drm/amdgpu: rework lock handling for flush_tlb v2

Instead of each implementation doing this more or less correctly
move taking the reset lock at a higher level.

v2: fix typo

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@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_gmc.c

@@ -596,8 +596,17 @@ void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 	    !adev->mman.buffer_funcs_enabled ||
 	    !adev->ib_pool_ready || amdgpu_in_reset(adev) ||
 	    !ring->sched.ready) {
+
+		/*
+		 * A GPU reset should flush all TLBs anyway, so no need to do
+		 * this while one is ongoing.
+		 */
+		if (!down_read_trylock(&adev->reset_domain->sem))
+			return;
+
 		adev->gmc.gmc_funcs->flush_gpu_tlb(adev, vmid, vmhub,
 						   flush_type);
+		up_read(&adev->reset_domain->sem);
 		return;
 	}
 

drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c

@@ -51,8 +51,6 @@
 #include "athub_v2_0.h"
 #include "athub_v2_1.h"
 
-#include "amdgpu_reset.h"
-
 static int gmc_v10_0_ecc_interrupt_state(struct amdgpu_device *adev,
 					 struct amdgpu_irq_src *src,
 					 unsigned int type,
@@ -265,11 +263,9 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 	 * Directly use kiq to do the vm invalidation instead
 	 */
 	if (adev->gfx.kiq[0].ring.sched.ready && !adev->enable_mes &&
-	    (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev)) &&
+	    (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev))) {
-	    down_read_trylock(&adev->reset_domain->sem)) {
 		amdgpu_virt_kiq_reg_write_reg_wait(adev, req, ack, inv_req,
 				1 << vmid);
-		up_read(&adev->reset_domain->sem);
 		return;
 	}
 

drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c

@@ -33,7 +33,6 @@
 #include "amdgpu_ucode.h"
 #include "amdgpu_amdkfd.h"
 #include "amdgpu_gem.h"
-#include "amdgpu_reset.h"
 
 #include "bif/bif_4_1_d.h"
 #include "bif/bif_4_1_sh_mask.h"
@@ -430,9 +429,6 @@ static void gmc_v7_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 	u32 mask = 0x0;
 	int vmid;
 
-	if (!down_read_trylock(&adev->reset_domain->sem))
-		return;
-
 	for (vmid = 1; vmid < 16; vmid++) {
 		u32 tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
 
@@ -443,7 +439,6 @@ static void gmc_v7_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 
 	WREG32(mmVM_INVALIDATE_REQUEST, mask);
 	RREG32(mmVM_INVALIDATE_RESPONSE);
-	up_read(&adev->reset_domain->sem);
 }
 
 /*

drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c

@@ -31,7 +31,6 @@
 #include "amdgpu_ucode.h"
 #include "amdgpu_amdkfd.h"
 #include "amdgpu_gem.h"
-#include "amdgpu_reset.h"
 
 #include "gmc/gmc_8_1_d.h"
 #include "gmc/gmc_8_1_sh_mask.h"
@@ -620,9 +619,6 @@ static void gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 	u32 mask = 0x0;
 	int vmid;
 
-	if (!down_read_trylock(&adev->reset_domain->sem))
-		return;
-
 	for (vmid = 1; vmid < 16; vmid++) {
 		u32 tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
 
@@ -633,7 +629,6 @@ static void gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 
 	WREG32(mmVM_INVALIDATE_REQUEST, mask);
 	RREG32(mmVM_INVALIDATE_RESPONSE);
-	up_read(&adev->reset_domain->sem);
 }
 
 /*

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c

@@ -65,8 +65,6 @@
 #include "amdgpu_ras.h"
 #include "amdgpu_xgmi.h"
 
-#include "amdgpu_reset.h"
-
 /* add these here since we already include dce12 headers and these are for DCN */
 #define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION                                                          0x055d
 #define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION_BASE_IDX                                                 2
@@ -851,8 +849,7 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 	 * as GFXOFF under bare metal
 	 */
 	if (adev->gfx.kiq[0].ring.sched.ready &&
-	    (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev)) &&
+	    (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev))) {
-	    down_read_trylock(&adev->reset_domain->sem)) {
 		uint32_t req = hub->vm_inv_eng0_req + hub->eng_distance * eng;
 		uint32_t ack = hub->vm_inv_eng0_ack + hub->eng_distance * eng;
 
@@ -862,7 +859,6 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 			amdgpu_virt_kiq_reg_write_reg_wait(adev, req, ack,
 							   inv_req2, 1 << vmid);
 
-		up_read(&adev->reset_domain->sem);
 		return;
 	}