From 58ab2c08d708ca8309768545b75741636c53a336 Mon Sep 17 00:00:00 2001 From: Christian König Date: Fri, 14 Jan 2022 16:49:44 +0100 Subject: drm/amdgpu: use VRAM|GTT for a bunch of kernel allocations MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Technically all of those can use GTT as well, no need to force things into VRAM. Signed-off-by: Christian König Signed-off-by: Luben Tuikov Acked-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index 23692e5d4d13..42a939cd2eac 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -372,8 +372,11 @@ int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev, * KIQ MQD no matter SRIOV or Bare-metal */ r = amdgpu_bo_create_kernel(adev, mqd_size, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, &ring->mqd_obj, - &ring->mqd_gpu_addr, &ring->mqd_ptr); + AMDGPU_GEM_DOMAIN_VRAM | + AMDGPU_GEM_DOMAIN_GTT, + &ring->mqd_obj, + &ring->mqd_gpu_addr, + &ring->mqd_ptr); if (r) { dev_warn(adev->dev, "failed to create ring mqd ob (%d)", r); return r; -- cgit From 89e4c448817835700de108edcb11f26a88414986 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Wed, 2 Nov 2022 14:07:35 +0800 Subject: drm/amdgpu: Add gfx ras function on gfx v11_0_3 Add gfx ras function on gfx v11_0_3. V2: 1. Add separate source files for gfx v11_0_3. 2. Create a common function to initialize gfx ras block. V3: 1. Rename amdgpu_gfx_ras_block_init to amdgpu_gfx_ras_sw_init. 2. Adjust the calling position of amdgpu_gfx_ras_sw_init. 3. Remove gfx_v11_0_3_ras_ops. V4: Revert changes in amdgpu_ras_interrupt_poison_consumption_handler. V5: 1. Remove invalid include file in gfx_v11_0_3.c. 2. Reduce the number of parameters of amdgpu_gfx_ras_sw_init. Signed-off-by: YiPeng Chai Reviewed-by: Hawking Zhang Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/Makefile | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 35 ++++++++++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 1 + drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 13 ++++++++++++ drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c | 27 ++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.h | 29 ++++++++++++++++++++++++++ 6 files changed, 106 insertions(+) create mode 100644 drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c create mode 100644 drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.h (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile index 332cf8bda7a2..5df603192cdc 100644 --- a/drivers/gpu/drm/amd/amdgpu/Makefile +++ b/drivers/gpu/drm/amd/amdgpu/Makefile @@ -137,6 +137,7 @@ amdgpu-y += \ gfx_v10_0.o \ imu_v11_0.o \ gfx_v11_0.o \ + gfx_v11_0_3.o \ imu_v11_0_3.o # add async DMA block diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index 42a939cd2eac..09c42c00e43c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -696,6 +696,41 @@ late_fini: return r; } +int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev) +{ + int err = 0; + struct amdgpu_gfx_ras *ras = NULL; + + /* adev->gfx.ras is NULL, which means gfx does not + * support ras function, then do nothing here. + */ + if (!adev->gfx.ras) + return 0; + + ras = adev->gfx.ras; + + err = amdgpu_ras_register_ras_block(adev, &ras->ras_block); + if (err) { + dev_err(adev->dev, "Failed to register gfx ras block!\n"); + return err; + } + + strcpy(ras->ras_block.ras_comm.name, "gfx"); + ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__GFX; + ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE; + adev->gfx.ras_if = &ras->ras_block.ras_comm; + + /* If not define special ras_late_init function, use gfx default ras_late_init */ + if (!ras->ras_block.ras_late_init) + ras->ras_block.ras_late_init = amdgpu_ras_block_late_init; + + /* If not defined special ras_cb function, use default ras_cb */ + if (!ras->ras_block.ras_cb) + ras->ras_block.ras_cb = amdgpu_gfx_process_ras_data_cb; + + return 0; +} + int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev, void *err_data, struct amdgpu_iv_entry *entry) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h index b3df4787877e..6b26597217ed 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h @@ -432,4 +432,5 @@ void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v); int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev); void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id); +int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c index 985fe704203e..37002f8cf2da 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c @@ -46,6 +46,7 @@ #include "clearstate_gfx11.h" #include "v11_structs.h" #include "gfx_v11_0.h" +#include "gfx_v11_0_3.h" #include "nbio_v4_3.h" #include "mes_v11_0.h" @@ -815,7 +816,14 @@ static int gfx_v11_0_gpu_early_init(struct amdgpu_device *adev) switch (adev->ip_versions[GC_HWIP][0]) { case IP_VERSION(11, 0, 0): case IP_VERSION(11, 0, 2): + adev->gfx.config.max_hw_contexts = 8; + adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; + adev->gfx.config.sc_prim_fifo_size_backend = 0x100; + adev->gfx.config.sc_hiz_tile_fifo_size = 0; + adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; + break; case IP_VERSION(11, 0, 3): + adev->gfx.ras = &gfx_v11_0_3_ras; adev->gfx.config.max_hw_contexts = 8; adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; adev->gfx.config.sc_prim_fifo_size_backend = 0x100; @@ -1380,6 +1388,11 @@ static int gfx_v11_0_sw_init(void *handle) if (r) return r; + if (amdgpu_gfx_ras_sw_init(adev)) { + dev_err(adev->dev, "Failed to initialize gfx ras block!\n"); + return -EINVAL; + } + return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c new file mode 100644 index 000000000000..5966d984a30a --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c @@ -0,0 +1,27 @@ +/* + * 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 "amdgpu.h" + + +struct amdgpu_gfx_ras gfx_v11_0_3_ras; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.h b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.h new file mode 100644 index 000000000000..672c7920b3d0 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.h @@ -0,0 +1,29 @@ +/* + * 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 __GFX_V11_0_3_H__ +#define __GFX_V11_0_3_H__ + +extern struct amdgpu_gfx_ras gfx_v11_0_3_ras; + +#endif -- cgit From ac7b25d92c6f967217c92a401734bf041187996f Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Wed, 4 Jan 2023 13:13:21 +0800 Subject: drm/amdgpu: Perform gpu reset after gfx finishes processing ras poison consumption on gfx_v11_0_3 Perform gpu reset after gfx finishes processing ras poison consumption on gfx_v11_0_3. V2: Move gfx poison consumption handler from hw_ops to ip function level. V3: Adjust the calling position of amdgpu_gfx_poison_consumation_handler. V4: Since gfx v11_0_3 does not have .hw_ops instance, the .hw_ops null pointer check in amdgpu_ras_interrupt_poison_consumption_handler needs to be adjusted. Signed-off-by: YiPeng Chai Reviewed-by: Hawking Zhang Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 9 +++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 4 ++++ drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 8 +++++--- drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c | 13 +++++++++++++ 4 files changed, 31 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index 09c42c00e43c..caf7fd3adcbd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -731,6 +731,15 @@ int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev) return 0; } +int amdgpu_gfx_poison_consumption_handler(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry) +{ + if (adev->gfx.ras && adev->gfx.ras->poison_consumption_handler) + return adev->gfx.ras->poison_consumption_handler(adev, entry); + + return 0; +} + int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev, void *err_data, struct amdgpu_iv_entry *entry) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h index 0b39fe3cd624..86ec9d0d12c8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h @@ -213,6 +213,8 @@ struct amdgpu_gfx_ras { int (*rlc_gc_fed_irq)(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry); + int (*poison_consumption_handler)(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry); }; struct amdgpu_gfx_funcs { @@ -437,4 +439,6 @@ int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev); void amdgpu_gfx_cp_init_microcode(struct amdgpu_device *adev, uint32_t ucode_id); int amdgpu_gfx_ras_sw_init(struct amdgpu_device *adev); +int amdgpu_gfx_poison_consumption_handler(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index d06beb884a16..0a95d1c1e7ab 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -1620,14 +1620,14 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager * struct amdgpu_ras_block_object *block_obj = amdgpu_ras_get_ras_block(adev, obj->head.block, 0); - if (!block_obj || !block_obj->hw_ops) + if (!block_obj) return; /* both query_poison_status and handle_poison_consumption are optional, * but at least one of them should be implemented if we need poison * consumption handler */ - if (block_obj->hw_ops->query_poison_status) { + if (block_obj->hw_ops && block_obj->hw_ops->query_poison_status) { poison_stat = block_obj->hw_ops->query_poison_status(adev); if (!poison_stat) { /* Not poison consumption interrupt, no need to handle it */ @@ -1641,7 +1641,7 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager * if (!adev->gmc.xgmi.connected_to_cpu) amdgpu_umc_poison_handler(adev, false); - if (block_obj->hw_ops->handle_poison_consumption) + if (block_obj->hw_ops && block_obj->hw_ops->handle_poison_consumption) poison_stat = block_obj->hw_ops->handle_poison_consumption(adev); /* gpu reset is fallback for failed and default cases */ @@ -1649,6 +1649,8 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager * dev_info(adev->dev, "GPU reset for %s RAS poison consumption is issued!\n", block_obj->ras_comm.name); amdgpu_ras_reset_gpu(adev); + } else { + amdgpu_gfx_poison_consumption_handler(adev, entry); } } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c index a18e09de31dd..b07a72ca25d9 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0_3.c @@ -70,6 +70,19 @@ static int gfx_v11_0_3_rlc_gc_fed_irq(struct amdgpu_device *adev, return 0; } +static int gfx_v11_0_3_poison_consumption_handler(struct amdgpu_device *adev, + struct amdgpu_iv_entry *entry) +{ + /* Workaround: when vmid and pasid are both zero, trigger gpu reset in KGD. */ + if (entry && (entry->client_id == SOC21_IH_CLIENTID_GFX) && + (entry->src_id == GFX_11_0_0__SRCID__RLC_GC_FED_INTERRUPT) && + !entry->vmid && !entry->pasid) + amdgpu_ras_reset_gpu(adev); + + return 0; +} + struct amdgpu_gfx_ras gfx_v11_0_3_ras = { .rlc_gc_fed_irq = gfx_v11_0_3_rlc_gc_fed_irq, + .poison_consumption_handler = gfx_v11_0_3_poison_consumption_handler, }; -- cgit From 25959dd67ddea75595af2b2d2217191aba1e6d71 Mon Sep 17 00:00:00 2001 From: Lang Yu Date: Wed, 11 Jan 2023 09:32:15 +0800 Subject: drm/amdgpu: allow multipipe policy on ASICs with one MEC Always enable multipipe policy on ASICs with GC VERSION > 9.0.0 instead of MEC number > 1. This will allow multipipe policy on ASICs with one MEC, e.g., gfx11 APUs. Signed-off-by: Lang Yu Reviewed-by: Aaron Liu Reviewed-by: Yifan Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index caf7fd3adcbd..35ed46b9249c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -156,6 +156,9 @@ static bool amdgpu_gfx_is_compute_multipipe_capable(struct amdgpu_device *adev) return amdgpu_compute_multipipe == 1; } + if (adev->ip_versions[GC_HWIP][0] > IP_VERSION(9, 0, 0)) + return true; + /* FIXME: spreading the queues across pipes causes perf regressions * on POLARIS11 compute workloads */ if (adev->asic_type == CHIP_POLARIS11) -- cgit