1 files changed, 29 insertions, 6 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index 2382921710ec..b9674c57c436 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -29,6 +29,7 @@
 #include "amdgpu_rlc.h"
 #include "amdgpu_ras.h"
 #include "amdgpu_xcp.h"
+#include "amdgpu_xgmi.h"
 
 /* delay 0.1 second to enable gfx off feature */
 #define GFX_OFF_DELAY_ENABLE         msecs_to_jiffies(100)
@@ -158,7 +159,7 @@ 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))
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) > IP_VERSION(9, 0, 0))
 		return true;
 
 	/* FIXME: spreading the queues across pipes causes perf regressions
@@ -384,9 +385,11 @@ int amdgpu_gfx_mqd_sw_init(struct amdgpu_device *adev,
 	struct amdgpu_ring *ring = &kiq->ring;
 	u32 domain = AMDGPU_GEM_DOMAIN_GTT;
 
+#if !defined(CONFIG_ARM) && !defined(CONFIG_ARM64)
 	/* Only enable on gfx10 and 11 for now to avoid changing behavior on older chips */
-	if (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(10, 0, 0))
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(10, 0, 0))
 		domain |= AMDGPU_GEM_DOMAIN_VRAM;
+#endif
 
 	/* create MQD for KIQ */
 	if (!adev->enable_mes_kiq && !ring->mqd_obj) {
@@ -501,6 +504,9 @@ int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id)
 {
 	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
 	struct amdgpu_ring *kiq_ring = &kiq->ring;
+	struct amdgpu_hive_info *hive;
+	struct amdgpu_ras *ras;
+	int hive_ras_recovery = 0;
 	int i, r = 0;
 	int j;
 
@@ -521,6 +527,23 @@ int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id)
 					   RESET_QUEUES, 0, 0);
 	}
 
+	/**
+	 * This is workaround: only skip kiq_ring test
+	 * during ras recovery in suspend stage for gfx9.4.3
+	 */
+	hive = amdgpu_get_xgmi_hive(adev);
+	if (hive) {
+		hive_ras_recovery = atomic_read(&hive->ras_recovery);
+		amdgpu_put_xgmi_hive(hive);
+	}
+
+	ras = amdgpu_ras_get_context(adev);
+	if ((amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3)) &&
+		ras && (atomic_read(&ras->in_recovery) || hive_ras_recovery)) {
+		spin_unlock(&kiq->ring_lock);
+		return 0;
+	}
+
 	if (kiq_ring->sched.ready && !adev->job_hang)
 		r = amdgpu_ring_test_helper(kiq_ring);
 	spin_unlock(&kiq->ring_lock);
@@ -908,12 +931,12 @@ void amdgpu_gfx_ras_error_func(struct amdgpu_device *adev,
 		func(adev, ras_error_status, i);
 }
 
-uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg)
+uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg, uint32_t xcc_id)
 {
 	signed long r, cnt = 0;
 	unsigned long flags;
 	uint32_t seq, reg_val_offs = 0, value = 0;
-	struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
 	struct amdgpu_ring *ring = &kiq->ring;
 
 	if (amdgpu_device_skip_hw_access(adev))
@@ -976,12 +999,12 @@ failed_kiq_read:
 	return ~0;
 }
 
-void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v)
+void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, uint32_t xcc_id)
 {
 	signed long r, cnt = 0;
 	unsigned long flags;
 	uint32_t seq;
-	struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq[xcc_id];
 	struct amdgpu_ring *ring = &kiq->ring;
 
 	BUG_ON(!ring->funcs->emit_wreg);