
@@ -48,8 +48,7 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd,
 	struct v10_compute_mqd *m;
 	uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
 
-	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
-	    !minfo->cu_mask.ptr)
+	if (!minfo || !minfo->cu_mask.ptr)
 		return;
 
 	mqd_symmetrically_map_cu_mask(mm,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c
index 3a48bbc589fe..9a9b4e853516 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c
@@ -46,15 +46,33 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd,
 {
 	struct v11_compute_mqd *m;
 	uint32_t se_mask[KFD_MAX_NUM_SE] = {0};
+	bool has_wa_flag = minfo && (minfo->update_flag & (UPDATE_FLAG_DBG_WA_ENABLE |
+			UPDATE_FLAG_DBG_WA_DISABLE));
 
-	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
-	    !minfo->cu_mask.ptr)
+	if (!minfo || !(has_wa_flag || minfo->cu_mask.ptr))
 		return;
 
+	m = get_mqd(mqd);
+
+	if (has_wa_flag) {
+		uint32_t wa_mask = minfo->update_flag == UPDATE_FLAG_DBG_WA_ENABLE ?
+						0xffff : 0xffffffff;
+
+		m->compute_static_thread_mgmt_se0 = wa_mask;
+		m->compute_static_thread_mgmt_se1 = wa_mask;
+		m->compute_static_thread_mgmt_se2 = wa_mask;
+		m->compute_static_thread_mgmt_se3 = wa_mask;
+		m->compute_static_thread_mgmt_se4 = wa_mask;
+		m->compute_static_thread_mgmt_se5 = wa_mask;
+		m->compute_static_thread_mgmt_se6 = wa_mask;
+		m->compute_static_thread_mgmt_se7 = wa_mask;
+
+		return;
+	}
+
 	mqd_symmetrically_map_cu_mask(mm,
 		minfo->cu_mask.ptr, minfo->cu_mask.count, se_mask);
 
-	m = get_mqd(mqd);
 	m->compute_static_thread_mgmt_se0 = se_mask[0];
 	m->compute_static_thread_mgmt_se1 = se_mask[1];
 	m->compute_static_thread_mgmt_se2 = se_mask[2];
@@ -109,6 +127,7 @@ static void init_mqd(struct mqd_manager *mm, void **mqd,
 	uint64_t addr;
 	struct v11_compute_mqd *m;
 	int size;
+	uint32_t wa_mask = q->is_dbg_wa ? 0xffff : 0xffffffff;
 
 	m = (struct v11_compute_mqd *) mqd_mem_obj->cpu_ptr;
 	addr = mqd_mem_obj->gpu_addr;
@@ -122,14 +141,15 @@ static void init_mqd(struct mqd_manager *mm, void **mqd,
 
 	m->header = 0xC0310800;
 	m->compute_pipelinestat_enable = 1;
-	m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se4 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se5 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se6 = 0xFFFFFFFF;
-	m->compute_static_thread_mgmt_se7 = 0xFFFFFFFF;
+
+	m->compute_static_thread_mgmt_se0 = wa_mask;
+	m->compute_static_thread_mgmt_se1 = wa_mask;
+	m->compute_static_thread_mgmt_se2 = wa_mask;
+	m->compute_static_thread_mgmt_se3 = wa_mask;
+	m->compute_static_thread_mgmt_se4 = wa_mask;
+	m->compute_static_thread_mgmt_se5 = wa_mask;
+	m->compute_static_thread_mgmt_se6 = wa_mask;
+	m->compute_static_thread_mgmt_se7 = wa_mask;
 
 	m->cp_hqd_persistent_state = CP_HQD_PERSISTENT_STATE__PRELOAD_REQ_MASK |
 			0x55 << CP_HQD_PERSISTENT_STATE__PRELOAD_SIZE__SHIFT;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
index b7c95158d4a0..5b87c244e909 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
@@ -65,8 +65,7 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd,
 	struct v9_mqd *m;
 	uint32_t se_mask[KFD_MAX_NUM_SE] = {0};
 
-	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
-	    !minfo->cu_mask.ptr)
+	if (!minfo || !minfo->cu_mask.ptr)
 		return;
 
 	mqd_symmetrically_map_cu_mask(mm,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
index fe69492b1bb3..d1e962da51dd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
@@ -51,8 +51,7 @@ static void update_cu_mask(struct mqd_manager *mm, void *mqd,
 	struct vi_mqd *m;
 	uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
 
-	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
-	    !minfo->cu_mask.ptr)
+	if (!minfo || !minfo->cu_mask.ptr)
 		return;
 
 	mqd_symmetrically_map_cu_mask(mm,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 8fca7175daab..f0a45d184c8f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -513,6 +513,8 @@ struct queue_properties {
 	bool is_active;
 	bool is_gws;
 	uint32_t pm4_target_xcc;
+	bool is_dbg_wa;
+	bool is_user_cu_masked;
 	/* Not relevant for user mode queues in cp scheduling */
 	unsigned int vmid;
 	/* Relevant only for sdma queues*/
@@ -535,7 +537,8 @@ struct queue_properties {
 			    !(q).is_evicted)
 
 enum mqd_update_flag {
-	UPDATE_FLAG_CU_MASK = 0,
+	UPDATE_FLAG_DBG_WA_ENABLE = 1,
+	UPDATE_FLAG_DBG_WA_DISABLE = 2,
 };
 
 struct mqd_update_info {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index b100933340d2..43d432b5c5bc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -506,8 +506,12 @@ int pqm_update_mqd(struct process_queue_manager *pqm,
 		return -EFAULT;
 	}
 
+	/* CUs are masked for debugger requirements so deny user mask  */
+	if (pqn->q->properties.is_dbg_wa && minfo && minfo->cu_mask.ptr)
+		return -EBUSY;
+
 	/* ASICs that have WGPs must enforce pairwise enabled mask checks. */
-	if (minfo && minfo->update_flag == UPDATE_FLAG_CU_MASK && minfo->cu_mask.ptr &&
+	if (minfo && minfo->cu_mask.ptr &&
 			KFD_GC_VERSION(pqn->q->device) >= IP_VERSION(10, 0, 0)) {
 		int i;
 
@@ -526,6 +530,9 @@ int pqm_update_mqd(struct process_queue_manager *pqm,
 	if (retval != 0)
 		return retval;
 
+	if (minfo && minfo->cu_mask.ptr)
+		pqn->q->properties.is_user_cu_masked = true;
+
 	return 0;
 }
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index 3def25b2bdbb..faa7939f35bd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -1863,10 +1863,13 @@ static void kfd_topology_set_dbg_firmware_support(struct kfd_topology_device *de
 {
 	bool firmware_supported = true;
 
+	/*
+	 * FIXME: GFX11 FW currently not sufficient to deal with CWSR WA.
+	 * Updated FW with API changes coming soon.
+	 */
 	if (KFD_GC_VERSION(dev->gpu) >= IP_VERSION(11, 0, 0) &&
 			KFD_GC_VERSION(dev->gpu) < IP_VERSION(12, 0, 0)) {
-		firmware_supported =
-			(dev->gpu->adev->mes.sched_version & AMDGPU_MES_VERSION_MASK) >= 9;
+		firmware_supported = false;
 		goto out;
 	}
 
-- 
cgit 


From 44b87bb0836c65d1b9d21b01503eb6e9b9297771 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Wed, 6 Apr 2022 12:03:31 -0400
Subject: drm/amdkfd: add raise exception event function

Exception events can be generated from interrupts or queue activitity.

The raise event function will save exception status of a queue, device
or process then notify the debugger of the status change by writing to
a debugger polled file descriptor that the debugger provides during
debug attach.

For memory violation exceptions, extra exception data will be saved.

The debugger will be able to query the saved exception states by query
operation that will be provided by follow up patches.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 104 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |   7 +++
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h    |  10 +++
 drivers/gpu/drm/amd/amdkfd/kfd_process.c |   2 +
 4 files changed, 123 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 5e2ee2d1acc4..dccb27fc764b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -24,6 +24,107 @@
 #include "kfd_device_queue_manager.h"
 #include <linux/file.h>
 
+void debug_event_write_work_handler(struct work_struct *work)
+{
+	struct kfd_process *process;
+
+	static const char write_data = '.';
+	loff_t pos = 0;
+
+	process = container_of(work,
+			struct kfd_process,
+			debug_event_workarea);
+
+	kernel_write(process->dbg_ev_file, &write_data, 1, &pos);
+}
+
+/* update process/device/queue exception status, write to descriptor
+ * only if exception_status is enabled.
+ */
+bool kfd_dbg_ev_raise(uint64_t event_mask,
+			struct kfd_process *process, struct kfd_node *dev,
+			unsigned int source_id, bool use_worker,
+			void *exception_data, size_t exception_data_size)
+{
+	struct process_queue_manager *pqm;
+	struct process_queue_node *pqn;
+	int i;
+	static const char write_data = '.';
+	loff_t pos = 0;
+	bool is_subscribed = true;
+
+	if (!(process && process->debug_trap_enabled))
+		return false;
+
+	mutex_lock(&process->event_mutex);
+
+	if (event_mask & KFD_EC_MASK_DEVICE) {
+		for (i = 0; i < process->n_pdds; i++) {
+			struct kfd_process_device *pdd = process->pdds[i];
+
+			if (pdd->dev != dev)
+				continue;
+
+			pdd->exception_status |= event_mask & KFD_EC_MASK_DEVICE;
+
+			if (event_mask & KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION)) {
+				if (!pdd->vm_fault_exc_data) {
+					pdd->vm_fault_exc_data = kmemdup(
+							exception_data,
+							exception_data_size,
+							GFP_KERNEL);
+					if (!pdd->vm_fault_exc_data)
+						pr_debug("Failed to allocate exception data memory");
+				} else {
+					pr_debug("Debugger exception data not saved\n");
+					print_hex_dump_bytes("exception data: ",
+							DUMP_PREFIX_OFFSET,
+							exception_data,
+							exception_data_size);
+				}
+			}
+			break;
+		}
+	} else if (event_mask & KFD_EC_MASK_PROCESS) {
+		process->exception_status |= event_mask & KFD_EC_MASK_PROCESS;
+	} else {
+		pqm = &process->pqm;
+		list_for_each_entry(pqn, &pqm->queues,
+				process_queue_list) {
+			int target_id;
+
+			if (!pqn->q)
+				continue;
+
+			target_id = event_mask & KFD_EC_MASK(EC_QUEUE_NEW) ?
+					pqn->q->properties.queue_id :
+							pqn->q->doorbell_id;
+
+			if (pqn->q->device != dev || target_id != source_id)
+				continue;
+
+			pqn->q->properties.exception_status |= event_mask;
+			break;
+		}
+	}
+
+	if (process->exception_enable_mask & event_mask) {
+		if (use_worker)
+			schedule_work(&process->debug_event_workarea);
+		else
+			kernel_write(process->dbg_ev_file,
+					&write_data,
+					1,
+					&pos);
+	} else {
+		is_subscribed = false;
+	}
+
+	mutex_unlock(&process->event_mutex);
+
+	return is_subscribed;
+}
+
 static int kfd_dbg_set_queue_workaround(struct queue *q, bool enable)
 {
 	struct mqd_update_info minfo = {0};
@@ -99,6 +200,9 @@ static void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int
 {
 	int i;
 
+	if (!unwind)
+		cancel_work_sync(&target->debug_event_workarea);
+
 	for (i = 0; i < target->n_pdds; i++) {
 		struct kfd_process_device *pdd = target->pdds[i];
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 3e56225f6ef6..66ee7b95d08a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -25,6 +25,11 @@
 
 #include "kfd_priv.h"
 
+bool kfd_dbg_ev_raise(uint64_t event_mask,
+			struct kfd_process *process, struct kfd_node *dev,
+			unsigned int source_id, bool use_worker,
+			void *exception_data,
+			size_t exception_data_size);
 int kfd_dbg_trap_disable(struct kfd_process *target);
 int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 			void __user *runtime_info,
@@ -35,6 +40,8 @@ static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
 	       KFD_GC_VERSION(dev) >= IP_VERSION(11, 0, 0);
 }
 
+void debug_event_write_work_handler(struct work_struct *work);
+
 /*
  * If GFX off is enabled, chips that do not support RLC restore for the debug
  * registers will disable GFX off temporarily for the entire debug session.
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index f0a45d184c8f..b18cd4bf76bf 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -529,6 +529,7 @@ struct queue_properties {
 	uint32_t ctl_stack_size;
 	uint64_t tba_addr;
 	uint64_t tma_addr;
+	uint64_t exception_status;
 };
 
 #define QUEUE_IS_ACTIVE(q) ((q).queue_size > 0 &&	\
@@ -820,6 +821,11 @@ struct kfd_process_device {
 	uint64_t page_in;
 	uint64_t page_out;
 
+	/* Exception code status*/
+	uint64_t exception_status;
+	void *vm_fault_exc_data;
+	size_t vm_fault_exc_data_size;
+
 	/* Tracks debug per-vmid request settings */
 	uint32_t spi_dbg_override;
 	uint32_t spi_dbg_launch_mode;
@@ -955,12 +961,16 @@ struct kfd_process {
 
 	/* Exception code enable mask and status */
 	uint64_t exception_enable_mask;
+	uint64_t exception_status;
 
 	/* shared virtual memory registered by this process */
 	struct svm_range_list svms;
 
 	bool xnack_enabled;
 
+	/* Work area for debugger event writer worker. */
+	struct work_struct debug_event_workarea;
+
 	/* Tracks debug per-vmid request for debug flags */
 	bool dbg_flags;
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index e77cadadb09b..f904d6d6e01c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -1509,6 +1509,8 @@ static struct kfd_process *create_process(const struct task_struct *thread)
 	kfd_unref_process(process);
 	get_task_struct(process->lead_thread);
 
+	INIT_WORK(&process->debug_event_workarea, debug_event_write_work_handler);
+
 	return process;
 
 err_register_notifier:
-- 
cgit 


From c2d2588c702364ff53916ddd97e2b26fd4f4a317 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Fri, 8 Apr 2022 12:49:48 -0400
Subject: drm/amdkfd: add send exception operation

Add a debug operation that allows the debugger to send an exception
directly to runtime through a payload address.

For memory violations, normal vmfault signals will be applied to
notify runtime instead after passing in the saved exception data
when a memory violation was raised to the debugger.

For runtime exceptions, this will unblock the runtime enable
function which will be explained and implemented in a follow up
patch.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c |  4 +-
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c         |  5 ++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c           | 43 ++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h           |  6 ++
 drivers/gpu/drm/amd/amdkfd/kfd_events.c          |  3 +-
 drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c  |  2 +-
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h            |  7 ++-
 drivers/gpu/drm/amd/amdkfd/kfd_process.c         | 71 +++++++++++++++++++++++-
 8 files changed, 135 insertions(+), 6 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c
index 4ebfff6b6c55..795382b55e0a 100644
--- a/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c
+++ b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c
@@ -118,9 +118,9 @@ static void cik_event_interrupt_wq(struct kfd_node *dev,
 			return;
 
 		if (info.vmid == vmid)
-			kfd_signal_vm_fault_event(dev, pasid, &info);
+			kfd_signal_vm_fault_event(dev, pasid, &info, NULL);
 		else
-			kfd_signal_vm_fault_event(dev, pasid, NULL);
+			kfd_signal_vm_fault_event(dev, pasid, NULL, NULL);
 	}
 }
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index d4df424e4514..5e57b3e96ff9 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -2833,6 +2833,11 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 		r = kfd_dbg_trap_disable(target);
 		break;
 	case KFD_IOC_DBG_TRAP_SEND_RUNTIME_EVENT:
+		r = kfd_dbg_send_exception_to_runtime(target,
+				args->send_runtime_event.gpu_id,
+				args->send_runtime_event.queue_id,
+				args->send_runtime_event.exception_mask);
+		break;
 	case KFD_IOC_DBG_TRAP_SET_EXCEPTIONS_ENABLED:
 	case KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_OVERRIDE:
 	case KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_MODE:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index dccb27fc764b..61098975bb0e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -125,6 +125,49 @@ bool kfd_dbg_ev_raise(uint64_t event_mask,
 	return is_subscribed;
 }
 
+int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
+					unsigned int dev_id,
+					unsigned int queue_id,
+					uint64_t error_reason)
+{
+	if (error_reason & KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION)) {
+		struct kfd_process_device *pdd = NULL;
+		struct kfd_hsa_memory_exception_data *data;
+		int i;
+
+		for (i = 0; i < p->n_pdds; i++) {
+			if (p->pdds[i]->dev->id == dev_id) {
+				pdd = p->pdds[i];
+				break;
+			}
+		}
+
+		if (!pdd)
+			return -ENODEV;
+
+		data = (struct kfd_hsa_memory_exception_data *)
+						pdd->vm_fault_exc_data;
+
+		kfd_dqm_evict_pasid(pdd->dev->dqm, p->pasid);
+		kfd_signal_vm_fault_event(pdd->dev, p->pasid, NULL, data);
+		error_reason &= ~KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION);
+	}
+
+	if (error_reason & (KFD_EC_MASK(EC_PROCESS_RUNTIME))) {
+		/*
+		 * block should only happen after the debugger receives runtime
+		 * enable notice.
+		 */
+		up(&p->runtime_enable_sema);
+		error_reason &= ~KFD_EC_MASK(EC_PROCESS_RUNTIME);
+	}
+
+	if (error_reason)
+		return kfd_send_exception_to_runtime(p, queue_id, error_reason);
+
+	return 0;
+}
+
 static int kfd_dbg_set_queue_workaround(struct queue *q, bool enable)
 {
 	struct mqd_update_info minfo = {0};
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 66ee7b95d08a..2c6866bb8850 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -34,6 +34,12 @@ int kfd_dbg_trap_disable(struct kfd_process *target);
 int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 			void __user *runtime_info,
 			uint32_t *runtime_info_size);
+
+int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
+					unsigned int dev_id,
+					unsigned int queue_id,
+					uint64_t error_reason);
+
 static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
 {
 	return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2) ||
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
index 9926186f88a6..0a5e7b172a64 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
@@ -1222,7 +1222,8 @@ void kfd_signal_hw_exception_event(u32 pasid)
 }
 
 void kfd_signal_vm_fault_event(struct kfd_node *dev, u32 pasid,
-			       struct kfd_vm_fault_info *info)
+				struct kfd_vm_fault_info *info,
+				struct kfd_hsa_memory_exception_data *data)
 {
 	struct kfd_event *ev;
 	uint32_t id;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
index 861bccb1e9dc..8cf58be80f4e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
@@ -362,7 +362,7 @@ static void event_interrupt_wq_v9(struct kfd_node *dev,
 
 		kfd_smi_event_update_vmfault(dev, pasid);
 		kfd_dqm_evict_pasid(dev->dqm, pasid);
-		kfd_signal_vm_fault_event(dev, pasid, &info);
+		kfd_signal_vm_fault_event(dev, pasid, &info, NULL);
 	}
 }
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index b18cd4bf76bf..58b82fa59584 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -979,6 +979,7 @@ struct kfd_process {
 	bool queues_paused;
 
 	/* Tracks runtime enable status */
+	struct semaphore runtime_enable_sema;
 	struct kfd_runtime_info runtime_info;
 
 };
@@ -1447,7 +1448,8 @@ int kfd_get_num_events(struct kfd_process *p);
 int kfd_event_destroy(struct kfd_process *p, uint32_t event_id);
 
 void kfd_signal_vm_fault_event(struct kfd_node *dev, u32 pasid,
-				struct kfd_vm_fault_info *info);
+				struct kfd_vm_fault_info *info,
+				struct kfd_hsa_memory_exception_data *data);
 
 void kfd_signal_reset_event(struct kfd_node *dev);
 
@@ -1463,6 +1465,9 @@ static inline bool kfd_flush_tlb_after_unmap(struct kfd_dev *dev)
 	       KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 0);
 }
 
+int kfd_send_exception_to_runtime(struct kfd_process *p,
+				unsigned int queue_id,
+				uint64_t error_reason);
 bool kfd_is_locked(void);
 
 /* Compute profile */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index f904d6d6e01c..5cbfcaf08c8f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -1462,6 +1462,7 @@ static struct kfd_process *create_process(const struct task_struct *thread)
 	process->debugger_process = NULL;
 	process->exception_enable_mask = 0;
 	atomic_set(&process->debugged_process_count, 0);
+	sema_init(&process->runtime_enable_sema, 0);
 
 	process->pasid = kfd_pasid_alloc();
 	if (process->pasid == 0) {
@@ -2120,6 +2121,75 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type)
 	}
 }
 
+struct send_exception_work_handler_workarea {
+	struct work_struct work;
+	struct kfd_process *p;
+	unsigned int queue_id;
+	uint64_t error_reason;
+};
+
+static void send_exception_work_handler(struct work_struct *work)
+{
+	struct send_exception_work_handler_workarea *workarea;
+	struct kfd_process *p;
+	struct queue *q;
+	struct mm_struct *mm;
+	struct kfd_context_save_area_header __user *csa_header;
+	uint64_t __user *err_payload_ptr;
+	uint64_t cur_err;
+	uint32_t ev_id;
+
+	workarea = container_of(work,
+				struct send_exception_work_handler_workarea,
+				work);
+	p = workarea->p;
+
+	mm = get_task_mm(p->lead_thread);
+
+	if (!mm)
+		return;
+
+	kthread_use_mm(mm);
+
+	q = pqm_get_user_queue(&p->pqm, workarea->queue_id);
+
+	if (!q)
+		goto out;
+
+	csa_header = (void __user *)q->properties.ctx_save_restore_area_address;
+
+	get_user(err_payload_ptr, (uint64_t __user **)&csa_header->err_payload_addr);
+	get_user(cur_err, err_payload_ptr);
+	cur_err |= workarea->error_reason;
+	put_user(cur_err, err_payload_ptr);
+	get_user(ev_id, &csa_header->err_event_id);
+
+	kfd_set_event(p, ev_id);
+
+out:
+	kthread_unuse_mm(mm);
+	mmput(mm);
+}
+
+int kfd_send_exception_to_runtime(struct kfd_process *p,
+			unsigned int queue_id,
+			uint64_t error_reason)
+{
+	struct send_exception_work_handler_workarea worker;
+
+	INIT_WORK_ONSTACK(&worker.work, send_exception_work_handler);
+
+	worker.p = p;
+	worker.queue_id = queue_id;
+	worker.error_reason = error_reason;
+
+	schedule_work(&worker.work);
+	flush_work(&worker.work);
+	destroy_work_on_stack(&worker.work);
+
+	return 0;
+}
+
 struct kfd_process_device *kfd_process_device_data_by_id(struct kfd_process *p, uint32_t gpu_id)
 {
 	int i;
@@ -2179,4 +2249,3 @@ int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data)
 }
 
 #endif
-
-- 
cgit 


From 455227c4642c5e1867213cea73a527e431779060 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Fri, 8 Apr 2022 13:12:24 -0400
Subject: drm/amdkfd: add runtime enable operation

The debugger can attach to a process prior to HSA enablement (i.e.
inferior is spawned by the debugger and attached to immediately before
target process has been enabled for HSA dispatches) or it
can attach to a running target that is already HSA enabled.  Either
way, the debugger needs to know the enablement status to know when
it can inspect queues.

For the scenario where the debugger spawns the target process,
it will have to wait for ROCr's runtime enable request from the target.
The runtime enable request will be able to see that its process has been
debug attached.  ROCr raises an EC_PROCESS_RUNTIME signal to the
debugger then blocks the target process while waiting the debugger's
response. Once the debugger has received the runtime signal, it will
unblock the target process.

For the scenario where the debugger attaches to a running target
process, ROCr will set the target process' runtime status as enabled so
that on an attach request, the debugger will be able to see this
status and will continue with debug enablement as normal.

A secondary requirement is to conditionally enable the trap tempories only
if the user requests it (env var HSA_ENABLE_DEBUG=1) or if the debugger
attaches with HSA runtime enabled.  This is because setting up the trap
temporaries incurs a performance overhead that is unacceptable for
microbench performance in normal mode for certain customers.

In the scenario where the debugger spawns the target process, when ROCr
detects that the debugger has attached during the runtime enable
request, it will enable the trap temporaries before it blocks the target
process while waiting for the debugger to respond.

In the scenario where the debugger attaches to a running target process,
it will enable to trap temporaries itself.

Finally, there is an additional restriction that is required to be
enforced with runtime enable and HW debug mode setting. The debugger must
first ensure that HW debug mode has been enabled before permitting HW debug
mode operations.

With single process debug devices, allowing the debugger to set debug
HW modes prior to trap activation means that debug HW mode setting can
occur before the KFD has reserved the debug VMID (0xf) from the hardware
scheduler's VMID allocation resource pool.  This can result in the
hardware scheduler assigning VMID 0xf to a non-debugged process and
having that process inherit debug HW mode settings intended for the
debugged target process instead, which is both incorrect and potentially
fatal for normal mode operation.

With multi process debug devices, allowing the debugger to set debug
HW modes prior to trap activation means that non-debugged processes
migrating to a new VMID could inherit unintended debug settings.

All debug operations that touch HW settings must require trap activation
where trap activation is triggered by both debug attach and runtime
enablement (target has KFD opened and is ready to dispatch work).

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 143 ++++++++++++++++++++++++++++++-
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c   |   6 +-
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |   4 +
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h    |   1 +
 4 files changed, 150 insertions(+), 4 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 5e57b3e96ff9..615fa9ab36b7 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -2738,11 +2738,140 @@ static int kfd_ioctl_criu(struct file *filep, struct kfd_process *p, void *data)
 	return ret;
 }
 
-static int kfd_ioctl_runtime_enable(struct file *filep, struct kfd_process *p, void *data)
+static int runtime_enable(struct kfd_process *p, uint64_t r_debug,
+			bool enable_ttmp_setup)
+{
+	int i = 0, ret = 0;
+
+	if (p->is_runtime_retry)
+		goto retry;
+
+	if (p->runtime_info.runtime_state != DEBUG_RUNTIME_STATE_DISABLED)
+		return -EBUSY;
+
+	for (i = 0; i < p->n_pdds; i++) {
+		struct kfd_process_device *pdd = p->pdds[i];
+
+		if (pdd->qpd.queue_count)
+			return -EEXIST;
+	}
+
+	p->runtime_info.runtime_state = DEBUG_RUNTIME_STATE_ENABLED;
+	p->runtime_info.r_debug = r_debug;
+	p->runtime_info.ttmp_setup = enable_ttmp_setup;
+
+	if (p->runtime_info.ttmp_setup) {
+		for (i = 0; i < p->n_pdds; i++) {
+			struct kfd_process_device *pdd = p->pdds[i];
+
+			if (!kfd_dbg_is_rlc_restore_supported(pdd->dev)) {
+				amdgpu_gfx_off_ctrl(pdd->dev->adev, false);
+				pdd->dev->kfd2kgd->enable_debug_trap(
+						pdd->dev->adev,
+						true,
+						pdd->dev->vm_info.last_vmid_kfd);
+			} else if (kfd_dbg_is_per_vmid_supported(pdd->dev)) {
+				pdd->spi_dbg_override = pdd->dev->kfd2kgd->enable_debug_trap(
+						pdd->dev->adev,
+						false,
+						0);
+			}
+		}
+	}
+
+retry:
+	if (p->debug_trap_enabled) {
+		if (!p->is_runtime_retry) {
+			kfd_dbg_trap_activate(p);
+			kfd_dbg_ev_raise(KFD_EC_MASK(EC_PROCESS_RUNTIME),
+					p, NULL, 0, false, NULL, 0);
+		}
+
+		mutex_unlock(&p->mutex);
+		ret = down_interruptible(&p->runtime_enable_sema);
+		mutex_lock(&p->mutex);
+
+		p->is_runtime_retry = !!ret;
+	}
+
+	return ret;
+}
+
+static int runtime_disable(struct kfd_process *p)
 {
+	int i = 0, ret;
+	bool was_enabled = p->runtime_info.runtime_state == DEBUG_RUNTIME_STATE_ENABLED;
+
+	p->runtime_info.runtime_state = DEBUG_RUNTIME_STATE_DISABLED;
+	p->runtime_info.r_debug = 0;
+
+	if (p->debug_trap_enabled) {
+		if (was_enabled)
+			kfd_dbg_trap_deactivate(p, false, 0);
+
+		if (!p->is_runtime_retry)
+			kfd_dbg_ev_raise(KFD_EC_MASK(EC_PROCESS_RUNTIME),
+					p, NULL, 0, false, NULL, 0);
+
+		mutex_unlock(&p->mutex);
+		ret = down_interruptible(&p->runtime_enable_sema);
+		mutex_lock(&p->mutex);
+
+		p->is_runtime_retry = !!ret;
+		if (ret)
+			return ret;
+	}
+
+	if (was_enabled && p->runtime_info.ttmp_setup) {
+		for (i = 0; i < p->n_pdds; i++) {
+			struct kfd_process_device *pdd = p->pdds[i];
+
+			if (!kfd_dbg_is_rlc_restore_supported(pdd->dev))
+				amdgpu_gfx_off_ctrl(pdd->dev->adev, true);
+		}
+	}
+
+	p->runtime_info.ttmp_setup = false;
+
+	/* disable ttmp setup */
+	for (i = 0; i < p->n_pdds; i++) {
+		struct kfd_process_device *pdd = p->pdds[i];
+
+		if (kfd_dbg_is_per_vmid_supported(pdd->dev)) {
+			pdd->spi_dbg_override =
+					pdd->dev->kfd2kgd->disable_debug_trap(
+					pdd->dev->adev,
+					false,
+					pdd->dev->vm_info.last_vmid_kfd);
+
+			if (!pdd->dev->kfd->shared_resources.enable_mes)
+				debug_refresh_runlist(pdd->dev->dqm);
+			else
+				kfd_dbg_set_mes_debug_mode(pdd);
+		}
+	}
+
 	return 0;
 }
 
+static int kfd_ioctl_runtime_enable(struct file *filep, struct kfd_process *p, void *data)
+{
+	struct kfd_ioctl_runtime_enable_args *args = data;
+	int r;
+
+	mutex_lock(&p->mutex);
+
+	if (args->mode_mask & KFD_RUNTIME_ENABLE_MODE_ENABLE_MASK)
+		r = runtime_enable(p, args->r_debug,
+				!!(args->mode_mask & KFD_RUNTIME_ENABLE_MODE_TTMP_SAVE_MASK));
+	else
+		r = runtime_disable(p);
+
+	mutex_unlock(&p->mutex);
+
+	return r;
+}
+
 static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, void *data)
 {
 	struct kfd_ioctl_dbg_trap_args *args = data;
@@ -2815,6 +2944,18 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 		goto unlock_out;
 	}
 
+	if (target->runtime_info.runtime_state != DEBUG_RUNTIME_STATE_ENABLED &&
+			(args->op == KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_OVERRIDE ||
+			 args->op == KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_MODE ||
+			 args->op == KFD_IOC_DBG_TRAP_SUSPEND_QUEUES ||
+			 args->op == KFD_IOC_DBG_TRAP_RESUME_QUEUES ||
+			 args->op == KFD_IOC_DBG_TRAP_SET_NODE_ADDRESS_WATCH ||
+			 args->op == KFD_IOC_DBG_TRAP_CLEAR_NODE_ADDRESS_WATCH ||
+			 args->op == KFD_IOC_DBG_TRAP_SET_FLAGS)) {
+		r = -EPERM;
+		goto unlock_out;
+	}
+
 	switch (args->op) {
 	case KFD_IOC_DBG_TRAP_ENABLE:
 		if (target != p)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 61098975bb0e..a19c21d04438 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -219,7 +219,7 @@ unwind:
 	return r;
 }
 
-static int kfd_dbg_set_mes_debug_mode(struct kfd_process_device *pdd)
+int kfd_dbg_set_mes_debug_mode(struct kfd_process_device *pdd)
 {
 	uint32_t spi_dbg_cntl = pdd->spi_dbg_override | pdd->spi_dbg_launch_mode;
 	uint32_t flags = pdd->process->dbg_flags;
@@ -239,7 +239,7 @@ static int kfd_dbg_set_mes_debug_mode(struct kfd_process_device *pdd)
  *				to unwind
  *		else: ignored
  */
-static void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind_count)
+void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind_count)
 {
 	int i;
 
@@ -307,7 +307,7 @@ int kfd_dbg_trap_disable(struct kfd_process *target)
 	return 0;
 }
 
-static int kfd_dbg_trap_activate(struct kfd_process *target)
+int kfd_dbg_trap_activate(struct kfd_process *target)
 {
 	int i, r = 0;
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 2c6866bb8850..fca928564948 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -25,6 +25,8 @@
 
 #include "kfd_priv.h"
 
+void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind_count);
+int kfd_dbg_trap_activate(struct kfd_process *target);
 bool kfd_dbg_ev_raise(uint64_t event_mask,
 			struct kfd_process *process, struct kfd_node *dev,
 			unsigned int source_id, bool use_worker,
@@ -77,4 +79,6 @@ static inline bool kfd_dbg_has_gws_support(struct kfd_node *dev)
 	/* Assume debugging and cooperative launch supported otherwise. */
 	return true;
 }
+
+int kfd_dbg_set_mes_debug_mode(struct kfd_process_device *pdd);
 #endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 58b82fa59584..4b80f74b9de0 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -980,6 +980,7 @@ struct kfd_process {
 
 	/* Tracks runtime enable status */
 	struct semaphore runtime_enable_sema;
+	bool is_runtime_retry;
 	struct kfd_runtime_info runtime_info;
 
 };
-- 
cgit 


From 12fb1ad70d65edc3405884792d044fa79df7244f Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Fri, 22 Apr 2022 12:26:18 -0400
Subject: drm/amdkfd: update process interrupt handling for debug events

The debugger must be notified by any debugger subscribed exception
that comes from hardware interrupts.

If a debugger session exits, any exceptions it subscribed to may still
have interrupts in the interrupt ring buffer or KGD/KFD pipeline.
To prevent a new session from inheriting stale interrupts, when a new
queue is created, open an interrupt drain and allow the IH ring to drain
from a timestamped checkpoint.  Then inject a custom IV so that once
the custom IV is picked up by the KFD, it's safe to close the drain
and proceed with queue creation.

The drain must also be on debug disable as SW interrupts may still
be processed.  Drain at this time and clear all the exception status.

The debugger may also not be attached nor subscibed to certain
exceptions so forward them directly to the runtime.

GFX10 also requires its own IV processing, hence the creation of
kfd_int_process_v10.c.  This is because the IV from SQ interrupts are
packed into a new continguous format unlike GFX9. To make this clear,
a separate interrupting handling code file was created.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c         |  16 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h         |   2 +
 drivers/gpu/drm/amd/amdkfd/Makefile                |   1 +
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c             |  84 +++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h             |   6 +
 drivers/gpu/drm/amd/amdkfd/kfd_device.c            |   4 +-
 drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c   | 405 +++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c   |  26 +-
 drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c    |  98 ++++-
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h              |  12 +
 drivers/gpu/drm/amd/amdkfd/kfd_process.c           |  47 +++
 .../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c |   4 +
 12 files changed, 686 insertions(+), 19 deletions(-)
 create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 66f80b9ab0c5..98cd52bb005f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -777,6 +777,22 @@ void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, bo
 	amdgpu_umc_poison_handler(adev, reset);
 }
 
+int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev,
+					uint32_t *payload)
+{
+	int ret;
+
+	/* Device or IH ring is not ready so bail. */
+	ret = amdgpu_ih_wait_on_checkpoint_process_ts(adev, &adev->irq.ih);
+	if (ret)
+		return ret;
+
+	/* Send payload to fence KFD interrupts */
+	amdgpu_amdkfd_interrupt(adev, payload);
+
+	return 0;
+}
+
 bool amdgpu_amdkfd_ras_query_utcl2_poison_status(struct amdgpu_device *adev)
 {
 	if (adev->gfx.ras && adev->gfx.ras->query_utcl2_poison_status)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 94cc456761e5..dd740e64e6e1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -250,6 +250,8 @@ int amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(struct amdgpu_device *dst,
 					    struct amdgpu_device *src,
 					    bool is_min);
 int amdgpu_amdkfd_get_pcie_bandwidth_mbytes(struct amdgpu_device *adev, bool is_min);
+int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev,
+					uint32_t *payload);
 
 /* Read user wptr from a specified user address space with page fault
  * disabled. The memory must be pinned and mapped to the hardware when
diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile b/drivers/gpu/drm/amd/amdkfd/Makefile
index 747754428073..2ec8f27c5366 100644
--- a/drivers/gpu/drm/amd/amdkfd/Makefile
+++ b/drivers/gpu/drm/amd/amdkfd/Makefile
@@ -53,6 +53,7 @@ AMDKFD_FILES	:= $(AMDKFD_PATH)/kfd_module.o \
 		$(AMDKFD_PATH)/kfd_events.o \
 		$(AMDKFD_PATH)/cik_event_interrupt.o \
 		$(AMDKFD_PATH)/kfd_int_process_v9.o \
+		$(AMDKFD_PATH)/kfd_int_process_v10.o \
 		$(AMDKFD_PATH)/kfd_int_process_v11.o \
 		$(AMDKFD_PATH)/kfd_smi_events.o \
 		$(AMDKFD_PATH)/kfd_crat.o \
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 17e8e9edccbf..68b657398d41 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -125,6 +125,64 @@ bool kfd_dbg_ev_raise(uint64_t event_mask,
 	return is_subscribed;
 }
 
+/* set pending event queue entry from ring entry  */
+bool kfd_set_dbg_ev_from_interrupt(struct kfd_node *dev,
+				   unsigned int pasid,
+				   uint32_t doorbell_id,
+				   uint64_t trap_mask,
+				   void *exception_data,
+				   size_t exception_data_size)
+{
+	struct kfd_process *p;
+	bool signaled_to_debugger_or_runtime = false;
+
+	p = kfd_lookup_process_by_pasid(pasid);
+
+	if (!p)
+		return false;
+
+	if (!kfd_dbg_ev_raise(trap_mask, p, dev, doorbell_id, true,
+			      exception_data, exception_data_size)) {
+		struct process_queue_manager *pqm;
+		struct process_queue_node *pqn;
+
+		if (!!(trap_mask & KFD_EC_MASK_QUEUE) &&
+		       p->runtime_info.runtime_state == DEBUG_RUNTIME_STATE_ENABLED) {
+			mutex_lock(&p->mutex);
+
+			pqm = &p->pqm;
+			list_for_each_entry(pqn, &pqm->queues,
+							process_queue_list) {
+
+				if (!(pqn->q && pqn->q->device == dev &&
+				      pqn->q->doorbell_id == doorbell_id))
+					continue;
+
+				kfd_send_exception_to_runtime(p, pqn->q->properties.queue_id,
+							      trap_mask);
+
+				signaled_to_debugger_or_runtime = true;
+
+				break;
+			}
+
+			mutex_unlock(&p->mutex);
+		} else if (trap_mask & KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION)) {
+			kfd_dqm_evict_pasid(dev->dqm, p->pasid);
+			kfd_signal_vm_fault_event(dev, p->pasid, NULL,
+							exception_data);
+
+			signaled_to_debugger_or_runtime = true;
+		}
+	} else {
+		signaled_to_debugger_or_runtime = true;
+	}
+
+	kfd_unref_process(p);
+
+	return signaled_to_debugger_or_runtime;
+}
+
 int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
 					unsigned int dev_id,
 					unsigned int queue_id,
@@ -281,6 +339,31 @@ void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind
 	kfd_dbg_set_workaround(target, false);
 }
 
+static void kfd_dbg_clean_exception_status(struct kfd_process *target)
+{
+	struct process_queue_manager *pqm;
+	struct process_queue_node *pqn;
+	int i;
+
+	for (i = 0; i < target->n_pdds; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+
+		kfd_process_drain_interrupts(pdd);
+
+		pdd->exception_status = 0;
+	}
+
+	pqm = &target->pqm;
+	list_for_each_entry(pqn, &pqm->queues, process_queue_list) {
+		if (!pqn->q)
+			continue;
+
+		pqn->q->properties.exception_status = 0;
+	}
+
+	target->exception_status = 0;
+}
+
 int kfd_dbg_trap_disable(struct kfd_process *target)
 {
 	if (!target->debug_trap_enabled)
@@ -304,6 +387,7 @@ int kfd_dbg_trap_disable(struct kfd_process *target)
 	}
 
 	target->debug_trap_enabled = false;
+	kfd_dbg_clean_exception_status(target);
 	kfd_unref_process(target);
 
 	return 0;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index fca928564948..5153ccbd7fd1 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -27,6 +27,12 @@
 
 void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind_count);
 int kfd_dbg_trap_activate(struct kfd_process *target);
+bool kfd_set_dbg_ev_from_interrupt(struct kfd_node *dev,
+				   unsigned int pasid,
+				   uint32_t doorbell_id,
+				   uint64_t trap_mask,
+				   void *exception_data,
+				   size_t exception_data_size);
 bool kfd_dbg_ev_raise(uint64_t event_mask,
 			struct kfd_process *process, struct kfd_node *dev,
 			unsigned int source_id, bool use_worker,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index f0ed6e6416c3..2c36bb578633 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -140,6 +140,8 @@ static void kfd_device_info_set_event_interrupt_class(struct kfd_dev *kfd)
 	case IP_VERSION(9, 4, 1): /* ARCTURUS */
 	case IP_VERSION(9, 4, 2): /* ALDEBARAN */
 	case IP_VERSION(9, 4, 3): /* GC 9.4.3 */
+		kfd->device_info.event_interrupt_class = &event_interrupt_class_v9;
+		break;
 	case IP_VERSION(10, 3, 1): /* VANGOGH */
 	case IP_VERSION(10, 3, 3): /* YELLOW_CARP */
 	case IP_VERSION(10, 3, 6): /* GC 10.3.6 */
@@ -153,7 +155,7 @@ static void kfd_device_info_set_event_interrupt_class(struct kfd_dev *kfd)
 	case IP_VERSION(10, 3, 2): /* NAVY_FLOUNDER */
 	case IP_VERSION(10, 3, 4): /* DIMGREY_CAVEFISH */
 	case IP_VERSION(10, 3, 5): /* BEIGE_GOBY */
-		kfd->device_info.event_interrupt_class = &event_interrupt_class_v9;
+		kfd->device_info.event_interrupt_class = &event_interrupt_class_v10;
 		break;
 	case IP_VERSION(11, 0, 0):
 	case IP_VERSION(11, 0, 1):
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c
new file mode 100644
index 000000000000..c7991e07b6be
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c
@@ -0,0 +1,405 @@
+/*
+ * Copyright 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include "kfd_events.h"
+#include "kfd_debug.h"
+#include "soc15_int.h"
+#include "kfd_device_queue_manager.h"
+
+/*
+ * GFX10 SQ Interrupts
+ *
+ * There are 3 encoding types of interrupts sourced from SQ sent as a 44-bit
+ * packet to the Interrupt Handler:
+ * Auto - Generated by the SQG (various cmd overflows, timestamps etc)
+ * Wave - Generated by S_SENDMSG through a shader program
+ * Error - HW generated errors (Illegal instructions, Memviols, EDC etc)
+ *
+ * The 44-bit packet is mapped as {context_id1[7:0],context_id0[31:0]} plus
+ * 4-bits for VMID (SOC15_VMID_FROM_IH_ENTRY) as such:
+ *
+ * - context_id1[7:6]
+ * Encoding type (0 = Auto, 1 = Wave, 2 = Error)
+ *
+ * - context_id0[24]
+ * PRIV bit indicates that Wave S_SEND or error occurred within trap
+ *
+ * - context_id0[22:0]
+ * 23-bit data with the following layout per encoding type:
+ * Auto - only context_id0[8:0] is used, which reports various interrupts
+ * generated by SQG.  The rest is 0.
+ * Wave - user data sent from m0 via S_SENDMSG
+ * Error - Error type (context_id0[22:19]), Error Details (rest of bits)
+ *
+ * The other context_id bits show coordinates (SE/SH/CU/SIMD/WGP) for wave
+ * S_SENDMSG and Errors.  These are 0 for Auto.
+ */
+
+enum SQ_INTERRUPT_WORD_ENCODING {
+	SQ_INTERRUPT_WORD_ENCODING_AUTO = 0x0,
+	SQ_INTERRUPT_WORD_ENCODING_INST,
+	SQ_INTERRUPT_WORD_ENCODING_ERROR,
+};
+
+enum SQ_INTERRUPT_ERROR_TYPE {
+	SQ_INTERRUPT_ERROR_TYPE_EDC_FUE = 0x0,
+	SQ_INTERRUPT_ERROR_TYPE_ILLEGAL_INST,
+	SQ_INTERRUPT_ERROR_TYPE_MEMVIOL,
+	SQ_INTERRUPT_ERROR_TYPE_EDC_FED,
+};
+
+/* SQ_INTERRUPT_WORD_AUTO_CTXID */
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__THREAD_TRACE__SHIFT 0
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__WLT__SHIFT 1
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__THREAD_TRACE_BUF0_FULL__SHIFT 2
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__THREAD_TRACE_BUF1_FULL__SHIFT 3
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__THREAD_TRACE_UTC_ERROR__SHIFT 7
+#define SQ_INTERRUPT_WORD_AUTO_CTXID1__SE_ID__SHIFT 4
+#define SQ_INTERRUPT_WORD_AUTO_CTXID1__ENCODING__SHIFT 6
+
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__THREAD_TRACE_MASK 0x00000001
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__WLT_MASK 0x00000002
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__THREAD_TRACE_BUF0_FULL_MASK 0x00000004
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__THREAD_TRACE_BUF1_FULL_MASK 0x00000008
+#define SQ_INTERRUPT_WORD_AUTO_CTXID0__THREAD_TRACE_UTC_ERROR_MASK 0x00000080
+#define SQ_INTERRUPT_WORD_AUTO_CTXID1__SE_ID_MASK 0x030
+#define SQ_INTERRUPT_WORD_AUTO_CTXID1__ENCODING_MASK 0x0c0
+
+/* SQ_INTERRUPT_WORD_WAVE_CTXID */
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__DATA__SHIFT 0
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__SA_ID__SHIFT 23
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__PRIV__SHIFT 24
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__WAVE_ID__SHIFT 25
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__SIMD_ID__SHIFT 30
+#define SQ_INTERRUPT_WORD_WAVE_CTXID1__WGP_ID__SHIFT 0
+#define SQ_INTERRUPT_WORD_WAVE_CTXID1__SE_ID__SHIFT 4
+#define SQ_INTERRUPT_WORD_WAVE_CTXID1__ENCODING__SHIFT 6
+
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__DATA_MASK 0x000007fffff
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__SA_ID_MASK 0x0000800000
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__PRIV_MASK 0x00001000000
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__WAVE_ID_MASK 0x0003e000000
+#define SQ_INTERRUPT_WORD_WAVE_CTXID0__SIMD_ID_MASK 0x000c0000000
+#define SQ_INTERRUPT_WORD_WAVE_CTXID1__WGP_ID_MASK 0x00f
+#define SQ_INTERRUPT_WORD_WAVE_CTXID1__SE_ID_MASK 0x030
+#define SQ_INTERRUPT_WORD_WAVE_CTXID1__ENCODING_MASK 0x0c0
+
+#define KFD_CTXID0__ERR_TYPE_MASK 0x780000
+#define KFD_CTXID0__ERR_TYPE__SHIFT 19
+
+/* GFX10 SQ interrupt ENC type bit (context_id1[7:6]) for wave s_sendmsg */
+#define KFD_CONTEXT_ID1_ENC_TYPE_WAVE_MASK	0x40
+/* GFX10 SQ interrupt PRIV bit (context_id0[24]) for s_sendmsg inside trap */
+#define KFD_CONTEXT_ID0_PRIV_MASK		0x1000000
+/*
+ * The debugger will send user data(m0) with PRIV=1 to indicate it requires
+ * notification from the KFD with the following queue id (DOORBELL_ID) and
+ * trap code (TRAP_CODE).
+ */
+#define KFD_CONTEXT_ID0_DEBUG_DOORBELL_MASK	0x0003ff
+#define KFD_CONTEXT_ID0_DEBUG_TRAP_CODE_SHIFT	10
+#define KFD_CONTEXT_ID0_DEBUG_TRAP_CODE_MASK	0x07fc00
+#define KFD_DEBUG_DOORBELL_ID(ctxid0)	((ctxid0) &	\
+				KFD_CONTEXT_ID0_DEBUG_DOORBELL_MASK)
+#define KFD_DEBUG_TRAP_CODE(ctxid0)	(((ctxid0) &	\
+				KFD_CONTEXT_ID0_DEBUG_TRAP_CODE_MASK)	\
+				>> KFD_CONTEXT_ID0_DEBUG_TRAP_CODE_SHIFT)
+#define KFD_DEBUG_CP_BAD_OP_ECODE_MASK		0x3fffc00
+#define KFD_DEBUG_CP_BAD_OP_ECODE_SHIFT		10
+#define KFD_DEBUG_CP_BAD_OP_ECODE(ctxid0) (((ctxid0) &			\
+				KFD_DEBUG_CP_BAD_OP_ECODE_MASK)		\
+				>> KFD_DEBUG_CP_BAD_OP_ECODE_SHIFT)
+
+static void event_interrupt_poison_consumption(struct kfd_node *dev,
+				uint16_t pasid, uint16_t client_id)
+{
+	int old_poison, ret = -EINVAL;
+	struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
+
+	if (!p)
+		return;
+
+	/* all queues of a process will be unmapped in one time */
+	old_poison = atomic_cmpxchg(&p->poison, 0, 1);
+	kfd_unref_process(p);
+	if (old_poison)
+		return;
+
+	switch (client_id) {
+	case SOC15_IH_CLIENTID_SE0SH:
+	case SOC15_IH_CLIENTID_SE1SH:
+	case SOC15_IH_CLIENTID_SE2SH:
+	case SOC15_IH_CLIENTID_SE3SH:
+	case SOC15_IH_CLIENTID_UTCL2:
+		ret = kfd_dqm_evict_pasid(dev->dqm, pasid);
+		break;
+	case SOC15_IH_CLIENTID_SDMA0:
+	case SOC15_IH_CLIENTID_SDMA1:
+	case SOC15_IH_CLIENTID_SDMA2:
+	case SOC15_IH_CLIENTID_SDMA3:
+	case SOC15_IH_CLIENTID_SDMA4:
+		break;
+	default:
+		break;
+	}
+
+	kfd_signal_poison_consumed_event(dev, pasid);
+
+	/* resetting queue passes, do page retirement without gpu reset
+	 * resetting queue fails, fallback to gpu reset solution
+	 */
+	if (!ret) {
+		dev_warn(dev->adev->dev,
+			"RAS poison consumption, unmap queue flow succeeded: client id %d\n",
+			client_id);
+		amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, false);
+	} else {
+		dev_warn(dev->adev->dev,
+			"RAS poison consumption, fall back to gpu reset flow: client id %d\n",
+			client_id);
+		amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, true);
+	}
+}
+
+static bool event_interrupt_isr_v10(struct kfd_node *dev,
+					const uint32_t *ih_ring_entry,
+					uint32_t *patched_ihre,
+					bool *patched_flag)
+{
+	uint16_t source_id, client_id, pasid, vmid;
+	const uint32_t *data = ih_ring_entry;
+
+	source_id = SOC15_SOURCE_ID_FROM_IH_ENTRY(ih_ring_entry);
+	client_id = SOC15_CLIENT_ID_FROM_IH_ENTRY(ih_ring_entry);
+
+	/* Only handle interrupts from KFD VMIDs */
+	vmid = SOC15_VMID_FROM_IH_ENTRY(ih_ring_entry);
+	if (!KFD_IRQ_IS_FENCE(client_id, source_id) &&
+	   (vmid < dev->vm_info.first_vmid_kfd ||
+	    vmid > dev->vm_info.last_vmid_kfd))
+		return false;
+
+	pasid = SOC15_PASID_FROM_IH_ENTRY(ih_ring_entry);
+
+	/* Only handle clients we care about */
+	if (client_id != SOC15_IH_CLIENTID_GRBM_CP &&
+	    client_id != SOC15_IH_CLIENTID_SDMA0 &&
+	    client_id != SOC15_IH_CLIENTID_SDMA1 &&
+	    client_id != SOC15_IH_CLIENTID_SDMA2 &&
+	    client_id != SOC15_IH_CLIENTID_SDMA3 &&
+	    client_id != SOC15_IH_CLIENTID_SDMA4 &&
+	    client_id != SOC15_IH_CLIENTID_SDMA5 &&
+	    client_id != SOC15_IH_CLIENTID_SDMA6 &&
+	    client_id != SOC15_IH_CLIENTID_SDMA7 &&
+	    client_id != SOC15_IH_CLIENTID_VMC &&
+	    client_id != SOC15_IH_CLIENTID_VMC1 &&
+	    client_id != SOC15_IH_CLIENTID_UTCL2 &&
+	    client_id != SOC15_IH_CLIENTID_SE0SH &&
+	    client_id != SOC15_IH_CLIENTID_SE1SH &&
+	    client_id != SOC15_IH_CLIENTID_SE2SH &&
+	    client_id != SOC15_IH_CLIENTID_SE3SH)
+		return false;
+
+	pr_debug("client id 0x%x, source id %d, vmid %d, pasid 0x%x. raw data:\n",
+		 client_id, source_id, vmid, pasid);
+	pr_debug("%8X, %8X, %8X, %8X, %8X, %8X, %8X, %8X.\n",
+		 data[0], data[1], data[2], data[3],
+		 data[4], data[5], data[6], data[7]);
+
+	/* If there is no valid PASID, it's likely a bug */
+	if (WARN_ONCE(pasid == 0, "Bug: No PASID in KFD interrupt"))
+		return 0;
+
+	/* Interrupt types we care about: various signals and faults.
+	 * They will be forwarded to a work queue (see below).
+	 */
+	return source_id == SOC15_INTSRC_CP_END_OF_PIPE ||
+		source_id == SOC15_INTSRC_SDMA_TRAP ||
+		source_id == SOC15_INTSRC_SQ_INTERRUPT_MSG ||
+		source_id == SOC15_INTSRC_CP_BAD_OPCODE ||
+		client_id == SOC15_IH_CLIENTID_VMC ||
+		client_id == SOC15_IH_CLIENTID_VMC1 ||
+		client_id == SOC15_IH_CLIENTID_UTCL2 ||
+		KFD_IRQ_IS_FENCE(client_id, source_id);
+}
+
+static void event_interrupt_wq_v10(struct kfd_node *dev,
+					const uint32_t *ih_ring_entry)
+{
+	uint16_t source_id, client_id, pasid, vmid;
+	uint32_t context_id0, context_id1;
+	uint32_t encoding, sq_intr_err_type;
+
+	source_id = SOC15_SOURCE_ID_FROM_IH_ENTRY(ih_ring_entry);
+	client_id = SOC15_CLIENT_ID_FROM_IH_ENTRY(ih_ring_entry);
+	pasid = SOC15_PASID_FROM_IH_ENTRY(ih_ring_entry);
+	vmid = SOC15_VMID_FROM_IH_ENTRY(ih_ring_entry);
+	context_id0 = SOC15_CONTEXT_ID0_FROM_IH_ENTRY(ih_ring_entry);
+	context_id1 = SOC15_CONTEXT_ID1_FROM_IH_ENTRY(ih_ring_entry);
+
+	if (client_id == SOC15_IH_CLIENTID_GRBM_CP ||
+	    client_id == SOC15_IH_CLIENTID_SE0SH ||
+	    client_id == SOC15_IH_CLIENTID_SE1SH ||
+	    client_id == SOC15_IH_CLIENTID_SE2SH ||
+	    client_id == SOC15_IH_CLIENTID_SE3SH) {
+		if (source_id == SOC15_INTSRC_CP_END_OF_PIPE)
+			kfd_signal_event_interrupt(pasid, context_id0, 32);
+		else if (source_id == SOC15_INTSRC_SQ_INTERRUPT_MSG) {
+			encoding = REG_GET_FIELD(context_id1,
+						SQ_INTERRUPT_WORD_WAVE_CTXID1, ENCODING);
+			switch (encoding) {
+			case SQ_INTERRUPT_WORD_ENCODING_AUTO:
+				pr_debug(
+					"sq_intr: auto, se %d, ttrace %d, wlt %d, ttrac_buf0_full %d, ttrac_buf1_full %d, ttrace_utc_err %d\n",
+					REG_GET_FIELD(context_id1, SQ_INTERRUPT_WORD_AUTO_CTXID1,
+							SE_ID),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_AUTO_CTXID0,
+							THREAD_TRACE),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_AUTO_CTXID0,
+							WLT),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_AUTO_CTXID0,
+							THREAD_TRACE_BUF0_FULL),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_AUTO_CTXID0,
+							THREAD_TRACE_BUF1_FULL),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_AUTO_CTXID0,
+							THREAD_TRACE_UTC_ERROR));
+				break;
+			case SQ_INTERRUPT_WORD_ENCODING_INST:
+				pr_debug("sq_intr: inst, se %d, data 0x%x, sa %d, priv %d, wave_id %d, simd_id %d, wgp_id %d\n",
+					REG_GET_FIELD(context_id1, SQ_INTERRUPT_WORD_WAVE_CTXID1,
+							SE_ID),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							DATA),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							SA_ID),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							PRIV),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							WAVE_ID),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							SIMD_ID),
+					REG_GET_FIELD(context_id1, SQ_INTERRUPT_WORD_WAVE_CTXID1,
+							WGP_ID));
+				if (context_id0 & SQ_INTERRUPT_WORD_WAVE_CTXID0__PRIV_MASK) {
+					if (kfd_set_dbg_ev_from_interrupt(dev, pasid,
+							KFD_DEBUG_DOORBELL_ID(context_id0),
+							KFD_DEBUG_TRAP_CODE(context_id0),
+							NULL, 0))
+						return;
+				}
+				break;
+			case SQ_INTERRUPT_WORD_ENCODING_ERROR:
+				sq_intr_err_type = REG_GET_FIELD(context_id0, KFD_CTXID0,
+								ERR_TYPE);
+				pr_warn("sq_intr: error, se %d, data 0x%x, sa %d, priv %d, wave_id %d, simd_id %d, wgp_id %d, err_type %d\n",
+					REG_GET_FIELD(context_id1, SQ_INTERRUPT_WORD_WAVE_CTXID1,
+							SE_ID),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							DATA),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							SA_ID),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							PRIV),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							WAVE_ID),
+					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID0,
+							SIMD_ID),
+					REG_GET_FIELD(context_id1, SQ_INTERRUPT_WORD_WAVE_CTXID1,
+							WGP_ID),
+					sq_intr_err_type);
+				if (sq_intr_err_type != SQ_INTERRUPT_ERROR_TYPE_ILLEGAL_INST &&
+					sq_intr_err_type != SQ_INTERRUPT_ERROR_TYPE_MEMVIOL) {
+					event_interrupt_poison_consumption(dev, pasid, source_id);
+					return;
+				}
+				break;
+			default:
+				break;
+			}
+			kfd_signal_event_interrupt(pasid, context_id0 & 0x7fffff, 23);
+		} else if (source_id == SOC15_INTSRC_CP_BAD_OPCODE) {
+			kfd_set_dbg_ev_from_interrupt(dev, pasid,
+				KFD_DEBUG_DOORBELL_ID(context_id0),
+				KFD_EC_MASK(KFD_DEBUG_CP_BAD_OP_ECODE(context_id0)),
+				NULL,
+				0);
+		}
+	} else if (client_id == SOC15_IH_CLIENTID_SDMA0 ||
+		   client_id == SOC15_IH_CLIENTID_SDMA1 ||
+		   client_id == SOC15_IH_CLIENTID_SDMA2 ||
+		   client_id == SOC15_IH_CLIENTID_SDMA3 ||
+		   (client_id == SOC15_IH_CLIENTID_SDMA3_Sienna_Cichlid &&
+		    KFD_GC_VERSION(dev) == IP_VERSION(10, 3, 0)) ||
+		   client_id == SOC15_IH_CLIENTID_SDMA4 ||
+		   client_id == SOC15_IH_CLIENTID_SDMA5 ||
+		   client_id == SOC15_IH_CLIENTID_SDMA6 ||
+		   client_id == SOC15_IH_CLIENTID_SDMA7) {
+		if (source_id == SOC15_INTSRC_SDMA_TRAP) {
+			kfd_signal_event_interrupt(pasid, context_id0 & 0xfffffff, 28);
+		} else if (source_id == SOC15_INTSRC_SDMA_ECC) {
+			event_interrupt_poison_consumption(dev, pasid, source_id);
+			return;
+		}
+	} else if (client_id == SOC15_IH_CLIENTID_VMC ||
+		   client_id == SOC15_IH_CLIENTID_VMC1 ||
+		   client_id == SOC15_IH_CLIENTID_UTCL2) {
+		struct kfd_vm_fault_info info = {0};
+		uint16_t ring_id = SOC15_RING_ID_FROM_IH_ENTRY(ih_ring_entry);
+		struct kfd_hsa_memory_exception_data exception_data;
+
+		if (client_id == SOC15_IH_CLIENTID_UTCL2 &&
+				amdgpu_amdkfd_ras_query_utcl2_poison_status(dev->adev)) {
+			event_interrupt_poison_consumption(dev, pasid, client_id);
+			return;
+		}
+
+		info.vmid = vmid;
+		info.mc_id = client_id;
+		info.page_addr = ih_ring_entry[4] |
+			(uint64_t)(ih_ring_entry[5] & 0xf) << 32;
+		info.prot_valid = ring_id & 0x08;
+		info.prot_read  = ring_id & 0x10;
+		info.prot_write = ring_id & 0x20;
+
+		memset(&exception_data, 0, sizeof(exception_data));
+		exception_data.gpu_id = dev->id;
+		exception_data.va = (info.page_addr) << PAGE_SHIFT;
+		exception_data.failure.NotPresent = info.prot_valid ? 1 : 0;
+		exception_data.failure.NoExecute = info.prot_exec ? 1 : 0;
+		exception_data.failure.ReadOnly = info.prot_write ? 1 : 0;
+		exception_data.failure.imprecise = 0;
+
+		kfd_set_dbg_ev_from_interrupt(dev,
+						pasid,
+						-1,
+						KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION),
+						&exception_data,
+						sizeof(exception_data));
+	} else if (KFD_IRQ_IS_FENCE(client_id, source_id)) {
+		kfd_process_close_interrupt_drain(pasid);
+	}
+}
+
+const struct kfd_event_interrupt_class event_interrupt_class_v10 = {
+	.interrupt_isr = event_interrupt_isr_v10,
+	.interrupt_wq = event_interrupt_wq_v10,
+};
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c
index c2166bf964ef..f933bd231fb9 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c
@@ -26,6 +26,7 @@
 #include "kfd_device_queue_manager.h"
 #include "ivsrcid/vmc/irqsrcs_vmc_1_0.h"
 #include "kfd_smi_events.h"
+#include "kfd_debug.h"
 
 /*
  * GFX11 SQ Interrupts
@@ -238,7 +239,7 @@ static bool event_interrupt_isr_v11(struct kfd_node *dev,
 	client_id = SOC15_CLIENT_ID_FROM_IH_ENTRY(ih_ring_entry);
 	/* Only handle interrupts from KFD VMIDs */
 	vmid = SOC15_VMID_FROM_IH_ENTRY(ih_ring_entry);
-	if (/*!KFD_IRQ_IS_FENCE(client_id, source_id) &&*/
+	if (!KFD_IRQ_IS_FENCE(client_id, source_id) &&
 	    (vmid < dev->vm_info.first_vmid_kfd ||
 	    vmid > dev->vm_info.last_vmid_kfd))
 		return false;
@@ -267,7 +268,7 @@ static bool event_interrupt_isr_v11(struct kfd_node *dev,
 		source_id == SOC15_INTSRC_SQ_INTERRUPT_MSG ||
 		source_id == SOC15_INTSRC_CP_BAD_OPCODE ||
 		source_id == SOC21_INTSRC_SDMA_TRAP ||
-		/* KFD_IRQ_IS_FENCE(client_id, source_id) || */
+		KFD_IRQ_IS_FENCE(client_id, source_id) ||
 		(((client_id == SOC21_IH_CLIENTID_VMC) ||
 		 ((client_id == SOC21_IH_CLIENTID_GFX) &&
 		  (source_id == UTCL2_1_0__SRCID__FAULT))) &&
@@ -279,7 +280,7 @@ static void event_interrupt_wq_v11(struct kfd_node *dev,
 {
 	uint16_t source_id, client_id, ring_id, pasid, vmid;
 	uint32_t context_id0, context_id1;
-	uint8_t sq_int_enc, sq_int_errtype;
+	uint8_t sq_int_enc, sq_int_priv, sq_int_errtype;
 	struct kfd_vm_fault_info info = {0};
 	struct kfd_hsa_memory_exception_data exception_data;
 
@@ -312,9 +313,9 @@ static void event_interrupt_wq_v11(struct kfd_node *dev,
 		exception_data.failure.ReadOnly = info.prot_write ? 1 : 0;
 		exception_data.failure.imprecise = 0;
 
-		/*kfd_set_dbg_ev_from_interrupt(dev, pasid, -1,
+		kfd_set_dbg_ev_from_interrupt(dev, pasid, -1,
 					      KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION),
-					      &exception_data, sizeof(exception_data));*/
+					      &exception_data, sizeof(exception_data));
 		kfd_smi_event_update_vmfault(dev, pasid);
 
 	/* GRBM, SDMA, SE, PMM */
@@ -324,11 +325,11 @@ static void event_interrupt_wq_v11(struct kfd_node *dev,
 		/* CP */
 		if (source_id == SOC15_INTSRC_CP_END_OF_PIPE)
 			kfd_signal_event_interrupt(pasid, context_id0, 32);
-		/*else if (source_id == SOC15_INTSRC_CP_BAD_OPCODE)
+		else if (source_id == SOC15_INTSRC_CP_BAD_OPCODE)
 			kfd_set_dbg_ev_from_interrupt(dev, pasid,
 				KFD_CTXID0_DOORBELL_ID(context_id0),
 				KFD_EC_MASK(KFD_CTXID0_CP_BAD_OP_ECODE(context_id0)),
-				NULL, 0);*/
+				NULL, 0);
 
 		/* SDMA */
 		else if (source_id == SOC21_INTSRC_SDMA_TRAP)
@@ -348,6 +349,13 @@ static void event_interrupt_wq_v11(struct kfd_node *dev,
 				break;
 			case SQ_INTERRUPT_WORD_ENCODING_INST:
 				print_sq_intr_info_inst(context_id0, context_id1);
+				sq_int_priv = REG_GET_FIELD(context_id0,
+						SQ_INTERRUPT_WORD_WAVE_CTXID0, PRIV);
+				if (sq_int_priv && (kfd_set_dbg_ev_from_interrupt(dev, pasid,
+						KFD_CTXID0_DOORBELL_ID(context_id0),
+						KFD_CTXID0_TRAP_CODE(context_id0),
+						NULL, 0)))
+					return;
 				break;
 			case SQ_INTERRUPT_WORD_ENCODING_ERROR:
 				print_sq_intr_info_error(context_id0, context_id1);
@@ -366,8 +374,8 @@ static void event_interrupt_wq_v11(struct kfd_node *dev,
 			kfd_signal_event_interrupt(pasid, context_id0 & 0xffffff, 24);
 		}
 
-	/*} else if (KFD_IRQ_IS_FENCE(client_id, source_id)) {
-		kfd_process_close_interrupt_drain(pasid);*/
+	} else if (KFD_IRQ_IS_FENCE(client_id, source_id)) {
+		kfd_process_close_interrupt_drain(pasid);
 	}
 }
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
index 8cf58be80f4e..d5c9f30552e3 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c
@@ -23,10 +23,40 @@
 
 #include "kfd_priv.h"
 #include "kfd_events.h"
+#include "kfd_debug.h"
 #include "soc15_int.h"
 #include "kfd_device_queue_manager.h"
 #include "kfd_smi_events.h"
 
+/*
+ * GFX9 SQ Interrupts
+ *
+ * There are 3 encoding types of interrupts sourced from SQ sent as a 44-bit
+ * packet to the Interrupt Handler:
+ * Auto - Generated by the SQG (various cmd overflows, timestamps etc)
+ * Wave - Generated by S_SENDMSG through a shader program
+ * Error - HW generated errors (Illegal instructions, Memviols, EDC etc)
+ *
+ * The 44-bit packet is mapped as {context_id1[7:0],context_id0[31:0]} plus
+ * 4-bits for VMID (SOC15_VMID_FROM_IH_ENTRY) as such:
+ *
+ * - context_id0[27:26]
+ * Encoding type (0 = Auto, 1 = Wave, 2 = Error)
+ *
+ * - context_id0[13]
+ * PRIV bit indicates that Wave S_SEND or error occurred within trap
+ *
+ * - {context_id1[7:0],context_id0[31:28],context_id0[11:0]}
+ * 24-bit data with the following layout per encoding type:
+ * Auto - only context_id0[8:0] is used, which reports various interrupts
+ * generated by SQG.  The rest is 0.
+ * Wave - user data sent from m0 via S_SENDMSG
+ * Error - Error type (context_id1[7:4]), Error Details (rest of bits)
+ *
+ * The other context_id bits show coordinates (SE/SH/CU/SIMD/WAVE) for wave
+ * S_SENDMSG and Errors.  These are 0 for Auto.
+ */
+
 enum SQ_INTERRUPT_WORD_ENCODING {
 	SQ_INTERRUPT_WORD_ENCODING_AUTO = 0x0,
 	SQ_INTERRUPT_WORD_ENCODING_INST,
@@ -84,12 +114,32 @@ enum SQ_INTERRUPT_ERROR_TYPE {
 #define SQ_INTERRUPT_WORD_WAVE_CTXID__SE_ID_MASK 0x03000000
 #define SQ_INTERRUPT_WORD_WAVE_CTXID__ENCODING_MASK 0x0c000000
 
+/* GFX9 SQ interrupt 24-bit data from context_id<0,1> */
 #define KFD_CONTEXT_ID_GET_SQ_INT_DATA(ctx0, ctx1)                             \
 	((ctx0 & 0xfff) | ((ctx0 >> 16) & 0xf000) | ((ctx1 << 16) & 0xff0000))
 
 #define KFD_SQ_INT_DATA__ERR_TYPE_MASK 0xF00000
 #define KFD_SQ_INT_DATA__ERR_TYPE__SHIFT 20
 
+/*
+ * The debugger will send user data(m0) with PRIV=1 to indicate it requires
+ * notification from the KFD with the following queue id (DOORBELL_ID) and
+ * trap code (TRAP_CODE).
+ */
+#define KFD_INT_DATA_DEBUG_DOORBELL_MASK	0x0003ff
+#define KFD_INT_DATA_DEBUG_TRAP_CODE_SHIFT	10
+#define KFD_INT_DATA_DEBUG_TRAP_CODE_MASK	0x07fc00
+#define KFD_DEBUG_DOORBELL_ID(sq_int_data)	((sq_int_data) &	\
+				KFD_INT_DATA_DEBUG_DOORBELL_MASK)
+#define KFD_DEBUG_TRAP_CODE(sq_int_data)	(((sq_int_data) &	\
+				KFD_INT_DATA_DEBUG_TRAP_CODE_MASK)	\
+				>> KFD_INT_DATA_DEBUG_TRAP_CODE_SHIFT)
+#define KFD_DEBUG_CP_BAD_OP_ECODE_MASK		0x3fffc00
+#define KFD_DEBUG_CP_BAD_OP_ECODE_SHIFT		10
+#define KFD_DEBUG_CP_BAD_OP_ECODE(ctxid0)	(((ctxid0) &		\
+				KFD_DEBUG_CP_BAD_OP_ECODE_MASK)		\
+				>> KFD_DEBUG_CP_BAD_OP_ECODE_SHIFT)
+
 static void event_interrupt_poison_consumption_v9(struct kfd_node *dev,
 				uint16_t pasid, uint16_t client_id)
 {
@@ -168,14 +218,16 @@ static bool event_interrupt_isr_v9(struct kfd_node *dev,
 	uint16_t source_id, client_id, pasid, vmid;
 	const uint32_t *data = ih_ring_entry;
 
+	source_id = SOC15_SOURCE_ID_FROM_IH_ENTRY(ih_ring_entry);
+	client_id = SOC15_CLIENT_ID_FROM_IH_ENTRY(ih_ring_entry);
+
 	/* Only handle interrupts from KFD VMIDs */
 	vmid = SOC15_VMID_FROM_IH_ENTRY(ih_ring_entry);
-	if (vmid < dev->vm_info.first_vmid_kfd ||
-	    vmid > dev->vm_info.last_vmid_kfd)
+	if (!KFD_IRQ_IS_FENCE(client_id, source_id) &&
+	   (vmid < dev->vm_info.first_vmid_kfd ||
+	    vmid > dev->vm_info.last_vmid_kfd))
 		return false;
 
-	source_id = SOC15_SOURCE_ID_FROM_IH_ENTRY(ih_ring_entry);
-	client_id = SOC15_CLIENT_ID_FROM_IH_ENTRY(ih_ring_entry);
 	pasid = SOC15_PASID_FROM_IH_ENTRY(ih_ring_entry);
 
 	/* Only handle clients we care about */
@@ -194,7 +246,8 @@ static bool event_interrupt_isr_v9(struct kfd_node *dev,
 	    client_id != SOC15_IH_CLIENTID_SE0SH &&
 	    client_id != SOC15_IH_CLIENTID_SE1SH &&
 	    client_id != SOC15_IH_CLIENTID_SE2SH &&
-	    client_id != SOC15_IH_CLIENTID_SE3SH)
+	    client_id != SOC15_IH_CLIENTID_SE3SH &&
+	    !KFD_IRQ_IS_FENCE(client_id, source_id))
 		return false;
 
 	/* This is a known issue for gfx9. Under non HWS, pasid is not set
@@ -247,6 +300,7 @@ static bool event_interrupt_isr_v9(struct kfd_node *dev,
 		source_id == SOC15_INTSRC_SDMA_ECC ||
 		source_id == SOC15_INTSRC_SQ_INTERRUPT_MSG ||
 		source_id == SOC15_INTSRC_CP_BAD_OPCODE ||
+		KFD_IRQ_IS_FENCE(client_id, source_id) ||
 		((client_id == SOC15_IH_CLIENTID_VMC ||
 		client_id == SOC15_IH_CLIENTID_VMC1 ||
 		client_id == SOC15_IH_CLIENTID_UTCL2) &&
@@ -302,6 +356,13 @@ static void event_interrupt_wq_v9(struct kfd_node *dev,
 					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID, SIMD_ID),
 					REG_GET_FIELD(context_id0, SQ_INTERRUPT_WORD_WAVE_CTXID, CU_ID),
 					sq_int_data);
+				if (context_id0 & SQ_INTERRUPT_WORD_WAVE_CTXID__PRIV_MASK) {
+					if (kfd_set_dbg_ev_from_interrupt(dev, pasid,
+							KFD_DEBUG_DOORBELL_ID(sq_int_data),
+							KFD_DEBUG_TRAP_CODE(sq_int_data),
+							NULL, 0))
+						return;
+				}
 				break;
 			case SQ_INTERRUPT_WORD_ENCODING_ERROR:
 				sq_intr_err = REG_GET_FIELD(sq_int_data, KFD_SQ_INT_DATA, ERR_TYPE);
@@ -324,8 +385,12 @@ static void event_interrupt_wq_v9(struct kfd_node *dev,
 				break;
 			}
 			kfd_signal_event_interrupt(pasid, context_id0 & 0xffffff, 24);
-		} else if (source_id == SOC15_INTSRC_CP_BAD_OPCODE)
-			kfd_signal_hw_exception_event(pasid);
+		} else if (source_id == SOC15_INTSRC_CP_BAD_OPCODE) {
+			kfd_set_dbg_ev_from_interrupt(dev, pasid,
+				KFD_DEBUG_DOORBELL_ID(context_id0),
+				KFD_EC_MASK(KFD_DEBUG_CP_BAD_OP_ECODE(context_id0)),
+				NULL, 0);
+		}
 	} else if (client_id == SOC15_IH_CLIENTID_SDMA0 ||
 		   client_id == SOC15_IH_CLIENTID_SDMA1 ||
 		   client_id == SOC15_IH_CLIENTID_SDMA2 ||
@@ -345,6 +410,7 @@ static void event_interrupt_wq_v9(struct kfd_node *dev,
 		   client_id == SOC15_IH_CLIENTID_UTCL2) {
 		struct kfd_vm_fault_info info = {0};
 		uint16_t ring_id = SOC15_RING_ID_FROM_IH_ENTRY(ih_ring_entry);
+		struct kfd_hsa_memory_exception_data exception_data;
 
 		if (client_id == SOC15_IH_CLIENTID_UTCL2 &&
 		    amdgpu_amdkfd_ras_query_utcl2_poison_status(dev->adev)) {
@@ -360,9 +426,23 @@ static void event_interrupt_wq_v9(struct kfd_node *dev,
 		info.prot_read  = ring_id & 0x10;
 		info.prot_write = ring_id & 0x20;
 
+		memset(&exception_data, 0, sizeof(exception_data));
+		exception_data.gpu_id = dev->id;
+		exception_data.va = (info.page_addr) << PAGE_SHIFT;
+		exception_data.failure.NotPresent = info.prot_valid ? 1 : 0;
+		exception_data.failure.NoExecute = info.prot_exec ? 1 : 0;
+		exception_data.failure.ReadOnly = info.prot_write ? 1 : 0;
+		exception_data.failure.imprecise = 0;
+
+		kfd_set_dbg_ev_from_interrupt(dev,
+						pasid,
+						-1,
+						KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION),
+						&exception_data,
+						sizeof(exception_data));
 		kfd_smi_event_update_vmfault(dev, pasid);
-		kfd_dqm_evict_pasid(dev->dqm, pasid);
-		kfd_signal_vm_fault_event(dev, pasid, &info, NULL);
+	} else if (KFD_IRQ_IS_FENCE(client_id, source_id)) {
+		kfd_process_close_interrupt_drain(pasid);
 	}
 }
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index a02fb939614a..cd2d56e5cdf0 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -963,6 +963,10 @@ struct kfd_process {
 	uint64_t exception_enable_mask;
 	uint64_t exception_status;
 
+	/* Used to drain stale interrupts */
+	wait_queue_head_t wait_irq_drain;
+	bool irq_drain_is_open;
+
 	/* shared virtual memory registered by this process */
 	struct svm_range_list svms;
 
@@ -1144,12 +1148,19 @@ int kfd_numa_node_to_apic_id(int numa_node_id);
 void kfd_double_confirm_iommu_support(struct kfd_dev *gpu);
 
 /* Interrupts */
+#define	KFD_IRQ_FENCE_CLIENTID	0xff
+#define	KFD_IRQ_FENCE_SOURCEID	0xff
+#define	KFD_IRQ_IS_FENCE(client, source)				\
+				((client) == KFD_IRQ_FENCE_CLIENTID &&	\
+				(source) == KFD_IRQ_FENCE_SOURCEID)
 int kfd_interrupt_init(struct kfd_node *dev);
 void kfd_interrupt_exit(struct kfd_node *dev);
 bool enqueue_ih_ring_entry(struct kfd_node *kfd, const void *ih_ring_entry);
 bool interrupt_is_wanted(struct kfd_node *dev,
 				const uint32_t *ih_ring_entry,
 				uint32_t *patched_ihre, bool *flag);
+int kfd_process_drain_interrupts(struct kfd_process_device *pdd);
+void kfd_process_close_interrupt_drain(unsigned int pasid);
 
 /* amdkfd Apertures */
 int kfd_init_apertures(struct kfd_process *process);
@@ -1421,6 +1432,7 @@ uint64_t kfd_get_number_elems(struct kfd_dev *kfd);
 /* Events */
 extern const struct kfd_event_interrupt_class event_interrupt_class_cik;
 extern const struct kfd_event_interrupt_class event_interrupt_class_v9;
+extern const struct kfd_event_interrupt_class event_interrupt_class_v10;
 extern const struct kfd_event_interrupt_class event_interrupt_class_v11;
 
 extern const struct kfd_device_global_init_class device_global_init_class_cik;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 3b7f219c9d06..3d3611705d41 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -862,6 +862,8 @@ struct kfd_process *kfd_create_process(struct task_struct *thread)
 		kfd_procfs_add_sysfs_stats(process);
 		kfd_procfs_add_sysfs_files(process);
 		kfd_procfs_add_sysfs_counters(process);
+
+		init_waitqueue_head(&process->wait_irq_drain);
 	}
 out:
 	if (!IS_ERR(process))
@@ -2136,6 +2138,51 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type)
 	}
 }
 
+/* assumes caller holds process lock. */
+int kfd_process_drain_interrupts(struct kfd_process_device *pdd)
+{
+	uint32_t irq_drain_fence[8];
+	int r = 0;
+
+	if (!KFD_IS_SOC15(pdd->dev))
+		return 0;
+
+	pdd->process->irq_drain_is_open = true;
+
+	memset(irq_drain_fence, 0, sizeof(irq_drain_fence));
+	irq_drain_fence[0] = (KFD_IRQ_FENCE_SOURCEID << 8) |
+							KFD_IRQ_FENCE_CLIENTID;
+	irq_drain_fence[3] = pdd->process->pasid;
+
+	/* ensure stale irqs scheduled KFD interrupts and send drain fence. */
+	if (amdgpu_amdkfd_send_close_event_drain_irq(pdd->dev->adev,
+						     irq_drain_fence)) {
+		pdd->process->irq_drain_is_open = false;
+		return 0;
+	}
+
+	r = wait_event_interruptible(pdd->process->wait_irq_drain,
+				     !READ_ONCE(pdd->process->irq_drain_is_open));
+	if (r)
+		pdd->process->irq_drain_is_open = false;
+
+	return r;
+}
+
+void kfd_process_close_interrupt_drain(unsigned int pasid)
+{
+	struct kfd_process *p;
+
+	p = kfd_lookup_process_by_pasid(pasid);
+
+	if (!p)
+		return;
+
+	WRITE_ONCE(p->irq_drain_is_open, false);
+	wake_up_all(&p->wait_irq_drain);
+	kfd_unref_process(p);
+}
+
 struct send_exception_work_handler_workarea {
 	struct work_struct work;
 	struct kfd_process *p;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index 43d432b5c5bc..70852a200d8f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -337,6 +337,10 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 		kq->queue->properties.queue_id = *qid;
 		pqn->kq = kq;
 		pqn->q = NULL;
+		retval = kfd_process_drain_interrupts(pdd);
+		if (retval)
+			break;
+
 		retval = dev->dqm->ops.create_kernel_queue(dev->dqm,
 							kq, &pdd->qpd);
 		break;
-- 
cgit 


From e90bf919f714ae2a658cdfd03238e7be9ce9185c Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Wed, 27 Apr 2022 10:24:37 -0400
Subject: drm/amdkfd: add debug set exceptions enabled operation

The debugger subscibes to nofication for requested exceptions on attach.
Allow the debugger to change its subsciption later on.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c |  3 +++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 36 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |  2 ++
 3 files changed, 41 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 615fa9ab36b7..dcf4b5f3886e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -2980,6 +2980,9 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 				args->send_runtime_event.exception_mask);
 		break;
 	case KFD_IOC_DBG_TRAP_SET_EXCEPTIONS_ENABLED:
+		kfd_dbg_set_enabled_debug_exception_mask(target,
+				args->set_exceptions_enabled.exception_mask);
+		break;
 	case KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_OVERRIDE:
 	case KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_MODE:
 	case KFD_IOC_DBG_TRAP_SUSPEND_QUEUES:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 68b657398d41..48a4e3cc2234 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -521,3 +521,39 @@ int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 
 	return r;
 }
+
+void kfd_dbg_set_enabled_debug_exception_mask(struct kfd_process *target,
+					uint64_t exception_set_mask)
+{
+	uint64_t found_mask = 0;
+	struct process_queue_manager *pqm;
+	struct process_queue_node *pqn;
+	static const char write_data = '.';
+	loff_t pos = 0;
+	int i;
+
+	mutex_lock(&target->event_mutex);
+
+	found_mask |= target->exception_status;
+
+	pqm = &target->pqm;
+	list_for_each_entry(pqn, &pqm->queues, process_queue_list) {
+		if (!pqn)
+			continue;
+
+		found_mask |= pqn->q->properties.exception_status;
+	}
+
+	for (i = 0; i < target->n_pdds; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+
+		found_mask |= pdd->exception_status;
+	}
+
+	if (exception_set_mask & found_mask)
+		kernel_write(target->dbg_ev_file, &write_data, 1, &pos);
+
+	target->exception_enable_mask = exception_set_mask;
+
+	mutex_unlock(&target->event_mutex);
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 5153ccbd7fd1..6c1054a08872 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -56,6 +56,8 @@ static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
 
 void debug_event_write_work_handler(struct work_struct *work);
 
+void kfd_dbg_set_enabled_debug_exception_mask(struct kfd_process *target,
+					uint64_t exception_set_mask);
 /*
  * If GFX off is enabled, chips that do not support RLC restore for the debug
  * registers will disable GFX off temporarily for the entire debug session.
-- 
cgit 


From 101827e13026a981e887527620fe9710adc0e481 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Wed, 27 Apr 2022 13:18:10 -0400
Subject: drm/amdkfd: add debug wave launch override operation

This operation allows the debugger to override the enabled HW
exceptions on the device.

On debug devices that only support the debugging of a single process,
the HW exceptions are global and set through the SPI_GDBG_TRAP_MASK
register.
Because they are global, only address watch exceptions are allowed to
be enabled.  In other words, the debugger must preserve all non-address
watch exception states in normal mode operation by barring a full
replacement override or a non-address watch override request.

For multi-process debugging, all HW exception overrides are per-VMID so
all exceptions can be overridden or fully replaced.

In order for the debugger to know what is permissible, returned the
supported override mask back to the debugger along with the previously
enable overrides.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c   | 47 ++++++++++++
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c    |  2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c | 55 ++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h | 10 +++
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c   |  5 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c | 87 +++++++++++++++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c  | 55 ++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h  | 10 +++
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c           |  7 ++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c             | 69 +++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h             |  6 ++
 11 files changed, 351 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
index b811a0985050..d7881bbd828d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
@@ -25,6 +25,7 @@
 #include "amdgpu_amdkfd_gfx_v9.h"
 #include "gc/gc_9_4_2_offset.h"
 #include "gc/gc_9_4_2_sh_mask.h"
+#include <uapi/linux/kfd_ioctl.h>
 
 /*
  * Returns TRAP_EN, EXCP_EN and EXCP_REPLACE.
@@ -62,6 +63,50 @@ static uint32_t kgd_aldebaran_disable_debug_trap(struct amdgpu_device *adev,
 	return data;
 }
 
+static int kgd_aldebaran_validate_trap_override_request(struct amdgpu_device *adev,
+							uint32_t trap_override,
+							uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_FP_INVALID |
+				KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+				KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+				KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+				KFD_DBG_TRAP_MASK_FP_INEXACT |
+				KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+				KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION;
+
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR &&
+			trap_override != KFD_DBG_TRAP_OVERRIDE_REPLACE)
+		return -EPERM;
+
+	return 0;
+}
+
+/* returns TRAP_EN, EXCP_EN and EXCP_RPLACE. */
+static uint32_t kgd_aldebaran_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					uint32_t vmid,
+					uint32_t trap_override,
+					uint32_t trap_mask_bits,
+					uint32_t trap_mask_request,
+					uint32_t *trap_mask_prev,
+					uint32_t kfd_dbg_trap_cntl_prev)
+
+{
+	uint32_t data = 0;
+
+	*trap_mask_prev = REG_GET_FIELD(kfd_dbg_trap_cntl_prev, SPI_GDBG_PER_VMID_CNTL, EXCP_EN);
+	trap_mask_bits = (trap_mask_bits & trap_mask_request) |
+		(*trap_mask_prev & ~trap_mask_request);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, trap_mask_bits);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, trap_override);
+
+	return data;
+}
+
 const struct kfd2kgd_calls aldebaran_kfd2kgd = {
 	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
 	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
@@ -82,6 +127,8 @@ const struct kfd2kgd_calls aldebaran_kfd2kgd = {
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
 	.enable_debug_trap = kgd_aldebaran_enable_debug_trap,
 	.disable_debug_trap = kgd_aldebaran_disable_debug_trap,
+	.validate_trap_override_request = kgd_aldebaran_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_aldebaran_set_wave_launch_trap_override,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
index a62bd0068515..ec2587664001 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
@@ -410,6 +410,8 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = {
 				kgd_gfx_v9_set_vm_context_page_table_base,
 	.enable_debug_trap = kgd_arcturus_enable_debug_trap,
 	.disable_debug_trap = kgd_arcturus_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
index 98006c7021dd..7ea0362dcab3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
@@ -32,6 +32,7 @@
 #include "v10_structs.h"
 #include "nv.h"
 #include "nvd.h"
+#include <uapi/linux/kfd_ioctl.h>
 
 enum hqd_dequeue_request_type {
 	NO_ACTION = 0,
@@ -803,6 +804,58 @@ uint32_t kgd_gfx_v10_disable_debug_trap(struct amdgpu_device *adev,
 	return 0;
 }
 
+int kgd_gfx_v10_validate_trap_override_request(struct amdgpu_device *adev,
+					      uint32_t trap_override,
+					      uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH;
+
+	/* The SPI_GDBG_TRAP_MASK register is global and affects all
+	 * processes. Only allow OR-ing the address-watch bit, since
+	 * this only affects processes under the debugger. Other bits
+	 * should stay 0 to avoid the debugger interfering with other
+	 * processes.
+	 */
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR)
+		return -EINVAL;
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					      uint32_t vmid,
+					      uint32_t trap_override,
+					      uint32_t trap_mask_bits,
+					      uint32_t trap_mask_request,
+					      uint32_t *trap_mask_prev,
+					      uint32_t kfd_dbg_trap_cntl_prev)
+{
+	uint32_t data, wave_cntl_prev;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	wave_cntl_prev = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
+
+	data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK));
+	*trap_mask_prev = REG_GET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN);
+
+	trap_mask_bits = (trap_mask_bits & trap_mask_request) |
+		(*trap_mask_prev & ~trap_mask_request);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN, trap_mask_bits);
+	data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, REPLACE, trap_override);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), data);
+
+	/* We need to preserve wave launch mode stall settings. */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), wave_cntl_prev);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
 /* kgd_gfx_v10_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
  * The values read are:
  *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
@@ -889,6 +942,8 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = {
 	.set_vm_context_page_table_base = set_vm_context_page_table_base,
 	.enable_debug_trap = kgd_gfx_v10_enable_debug_trap,
 	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
 	.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v10_build_grace_period_packet_info,
 	.program_trap_handler_settings = program_trap_handler_settings,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
index 1e993a213646..57339fa12807 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
@@ -26,6 +26,16 @@ uint32_t kgd_gfx_v10_enable_debug_trap(struct amdgpu_device *adev,
 uint32_t kgd_gfx_v10_disable_debug_trap(struct amdgpu_device *adev,
 					bool keep_trap_enabled,
 					uint32_t vmid);
+int kgd_gfx_v10_validate_trap_override_request(struct amdgpu_device *adev,
+					     uint32_t trap_override,
+					     uint32_t *trap_mask_supported);
+uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					     uint32_t vmid,
+					     uint32_t trap_override,
+					     uint32_t trap_mask_bits,
+					     uint32_t trap_mask_request,
+					     uint32_t *trap_mask_prev,
+					     uint32_t kfd_dbg_trap_cntl_prev);
 void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev, uint32_t *wait_times);
 void kgd_gfx_v10_build_grace_period_packet_info(struct amdgpu_device *adev,
 					       uint32_t wait_times,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
index 387bdf4823c9..7120927fed15 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
@@ -675,5 +675,8 @@ const struct kfd2kgd_calls gfx_v10_3_kfd2kgd = {
 	.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v10_build_grace_period_packet_info,
 	.enable_debug_trap = kgd_gfx_v10_enable_debug_trap,
-	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap
+	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override
+
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
index cc954cf248ca..ae0c4707919f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
@@ -30,6 +30,7 @@
 #include "soc15d.h"
 #include "v11_structs.h"
 #include "soc21.h"
+#include <uapi/linux/kfd_ioctl.h>
 
 enum hqd_dequeue_request_type {
 	NO_ACTION = 0,
@@ -643,6 +644,88 @@ static uint32_t kgd_gfx_v11_disable_debug_trap(struct amdgpu_device *adev,
 	return data;
 }
 
+static int kgd_gfx_v11_validate_trap_override_request(struct amdgpu_device *adev,
+							uint32_t trap_override,
+							uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_FP_INVALID |
+				KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+				KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+				KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+				KFD_DBG_TRAP_MASK_FP_INEXACT |
+				KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+				KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+				KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION;
+
+	if (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(11, 0, 4))
+		*trap_mask_supported |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START |
+					KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
+
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR &&
+			trap_override != KFD_DBG_TRAP_OVERRIDE_REPLACE)
+		return -EPERM;
+
+	return 0;
+}
+
+static uint32_t trap_mask_map_sw_to_hw(uint32_t mask)
+{
+	uint32_t trap_on_start = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START) ? 1 : 0;
+	uint32_t trap_on_end = (mask & KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END) ? 1 : 0;
+	uint32_t excp_en = mask & (KFD_DBG_TRAP_MASK_FP_INVALID |
+			KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL |
+			KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO |
+			KFD_DBG_TRAP_MASK_FP_OVERFLOW |
+			KFD_DBG_TRAP_MASK_FP_UNDERFLOW |
+			KFD_DBG_TRAP_MASK_FP_INEXACT |
+			KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO |
+			KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH |
+			KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION);
+	uint32_t ret;
+
+	ret = REG_SET_FIELD(0, SPI_GDBG_PER_VMID_CNTL, EXCP_EN, excp_en);
+	ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START, trap_on_start);
+	ret = REG_SET_FIELD(ret, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END, trap_on_end);
+
+	return ret;
+}
+
+static uint32_t trap_mask_map_hw_to_sw(uint32_t mask)
+{
+	uint32_t ret = REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, EXCP_EN);
+
+	if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_START))
+		ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START;
+
+	if (REG_GET_FIELD(mask, SPI_GDBG_PER_VMID_CNTL, TRAP_ON_END))
+		ret |= KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END;
+
+	return ret;
+}
+
+/* Returns TRAP_EN, EXCP_EN and EXCP_REPLACE. */
+static uint32_t kgd_gfx_v11_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					uint32_t vmid,
+					uint32_t trap_override,
+					uint32_t trap_mask_bits,
+					uint32_t trap_mask_request,
+					uint32_t *trap_mask_prev,
+					uint32_t kfd_dbg_trap_cntl_prev)
+{
+	uint32_t data = 0;
+
+	*trap_mask_prev = trap_mask_map_hw_to_sw(kfd_dbg_trap_cntl_prev);
+
+	data = (trap_mask_bits & trap_mask_request) | (*trap_mask_prev & ~trap_mask_request);
+	data = trap_mask_map_sw_to_hw(data);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, EXCP_REPLACE, trap_override);
+
+	return data;
+}
+
 const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
 	.program_sh_mem_settings = program_sh_mem_settings_v11,
 	.set_pasid_vmid_mapping = set_pasid_vmid_mapping_v11,
@@ -660,5 +743,7 @@ const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
 	.get_atc_vmid_pasid_mapping_info = NULL,
 	.set_vm_context_page_table_base = set_vm_context_page_table_base_v11,
 	.enable_debug_trap = kgd_gfx_v11_enable_debug_trap,
-	.disable_debug_trap = kgd_gfx_v11_disable_debug_trap
+	.disable_debug_trap = kgd_gfx_v11_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v11_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v11_set_wave_launch_trap_override
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
index 829ee720cc44..ce5c256f0517 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -38,6 +38,7 @@
 #include "soc15d.h"
 #include "gfx_v9_0.h"
 #include "amdgpu_amdkfd_gfx_v9.h"
+#include <uapi/linux/kfd_ioctl.h>
 
 enum hqd_dequeue_request_type {
 	NO_ACTION = 0,
@@ -739,6 +740,58 @@ uint32_t kgd_gfx_v9_disable_debug_trap(struct amdgpu_device *adev,
 	return 0;
 }
 
+int kgd_gfx_v9_validate_trap_override_request(struct amdgpu_device *adev,
+					uint32_t trap_override,
+					uint32_t *trap_mask_supported)
+{
+	*trap_mask_supported &= KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH;
+
+	/* The SPI_GDBG_TRAP_MASK register is global and affects all
+	 * processes. Only allow OR-ing the address-watch bit, since
+	 * this only affects processes under the debugger. Other bits
+	 * should stay 0 to avoid the debugger interfering with other
+	 * processes.
+	 */
+	if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR)
+		return -EINVAL;
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v9_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					     uint32_t vmid,
+					     uint32_t trap_override,
+					     uint32_t trap_mask_bits,
+					     uint32_t trap_mask_request,
+					     uint32_t *trap_mask_prev,
+					     uint32_t kfd_dbg_cntl_prev)
+{
+	uint32_t data, wave_cntl_prev;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	wave_cntl_prev = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL));
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
+
+	data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK));
+	*trap_mask_prev = REG_GET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN);
+
+	trap_mask_bits = (trap_mask_bits & trap_mask_request) |
+		(*trap_mask_prev & ~trap_mask_request);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN, trap_mask_bits);
+	data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, REPLACE, trap_override);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), data);
+
+	/* We need to preserve wave launch mode stall settings. */
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), wave_cntl_prev);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
 /* kgd_gfx_v9_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
  * The values read are:
  *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
@@ -1010,6 +1063,8 @@ const struct kfd2kgd_calls gfx_v9_kfd2kgd = {
 	.set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
 	.enable_debug_trap = kgd_gfx_v9_enable_debug_trap,
 	.disable_debug_trap = kgd_gfx_v9_disable_debug_trap,
+	.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
+	.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
index fed5b7f18b1a..76812ddd35b1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
@@ -71,6 +71,16 @@ uint32_t kgd_gfx_v9_enable_debug_trap(struct amdgpu_device *adev,
 uint32_t kgd_gfx_v9_disable_debug_trap(struct amdgpu_device *adev,
 					bool keep_trap_enabled,
 					uint32_t vmid);
+int kgd_gfx_v9_validate_trap_override_request(struct amdgpu_device *adev,
+					     uint32_t trap_override,
+					     uint32_t *trap_mask_supported);
+uint32_t kgd_gfx_v9_set_wave_launch_trap_override(struct amdgpu_device *adev,
+					     uint32_t vmid,
+					     uint32_t trap_override,
+					     uint32_t trap_mask_bits,
+					     uint32_t trap_mask_request,
+					     uint32_t *trap_mask_prev,
+					     uint32_t kfd_dbg_trap_cntl_prev);
 void kgd_gfx_v9_get_iq_wait_times(struct amdgpu_device *adev, uint32_t *wait_times);
 void kgd_gfx_v9_build_grace_period_packet_info(struct amdgpu_device *adev,
 					       uint32_t wait_times,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index dcf4b5f3886e..7fa249807671 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -2984,6 +2984,13 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 				args->set_exceptions_enabled.exception_mask);
 		break;
 	case KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_OVERRIDE:
+		r = kfd_dbg_trap_set_wave_launch_override(target,
+				args->launch_override.override_mode,
+				args->launch_override.enable_mask,
+				args->launch_override.support_request_mask,
+				&args->launch_override.enable_mask,
+				&args->launch_override.support_request_mask);
+		break;
 	case KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_MODE:
 	case KFD_IOC_DBG_TRAP_SUSPEND_QUEUES:
 	case KFD_IOC_DBG_TRAP_RESUME_QUEUES:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 48a4e3cc2234..733390fb2459 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -522,6 +522,75 @@ int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 	return r;
 }
 
+static int kfd_dbg_validate_trap_override_request(struct kfd_process *p,
+						uint32_t trap_override,
+						uint32_t trap_mask_request,
+						uint32_t *trap_mask_supported)
+{
+	int i = 0;
+
+	*trap_mask_supported = 0xffffffff;
+
+	for (i = 0; i < p->n_pdds; i++) {
+		struct kfd_process_device *pdd = p->pdds[i];
+		int err = pdd->dev->kfd2kgd->validate_trap_override_request(
+								pdd->dev->adev,
+								trap_override,
+								trap_mask_supported);
+
+		if (err)
+			return err;
+	}
+
+	if (trap_mask_request & ~*trap_mask_supported)
+		return -EACCES;
+
+	return 0;
+}
+
+int kfd_dbg_trap_set_wave_launch_override(struct kfd_process *target,
+					uint32_t trap_override,
+					uint32_t trap_mask_bits,
+					uint32_t trap_mask_request,
+					uint32_t *trap_mask_prev,
+					uint32_t *trap_mask_supported)
+{
+	int r = 0, i;
+
+	r = kfd_dbg_validate_trap_override_request(target,
+						trap_override,
+						trap_mask_request,
+						trap_mask_supported);
+
+	if (r)
+		return r;
+
+	for (i = 0; i < target->n_pdds; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+
+		amdgpu_gfx_off_ctrl(pdd->dev->adev, false);
+		pdd->spi_dbg_override = pdd->dev->kfd2kgd->set_wave_launch_trap_override(
+				pdd->dev->adev,
+				pdd->dev->vm_info.last_vmid_kfd,
+				trap_override,
+				trap_mask_bits,
+				trap_mask_request,
+				trap_mask_prev,
+				pdd->spi_dbg_override);
+		amdgpu_gfx_off_ctrl(pdd->dev->adev, true);
+
+		if (!pdd->dev->kfd->shared_resources.enable_mes)
+			r = debug_refresh_runlist(pdd->dev->dqm);
+		else
+			r = kfd_dbg_set_mes_debug_mode(pdd);
+
+		if (r)
+			break;
+	}
+
+	return r;
+}
+
 void kfd_dbg_set_enabled_debug_exception_mask(struct kfd_process *target,
 					uint64_t exception_set_mask)
 {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 6c1054a08872..c9245221aa76 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -42,6 +42,12 @@ int kfd_dbg_trap_disable(struct kfd_process *target);
 int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
 			void __user *runtime_info,
 			uint32_t *runtime_info_size);
+int kfd_dbg_trap_set_wave_launch_override(struct kfd_process *target,
+					uint32_t trap_override,
+					uint32_t trap_mask_bits,
+					uint32_t trap_mask_request,
+					uint32_t *trap_mask_prev,
+					uint32_t *trap_mask_supported);
 
 int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
 					unsigned int dev_id,
-- 
cgit 


From aea1b4738bebd8092bd437ce0b03aa9587fc20a7 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Mon, 2 May 2022 11:45:05 -0400
Subject: drm/amdkfd: add debug wave launch mode operation

Allow the debugger to set wave behaviour on to either normally operate,
halt at launch, trap on every instruction, terminate immediately or
stall on allocation.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c   | 12 ++++++++
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c    |  1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c | 25 +++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h |  3 ++
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c   |  3 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c | 14 ++++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c  | 25 +++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h  |  3 ++
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c           |  3 ++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c             | 36 +++++++++++++++++++++-
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h             |  2 ++
 11 files changed, 124 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
index d7881bbd828d..774ecfc3451a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
@@ -107,6 +107,17 @@ static uint32_t kgd_aldebaran_set_wave_launch_trap_override(struct amdgpu_device
 	return data;
 }
 
+static uint32_t kgd_aldebaran_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE, wave_launch_mode);
+
+	return data;
+}
+
 const struct kfd2kgd_calls aldebaran_kfd2kgd = {
 	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
 	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
@@ -129,6 +140,7 @@ const struct kfd2kgd_calls aldebaran_kfd2kgd = {
 	.disable_debug_trap = kgd_aldebaran_disable_debug_trap,
 	.validate_trap_override_request = kgd_aldebaran_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_aldebaran_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_aldebaran_set_wave_launch_mode,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
index ec2587664001..fbdc1b7b1e42 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
@@ -412,6 +412,7 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = {
 	.disable_debug_trap = kgd_arcturus_disable_debug_trap,
 	.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
index 7ea0362dcab3..a7a6edda557f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
@@ -856,6 +856,30 @@ uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev,
 	return 0;
 }
 
+uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+	bool is_mode_set = !!wave_launch_mode;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
+			VMID_MASK, is_mode_set ? 1 << vmid : 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
+			MODE, is_mode_set ? wave_launch_mode : 0);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data);
+
+	kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
 /* kgd_gfx_v10_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
  * The values read are:
  *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
@@ -944,6 +968,7 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = {
 	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
 	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode,
 	.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v10_build_grace_period_packet_info,
 	.program_trap_handler_settings = program_trap_handler_settings,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
index 57339fa12807..3a6aca2b0eaa 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
@@ -36,6 +36,9 @@ uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev,
 					     uint32_t trap_mask_request,
 					     uint32_t *trap_mask_prev,
 					     uint32_t kfd_dbg_trap_cntl_prev);
+uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev,
+					 uint8_t wave_launch_mode,
+					 uint32_t vmid);
 void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev, uint32_t *wait_times);
 void kgd_gfx_v10_build_grace_period_packet_info(struct amdgpu_device *adev,
 					       uint32_t wait_times,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
index 7120927fed15..ed36b433a48b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
@@ -677,6 +677,7 @@ const struct kfd2kgd_calls gfx_v10_3_kfd2kgd = {
 	.enable_debug_trap = kgd_gfx_v10_enable_debug_trap,
 	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
 	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
-	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override
+	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode
 
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
index ae0c4707919f..9711d5128d09 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
@@ -726,6 +726,17 @@ static uint32_t kgd_gfx_v11_set_wave_launch_trap_override(struct amdgpu_device *
 	return data;
 }
 
+static uint32_t kgd_gfx_v11_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+
+	data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE, wave_launch_mode);
+
+	return data;
+}
+
 const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
 	.program_sh_mem_settings = program_sh_mem_settings_v11,
 	.set_pasid_vmid_mapping = set_pasid_vmid_mapping_v11,
@@ -745,5 +756,6 @@ const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
 	.enable_debug_trap = kgd_gfx_v11_enable_debug_trap,
 	.disable_debug_trap = kgd_gfx_v11_disable_debug_trap,
 	.validate_trap_override_request = kgd_gfx_v11_validate_trap_override_request,
-	.set_wave_launch_trap_override = kgd_gfx_v11_set_wave_launch_trap_override
+	.set_wave_launch_trap_override = kgd_gfx_v11_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v11_set_wave_launch_mode
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
index ce5c256f0517..0acc0c18dfe6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -792,6 +792,30 @@ uint32_t kgd_gfx_v9_set_wave_launch_trap_override(struct amdgpu_device *adev,
 	return 0;
 }
 
+uint32_t kgd_gfx_v9_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid)
+{
+	uint32_t data = 0;
+	bool is_mode_set = !!wave_launch_mode;
+
+	mutex_lock(&adev->grbm_idx_mutex);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true);
+
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
+		VMID_MASK, is_mode_set ? 1 << vmid : 0);
+	data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2,
+		MODE, is_mode_set ? wave_launch_mode : 0);
+	WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data);
+
+	kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false);
+
+	mutex_unlock(&adev->grbm_idx_mutex);
+
+	return 0;
+}
+
 /* kgd_gfx_v9_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
  * The values read are:
  *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
@@ -1065,6 +1089,7 @@ const struct kfd2kgd_calls gfx_v9_kfd2kgd = {
 	.disable_debug_trap = kgd_gfx_v9_disable_debug_trap,
 	.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
+	.set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
index 76812ddd35b1..18f4970ac8e4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
@@ -74,6 +74,9 @@ uint32_t kgd_gfx_v9_disable_debug_trap(struct amdgpu_device *adev,
 int kgd_gfx_v9_validate_trap_override_request(struct amdgpu_device *adev,
 					     uint32_t trap_override,
 					     uint32_t *trap_mask_supported);
+uint32_t kgd_gfx_v9_set_wave_launch_mode(struct amdgpu_device *adev,
+					uint8_t wave_launch_mode,
+					uint32_t vmid);
 uint32_t kgd_gfx_v9_set_wave_launch_trap_override(struct amdgpu_device *adev,
 					     uint32_t vmid,
 					     uint32_t trap_override,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 7fa249807671..a6570b124b2b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -2992,6 +2992,9 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 				&args->launch_override.support_request_mask);
 		break;
 	case KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_MODE:
+		r = kfd_dbg_trap_set_wave_launch_mode(target,
+				args->launch_mode.launch_mode);
+		break;
 	case KFD_IOC_DBG_TRAP_SUSPEND_QUEUES:
 	case KFD_IOC_DBG_TRAP_RESUME_QUEUES:
 	case KFD_IOC_DBG_TRAP_SET_NODE_ADDRESS_WATCH:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 733390fb2459..53c3418562d4 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -301,8 +301,10 @@ void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind
 {
 	int i;
 
-	if (!unwind)
+	if (!unwind) {
 		cancel_work_sync(&target->debug_event_workarea);
+		kfd_dbg_trap_set_wave_launch_mode(target, 0);
+	}
 
 	for (i = 0; i < target->n_pdds; i++) {
 		struct kfd_process_device *pdd = target->pdds[i];
@@ -591,6 +593,38 @@ int kfd_dbg_trap_set_wave_launch_override(struct kfd_process *target,
 	return r;
 }
 
+int kfd_dbg_trap_set_wave_launch_mode(struct kfd_process *target,
+					uint8_t wave_launch_mode)
+{
+	int r = 0, i;
+
+	if (wave_launch_mode != KFD_DBG_TRAP_WAVE_LAUNCH_MODE_NORMAL &&
+			wave_launch_mode != KFD_DBG_TRAP_WAVE_LAUNCH_MODE_HALT &&
+			wave_launch_mode != KFD_DBG_TRAP_WAVE_LAUNCH_MODE_DEBUG)
+		return -EINVAL;
+
+	for (i = 0; i < target->n_pdds; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+
+		amdgpu_gfx_off_ctrl(pdd->dev->adev, false);
+		pdd->spi_dbg_launch_mode = pdd->dev->kfd2kgd->set_wave_launch_mode(
+				pdd->dev->adev,
+				wave_launch_mode,
+				pdd->dev->vm_info.last_vmid_kfd);
+		amdgpu_gfx_off_ctrl(pdd->dev->adev, true);
+
+		if (!pdd->dev->kfd->shared_resources.enable_mes)
+			r = debug_refresh_runlist(pdd->dev->dqm);
+		else
+			r = kfd_dbg_set_mes_debug_mode(pdd);
+
+		if (r)
+			break;
+	}
+
+	return r;
+}
+
 void kfd_dbg_set_enabled_debug_exception_mask(struct kfd_process *target,
 					uint64_t exception_set_mask)
 {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index c9245221aa76..cb17869437c5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -48,6 +48,8 @@ int kfd_dbg_trap_set_wave_launch_override(struct kfd_process *target,
 					uint32_t trap_mask_request,
 					uint32_t *trap_mask_prev,
 					uint32_t *trap_mask_supported);
+int kfd_dbg_trap_set_wave_launch_mode(struct kfd_process *target,
+					uint8_t wave_launch_mode);
 
 int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
 					unsigned int dev_id,
-- 
cgit 


From e0f85f4690d089cc1a60337decafb1acf7eec45e Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Fri, 6 May 2022 14:58:55 -0400
Subject: drm/amdkfd: add debug set and clear address watch points operation

Shader read, write and atomic memory operations can be alerted to the
debugger as an address watch exception.

Allow the debugger to pass in a watch point to a particular memory
address per device.

Note that there exists only 4 watch points per devices to date, so have
the KFD keep track of what watch points are allocated or not.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c   |  51 ++++++++
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c    |   2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c |  78 ++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h |   8 ++
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c   |   5 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c |  52 +++++++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c  |  77 ++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h  |   8 ++
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c           |  24 ++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c             | 136 +++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h             |   8 +-
 drivers/gpu/drm/amd/amdkfd/kfd_device.c            |   2 +
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h              |   6 +-
 13 files changed, 452 insertions(+), 5 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
index 774ecfc3451a..efd6a72aab4e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
@@ -118,6 +118,55 @@ static uint32_t kgd_aldebaran_set_wave_launch_mode(struct amdgpu_device *adev,
 	return data;
 }
 
+#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
+static uint32_t kgd_gfx_aldebaran_set_address_watch(
+					struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 6);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	return watch_address_cntl;
+}
+
+uint32_t kgd_gfx_aldebaran_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id)
+{
+	return 0;
+}
+
 const struct kfd2kgd_calls aldebaran_kfd2kgd = {
 	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
 	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
@@ -141,6 +190,8 @@ const struct kfd2kgd_calls aldebaran_kfd2kgd = {
 	.validate_trap_override_request = kgd_aldebaran_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_aldebaran_set_wave_launch_trap_override,
 	.set_wave_launch_mode = kgd_aldebaran_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_aldebaran_set_address_watch,
+	.clear_address_watch = kgd_gfx_aldebaran_clear_address_watch,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
index fbdc1b7b1e42..6df215aba4c4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
@@ -413,6 +413,8 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = {
 	.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
 	.set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v9_set_address_watch,
+	.clear_address_watch = kgd_gfx_v9_clear_address_watch,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
index a7a6edda557f..8ad7a7779e14 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
@@ -880,6 +880,82 @@ uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev,
 	return 0;
 }
 
+#define TCP_WATCH_STRIDE (mmTCP_WATCH1_ADDR_H - mmTCP_WATCH0_ADDR_H)
+uint32_t kgd_gfx_v10_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VMID,
+			debug_vmid);
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 7);
+
+	/* Turning off this watch point until we set all the registers */
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			0);
+
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	/* Enable the watch point */
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v10_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id)
+{
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+
+	WREG32((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	return 0;
+}
+
+
 /* kgd_gfx_v10_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
  * The values read are:
  *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
@@ -969,6 +1045,8 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = {
 	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
 	.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v10_set_address_watch,
+	.clear_address_watch = kgd_gfx_v10_clear_address_watch,
 	.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v10_build_grace_period_packet_info,
 	.program_trap_handler_settings = program_trap_handler_settings,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
index 3a6aca2b0eaa..e6b70196071a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h
@@ -39,6 +39,14 @@ uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev,
 uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev,
 					 uint8_t wave_launch_mode,
 					 uint32_t vmid);
+uint32_t kgd_gfx_v10_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid);
+uint32_t kgd_gfx_v10_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id);
 void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev, uint32_t *wait_times);
 void kgd_gfx_v10_build_grace_period_packet_info(struct amdgpu_device *adev,
 					       uint32_t wait_times,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
index ed36b433a48b..8c8437a4383f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c
@@ -678,6 +678,7 @@ const struct kfd2kgd_calls gfx_v10_3_kfd2kgd = {
 	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
 	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_gfx_v10_set_wave_launch_trap_override,
-	.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode
-
+	.set_wave_launch_mode = kgd_gfx_v10_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v10_set_address_watch,
+	.clear_address_watch = kgd_gfx_v10_clear_address_watch
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
index 9711d5128d09..52efa690a3c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c
@@ -737,6 +737,54 @@ static uint32_t kgd_gfx_v11_set_wave_launch_mode(struct amdgpu_device *adev,
 	return data;
 }
 
+#define TCP_WATCH_STRIDE (regTCP_WATCH1_ADDR_H - regTCP_WATCH0_ADDR_H)
+static uint32_t kgd_gfx_v11_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 7);
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, regTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	return watch_address_cntl;
+}
+
+uint32_t kgd_gfx_v11_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id)
+{
+	return 0;
+}
+
 const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
 	.program_sh_mem_settings = program_sh_mem_settings_v11,
 	.set_pasid_vmid_mapping = set_pasid_vmid_mapping_v11,
@@ -757,5 +805,7 @@ const struct kfd2kgd_calls gfx_v11_kfd2kgd = {
 	.disable_debug_trap = kgd_gfx_v11_disable_debug_trap,
 	.validate_trap_override_request = kgd_gfx_v11_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_gfx_v11_set_wave_launch_trap_override,
-	.set_wave_launch_mode = kgd_gfx_v11_set_wave_launch_mode
+	.set_wave_launch_mode = kgd_gfx_v11_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v11_set_address_watch,
+	.clear_address_watch = kgd_gfx_v11_clear_address_watch
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
index 0acc0c18dfe6..51d93fb13ea3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -816,6 +816,81 @@ uint32_t kgd_gfx_v9_set_wave_launch_mode(struct amdgpu_device *adev,
 	return 0;
 }
 
+#define TCP_WATCH_STRIDE (mmTCP_WATCH1_ADDR_H - mmTCP_WATCH0_ADDR_H)
+uint32_t kgd_gfx_v9_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid)
+{
+	uint32_t watch_address_high;
+	uint32_t watch_address_low;
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+
+	watch_address_low = lower_32_bits(watch_address);
+	watch_address_high = upper_32_bits(watch_address) & 0xffff;
+
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VMID,
+			debug_vmid);
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MODE,
+			watch_mode);
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			MASK,
+			watch_address_mask >> 6);
+
+	/* Turning off this watch point until we set all the registers */
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			0);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_H) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_high);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_L) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_low);
+
+	/* Enable the watch point */
+	watch_address_cntl = REG_SET_FIELD(watch_address_cntl,
+			TCP_WATCH0_CNTL,
+			VALID,
+			1);
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	return 0;
+}
+
+uint32_t kgd_gfx_v9_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id)
+{
+	uint32_t watch_address_cntl;
+
+	watch_address_cntl = 0;
+
+	WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) +
+			(watch_id * TCP_WATCH_STRIDE)),
+			watch_address_cntl);
+
+	return 0;
+}
+
 /* kgd_gfx_v9_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values
  * The values read are:
  *     ib_offload_wait_time     -- Wait Count for Indirect Buffer Offloads.
@@ -1090,6 +1165,8 @@ const struct kfd2kgd_calls gfx_v9_kfd2kgd = {
 	.validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request,
 	.set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override,
 	.set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode,
+	.set_address_watch = kgd_gfx_v9_set_address_watch,
+	.clear_address_watch = kgd_gfx_v9_clear_address_watch,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
index 18f4970ac8e4..5f54bff0db49 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
@@ -84,6 +84,14 @@ uint32_t kgd_gfx_v9_set_wave_launch_trap_override(struct amdgpu_device *adev,
 					     uint32_t trap_mask_request,
 					     uint32_t *trap_mask_prev,
 					     uint32_t kfd_dbg_trap_cntl_prev);
+uint32_t kgd_gfx_v9_set_address_watch(struct amdgpu_device *adev,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t watch_id,
+					uint32_t watch_mode,
+					uint32_t debug_vmid);
+uint32_t kgd_gfx_v9_clear_address_watch(struct amdgpu_device *adev,
+					uint32_t watch_id);
 void kgd_gfx_v9_get_iq_wait_times(struct amdgpu_device *adev, uint32_t *wait_times);
 void kgd_gfx_v9_build_grace_period_packet_info(struct amdgpu_device *adev,
 					       uint32_t wait_times,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 1fae97df7a1e..016724c82928 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -2880,6 +2880,7 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 	struct mm_struct *mm = NULL;
 	struct pid *pid = NULL;
 	struct kfd_process *target = NULL;
+	struct kfd_process_device *pdd = NULL;
 	int r = 0;
 
 	if (sched_policy == KFD_SCHED_POLICY_NO_HWS) {
@@ -2957,6 +2958,20 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 		goto unlock_out;
 	}
 
+	if (args->op == KFD_IOC_DBG_TRAP_SET_NODE_ADDRESS_WATCH ||
+	    args->op == KFD_IOC_DBG_TRAP_CLEAR_NODE_ADDRESS_WATCH) {
+		int user_gpu_id = kfd_process_get_user_gpu_id(target,
+				args->op == KFD_IOC_DBG_TRAP_SET_NODE_ADDRESS_WATCH ?
+					args->set_node_address_watch.gpu_id :
+					args->clear_node_address_watch.gpu_id);
+
+		pdd = kfd_process_device_data_by_id(target, user_gpu_id);
+		if (user_gpu_id == -EINVAL || !pdd) {
+			r = -ENODEV;
+			goto unlock_out;
+		}
+	}
+
 	switch (args->op) {
 	case KFD_IOC_DBG_TRAP_ENABLE:
 		if (target != p)
@@ -3009,7 +3024,16 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 				(uint32_t *)args->resume_queues.queue_array_ptr);
 		break;
 	case KFD_IOC_DBG_TRAP_SET_NODE_ADDRESS_WATCH:
+		r = kfd_dbg_trap_set_dev_address_watch(pdd,
+				args->set_node_address_watch.address,
+				args->set_node_address_watch.mask,
+				&args->set_node_address_watch.id,
+				args->set_node_address_watch.mode);
+		break;
 	case KFD_IOC_DBG_TRAP_CLEAR_NODE_ADDRESS_WATCH:
+		r = kfd_dbg_trap_clear_dev_address_watch(pdd,
+				args->clear_node_address_watch.id);
+		break;
 	case KFD_IOC_DBG_TRAP_SET_FLAGS:
 	case KFD_IOC_DBG_TRAP_QUERY_DEBUG_EVENT:
 	case KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index f4d3dfb35cb3..4b36cc8b5fb7 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -24,6 +24,8 @@
 #include "kfd_device_queue_manager.h"
 #include <linux/file.h>
 
+#define MAX_WATCH_ADDRESSES	4
+
 void debug_event_write_work_handler(struct work_struct *work)
 {
 	struct kfd_process *process;
@@ -289,6 +291,139 @@ int kfd_dbg_set_mes_debug_mode(struct kfd_process_device *pdd)
 						pdd->watch_points, flags);
 }
 
+#define KFD_DEBUGGER_INVALID_WATCH_POINT_ID -1
+static int kfd_dbg_get_dev_watch_id(struct kfd_process_device *pdd, int *watch_id)
+{
+	int i;
+
+	*watch_id = KFD_DEBUGGER_INVALID_WATCH_POINT_ID;
+
+	spin_lock(&pdd->dev->kfd->watch_points_lock);
+
+	for (i = 0; i < MAX_WATCH_ADDRESSES; i++) {
+		/* device watchpoint in use so skip */
+		if ((pdd->dev->kfd->alloc_watch_ids >> i) & 0x1)
+			continue;
+
+		pdd->alloc_watch_ids |= 0x1 << i;
+		pdd->dev->kfd->alloc_watch_ids |= 0x1 << i;
+		*watch_id = i;
+		spin_unlock(&pdd->dev->kfd->watch_points_lock);
+		return 0;
+	}
+
+	spin_unlock(&pdd->dev->kfd->watch_points_lock);
+
+	return -ENOMEM;
+}
+
+static void kfd_dbg_clear_dev_watch_id(struct kfd_process_device *pdd, int watch_id)
+{
+	spin_lock(&pdd->dev->kfd->watch_points_lock);
+
+	/* process owns device watch point so safe to clear */
+	if ((pdd->alloc_watch_ids >> watch_id) & 0x1) {
+		pdd->alloc_watch_ids &= ~(0x1 << watch_id);
+		pdd->dev->kfd->alloc_watch_ids &= ~(0x1 << watch_id);
+	}
+
+	spin_unlock(&pdd->dev->kfd->watch_points_lock);
+}
+
+static bool kfd_dbg_owns_dev_watch_id(struct kfd_process_device *pdd, int watch_id)
+{
+	bool owns_watch_id = false;
+
+	spin_lock(&pdd->dev->kfd->watch_points_lock);
+	owns_watch_id = watch_id < MAX_WATCH_ADDRESSES &&
+			((pdd->alloc_watch_ids >> watch_id) & 0x1);
+
+	spin_unlock(&pdd->dev->kfd->watch_points_lock);
+
+	return owns_watch_id;
+}
+
+int kfd_dbg_trap_clear_dev_address_watch(struct kfd_process_device *pdd,
+					uint32_t watch_id)
+{
+	int r;
+
+	if (!kfd_dbg_owns_dev_watch_id(pdd, watch_id))
+		return -EINVAL;
+
+	if (!pdd->dev->kfd->shared_resources.enable_mes) {
+		r = debug_lock_and_unmap(pdd->dev->dqm);
+		if (r)
+			return r;
+	}
+
+	amdgpu_gfx_off_ctrl(pdd->dev->adev, false);
+	pdd->watch_points[watch_id] = pdd->dev->kfd2kgd->clear_address_watch(
+							pdd->dev->adev,
+							watch_id);
+	amdgpu_gfx_off_ctrl(pdd->dev->adev, true);
+
+	if (!pdd->dev->kfd->shared_resources.enable_mes)
+		r = debug_map_and_unlock(pdd->dev->dqm);
+	else
+		r = kfd_dbg_set_mes_debug_mode(pdd);
+
+	kfd_dbg_clear_dev_watch_id(pdd, watch_id);
+
+	return r;
+}
+
+int kfd_dbg_trap_set_dev_address_watch(struct kfd_process_device *pdd,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t *watch_id,
+					uint32_t watch_mode)
+{
+	int r = kfd_dbg_get_dev_watch_id(pdd, watch_id);
+
+	if (r)
+		return r;
+
+	if (!pdd->dev->kfd->shared_resources.enable_mes) {
+		r = debug_lock_and_unmap(pdd->dev->dqm);
+		if (r) {
+			kfd_dbg_clear_dev_watch_id(pdd, *watch_id);
+			return r;
+		}
+	}
+
+	amdgpu_gfx_off_ctrl(pdd->dev->adev, false);
+	pdd->watch_points[*watch_id] = pdd->dev->kfd2kgd->set_address_watch(
+				pdd->dev->adev,
+				watch_address,
+				watch_address_mask,
+				*watch_id,
+				watch_mode,
+				pdd->dev->vm_info.last_vmid_kfd);
+	amdgpu_gfx_off_ctrl(pdd->dev->adev, true);
+
+	if (!pdd->dev->kfd->shared_resources.enable_mes)
+		r = debug_map_and_unlock(pdd->dev->dqm);
+	else
+		r = kfd_dbg_set_mes_debug_mode(pdd);
+
+	/* HWS is broken so no point in HW rollback but release the watchpoint anyways */
+	if (r)
+		kfd_dbg_clear_dev_watch_id(pdd, *watch_id);
+
+	return 0;
+}
+
+static void kfd_dbg_clear_process_address_watch(struct kfd_process *target)
+{
+	int i, j;
+
+	for (i = 0; i < target->n_pdds; i++)
+		for (j = 0; j < MAX_WATCH_ADDRESSES; j++)
+			kfd_dbg_trap_clear_dev_address_watch(target->pdds[i], j);
+}
+
+
 /* kfd_dbg_trap_deactivate:
  *	target: target process
  *	unwind: If this is unwinding a failed kfd_dbg_trap_enable()
@@ -303,6 +438,7 @@ void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind
 
 	if (!unwind) {
 		cancel_work_sync(&target->debug_event_workarea);
+		kfd_dbg_clear_process_address_watch(target);
 		kfd_dbg_trap_set_wave_launch_mode(target, 0);
 	}
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index cb17869437c5..7f0757c2af2c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -50,7 +50,13 @@ int kfd_dbg_trap_set_wave_launch_override(struct kfd_process *target,
 					uint32_t *trap_mask_supported);
 int kfd_dbg_trap_set_wave_launch_mode(struct kfd_process *target,
 					uint8_t wave_launch_mode);
-
+int kfd_dbg_trap_clear_dev_address_watch(struct kfd_process_device *pdd,
+					uint32_t watch_id);
+int kfd_dbg_trap_set_dev_address_watch(struct kfd_process_device *pdd,
+					uint64_t watch_address,
+					uint32_t watch_address_mask,
+					uint32_t *watch_id,
+					uint32_t watch_mode);
 int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
 					unsigned int dev_id,
 					unsigned int queue_id,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 2c36bb578633..9fc9d32cb579 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -811,6 +811,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	if (kfd_resume_iommu(kfd))
 		goto kfd_resume_iommu_error;
 
+	spin_lock_init(&kfd->watch_points_lock);
+
 	kfd->init_complete = true;
 	dev_info(kfd_device, "added device %x:%x\n", kfd->adev->pdev->vendor,
 		 kfd->adev->pdev->device);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 05da43bf233a..8ec87bc8ba82 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -381,6 +381,10 @@ struct kfd_dev {
 
 	struct kfd_node *nodes[MAX_KFD_NODES];
 	unsigned int num_nodes;
+
+	/* Track per device allocated watch points */
+	uint32_t alloc_watch_ids;
+	spinlock_t watch_points_lock;
 };
 
 enum kfd_mempool {
@@ -833,6 +837,7 @@ struct kfd_process_device {
 	uint32_t spi_dbg_override;
 	uint32_t spi_dbg_launch_mode;
 	uint32_t watch_points[4];
+	uint32_t alloc_watch_ids;
 
 	/*
 	 * If this process has been checkpointed before, then the user
@@ -989,7 +994,6 @@ struct kfd_process {
 	struct semaphore runtime_enable_sema;
 	bool is_runtime_retry;
 	struct kfd_runtime_info runtime_info;
-
 };
 
 #define KFD_PROCESS_TABLE_SIZE 5 /* bits: 32 entries */
-- 
cgit 


From 103d5f08ff42b666c61c350be2c3e724c1646918 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Mon, 9 May 2022 10:51:56 -0400
Subject: drm/amdkfd: add debug set flags operation

Allow the debugger to set single memory and single ALU operations.

Some exceptions are imprecise (memory violations, address watch) in the
sense that a trap occurs only when the exception interrupt occurs and
not at the non-halting faulty instruction.  Trap temporaries 0 & 1 save
the program counter address, which means that these values will not point
to the faulty instruction address but to whenever the interrupt was
raised.

Setting the Single Memory Operations flag will inject an automatic wait
on every memory operation instruction forcing imprecise memory exceptions
to become precise at the cost of performance.  This setting is not
permitted on debug devices that support only a global setting of this
option.

Return the previous set flags to the debugger as well.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c |  2 ++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 58 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |  1 +
 3 files changed, 61 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 016724c82928..5ee38614ed9b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -3035,6 +3035,8 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 				args->clear_node_address_watch.id);
 		break;
 	case KFD_IOC_DBG_TRAP_SET_FLAGS:
+		r = kfd_dbg_trap_set_flags(target, &args->set_flags.flags);
+		break;
 	case KFD_IOC_DBG_TRAP_QUERY_DEBUG_EVENT:
 	case KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO:
 	case KFD_IOC_DBG_TRAP_GET_QUEUE_SNAPSHOT:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 4b36cc8b5fb7..43c3170998d3 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -23,6 +23,7 @@
 #include "kfd_debug.h"
 #include "kfd_device_queue_manager.h"
 #include <linux/file.h>
+#include <uapi/linux/kfd_ioctl.h>
 
 #define MAX_WATCH_ADDRESSES	4
 
@@ -423,6 +424,59 @@ static void kfd_dbg_clear_process_address_watch(struct kfd_process *target)
 			kfd_dbg_trap_clear_dev_address_watch(target->pdds[i], j);
 }
 
+int kfd_dbg_trap_set_flags(struct kfd_process *target, uint32_t *flags)
+{
+	uint32_t prev_flags = target->dbg_flags;
+	int i, r = 0, rewind_count = 0;
+
+	for (i = 0; i < target->n_pdds; i++) {
+		if (!kfd_dbg_is_per_vmid_supported(target->pdds[i]->dev) &&
+			(*flags & KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP)) {
+			*flags = prev_flags;
+			return -EACCES;
+		}
+	}
+
+	target->dbg_flags = *flags & KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP;
+	*flags = prev_flags;
+	for (i = 0; i < target->n_pdds; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+
+		if (!kfd_dbg_is_per_vmid_supported(pdd->dev))
+			continue;
+
+		if (!pdd->dev->kfd->shared_resources.enable_mes)
+			r = debug_refresh_runlist(pdd->dev->dqm);
+		else
+			r = kfd_dbg_set_mes_debug_mode(pdd);
+
+		if (r) {
+			target->dbg_flags = prev_flags;
+			break;
+		}
+
+		rewind_count++;
+	}
+
+	/* Rewind flags */
+	if (r) {
+		target->dbg_flags = prev_flags;
+
+		for (i = 0; i < rewind_count; i++) {
+			struct kfd_process_device *pdd = target->pdds[i];
+
+			if (!kfd_dbg_is_per_vmid_supported(pdd->dev))
+				continue;
+
+			if (!pdd->dev->kfd->shared_resources.enable_mes)
+				debug_refresh_runlist(pdd->dev->dqm);
+			else
+				kfd_dbg_set_mes_debug_mode(pdd);
+		}
+	}
+
+	return r;
+}
 
 /* kfd_dbg_trap_deactivate:
  *	target: target process
@@ -437,9 +491,13 @@ void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind
 	int i;
 
 	if (!unwind) {
+		uint32_t flags = 0;
+
 		cancel_work_sync(&target->debug_event_workarea);
 		kfd_dbg_clear_process_address_watch(target);
 		kfd_dbg_trap_set_wave_launch_mode(target, 0);
+
+		kfd_dbg_trap_set_flags(target, &flags);
 	}
 
 	for (i = 0; i < target->n_pdds; i++) {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 7f0757c2af2c..ef8e9f7f1716 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -57,6 +57,7 @@ int kfd_dbg_trap_set_dev_address_watch(struct kfd_process_device *pdd,
 					uint32_t watch_address_mask,
 					uint32_t *watch_id,
 					uint32_t watch_mode);
+int kfd_dbg_trap_set_flags(struct kfd_process *target, uint32_t *flags);
 int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
 					unsigned int dev_id,
 					unsigned int queue_id,
-- 
cgit 


From 5bc20c224bcb863571e8831cdbba23cd61b10ac3 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Mon, 9 May 2022 11:10:32 -0400
Subject: drm/amdkfd: add debug query event operation

Allow the debugger to query a single queue, device and process
exception.
The KFD should also return the GPU or Queue id of the exception.
The debugger also has the option of clearing exceptions after
being queried.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c |  6 +++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 64 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |  5 +++
 3 files changed, 75 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 5ee38614ed9b..498859259b55 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -3038,6 +3038,12 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 		r = kfd_dbg_trap_set_flags(target, &args->set_flags.flags);
 		break;
 	case KFD_IOC_DBG_TRAP_QUERY_DEBUG_EVENT:
+		r = kfd_dbg_ev_query_debug_event(target,
+				&args->query_debug_event.queue_id,
+				&args->query_debug_event.gpu_id,
+				args->query_debug_event.exception_mask,
+				&args->query_debug_event.exception_mask);
+		break;
 	case KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO:
 	case KFD_IOC_DBG_TRAP_GET_QUEUE_SNAPSHOT:
 	case KFD_IOC_DBG_TRAP_GET_DEVICE_SNAPSHOT:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 43c3170998d3..e9530e682e85 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -27,6 +27,70 @@
 
 #define MAX_WATCH_ADDRESSES	4
 
+int kfd_dbg_ev_query_debug_event(struct kfd_process *process,
+		      unsigned int *queue_id,
+		      unsigned int *gpu_id,
+		      uint64_t exception_clear_mask,
+		      uint64_t *event_status)
+{
+	struct process_queue_manager *pqm;
+	struct process_queue_node *pqn;
+	int i;
+
+	if (!(process && process->debug_trap_enabled))
+		return -ENODATA;
+
+	mutex_lock(&process->event_mutex);
+	*event_status = 0;
+	*queue_id = 0;
+	*gpu_id = 0;
+
+	/* find and report queue events */
+	pqm = &process->pqm;
+	list_for_each_entry(pqn, &pqm->queues, process_queue_list) {
+		uint64_t tmp = process->exception_enable_mask;
+
+		if (!pqn->q)
+			continue;
+
+		tmp &= pqn->q->properties.exception_status;
+
+		if (!tmp)
+			continue;
+
+		*event_status = pqn->q->properties.exception_status;
+		*queue_id = pqn->q->properties.queue_id;
+		*gpu_id = pqn->q->device->id;
+		pqn->q->properties.exception_status &= ~exception_clear_mask;
+		goto out;
+	}
+
+	/* find and report device events */
+	for (i = 0; i < process->n_pdds; i++) {
+		struct kfd_process_device *pdd = process->pdds[i];
+		uint64_t tmp = process->exception_enable_mask
+						& pdd->exception_status;
+
+		if (!tmp)
+			continue;
+
+		*event_status = pdd->exception_status;
+		*gpu_id = pdd->dev->id;
+		pdd->exception_status &= ~exception_clear_mask;
+		goto out;
+	}
+
+	/* report process events */
+	if (process->exception_enable_mask & process->exception_status) {
+		*event_status = process->exception_status;
+		process->exception_status &= ~exception_clear_mask;
+	}
+
+out:
+	mutex_unlock(&process->event_mutex);
+	return *event_status ? 0 : -EAGAIN;
+}
+
 void debug_event_write_work_handler(struct work_struct *work)
 {
 	struct kfd_process *process;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index ef8e9f7f1716..e78f954c0684 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -27,6 +27,11 @@
 
 void kfd_dbg_trap_deactivate(struct kfd_process *target, bool unwind, int unwind_count);
 int kfd_dbg_trap_activate(struct kfd_process *target);
+int kfd_dbg_ev_query_debug_event(struct kfd_process *process,
+			unsigned int *queue_id,
+			unsigned int *gpu_id,
+			uint64_t exception_clear_mask,
+			uint64_t *event_status);
 bool kfd_set_dbg_ev_from_interrupt(struct kfd_node *dev,
 				   unsigned int pasid,
 				   uint32_t doorbell_id,
-- 
cgit 


From 2b36de971d25daa2ad287114ae3ca11a8f8d49d7 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Mon, 9 May 2022 13:37:36 -0400
Subject: drm/amdkfd: add debug query exception info operation

Allow the debugger to query additional info based on an exception code.
For device exceptions, it's currently only memory violation information.
For process exceptions, it's currently only runtime information.
Queue exception only report the queue exception status.

The debugger has the option of clearing the target exception on query.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c |   7 ++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 120 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |   6 ++
 3 files changed, 133 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 498859259b55..b7ee79b5220a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -3045,6 +3045,13 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 				&args->query_debug_event.exception_mask);
 		break;
 	case KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO:
+		r = kfd_dbg_trap_query_exception_info(target,
+				args->query_exception_info.source_id,
+				args->query_exception_info.exception_code,
+				args->query_exception_info.clear_exception,
+				(void __user *)args->query_exception_info.info_ptr,
+				&args->query_exception_info.info_size);
+		break;
 	case KFD_IOC_DBG_TRAP_GET_QUEUE_SNAPSHOT:
 	case KFD_IOC_DBG_TRAP_GET_DEVICE_SNAPSHOT:
 		pr_warn("Debug op %i not supported yet\n", args->op);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index e9530e682e85..24e2b285448a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -890,6 +890,126 @@ int kfd_dbg_trap_set_wave_launch_mode(struct kfd_process *target,
 	return r;
 }
 
+int kfd_dbg_trap_query_exception_info(struct kfd_process *target,
+		uint32_t source_id,
+		uint32_t exception_code,
+		bool clear_exception,
+		void __user *info,
+		uint32_t *info_size)
+{
+	bool found = false;
+	int r = 0;
+	uint32_t copy_size, actual_info_size = 0;
+	uint64_t *exception_status_ptr = NULL;
+
+	if (!target)
+		return -EINVAL;
+
+	if (!info || !info_size)
+		return -EINVAL;
+
+	mutex_lock(&target->event_mutex);
+
+	if (KFD_DBG_EC_TYPE_IS_QUEUE(exception_code)) {
+		/* Per queue exceptions */
+		struct queue *queue = NULL;
+		int i;
+
+		for (i = 0; i < target->n_pdds; i++) {
+			struct kfd_process_device *pdd = target->pdds[i];
+			struct qcm_process_device *qpd = &pdd->qpd;
+
+			list_for_each_entry(queue, &qpd->queues_list, list) {
+				if (!found && queue->properties.queue_id == source_id) {
+					found = true;
+					break;
+				}
+			}
+			if (found)
+				break;
+		}
+
+		if (!found) {
+			r = -EINVAL;
+			goto out;
+		}
+
+		if (!(queue->properties.exception_status & KFD_EC_MASK(exception_code))) {
+			r = -ENODATA;
+			goto out;
+		}
+		exception_status_ptr = &queue->properties.exception_status;
+	} else if (KFD_DBG_EC_TYPE_IS_DEVICE(exception_code)) {
+		/* Per device exceptions */
+		struct kfd_process_device *pdd = NULL;
+		int i;
+
+		for (i = 0; i < target->n_pdds; i++) {
+			pdd = target->pdds[i];
+			if (pdd->dev->id == source_id) {
+				found = true;
+				break;
+			}
+		}
+
+		if (!found) {
+			r = -EINVAL;
+			goto out;
+		}
+
+		if (!(pdd->exception_status & KFD_EC_MASK(exception_code))) {
+			r = -ENODATA;
+			goto out;
+		}
+
+		if (exception_code == EC_DEVICE_MEMORY_VIOLATION) {
+			copy_size = min((size_t)(*info_size), pdd->vm_fault_exc_data_size);
+
+			if (copy_to_user(info, pdd->vm_fault_exc_data, copy_size)) {
+				r = -EFAULT;
+				goto out;
+			}
+			actual_info_size = pdd->vm_fault_exc_data_size;
+			if (clear_exception) {
+				kfree(pdd->vm_fault_exc_data);
+				pdd->vm_fault_exc_data = NULL;
+				pdd->vm_fault_exc_data_size = 0;
+			}
+		}
+		exception_status_ptr = &pdd->exception_status;
+	} else if (KFD_DBG_EC_TYPE_IS_PROCESS(exception_code)) {
+		/* Per process exceptions */
+		if (!(target->exception_status & KFD_EC_MASK(exception_code))) {
+			r = -ENODATA;
+			goto out;
+		}
+
+		if (exception_code == EC_PROCESS_RUNTIME) {
+			copy_size = min((size_t)(*info_size), sizeof(target->runtime_info));
+
+			if (copy_to_user(info, (void *)&target->runtime_info, copy_size)) {
+				r = -EFAULT;
+				goto out;
+			}
+
+			actual_info_size = sizeof(target->runtime_info);
+		}
+
+		exception_status_ptr = &target->exception_status;
+	} else {
+		pr_debug("Bad exception type [%i]\n", exception_code);
+		r = -EINVAL;
+		goto out;
+	}
+
+	*info_size = actual_info_size;
+	if (clear_exception)
+		*exception_status_ptr &= ~KFD_EC_MASK(exception_code);
+out:
+	mutex_unlock(&target->event_mutex);
+	return r;
+}
+
 void kfd_dbg_set_enabled_debug_exception_mask(struct kfd_process *target,
 					uint64_t exception_set_mask)
 {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index e78f954c0684..234e2ccafa87 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -63,6 +63,12 @@ int kfd_dbg_trap_set_dev_address_watch(struct kfd_process_device *pdd,
 					uint32_t *watch_id,
 					uint32_t watch_mode);
 int kfd_dbg_trap_set_flags(struct kfd_process *target, uint32_t *flags);
+int kfd_dbg_trap_query_exception_info(struct kfd_process *target,
+		uint32_t source_id,
+		uint32_t exception_code,
+		bool clear_exception,
+		void __user *info,
+		uint32_t *info_size);
 int kfd_dbg_send_exception_to_runtime(struct kfd_process *p,
 					unsigned int dev_id,
 					unsigned int queue_id,
-- 
cgit 


From 12976e6a5ab8fc3766c0304d72f7eec81a109b55 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Tue, 10 May 2022 12:47:45 -0400
Subject: drm/amdkfd: add debug device snapshot operation

Similar to queue snapshot, return an array of device information using
an entry_size check and return.
Unlike queue snapshots, the debugger needs to pass to correct number of
devices that exist.  If it fails to do so, the KFD will return the
number of actual devices so that the debugger can make a subsequent
successful call.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c |  7 ++-
 drivers/gpu/drm/amd/amdkfd/kfd_debug.c   | 73 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_debug.h   |  5 +++
 3 files changed, 83 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_debug.h')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 24066756e478..f54ff5c3387d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -3060,8 +3060,11 @@ static int kfd_ioctl_set_debug_trap(struct file *filep, struct kfd_process *p, v
 				&args->queue_snapshot.entry_size);
 		break;
 	case KFD_IOC_DBG_TRAP_GET_DEVICE_SNAPSHOT:
-		pr_warn("Debug op %i not supported yet\n", args->op);
-		r = -EACCES;
+		r = kfd_dbg_trap_device_snapshot(target,
+				args->device_snapshot.exception_mask,
+				(void __user *)args->device_snapshot.snapshot_buf_ptr,
+				&args->device_snapshot.num_devices,
+				&args->device_snapshot.entry_size);
 		break;
 	default:
 		pr_err("Invalid option: %i\n", args->op);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
index 24e2b285448a..125274445f43 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.c
@@ -22,6 +22,7 @@
 
 #include "kfd_debug.h"
 #include "kfd_device_queue_manager.h"
+#include "kfd_topology.h"
 #include <linux/file.h>
 #include <uapi/linux/kfd_ioctl.h>
 
@@ -1010,6 +1011,78 @@ out:
 	return r;
 }
 
+int kfd_dbg_trap_device_snapshot(struct kfd_process *target,
+		uint64_t exception_clear_mask,
+		void __user *user_info,
+		uint32_t *number_of_device_infos,
+		uint32_t *entry_size)
+{
+	struct kfd_dbg_device_info_entry device_info;
+	uint32_t tmp_entry_size = *entry_size, tmp_num_devices;
+	int i, r = 0;
+
+	if (!(target && user_info && number_of_device_infos && entry_size))
+		return -EINVAL;
+
+	tmp_num_devices = min_t(size_t, *number_of_device_infos, target->n_pdds);
+	*number_of_device_infos = target->n_pdds;
+	*entry_size = min_t(size_t, *entry_size, sizeof(device_info));
+
+	if (!tmp_num_devices)
+		return 0;
+
+	memset(&device_info, 0, sizeof(device_info));
+
+	mutex_lock(&target->event_mutex);
+
+	/* Run over all pdd of the process */
+	for (i = 0; i < tmp_num_devices; i++) {
+		struct kfd_process_device *pdd = target->pdds[i];
+		struct kfd_topology_device *topo_dev = kfd_topology_device_by_id(pdd->dev->id);
+
+		device_info.gpu_id = pdd->dev->id;
+		device_info.exception_status = pdd->exception_status;
+		device_info.lds_base = pdd->lds_base;
+		device_info.lds_limit = pdd->lds_limit;
+		device_info.scratch_base = pdd->scratch_base;
+		device_info.scratch_limit = pdd->scratch_limit;
+		device_info.gpuvm_base = pdd->gpuvm_base;
+		device_info.gpuvm_limit = pdd->gpuvm_limit;
+		device_info.location_id = topo_dev->node_props.location_id;
+		device_info.vendor_id = topo_dev->node_props.vendor_id;
+		device_info.device_id = topo_dev->node_props.device_id;
+		device_info.revision_id = pdd->dev->adev->pdev->revision;
+		device_info.subsystem_vendor_id = pdd->dev->adev->pdev->subsystem_vendor;
+		device_info.subsystem_device_id = pdd->dev->adev->pdev->subsystem_device;
+		device_info.fw_version = pdd->dev->kfd->mec_fw_version;
+		device_info.gfx_target_version =
+			topo_dev->node_props.gfx_target_version;
+		device_info.simd_count = topo_dev->node_props.simd_count;
+		device_info.max_waves_per_simd =
+			topo_dev->node_props.max_waves_per_simd;
+		device_info.array_count = topo_dev->node_props.array_count;
+		device_info.simd_arrays_per_engine =
+			topo_dev->node_props.simd_arrays_per_engine;
+		device_info.num_xcc = NUM_XCC(pdd->dev->xcc_mask);
+		device_info.capability = topo_dev->node_props.capability;
+		device_info.debug_prop = topo_dev->node_props.debug_prop;
+
+		if (exception_clear_mask)
+			pdd->exception_status &= ~exception_clear_mask;
+
+		if (copy_to_user(user_info, &device_info, *entry_size)) {
+			r = -EFAULT;
+			break;
+		}
+
+		user_info += tmp_entry_size;
+	}
+
+	mutex_unlock(&target->event_mutex);
+
+	return r;
+}
+
 void kfd_dbg_set_enabled_debug_exception_mask(struct kfd_process *target,
 					uint64_t exception_set_mask)
 {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 234e2ccafa87..a289e59ceb79 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -81,6 +81,11 @@ static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_node *dev)
 }
 
 void debug_event_write_work_handler(struct work_struct *work);
+int kfd_dbg_trap_device_snapshot(struct kfd_process *target,
+		uint64_t exception_clear_mask,
+		void __user *user_info,
+		uint32_t *number_of_device_infos,
+		uint32_t *entry_size);
 
 void kfd_dbg_set_enabled_debug_exception_mask(struct kfd_process *target,
 					uint64_t exception_set_mask);
-- 
cgit