Merge tag 'drm-amdkfd-next-2017-12-24' of git://people.freedesktop.org/~gabbayo/linux into drm-next

- Add CWSR (compute wave save restore) support for GFX8 (Carrizo)
- Fix SDMA user-mode queues support for GFX7 (Kaveri)
- Add SDMA user-mode queues support for GFX8 (Carrizo)
- Allow HWS (hardware scheduling) to schedule multiple processes concurrently
- Add debugfs support
- Simplify process locking and lock dependencies
- Refactoring topology code to prepare for dGPU support + fixes to that code
  - Add option to generate dummy/virtual CRAT table when its missing or deformed
  - Recognize CPUs other then APUs as compute entities
- Various clean ups and bug fixes

I have not yet sent the dGPU topology code because it depends on a patch
for the PCI subsystem that adds PCIe atomics support. Once that patch is
upstreamed we can continue with the rest of the dGPU code.

* tag 'drm-amdkfd-next-2017-12-24' of git://people.freedesktop.org/~gabbayo/linux: (53 commits)
  drm/amdgpu: Add support for reporting VRAM usage
  drm/amdkfd: Ignore ACPI CRAT for non-APU systems
  drm/amdkfd: Module option to disable CRAT table
  drm/amdkfd: Add AQL Queue Memory flag on topology
  drm/amdkfd: Fixup incorrect info in the CZ CRAT table
  drm/amdkfd: Add perf counters to topology
  drm/amdkfd: Add topology support for dGPUs
  drm/amdkfd: Add topology support for CPUs
  drm/amdkfd: Fix sibling_map[] size
  drm/amdkfd: Simplify counting of memory banks
  drm/amdkfd: Turn verbose topology messages into pr_debug
  drm/amdkfd: sync IOLINK defines to thunk spec
  drm/amdkfd: Support enumerating non-GPU devices
  drm/amdkfd: Decouple CRAT parsing from device list update
  drm/amdkfd: Reorganize CRAT fetching from ACPI
  drm/amdkfd: Group up CRAT related functions
  drm/amdkfd: Fix memory leaks in kfd topology
  drm/amdkfd: Topology: Fix location_id
  drm/amdkfd: Update number of compute unit from KGD
  drm/amd: Remove get_vmem_size from KGD-KFD interface
  ...

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -959,6 +959,7 @@ struct amdgpu_gfx_config {
 };
 
 struct amdgpu_cu_info {
+	uint32_t simd_per_cu;
 	uint32_t max_waves_per_simd;
 	uint32_t wave_front_size;
 	uint32_t max_scratch_slots_per_cu;

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

@@ -275,14 +275,34 @@ void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj)
 	kfree(mem);
 }
 
-uint64_t get_vmem_size(struct kgd_dev *kgd)
+void get_local_mem_info(struct kgd_dev *kgd,
+			struct kfd_local_mem_info *mem_info)
 {
-	struct amdgpu_device *adev =
+	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
-		(struct amdgpu_device *)kgd;
+	uint64_t address_mask = adev->dev->dma_mask ? ~*adev->dev->dma_mask :
+					     ~((1ULL << 32) - 1);
+	resource_size_t aper_limit = adev->mc.aper_base + adev->mc.aper_size;
 
-	BUG_ON(kgd == NULL);
+	memset(mem_info, 0, sizeof(*mem_info));
+	if (!(adev->mc.aper_base & address_mask || aper_limit & address_mask)) {
+		mem_info->local_mem_size_public = adev->mc.visible_vram_size;
+		mem_info->local_mem_size_private = adev->mc.real_vram_size -
+				adev->mc.visible_vram_size;
+	} else {
+		mem_info->local_mem_size_public = 0;
+		mem_info->local_mem_size_private = adev->mc.real_vram_size;
+	}
+	mem_info->vram_width = adev->mc.vram_width;
 
-	return adev->mc.real_vram_size;
+	pr_debug("Address base: 0x%llx limit 0x%llx public 0x%llx private 0x%llx\n",
+			adev->mc.aper_base, aper_limit,
+			mem_info->local_mem_size_public,
+			mem_info->local_mem_size_private);
+
+	if (amdgpu_sriov_vf(adev))
+		mem_info->mem_clk_max = adev->clock.default_mclk / 100;
+	else
+		mem_info->mem_clk_max = amdgpu_dpm_get_mclk(adev, false) / 100;
 }
 
 uint64_t get_gpu_clock_counter(struct kgd_dev *kgd)
@@ -298,6 +318,39 @@ uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
 
-	/* The sclk is in quantas of 10kHz */
+	/* the sclk is in quantas of 10kHz */
-	return adev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100;
+	if (amdgpu_sriov_vf(adev))
+		return adev->clock.default_sclk / 100;
+
+	return amdgpu_dpm_get_sclk(adev, false) / 100;
+}
+
+void get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
+	struct amdgpu_cu_info acu_info = adev->gfx.cu_info;
+
+	memset(cu_info, 0, sizeof(*cu_info));
+	if (sizeof(cu_info->cu_bitmap) != sizeof(acu_info.bitmap))
+		return;
+
+	cu_info->cu_active_number = acu_info.number;
+	cu_info->cu_ao_mask = acu_info.ao_cu_mask;
+	memcpy(&cu_info->cu_bitmap[0], &acu_info.bitmap[0],
+	       sizeof(acu_info.bitmap));
+	cu_info->num_shader_engines = adev->gfx.config.max_shader_engines;
+	cu_info->num_shader_arrays_per_engine = adev->gfx.config.max_sh_per_se;
+	cu_info->num_cu_per_sh = adev->gfx.config.max_cu_per_sh;
+	cu_info->simd_per_cu = acu_info.simd_per_cu;
+	cu_info->max_waves_per_simd = acu_info.max_waves_per_simd;
+	cu_info->wave_front_size = acu_info.wave_front_size;
+	cu_info->max_scratch_slots_per_cu = acu_info.max_scratch_slots_per_cu;
+	cu_info->lds_size = acu_info.lds_size;
+}
+
+uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
+
+	return amdgpu_vram_mgr_usage(&adev->mman.bdev.man[TTM_PL_VRAM]);
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h

@@ -56,10 +56,13 @@ int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
 			void **mem_obj, uint64_t *gpu_addr,
 			void **cpu_ptr);
 void free_gtt_mem(struct kgd_dev *kgd, void *mem_obj);
-uint64_t get_vmem_size(struct kgd_dev *kgd);
+void get_local_mem_info(struct kgd_dev *kgd,
+			struct kfd_local_mem_info *mem_info);
 uint64_t get_gpu_clock_counter(struct kgd_dev *kgd);
 
 uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd);
+void get_cu_info(struct kgd_dev *kgd, struct kfd_cu_info *cu_info);
+uint64_t amdgpu_amdkfd_get_vram_usage(struct kgd_dev *kgd);
 
 #define read_user_wptr(mmptr, wptr, dst)				\
 	({								\

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c

@@ -105,7 +105,14 @@ static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
 			uint32_t queue_id, uint32_t __user *wptr,
 			uint32_t wptr_shift, uint32_t wptr_mask,
 			struct mm_struct *mm);
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
+static int kgd_hqd_dump(struct kgd_dev *kgd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs);
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm);
+static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs);
 static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
 				uint32_t pipe_id, uint32_t queue_id);
 
@@ -166,7 +173,7 @@ static int get_tile_config(struct kgd_dev *kgd,
 static const struct kfd2kgd_calls kfd2kgd = {
 	.init_gtt_mem_allocation = alloc_gtt_mem,
 	.free_gtt_mem = free_gtt_mem,
-	.get_vmem_size = get_vmem_size,
+	.get_local_mem_info = get_local_mem_info,
 	.get_gpu_clock_counter = get_gpu_clock_counter,
 	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
 	.alloc_pasid = amdgpu_vm_alloc_pasid,
@@ -177,6 +184,8 @@ static const struct kfd2kgd_calls kfd2kgd = {
 	.init_interrupts = kgd_init_interrupts,
 	.hqd_load = kgd_hqd_load,
 	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_dump = kgd_hqd_dump,
+	.hqd_sdma_dump = kgd_hqd_sdma_dump,
 	.hqd_is_occupied = kgd_hqd_is_occupied,
 	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
 	.hqd_destroy = kgd_hqd_destroy,
@@ -191,6 +200,8 @@ static const struct kfd2kgd_calls kfd2kgd = {
 	.get_fw_version = get_fw_version,
 	.set_scratch_backing_va = set_scratch_backing_va,
 	.get_tile_config = get_tile_config,
+	.get_cu_info = get_cu_info,
+	.get_vram_usage = amdgpu_amdkfd_get_vram_usage
 };
 
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void)
@@ -375,7 +386,44 @@ static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
 	return 0;
 }
 
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+static int kgd_hqd_dump(struct kgd_dev *kgd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t i = 0, reg;
+#define HQD_N_REGS (35+4)
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
+
+	*dump = kmalloc(HQD_N_REGS*2*sizeof(uint32_t), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE0);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE1);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE2);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE3);
+
+	for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_MQD_CONTROL; reg++)
+		DUMP_REG(reg);
+
+	release_queue(kgd);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct cik_sdma_rlc_registers *m;
@@ -410,10 +458,17 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
 		WREG32(mmSDMA0_GFX_CONTEXT_CNTL, data);
 	}
 
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL,
+	data = REG_SET_FIELD(m->sdma_rlc_doorbell, SDMA0_RLC0_DOORBELL,
-				m->sdma_rlc_doorbell);
+			     ENABLE, 1);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, data);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdma_rlc_rb_rptr);
+
+	if (read_user_wptr(mm, wptr, data))
+		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, data);
+	else
+		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdma_rlc_rb_rptr);
+
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_VIRTUAL_ADDR,
 				m->sdma_rlc_virtual_addr);
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdma_rlc_rb_base);
@@ -423,8 +478,37 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
 			m->sdma_rlc_rb_rptr_addr_lo);
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
 			m->sdma_rlc_rb_rptr_addr_hi);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+
-			m->sdma_rlc_rb_cntl);
+	data = REG_SET_FIELD(m->sdma_rlc_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t sdma_offset = engine_id * SDMA1_REGISTER_OFFSET +
+		queue_id * KFD_CIK_SDMA_QUEUE_OFFSET;
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+4)
+
+	*dump = kmalloc(HQD_N_REGS*2*sizeof(uint32_t), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_VIRTUAL_ADDR; reg <= mmSDMA0_RLC0_WATERMARK;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
 
 	return 0;
 }
@@ -575,7 +659,7 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
 	struct cik_sdma_rlc_registers *m;
 	uint32_t sdma_base_addr;
 	uint32_t temp;
-	int timeout = utimeout;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
 
 	m = get_sdma_mqd(mqd);
 	sdma_base_addr = get_sdma_base_addr(m);
@@ -588,10 +672,9 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
 		temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
 		if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT)
 			break;
-		if (timeout <= 0)
+		if (time_after(jiffies, end_jiffies))
 			return -ETIME;
-		msleep(20);
+		usleep_range(500, 1000);
-		timeout -= 20;
 	}
 
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
@@ -599,6 +682,8 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
 		RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) |
 		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
 
+	m->sdma_rlc_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR);
+
 	return 0;
 }
 

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c

@@ -45,7 +45,7 @@ enum hqd_dequeue_request_type {
 	RESET_WAVES
 };
 
-struct cik_sdma_rlc_registers;
+struct vi_sdma_mqd;
 
 /*
  * Register access functions
@@ -64,7 +64,14 @@ static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
 			uint32_t queue_id, uint32_t __user *wptr,
 			uint32_t wptr_shift, uint32_t wptr_mask,
 			struct mm_struct *mm);
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
+static int kgd_hqd_dump(struct kgd_dev *kgd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs);
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm);
+static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs);
 static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
 		uint32_t pipe_id, uint32_t queue_id);
 static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd);
@@ -125,7 +132,7 @@ static int get_tile_config(struct kgd_dev *kgd,
 static const struct kfd2kgd_calls kfd2kgd = {
 	.init_gtt_mem_allocation = alloc_gtt_mem,
 	.free_gtt_mem = free_gtt_mem,
-	.get_vmem_size = get_vmem_size,
+	.get_local_mem_info = get_local_mem_info,
 	.get_gpu_clock_counter = get_gpu_clock_counter,
 	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
 	.alloc_pasid = amdgpu_vm_alloc_pasid,
@@ -136,6 +143,8 @@ static const struct kfd2kgd_calls kfd2kgd = {
 	.init_interrupts = kgd_init_interrupts,
 	.hqd_load = kgd_hqd_load,
 	.hqd_sdma_load = kgd_hqd_sdma_load,
+	.hqd_dump = kgd_hqd_dump,
+	.hqd_sdma_dump = kgd_hqd_sdma_dump,
 	.hqd_is_occupied = kgd_hqd_is_occupied,
 	.hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
 	.hqd_destroy = kgd_hqd_destroy,
@@ -152,6 +161,8 @@ static const struct kfd2kgd_calls kfd2kgd = {
 	.get_fw_version = get_fw_version,
 	.set_scratch_backing_va = set_scratch_backing_va,
 	.get_tile_config = get_tile_config,
+	.get_cu_info = get_cu_info,
+	.get_vram_usage = amdgpu_amdkfd_get_vram_usage
 };
 
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void)
@@ -268,9 +279,15 @@ static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
 	return 0;
 }
 
-static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m)
+static inline uint32_t get_sdma_base_addr(struct vi_sdma_mqd *m)
 {
-	return 0;
+	uint32_t retval;
+
+	retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET +
+		m->sdma_queue_id * KFD_VI_SDMA_QUEUE_OFFSET;
+	pr_debug("kfd: sdma base address: 0x%x\n", retval);
+
+	return retval;
 }
 
 static inline struct vi_mqd *get_mqd(void *mqd)
@@ -278,9 +295,9 @@ static inline struct vi_mqd *get_mqd(void *mqd)
 	return (struct vi_mqd *)mqd;
 }
 
-static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+static inline struct vi_sdma_mqd *get_sdma_mqd(void *mqd)
 {
-	return (struct cik_sdma_rlc_registers *)mqd;
+	return (struct vi_sdma_mqd *)mqd;
 }
 
 static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
@@ -358,8 +375,138 @@ static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
 	return 0;
 }
 
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+static int kgd_hqd_dump(struct kgd_dev *kgd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs)
 {
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t i = 0, reg;
+#define HQD_N_REGS (54+4)
+#define DUMP_REG(addr) do {				\
+		if (WARN_ON_ONCE(i >= HQD_N_REGS))	\
+			break;				\
+		(*dump)[i][0] = (addr) << 2;		\
+		(*dump)[i++][1] = RREG32(addr);		\
+	} while (0)
+
+	*dump = kmalloc(HQD_N_REGS*2*sizeof(uint32_t), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE0);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE1);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE2);
+	DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE3);
+
+	for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_HQD_EOP_DONES; reg++)
+		DUMP_REG(reg);
+
+	release_queue(kgd);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	struct vi_sdma_mqd *m;
+	unsigned long end_jiffies;
+	uint32_t sdma_base_addr;
+	uint32_t data;
+
+	m = get_sdma_mqd(mqd);
+	sdma_base_addr = get_sdma_base_addr(m);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+		m->sdmax_rlcx_rb_cntl & (~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK));
+
+	end_jiffies = msecs_to_jiffies(2000) + jiffies;
+	while (true) {
+		data = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
+		if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
+			break;
+		if (time_after(jiffies, end_jiffies))
+			return -ETIME;
+		usleep_range(500, 1000);
+	}
+	if (m->sdma_engine_id) {
+		data = RREG32(mmSDMA1_GFX_CONTEXT_CNTL);
+		data = REG_SET_FIELD(data, SDMA1_GFX_CONTEXT_CNTL,
+				RESUME_CTX, 0);
+		WREG32(mmSDMA1_GFX_CONTEXT_CNTL, data);
+	} else {
+		data = RREG32(mmSDMA0_GFX_CONTEXT_CNTL);
+		data = REG_SET_FIELD(data, SDMA0_GFX_CONTEXT_CNTL,
+				RESUME_CTX, 0);
+		WREG32(mmSDMA0_GFX_CONTEXT_CNTL, data);
+	}
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, data);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, m->sdmax_rlcx_rb_rptr);
+
+	if (read_user_wptr(mm, wptr, data))
+		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, data);
+	else
+		WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR,
+		       m->sdmax_rlcx_rb_rptr);
+
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_VIRTUAL_ADDR,
+				m->sdmax_rlcx_virtual_addr);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI,
+			m->sdmax_rlcx_rb_base_hi);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO,
+			m->sdmax_rlcx_rb_rptr_addr_lo);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI,
+			m->sdmax_rlcx_rb_rptr_addr_hi);
+
+	data = REG_SET_FIELD(m->sdmax_rlcx_rb_cntl, SDMA0_RLC0_RB_CNTL,
+			     RB_ENABLE, 1);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, data);
+
+	return 0;
+}
+
+static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+	uint32_t sdma_offset = engine_id * SDMA1_REGISTER_OFFSET +
+		queue_id * KFD_VI_SDMA_QUEUE_OFFSET;
+	uint32_t i = 0, reg;
+#undef HQD_N_REGS
+#define HQD_N_REGS (19+4+2+3+7)
+
+	*dump = kmalloc(HQD_N_REGS*2*sizeof(uint32_t), GFP_KERNEL);
+	if (*dump == NULL)
+		return -ENOMEM;
+
+	for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_VIRTUAL_ADDR; reg <= mmSDMA0_RLC0_WATERMARK;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_CSA_ADDR_LO; reg <= mmSDMA0_RLC0_CSA_ADDR_HI;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN; reg <= mmSDMA0_RLC0_DUMMY_REG;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+	for (reg = mmSDMA0_RLC0_MIDCMD_DATA0; reg <= mmSDMA0_RLC0_MIDCMD_CNTL;
+	     reg++)
+		DUMP_REG(sdma_offset + reg);
+
+	WARN_ON_ONCE(i != HQD_N_REGS);
+	*n_regs = i;
+
 	return 0;
 }
 
@@ -388,7 +535,7 @@ static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
 static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct cik_sdma_rlc_registers *m;
+	struct vi_sdma_mqd *m;
 	uint32_t sdma_base_addr;
 	uint32_t sdma_rlc_rb_cntl;
 
@@ -509,10 +656,10 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
 				unsigned int utimeout)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct cik_sdma_rlc_registers *m;
+	struct vi_sdma_mqd *m;
 	uint32_t sdma_base_addr;
 	uint32_t temp;
-	int timeout = utimeout;
+	unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies;
 
 	m = get_sdma_mqd(mqd);
 	sdma_base_addr = get_sdma_base_addr(m);
@@ -523,18 +670,19 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
 
 	while (true) {
 		temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
-		if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT)
+		if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK)
 			break;
-		if (timeout <= 0)
+		if (time_after(jiffies, end_jiffies))
 			return -ETIME;
-		msleep(20);
+		usleep_range(500, 1000);
-		timeout -= 20;
 	}
 
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+		RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) |
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR);
 
 	return 0;
 }

drivers/gpu/drm/amd/amdgpu/cikd.h

@@ -562,7 +562,7 @@
 #define	PRIVATE_BASE(x)	((x) << 0) /* scratch */
 #define	SHARED_BASE(x)	((x) << 16) /* LDS */
 
-#define KFD_CIK_SDMA_QUEUE_OFFSET	0x200
+#define KFD_CIK_SDMA_QUEUE_OFFSET (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL)
 
 /* valid for both DEFAULT_MTYPE and APE1_MTYPE */
 enum {

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -48,6 +48,8 @@
 #include "oss/oss_2_0_d.h"
 #include "oss/oss_2_0_sh_mask.h"
 
+#define NUM_SIMD_PER_CU 0x4 /* missing from the gfx_7 IP headers */
+
 #define GFX7_NUM_GFX_RINGS     1
 #define GFX7_MEC_HPD_SIZE      2048
 
@@ -5277,6 +5279,11 @@ static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev)
 
 	cu_info->number = active_cu_number;
 	cu_info->ao_cu_mask = ao_cu_mask;
+	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
+	cu_info->max_waves_per_simd = 10;
+	cu_info->max_scratch_slots_per_cu = 32;
+	cu_info->wave_front_size = 64;
+	cu_info->lds_size = 64;
 }
 
 const struct amdgpu_ip_block_version gfx_v7_0_ip_block =

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -7116,6 +7116,11 @@ static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev)
 
 	cu_info->number = active_cu_number;
 	cu_info->ao_cu_mask = ao_cu_mask;
+	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
+	cu_info->max_waves_per_simd = 10;
+	cu_info->max_scratch_slots_per_cu = 32;
+	cu_info->wave_front_size = 64;
+	cu_info->lds_size = 64;
 }
 
 const struct amdgpu_ip_block_version gfx_v8_0_ip_block =

drivers/gpu/drm/amd/amdgpu/vid.h

@@ -27,6 +27,8 @@
 #define SDMA1_REGISTER_OFFSET                             0x200 /* not a register */
 #define SDMA_MAX_INSTANCE 2
 
+#define KFD_VI_SDMA_QUEUE_OFFSET                      0x80 /* not a register */
+
 /* crtc instance offsets */
 #define CRTC0_REGISTER_OFFSET                 (0x1b9c - 0x1b9c)
 #define CRTC1_REGISTER_OFFSET                 (0x1d9c - 0x1b9c)

drivers/gpu/drm/amd/amdkfd/Makefile

@@ -35,6 +35,8 @@ amdkfd-y	:= kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \
 		kfd_process_queue_manager.o kfd_device_queue_manager.o \
 		kfd_device_queue_manager_cik.o kfd_device_queue_manager_vi.o \
 		kfd_interrupt.o kfd_events.o cik_event_interrupt.o \
-		kfd_dbgdev.o kfd_dbgmgr.o
+		kfd_dbgdev.o kfd_dbgmgr.o kfd_crat.o
+
+amdkfd-$(CONFIG_DEBUG_FS) += kfd_debugfs.o
 
 obj-$(CONFIG_HSA_AMD)	+= amdkfd.o

drivers/gpu/drm/amd/amdkfd/kfd_chardev.c

@@ -117,7 +117,7 @@ static int kfd_open(struct inode *inode, struct file *filep)
 		return -EPERM;
 	}
 
-	process = kfd_create_process(current);
+	process = kfd_create_process(filep);
 	if (IS_ERR(process))
 		return PTR_ERR(process);
 
@@ -206,6 +206,7 @@ static int set_queue_properties_from_user(struct queue_properties *q_properties,
 	q_properties->ctx_save_restore_area_address =
 			args->ctx_save_restore_address;
 	q_properties->ctx_save_restore_area_size = args->ctx_save_restore_size;
+	q_properties->ctl_stack_size = args->ctl_stack_size;
 	if (args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE ||
 		args->queue_type == KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)
 		q_properties->type = KFD_QUEUE_TYPE_COMPUTE;
@@ -431,6 +432,38 @@ out:
 	return err;
 }
 
+static int kfd_ioctl_set_trap_handler(struct file *filep,
+					struct kfd_process *p, void *data)
+{
+	struct kfd_ioctl_set_trap_handler_args *args = data;
+	struct kfd_dev *dev;
+	int err = 0;
+	struct kfd_process_device *pdd;
+
+	dev = kfd_device_by_id(args->gpu_id);
+	if (dev == NULL)
+		return -EINVAL;
+
+	mutex_lock(&p->mutex);
+
+	pdd = kfd_bind_process_to_device(dev, p);
+	if (IS_ERR(pdd)) {
+		err = -ESRCH;
+		goto out;
+	}
+
+	if (dev->dqm->ops.set_trap_handler(dev->dqm,
+					&pdd->qpd,
+					args->tba_addr,
+					args->tma_addr))
+		err = -EINVAL;
+
+out:
+	mutex_unlock(&p->mutex);
+
+	return err;
+}
+
 static int kfd_ioctl_dbg_register(struct file *filep,
 				struct kfd_process *p, void *data)
 {
@@ -493,7 +526,7 @@ static int kfd_ioctl_dbg_unregister(struct file *filep,
 	long status;
 
 	dev = kfd_device_by_id(args->gpu_id);
-	if (!dev)
+	if (!dev || !dev->dbgmgr)
 		return -EINVAL;
 
 	if (dev->device_info->asic_family == CHIP_CARRIZO) {
@@ -979,7 +1012,10 @@ static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = {
 			kfd_ioctl_set_scratch_backing_va, 0),
 
 	AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_TILE_CONFIG,
-			kfd_ioctl_get_tile_config, 0)
+			kfd_ioctl_get_tile_config, 0),
+
+	AMDKFD_IOCTL_DEF(AMDKFD_IOC_SET_TRAP_HANDLER,
+			kfd_ioctl_set_trap_handler, 0),
 };
 
 #define AMDKFD_CORE_IOCTL_COUNT	ARRAY_SIZE(amdkfd_ioctls)
@@ -1088,6 +1124,10 @@ static int kfd_mmap(struct file *filp, struct vm_area_struct *vma)
 			KFD_MMAP_EVENTS_MASK) {
 		vma->vm_pgoff = vma->vm_pgoff ^ KFD_MMAP_EVENTS_MASK;
 		return kfd_event_mmap(process, vma);
+	} else if ((vma->vm_pgoff & KFD_MMAP_RESERVED_MEM_MASK) ==
+			KFD_MMAP_RESERVED_MEM_MASK) {
+		vma->vm_pgoff = vma->vm_pgoff ^ KFD_MMAP_RESERVED_MEM_MASK;
+		return kfd_reserved_mem_mmap(process, vma);
 	}
 
 	return -EFAULT;

drivers/gpu/drm/amd/amdkfd/kfd_crat.h

@@ -44,6 +44,10 @@
 
 #define CRAT_OEMID_64BIT_MASK ((1ULL << (CRAT_OEMID_LENGTH * 8)) - 1)
 
+/* Compute Unit flags */
+#define COMPUTE_UNIT_CPU	(1 << 0)  /* Create Virtual CRAT for CPU */
+#define COMPUTE_UNIT_GPU	(1 << 1)  /* Create Virtual CRAT for GPU */
+
 struct crat_header {
 	uint32_t	signature;
 	uint32_t	length;
@@ -105,7 +109,7 @@ struct crat_subtype_computeunit {
 	uint8_t		wave_front_size;
 	uint8_t		num_banks;
 	uint16_t	micro_engine_id;
-	uint8_t		num_arrays;
+	uint8_t		array_count;
 	uint8_t		num_cu_per_array;
 	uint8_t		num_simd_per_cu;
 	uint8_t		max_slots_scatch_cu;
@@ -127,13 +131,14 @@ struct crat_subtype_memory {
 	uint8_t		length;
 	uint16_t	reserved;
 	uint32_t	flags;
-	uint32_t	promixity_domain;
+	uint32_t	proximity_domain;
 	uint32_t	base_addr_low;
 	uint32_t	base_addr_high;
 	uint32_t	length_low;
 	uint32_t	length_high;
 	uint32_t	width;
-	uint8_t		reserved2[CRAT_MEMORY_RESERVED_LENGTH];
+	uint8_t		visibility_type; /* for virtual (dGPU) CRAT */
+	uint8_t		reserved2[CRAT_MEMORY_RESERVED_LENGTH - 1];
 };
 
 /*
@@ -222,9 +227,12 @@ struct crat_subtype_ccompute {
 /*
  * HSA IO Link Affinity structure and definitions
  */
-#define CRAT_IOLINK_FLAGS_ENABLED	0x00000001
+#define CRAT_IOLINK_FLAGS_ENABLED		(1 << 0)
-#define CRAT_IOLINK_FLAGS_COHERENCY	0x00000002
+#define CRAT_IOLINK_FLAGS_NON_COHERENT		(1 << 1)
-#define CRAT_IOLINK_FLAGS_RESERVED	0xfffffffc
+#define CRAT_IOLINK_FLAGS_NO_ATOMICS_32_BIT	(1 << 2)
+#define CRAT_IOLINK_FLAGS_NO_ATOMICS_64_BIT	(1 << 3)
+#define CRAT_IOLINK_FLAGS_NO_PEER_TO_PEER_DMA	(1 << 4)
+#define CRAT_IOLINK_FLAGS_RESERVED_MASK		0xffffffe0
 
 /*
  * IO interface types
@@ -232,10 +240,18 @@ struct crat_subtype_ccompute {
 #define CRAT_IOLINK_TYPE_UNDEFINED	0
 #define CRAT_IOLINK_TYPE_HYPERTRANSPORT	1
 #define CRAT_IOLINK_TYPE_PCIEXPRESS	2
-#define CRAT_IOLINK_TYPE_OTHER		3
+#define CRAT_IOLINK_TYPE_AMBA		3
+#define CRAT_IOLINK_TYPE_MIPI		4
+#define CRAT_IOLINK_TYPE_QPI_1_1	5
+#define CRAT_IOLINK_TYPE_RESERVED1	6
+#define CRAT_IOLINK_TYPE_RESERVED2	7
+#define CRAT_IOLINK_TYPE_RAPID_IO	8
+#define CRAT_IOLINK_TYPE_INFINIBAND	9
+#define CRAT_IOLINK_TYPE_RESERVED3	10
+#define CRAT_IOLINK_TYPE_OTHER		11
 #define CRAT_IOLINK_TYPE_MAX		255
 
-#define CRAT_IOLINK_RESERVED_LENGTH 24
+#define CRAT_IOLINK_RESERVED_LENGTH	24
 
 struct crat_subtype_iolink {
 	uint8_t		type;
@@ -291,4 +307,14 @@ struct cdit_header {
 
 #pragma pack()
 
+struct kfd_dev;
+
+int kfd_create_crat_image_acpi(void **crat_image, size_t *size);
+void kfd_destroy_crat_image(void *crat_image);
+int kfd_parse_crat_table(void *crat_image, struct list_head *device_list,
+			 uint32_t proximity_domain);
+int kfd_create_crat_image_virtual(void **crat_image, size_t *size,
+				  int flags, struct kfd_dev *kdev,
+				  uint32_t proximity_domain);
+
 #endif /* KFD_CRAT_H_INCLUDED */

drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c

@@ -95,7 +95,7 @@ static int dbgdev_diq_submit_ib(struct kfd_dbgdev *dbgdev,
 	ib_packet->bitfields3.ib_base_hi = largep->u.high_part;
 
 	ib_packet->control = (1 << 23) | (1 << 31) |
-			((size_in_bytes / sizeof(uint32_t)) & 0xfffff);
+			((size_in_bytes / 4) & 0xfffff);
 
 	ib_packet->bitfields5.pasid = pasid;
 
@@ -126,8 +126,7 @@ static int dbgdev_diq_submit_ib(struct kfd_dbgdev *dbgdev,
 
 	rm_packet->header.opcode = IT_RELEASE_MEM;
 	rm_packet->header.type = PM4_TYPE_3;
-	rm_packet->header.count = sizeof(struct pm4__release_mem) /
+	rm_packet->header.count = sizeof(struct pm4__release_mem) / 4 - 2;
-					sizeof(unsigned int) - 2;
 
 	rm_packet->bitfields2.event_type = CACHE_FLUSH_AND_INV_TS_EVENT;
 	rm_packet->bitfields2.event_index =
@@ -652,8 +651,7 @@ static int dbgdev_wave_control_diq(struct kfd_dbgdev *dbgdev,
 	packets_vec[0].header.opcode = IT_SET_UCONFIG_REG;
 	packets_vec[0].header.type = PM4_TYPE_3;
 	packets_vec[0].bitfields2.reg_offset =
-			GRBM_GFX_INDEX / (sizeof(uint32_t)) -
+			GRBM_GFX_INDEX / 4 - USERCONFIG_REG_BASE;
-				USERCONFIG_REG_BASE;
 
 	packets_vec[0].bitfields2.insert_vmid = 0;
 	packets_vec[0].reg_data[0] = reg_gfx_index.u32All;
@@ -661,8 +659,7 @@ static int dbgdev_wave_control_diq(struct kfd_dbgdev *dbgdev,
 	packets_vec[1].header.count = 1;
 	packets_vec[1].header.opcode = IT_SET_CONFIG_REG;
 	packets_vec[1].header.type = PM4_TYPE_3;
-	packets_vec[1].bitfields2.reg_offset = SQ_CMD / (sizeof(uint32_t)) -
+	packets_vec[1].bitfields2.reg_offset = SQ_CMD / 4 - AMD_CONFIG_REG_BASE;
-						AMD_CONFIG_REG_BASE;
 
 	packets_vec[1].bitfields2.vmid_shift = SQ_CMD_VMID_OFFSET;
 	packets_vec[1].bitfields2.insert_vmid = 1;
@@ -678,8 +675,7 @@ static int dbgdev_wave_control_diq(struct kfd_dbgdev *dbgdev,
 
 	packets_vec[2].ordinal1 = packets_vec[0].ordinal1;
 	packets_vec[2].bitfields2.reg_offset =
-				GRBM_GFX_INDEX / (sizeof(uint32_t)) -
+				GRBM_GFX_INDEX / 4 - USERCONFIG_REG_BASE;
-					USERCONFIG_REG_BASE;
 
 	packets_vec[2].bitfields2.insert_vmid = 0;
 	packets_vec[2].reg_data[0] = reg_gfx_index.u32All;

drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c

@@ -0,0 +1,75 @@
+/*
+ * Copyright 2016-2017 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 <linux/debugfs.h>
+#include "kfd_priv.h"
+
+static struct dentry *debugfs_root;
+
+static int kfd_debugfs_open(struct inode *inode, struct file *file)
+{
+	int (*show)(struct seq_file *, void *) = inode->i_private;
+
+	return single_open(file, show, NULL);
+}
+
+static const struct file_operations kfd_debugfs_fops = {
+	.owner = THIS_MODULE,
+	.open = kfd_debugfs_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = single_release,
+};
+
+void kfd_debugfs_init(void)
+{
+	struct dentry *ent;
+
+	debugfs_root = debugfs_create_dir("kfd", NULL);
+	if (!debugfs_root || debugfs_root == ERR_PTR(-ENODEV)) {
+		pr_warn("Failed to create kfd debugfs dir\n");
+		return;
+	}
+
+	ent = debugfs_create_file("mqds", S_IFREG | 0444, debugfs_root,
+				  kfd_debugfs_mqds_by_process,
+				  &kfd_debugfs_fops);
+	if (!ent)
+		pr_warn("Failed to create mqds in kfd debugfs\n");
+
+	ent = debugfs_create_file("hqds", S_IFREG | 0444, debugfs_root,
+				  kfd_debugfs_hqds_by_device,
+				  &kfd_debugfs_fops);
+	if (!ent)
+		pr_warn("Failed to create hqds in kfd debugfs\n");
+
+	ent = debugfs_create_file("rls", S_IFREG | 0444, debugfs_root,
+				  kfd_debugfs_rls_by_device,
+				  &kfd_debugfs_fops);
+	if (!ent)
+		pr_warn("Failed to create rls in kfd debugfs\n");
+}
+
+void kfd_debugfs_fini(void)
+{
+	debugfs_remove_recursive(debugfs_root);
+}

drivers/gpu/drm/amd/amdkfd/kfd_device.c

@@ -27,6 +27,7 @@
 #include "kfd_priv.h"
 #include "kfd_device_queue_manager.h"
 #include "kfd_pm4_headers_vi.h"
+#include "cwsr_trap_handler_gfx8.asm"
 
 #define MQD_SIZE_ALIGNED 768
 
@@ -38,7 +39,8 @@ static const struct kfd_device_info kaveri_device_info = {
 	.ih_ring_entry_size = 4 * sizeof(uint32_t),
 	.event_interrupt_class = &event_interrupt_class_cik,
 	.num_of_watch_points = 4,
-	.mqd_size_aligned = MQD_SIZE_ALIGNED
+	.mqd_size_aligned = MQD_SIZE_ALIGNED,
+	.supports_cwsr = false,
 };
 
 static const struct kfd_device_info carrizo_device_info = {
@@ -49,7 +51,8 @@ static const struct kfd_device_info carrizo_device_info = {
 	.ih_ring_entry_size = 4 * sizeof(uint32_t),
 	.event_interrupt_class = &event_interrupt_class_cik,
 	.num_of_watch_points = 4,
-	.mqd_size_aligned = MQD_SIZE_ALIGNED
+	.mqd_size_aligned = MQD_SIZE_ALIGNED,
+	.supports_cwsr = true,
 };
 
 struct kfd_deviceid {
@@ -212,6 +215,17 @@ static int iommu_invalid_ppr_cb(struct pci_dev *pdev, int pasid,
 	return AMD_IOMMU_INV_PRI_RSP_INVALID;
 }
 
+static void kfd_cwsr_init(struct kfd_dev *kfd)
+{
+	if (cwsr_enable && kfd->device_info->supports_cwsr) {
+		BUILD_BUG_ON(sizeof(cwsr_trap_gfx8_hex) > PAGE_SIZE);
+
+		kfd->cwsr_isa = cwsr_trap_gfx8_hex;
+		kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx8_hex);
+		kfd->cwsr_enabled = true;
+	}
+}
+
 bool kgd2kfd_device_init(struct kfd_dev *kfd,
 			 const struct kgd2kfd_shared_resources *gpu_resources)
 {
@@ -224,6 +238,17 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	kfd->vm_info.vmid_num_kfd = kfd->vm_info.last_vmid_kfd
 			- kfd->vm_info.first_vmid_kfd + 1;
 
+	/* Verify module parameters regarding mapped process number*/
+	if ((hws_max_conc_proc < 0)
+			|| (hws_max_conc_proc > kfd->vm_info.vmid_num_kfd)) {
+		dev_err(kfd_device,
+			"hws_max_conc_proc %d must be between 0 and %d, use %d instead\n",
+			hws_max_conc_proc, kfd->vm_info.vmid_num_kfd,
+			kfd->vm_info.vmid_num_kfd);
+		kfd->max_proc_per_quantum = kfd->vm_info.vmid_num_kfd;
+	} else
+		kfd->max_proc_per_quantum = hws_max_conc_proc;
+
 	/* calculate max size of mqds needed for queues */
 	size = max_num_of_queues_per_device *
 			kfd->device_info->mqd_size_aligned;
@@ -286,6 +311,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		goto device_iommu_pasid_error;
 	}
 
+	kfd_cwsr_init(kfd);
+
 	if (kfd_resume(kfd))
 		goto kfd_resume_error;
 

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c

@@ -149,8 +149,7 @@ static void deallocate_vmid(struct device_queue_manager *dqm,
 
 static int create_queue_nocpsch(struct device_queue_manager *dqm,
 				struct queue *q,
-				struct qcm_process_device *qpd,
+				struct qcm_process_device *qpd)
-				int *allocated_vmid)
 {
 	int retval;
 
@@ -170,9 +169,11 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm,
 		if (retval)
 			goto out_unlock;
 	}
-	*allocated_vmid = qpd->vmid;
 	q->properties.vmid = qpd->vmid;
 
+	q->properties.tba_addr = qpd->tba_addr;
+	q->properties.tma_addr = qpd->tma_addr;
+
 	if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE)
 		retval = create_compute_queue_nocpsch(dqm, q, qpd);
 	else if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
@@ -181,10 +182,8 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm,
 		retval = -EINVAL;
 
 	if (retval) {
-		if (list_empty(&qpd->queues_list)) {
+		if (list_empty(&qpd->queues_list))
 			deallocate_vmid(dqm, qpd, q);
-			*allocated_vmid = 0;
-		}
 		goto out_unlock;
 	}
 
@@ -809,16 +808,13 @@ static void destroy_kernel_queue_cpsch(struct device_queue_manager *dqm,
 }
 
 static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
-			struct qcm_process_device *qpd, int *allocate_vmid)
+			struct qcm_process_device *qpd)
 {
 	int retval;
 	struct mqd_manager *mqd;
 
 	retval = 0;
 
-	if (allocate_vmid)
-		*allocate_vmid = 0;
-
 	mutex_lock(&dqm->lock);
 
 	if (dqm->total_queue_count >= max_num_of_queues_per_device) {
@@ -846,6 +842,9 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
 	}
 
 	dqm->asic_ops.init_sdma_vm(dqm, q, qpd);
+
+	q->properties.tba_addr = qpd->tba_addr;
+	q->properties.tma_addr = qpd->tma_addr;
 	retval = mqd->init_mqd(mqd, &q->mqd, &q->mqd_mem_obj,
 				&q->gart_mqd_addr, &q->properties);
 	if (retval)
@@ -1110,6 +1109,26 @@ out:
 	return retval;
 }
 
+static int set_trap_handler(struct device_queue_manager *dqm,
+				struct qcm_process_device *qpd,
+				uint64_t tba_addr,
+				uint64_t tma_addr)
+{
+	uint64_t *tma;
+
+	if (dqm->dev->cwsr_enabled) {
+		/* Jump from CWSR trap handler to user trap */
+		tma = (uint64_t *)(qpd->cwsr_kaddr + KFD_CWSR_TMA_OFFSET);
+		tma[0] = tba_addr;
+		tma[1] = tma_addr;
+	} else {
+		qpd->tba_addr = tba_addr;
+		qpd->tma_addr = tma_addr;
+	}
+
+	return 0;
+}
+
 static int process_termination_nocpsch(struct device_queue_manager *dqm,
 		struct qcm_process_device *qpd)
 {
@@ -1241,6 +1260,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 		dqm->ops.create_kernel_queue = create_kernel_queue_cpsch;
 		dqm->ops.destroy_kernel_queue = destroy_kernel_queue_cpsch;
 		dqm->ops.set_cache_memory_policy = set_cache_memory_policy;
+		dqm->ops.set_trap_handler = set_trap_handler;
 		dqm->ops.process_termination = process_termination_cpsch;
 		break;
 	case KFD_SCHED_POLICY_NO_HWS:
@@ -1256,6 +1276,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
 		dqm->ops.initialize = initialize_nocpsch;
 		dqm->ops.uninitialize = uninitialize;
 		dqm->ops.set_cache_memory_policy = set_cache_memory_policy;
+		dqm->ops.set_trap_handler = set_trap_handler;
 		dqm->ops.process_termination = process_termination_nocpsch;
 		break;
 	default:
@@ -1290,3 +1311,74 @@ void device_queue_manager_uninit(struct device_queue_manager *dqm)
 	dqm->ops.uninitialize(dqm);
 	kfree(dqm);
 }
+
+#if defined(CONFIG_DEBUG_FS)
+
+static void seq_reg_dump(struct seq_file *m,
+			 uint32_t (*dump)[2], uint32_t n_regs)
+{
+	uint32_t i, count;
+
+	for (i = 0, count = 0; i < n_regs; i++) {
+		if (count == 0 ||
+		    dump[i-1][0] + sizeof(uint32_t) != dump[i][0]) {
+			seq_printf(m, "%s    %08x: %08x",
+				   i ? "\n" : "",
+				   dump[i][0], dump[i][1]);
+			count = 7;
+		} else {
+			seq_printf(m, " %08x", dump[i][1]);
+			count--;
+		}
+	}
+
+	seq_puts(m, "\n");
+}
+
+int dqm_debugfs_hqds(struct seq_file *m, void *data)
+{
+	struct device_queue_manager *dqm = data;
+	uint32_t (*dump)[2], n_regs;
+	int pipe, queue;
+	int r = 0;
+
+	for (pipe = 0; pipe < get_pipes_per_mec(dqm); pipe++) {
+		int pipe_offset = pipe * get_queues_per_pipe(dqm);
+
+		for (queue = 0; queue < get_queues_per_pipe(dqm); queue++) {
+			if (!test_bit(pipe_offset + queue,
+				      dqm->dev->shared_resources.queue_bitmap))
+				continue;
+
+			r = dqm->dev->kfd2kgd->hqd_dump(
+				dqm->dev->kgd, pipe, queue, &dump, &n_regs);
+			if (r)
+				break;
+
+			seq_printf(m, "  CP Pipe %d, Queue %d\n",
+				  pipe, queue);
+			seq_reg_dump(m, dump, n_regs);
+
+			kfree(dump);
+		}
+	}
+
+	for (pipe = 0; pipe < CIK_SDMA_ENGINE_NUM; pipe++) {
+		for (queue = 0; queue < CIK_SDMA_QUEUES_PER_ENGINE; queue++) {
+			r = dqm->dev->kfd2kgd->hqd_sdma_dump(
+				dqm->dev->kgd, pipe, queue, &dump, &n_regs);
+			if (r)
+				break;
+
+			seq_printf(m, "  SDMA Engine %d, RLC %d\n",
+				  pipe, queue);
+			seq_reg_dump(m, dump, n_regs);
+
+			kfree(dump);
+		}
+	}
+
+	return r;
+}
+
+#endif

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h

@@ -84,8 +84,7 @@ struct device_process_node {
 struct device_queue_manager_ops {
 	int	(*create_queue)(struct device_queue_manager *dqm,
 				struct queue *q,
-				struct qcm_process_device *qpd,
+				struct qcm_process_device *qpd);
-				int *allocate_vmid);
 
 	int	(*destroy_queue)(struct device_queue_manager *dqm,
 				struct qcm_process_device *qpd,
@@ -123,6 +122,11 @@ struct device_queue_manager_ops {
 					   void __user *alternate_aperture_base,
 					   uint64_t alternate_aperture_size);
 
+	int	(*set_trap_handler)(struct device_queue_manager *dqm,
+				    struct qcm_process_device *qpd,
+				    uint64_t tba_addr,
+				    uint64_t tma_addr);
+
 	int (*process_termination)(struct device_queue_manager *dqm,
 			struct qcm_process_device *qpd);
 };

drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c

@@ -116,8 +116,7 @@ int kfd_doorbell_init(struct kfd_dev *kfd)
 	pr_debug("doorbell aperture size  == 0x%08lX\n",
 			kfd->shared_resources.doorbell_aperture_size);
 
-	pr_debug("doorbell kernel address == 0x%08lX\n",
+	pr_debug("doorbell kernel address == %p\n", kfd->doorbell_kernel_ptr);
-			(uintptr_t)kfd->doorbell_kernel_ptr);
 
 	return 0;
 }
@@ -194,8 +193,8 @@ u32 __iomem *kfd_get_kernel_doorbell(struct kfd_dev *kfd,
 
 	pr_debug("Get kernel queue doorbell\n"
 			 "     doorbell offset   == 0x%08X\n"
-			 "     kernel address    == 0x%08lX\n",
+			 "     kernel address    == %p\n",
-		*doorbell_off, (uintptr_t)(kfd->doorbell_kernel_ptr + inx));
+		*doorbell_off, (kfd->doorbell_kernel_ptr + inx));
 
 	return kfd->doorbell_kernel_ptr + inx;
 }
@@ -215,7 +214,7 @@ inline void write_kernel_doorbell(u32 __iomem *db, u32 value)
 {
 	if (db) {
 		writel(value, db);
-		pr_debug("Writing %d to doorbell address 0x%p\n", value, db);
+		pr_debug("Writing %d to doorbell address %p\n", value, db);
 	}
 }
 

drivers/gpu/drm/amd/amdkfd/kfd_events.c

@@ -441,7 +441,7 @@ void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id,
 	/*
 	 * Because we are called from arbitrary context (workqueue) as opposed
 	 * to process context, kfd_process could attempt to exit while we are
-	 * running so the lookup function returns a locked process.
+	 * running so the lookup function increments the process ref count.
 	 */
 	struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
 
@@ -493,7 +493,7 @@ void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id,
 	}
 
 	mutex_unlock(&p->event_mutex);
-	mutex_unlock(&p->mutex);
+	kfd_unref_process(p);
 }
 
 static struct kfd_event_waiter *alloc_event_waiters(uint32_t num_events)
@@ -847,7 +847,7 @@ void kfd_signal_iommu_event(struct kfd_dev *dev, unsigned int pasid,
 	/*
 	 * Because we are called from arbitrary context (workqueue) as opposed
 	 * to process context, kfd_process could attempt to exit while we are
-	 * running so the lookup function returns a locked process.
+	 * running so the lookup function increments the process ref count.
 	 */
 	struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
 	struct mm_struct *mm;
@@ -860,7 +860,7 @@ void kfd_signal_iommu_event(struct kfd_dev *dev, unsigned int pasid,
 	 */
 	mm = get_task_mm(p->lead_thread);
 	if (!mm) {
-		mutex_unlock(&p->mutex);
+		kfd_unref_process(p);
 		return; /* Process is exiting */
 	}
 
@@ -903,7 +903,7 @@ void kfd_signal_iommu_event(struct kfd_dev *dev, unsigned int pasid,
 			&memory_exception_data);
 
 	mutex_unlock(&p->event_mutex);
-	mutex_unlock(&p->mutex);
+	kfd_unref_process(p);
 }
 
 void kfd_signal_hw_exception_event(unsigned int pasid)
@@ -911,7 +911,7 @@ void kfd_signal_hw_exception_event(unsigned int pasid)
 	/*
 	 * Because we are called from arbitrary context (workqueue) as opposed
 	 * to process context, kfd_process could attempt to exit while we are
-	 * running so the lookup function returns a locked process.
+	 * running so the lookup function increments the process ref count.
 	 */
 	struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
 
@@ -924,5 +924,5 @@ void kfd_signal_hw_exception_event(unsigned int pasid)
 	lookup_events_by_type_and_signal(p, KFD_EVENT_TYPE_HW_EXCEPTION, NULL);
 
 	mutex_unlock(&p->event_mutex);
-	mutex_unlock(&p->mutex);
+	kfd_unref_process(p);
 }

drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c

@@ -300,9 +300,14 @@ int kfd_init_apertures(struct kfd_process *process)
 	struct kfd_process_device *pdd;
 
 	/*Iterating over all devices*/
-	while ((dev = kfd_topology_enum_kfd_devices(id)) != NULL &&
+	while (kfd_topology_enum_kfd_devices(id, &dev) == 0 &&
 		id < NUM_OF_SUPPORTED_GPUS) {
 
+		if (!dev) {
+			id++; /* Skip non GPU devices */
+			continue;
+		}
+
 		pdd = kfd_create_process_device_data(dev, process);
 		if (!pdd) {
 			pr_err("Failed to create process device data\n");

drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c

@@ -218,7 +218,7 @@ static int acquire_packet_buffer(struct kernel_queue *kq,
 	rptr = *kq->rptr_kernel;
 	wptr = *kq->wptr_kernel;
 	queue_address = (unsigned int *)kq->pq_kernel_addr;
-	queue_size_dwords = kq->queue->properties.queue_size / sizeof(uint32_t);
+	queue_size_dwords = kq->queue->properties.queue_size / 4;
 
 	pr_debug("rptr: %d\n", rptr);
 	pr_debug("wptr: %d\n", wptr);

drivers/gpu/drm/amd/amdkfd/kfd_module.c

@@ -50,6 +50,15 @@ module_param(sched_policy, int, 0444);
 MODULE_PARM_DESC(sched_policy,
 	"Scheduling policy (0 = HWS (Default), 1 = HWS without over-subscription, 2 = Non-HWS (Used for debugging only)");
 
+int hws_max_conc_proc = 8;
+module_param(hws_max_conc_proc, int, 0444);
+MODULE_PARM_DESC(hws_max_conc_proc,
+	"Max # processes HWS can execute concurrently when sched_policy=0 (0 = no concurrency, #VMIDs for KFD = Maximum(default))");
+
+int cwsr_enable = 1;
+module_param(cwsr_enable, int, 0444);
+MODULE_PARM_DESC(cwsr_enable, "CWSR enable (0 = Off, 1 = On (Default))");
+
 int max_num_of_queues_per_device = KFD_MAX_NUM_OF_QUEUES_PER_DEVICE_DEFAULT;
 module_param(max_num_of_queues_per_device, int, 0444);
 MODULE_PARM_DESC(max_num_of_queues_per_device,
@@ -60,6 +69,11 @@ module_param(send_sigterm, int, 0444);
 MODULE_PARM_DESC(send_sigterm,
 	"Send sigterm to HSA process on unhandled exception (0 = disable, 1 = enable)");
 
+int ignore_crat;
+module_param(ignore_crat, int, 0444);
+MODULE_PARM_DESC(ignore_crat,
+	"Ignore CRAT table during KFD initialization (0 = use CRAT (default), 1 = ignore CRAT)");
+
 static int amdkfd_init_completed;
 
 int kgd2kfd_init(unsigned int interface_version,
@@ -114,6 +128,8 @@ static int __init kfd_module_init(void)
 
 	kfd_process_create_wq();
 
+	kfd_debugfs_init();
+
 	amdkfd_init_completed = 1;
 
 	dev_info(kfd_device, "Initialized module\n");
@@ -130,6 +146,7 @@ static void __exit kfd_module_exit(void)
 {
 	amdkfd_init_completed = 0;
 
+	kfd_debugfs_fini();
 	kfd_process_destroy_wq();
 	kfd_topology_shutdown();
 	kfd_chardev_exit();

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h

@@ -85,6 +85,10 @@ struct mqd_manager {
 				uint64_t queue_address,	uint32_t pipe_id,
 				uint32_t queue_id);
 
+#if defined(CONFIG_DEBUG_FS)
+	int	(*debugfs_show_mqd)(struct seq_file *m, void *data);
+#endif
+
 	struct mutex	mqd_mutex;
 	struct kfd_dev	*dev;
 };

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c

@@ -36,6 +36,11 @@ static inline struct cik_mqd *get_mqd(void *mqd)
 	return (struct cik_mqd *)mqd;
 }
 
+static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+{
+	return (struct cik_sdma_rlc_registers *)mqd;
+}
+
 static int init_mqd(struct mqd_manager *mm, void **mqd,
 		struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
 		struct queue_properties *q)
@@ -149,7 +154,7 @@ static int load_mqd(struct mqd_manager *mm, void *mqd, uint32_t pipe_id,
 {
 	/* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */
 	uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0);
-	uint32_t wptr_mask = (uint32_t)((p->queue_size / sizeof(uint32_t)) - 1);
+	uint32_t wptr_mask = (uint32_t)((p->queue_size / 4) - 1);
 
 	return mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id,
 					  (uint32_t __user *)p->write_ptr,
@@ -160,7 +165,9 @@ static int load_mqd_sdma(struct mqd_manager *mm, void *mqd,
 			 uint32_t pipe_id, uint32_t queue_id,
 			 struct queue_properties *p, struct mm_struct *mms)
 {
-	return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd);
+	return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd,
+					       (uint32_t __user *)p->write_ptr,
+					       mms);
 }
 
 static int update_mqd(struct mqd_manager *mm, void *mqd,
@@ -176,8 +183,7 @@ static int update_mqd(struct mqd_manager *mm, void *mqd,
 	 * Calculating queue size which is log base 2 of actual queue size -1
 	 * dwords and another -1 for ffs
 	 */
-	m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int))
+	m->cp_hqd_pq_control |= order_base_2(q->queue_size / 4) - 1;
-								- 1 - 1;
 	m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
 	m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
 	m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
@@ -202,7 +208,7 @@ static int update_mqd_sdma(struct mqd_manager *mm, void *mqd,
 	struct cik_sdma_rlc_registers *m;
 
 	m = get_sdma_mqd(mqd);
-	m->sdma_rlc_rb_cntl = (ffs(q->queue_size / sizeof(unsigned int)) - 1)
+	m->sdma_rlc_rb_cntl = order_base_2(q->queue_size / 4)
 			<< SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
 			q->vmid << SDMA0_RLC0_RB_CNTL__RB_VMID__SHIFT |
 			1 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_ENABLE__SHIFT |
@@ -343,8 +349,7 @@ static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
 	 * Calculating queue size which is log base 2 of actual queue
 	 * size -1 dwords
 	 */
-	m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int))
+	m->cp_hqd_pq_control |= order_base_2(q->queue_size / 4) - 1;
-								- 1 - 1;
 	m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
 	m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
 	m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
@@ -360,15 +365,25 @@ static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
 	return 0;
 }
 
-struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+#if defined(CONFIG_DEBUG_FS)
+
+static int debugfs_show_mqd(struct seq_file *m, void *data)
 {
-	struct cik_sdma_rlc_registers *m;
+	seq_hex_dump(m, "    ", DUMP_PREFIX_OFFSET, 32, 4,
-
+		     data, sizeof(struct cik_mqd), false);
-	m = (struct cik_sdma_rlc_registers *)mqd;
+	return 0;
-
-	return m;
 }
 
+static int debugfs_show_mqd_sdma(struct seq_file *m, void *data)
+{
+	seq_hex_dump(m, "    ", DUMP_PREFIX_OFFSET, 32, 4,
+		     data, sizeof(struct cik_sdma_rlc_registers), false);
+	return 0;
+}
+
+#endif
+
+
 struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
 		struct kfd_dev *dev)
 {
@@ -392,6 +407,9 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
 		mqd->update_mqd = update_mqd;
 		mqd->destroy_mqd = destroy_mqd;
 		mqd->is_occupied = is_occupied;
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
 		break;
 	case KFD_MQD_TYPE_HIQ:
 		mqd->init_mqd = init_mqd_hiq;
@@ -400,6 +418,9 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
 		mqd->update_mqd = update_mqd_hiq;
 		mqd->destroy_mqd = destroy_mqd;
 		mqd->is_occupied = is_occupied;
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
 		break;
 	case KFD_MQD_TYPE_SDMA:
 		mqd->init_mqd = init_mqd_sdma;
@@ -408,6 +429,9 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
 		mqd->update_mqd = update_mqd_sdma;
 		mqd->destroy_mqd = destroy_mqd_sdma;
 		mqd->is_occupied = is_occupied_sdma;
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
+#endif
 		break;
 	default:
 		kfree(mqd);

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c

@@ -30,7 +30,7 @@
 #include "vi_structs.h"
 #include "gca/gfx_8_0_sh_mask.h"
 #include "gca/gfx_8_0_enum.h"
-
+#include "oss/oss_3_0_sh_mask.h"
 #define CP_MQD_CONTROL__PRIV_STATE__SHIFT 0x8
 
 static inline struct vi_mqd *get_mqd(void *mqd)
@@ -38,6 +38,11 @@ static inline struct vi_mqd *get_mqd(void *mqd)
 	return (struct vi_mqd *)mqd;
 }
 
+static inline struct vi_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+	return (struct vi_sdma_mqd *)mqd;
+}
+
 static int init_mqd(struct mqd_manager *mm, void **mqd,
 			struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
 			struct queue_properties *q)
@@ -84,6 +89,28 @@ static int init_mqd(struct mqd_manager *mm, void **mqd,
 	if (q->format == KFD_QUEUE_FORMAT_AQL)
 		m->cp_hqd_iq_rptr = 1;
 
+	if (q->tba_addr) {
+		m->compute_tba_lo = lower_32_bits(q->tba_addr >> 8);
+		m->compute_tba_hi = upper_32_bits(q->tba_addr >> 8);
+		m->compute_tma_lo = lower_32_bits(q->tma_addr >> 8);
+		m->compute_tma_hi = upper_32_bits(q->tma_addr >> 8);
+		m->compute_pgm_rsrc2 |=
+			(1 << COMPUTE_PGM_RSRC2__TRAP_PRESENT__SHIFT);
+	}
+
+	if (mm->dev->cwsr_enabled && q->ctx_save_restore_area_address) {
+		m->cp_hqd_persistent_state |=
+			(1 << CP_HQD_PERSISTENT_STATE__QSWITCH_MODE__SHIFT);
+		m->cp_hqd_ctx_save_base_addr_lo =
+			lower_32_bits(q->ctx_save_restore_area_address);
+		m->cp_hqd_ctx_save_base_addr_hi =
+			upper_32_bits(q->ctx_save_restore_area_address);
+		m->cp_hqd_ctx_save_size = q->ctx_save_restore_area_size;
+		m->cp_hqd_cntl_stack_size = q->ctl_stack_size;
+		m->cp_hqd_cntl_stack_offset = q->ctl_stack_size;
+		m->cp_hqd_wg_state_offset = q->ctl_stack_size;
+	}
+
 	*mqd = m;
 	if (gart_addr)
 		*gart_addr = addr;
@@ -98,7 +125,7 @@ static int load_mqd(struct mqd_manager *mm, void *mqd,
 {
 	/* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */
 	uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0);
-	uint32_t wptr_mask = (uint32_t)((p->queue_size / sizeof(uint32_t)) - 1);
+	uint32_t wptr_mask = (uint32_t)((p->queue_size / 4) - 1);
 
 	return mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id,
 					  (uint32_t __user *)p->write_ptr,
@@ -116,8 +143,7 @@ static int __update_mqd(struct mqd_manager *mm, void *mqd,
 	m->cp_hqd_pq_control = 5 << CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE__SHIFT |
 			atc_bit << CP_HQD_PQ_CONTROL__PQ_ATC__SHIFT |
 			mtype << CP_HQD_PQ_CONTROL__MTYPE__SHIFT;
-	m->cp_hqd_pq_control |=
+	m->cp_hqd_pq_control |=	order_base_2(q->queue_size / 4) - 1;
-			ffs(q->queue_size / sizeof(unsigned int)) - 1 - 1;
 	pr_debug("cp_hqd_pq_control 0x%x\n", m->cp_hqd_pq_control);
 
 	m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
@@ -147,7 +173,7 @@ static int __update_mqd(struct mqd_manager *mm, void *mqd,
 	 * is safe, giving a maximum field value of 0xA.
 	 */
 	m->cp_hqd_eop_control |= min(0xA,
-		ffs(q->eop_ring_buffer_size / sizeof(unsigned int)) - 1 - 1);
+		order_base_2(q->eop_ring_buffer_size / 4) - 1);
 	m->cp_hqd_eop_base_addr_lo =
 			lower_32_bits(q->eop_ring_buffer_address >> 8);
 	m->cp_hqd_eop_base_addr_hi =
@@ -163,6 +189,11 @@ static int __update_mqd(struct mqd_manager *mm, void *mqd,
 				2 << CP_HQD_PQ_CONTROL__SLOT_BASED_WPTR__SHIFT;
 	}
 
+	if (mm->dev->cwsr_enabled && q->ctx_save_restore_area_address)
+		m->cp_hqd_ctx_save_control =
+			atc_bit << CP_HQD_CTX_SAVE_CONTROL__ATC__SHIFT |
+			mtype << CP_HQD_CTX_SAVE_CONTROL__MTYPE__SHIFT;
+
 	q->is_active = (q->queue_size > 0 &&
 			q->queue_address != 0 &&
 			q->queue_percent > 0);
@@ -234,6 +265,117 @@ static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
 	return retval;
 }
 
+static int init_mqd_sdma(struct mqd_manager *mm, void **mqd,
+		struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
+		struct queue_properties *q)
+{
+	int retval;
+	struct vi_sdma_mqd *m;
+
+
+	retval = kfd_gtt_sa_allocate(mm->dev,
+			sizeof(struct vi_sdma_mqd),
+			mqd_mem_obj);
+
+	if (retval != 0)
+		return -ENOMEM;
+
+	m = (struct vi_sdma_mqd *) (*mqd_mem_obj)->cpu_ptr;
+
+	memset(m, 0, sizeof(struct vi_sdma_mqd));
+
+	*mqd = m;
+	if (gart_addr != NULL)
+		*gart_addr = (*mqd_mem_obj)->gpu_addr;
+
+	retval = mm->update_mqd(mm, m, q);
+
+	return retval;
+}
+
+static void uninit_mqd_sdma(struct mqd_manager *mm, void *mqd,
+		struct kfd_mem_obj *mqd_mem_obj)
+{
+	kfd_gtt_sa_free(mm->dev, mqd_mem_obj);
+}
+
+static int load_mqd_sdma(struct mqd_manager *mm, void *mqd,
+		uint32_t pipe_id, uint32_t queue_id,
+		struct queue_properties *p, struct mm_struct *mms)
+{
+	return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd,
+					       (uint32_t __user *)p->write_ptr,
+					       mms);
+}
+
+static int update_mqd_sdma(struct mqd_manager *mm, void *mqd,
+		struct queue_properties *q)
+{
+	struct vi_sdma_mqd *m;
+
+	m = get_sdma_mqd(mqd);
+	m->sdmax_rlcx_rb_cntl = order_base_2(q->queue_size / 4)
+		<< SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
+		q->vmid << SDMA0_RLC0_RB_CNTL__RB_VMID__SHIFT |
+		1 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_ENABLE__SHIFT |
+		6 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_TIMER__SHIFT;
+
+	m->sdmax_rlcx_rb_base = lower_32_bits(q->queue_address >> 8);
+	m->sdmax_rlcx_rb_base_hi = upper_32_bits(q->queue_address >> 8);
+	m->sdmax_rlcx_rb_rptr_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
+	m->sdmax_rlcx_rb_rptr_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
+	m->sdmax_rlcx_doorbell =
+		q->doorbell_off << SDMA0_RLC0_DOORBELL__OFFSET__SHIFT;
+
+	m->sdmax_rlcx_virtual_addr = q->sdma_vm_addr;
+
+	m->sdma_engine_id = q->sdma_engine_id;
+	m->sdma_queue_id = q->sdma_queue_id;
+
+	q->is_active = (q->queue_size > 0 &&
+			q->queue_address != 0 &&
+			q->queue_percent > 0);
+
+	return 0;
+}
+
+/*
+ *  * preempt type here is ignored because there is only one way
+ *  * to preempt sdma queue
+ */
+static int destroy_mqd_sdma(struct mqd_manager *mm, void *mqd,
+		enum kfd_preempt_type type,
+		unsigned int timeout, uint32_t pipe_id,
+		uint32_t queue_id)
+{
+	return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->kgd, mqd, timeout);
+}
+
+static bool is_occupied_sdma(struct mqd_manager *mm, void *mqd,
+		uint64_t queue_address, uint32_t pipe_id,
+		uint32_t queue_id)
+{
+	return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->kgd, mqd);
+}
+
+#if defined(CONFIG_DEBUG_FS)
+
+static int debugfs_show_mqd(struct seq_file *m, void *data)
+{
+	seq_hex_dump(m, "    ", DUMP_PREFIX_OFFSET, 32, 4,
+		     data, sizeof(struct vi_mqd), false);
+	return 0;
+}
+
+static int debugfs_show_mqd_sdma(struct seq_file *m, void *data)
+{
+	seq_hex_dump(m, "    ", DUMP_PREFIX_OFFSET, 32, 4,
+		     data, sizeof(struct vi_sdma_mqd), false);
+	return 0;
+}
+
+#endif
+
 struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
 		struct kfd_dev *dev)
 {
@@ -257,6 +399,9 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
 		mqd->update_mqd = update_mqd;
 		mqd->destroy_mqd = destroy_mqd;
 		mqd->is_occupied = is_occupied;
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
 		break;
 	case KFD_MQD_TYPE_HIQ:
 		mqd->init_mqd = init_mqd_hiq;
@@ -265,8 +410,20 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
 		mqd->update_mqd = update_mqd_hiq;
 		mqd->destroy_mqd = destroy_mqd;
 		mqd->is_occupied = is_occupied;
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
 		break;
 	case KFD_MQD_TYPE_SDMA:
+		mqd->init_mqd = init_mqd_sdma;
+		mqd->uninit_mqd = uninit_mqd_sdma;
+		mqd->load_mqd = load_mqd_sdma;
+		mqd->update_mqd = update_mqd_sdma;
+		mqd->destroy_mqd = destroy_mqd_sdma;
+		mqd->is_occupied = is_occupied_sdma;
+#if defined(CONFIG_DEBUG_FS)
+		mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
+#endif
 		break;
 	default:
 		kfree(mqd);

drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c

@@ -45,7 +45,7 @@ static unsigned int build_pm4_header(unsigned int opcode, size_t packet_size)
 
 	header.u32All = 0;
 	header.opcode = opcode;
-	header.count = packet_size/sizeof(uint32_t) - 2;
+	header.count = packet_size / 4 - 2;
 	header.type = PM4_TYPE_3;
 
 	return header.u32All;
@@ -55,15 +55,27 @@ static void pm_calc_rlib_size(struct packet_manager *pm,
 				unsigned int *rlib_size,
 				bool *over_subscription)
 {
-	unsigned int process_count, queue_count;
+	unsigned int process_count, queue_count, compute_queue_count;
 	unsigned int map_queue_size;
+	unsigned int max_proc_per_quantum = 1;
+	struct kfd_dev *dev = pm->dqm->dev;
 
 	process_count = pm->dqm->processes_count;
 	queue_count = pm->dqm->queue_count;
+	compute_queue_count = queue_count - pm->dqm->sdma_queue_count;
 
-	/* check if there is over subscription*/
+	/* check if there is over subscription
+	 * Note: the arbitration between the number of VMIDs and
+	 * hws_max_conc_proc has been done in
+	 * kgd2kfd_device_init().
+	 */
 	*over_subscription = false;
-	if ((process_count > 1) || queue_count > get_queues_num(pm->dqm)) {
+
+	if (dev->max_proc_per_quantum > 1)
+		max_proc_per_quantum = dev->max_proc_per_quantum;
+
+	if ((process_count > max_proc_per_quantum) ||
+	    compute_queue_count > get_queues_num(pm->dqm)) {
 		*over_subscription = true;
 		pr_debug("Over subscribed runlist\n");
 	}
@@ -116,10 +128,24 @@ static int pm_create_runlist(struct packet_manager *pm, uint32_t *buffer,
 			uint64_t ib, size_t ib_size_in_dwords, bool chain)
 {
 	struct pm4_mes_runlist *packet;
+	int concurrent_proc_cnt = 0;
+	struct kfd_dev *kfd = pm->dqm->dev;
 
 	if (WARN_ON(!ib))
 		return -EFAULT;
 
+	/* Determine the number of processes to map together to HW:
+	 * it can not exceed the number of VMIDs available to the
+	 * scheduler, and it is determined by the smaller of the number
+	 * of processes in the runlist and kfd module parameter
+	 * hws_max_conc_proc.
+	 * Note: the arbitration between the number of VMIDs and
+	 * hws_max_conc_proc has been done in
+	 * kgd2kfd_device_init().
+	 */
+	concurrent_proc_cnt = min(pm->dqm->processes_count,
+			kfd->max_proc_per_quantum);
+
 	packet = (struct pm4_mes_runlist *)buffer;
 
 	memset(buffer, 0, sizeof(struct pm4_mes_runlist));
@@ -130,6 +156,7 @@ static int pm_create_runlist(struct packet_manager *pm, uint32_t *buffer,
 	packet->bitfields4.chain = chain ? 1 : 0;
 	packet->bitfields4.offload_polling = 0;
 	packet->bitfields4.valid = 1;
+	packet->bitfields4.process_cnt = concurrent_proc_cnt;
 	packet->ordinal2 = lower_32_bits(ib);
 	packet->bitfields3.ib_base_hi = upper_32_bits(ib);
 
@@ -251,6 +278,7 @@ static int pm_create_runlist_ib(struct packet_manager *pm,
 		return retval;
 
 	*rl_size_bytes = alloc_size_bytes;
+	pm->ib_size_bytes = alloc_size_bytes;
 
 	pr_debug("Building runlist ib process count: %d queues count %d\n",
 		pm->dqm->processes_count, pm->dqm->queue_count);
@@ -564,3 +592,26 @@ void pm_release_ib(struct packet_manager *pm)
 	}
 	mutex_unlock(&pm->lock);
 }
+
+#if defined(CONFIG_DEBUG_FS)
+
+int pm_debugfs_runlist(struct seq_file *m, void *data)
+{
+	struct packet_manager *pm = data;
+
+	mutex_lock(&pm->lock);
+
+	if (!pm->allocated) {
+		seq_puts(m, "  No active runlist\n");
+		goto out;
+	}
+
+	seq_hex_dump(m, "  ", DUMP_PREFIX_OFFSET, 32, 4,
+		     pm->ib_buffer_obj->cpu_ptr, pm->ib_size_bytes, false);
+
+out:
+	mutex_unlock(&pm->lock);
+	return 0;
+}
+
+#endif

drivers/gpu/drm/amd/amdkfd/kfd_pasid.c

@@ -59,7 +59,7 @@ unsigned int kfd_pasid_alloc(void)
 		struct kfd_dev *dev = NULL;
 		unsigned int i = 0;
 
-		while ((dev = kfd_topology_enum_kfd_devices(i)) != NULL) {
+		while ((kfd_topology_enum_kfd_devices(i, &dev)) == 0) {
 			if (dev && dev->kfd2kgd) {
 				kfd2kgd = dev->kfd2kgd;
 				break;

drivers/gpu/drm/amd/amdkfd/kfd_priv.h

@@ -33,6 +33,8 @@
 #include <linux/kfd_ioctl.h>
 #include <linux/idr.h>
 #include <linux/kfifo.h>
+#include <linux/seq_file.h>
+#include <linux/kref.h>
 #include <kgd_kfd_interface.h>
 
 #include "amd_shared.h"
@@ -41,6 +43,7 @@
 
 #define KFD_MMAP_DOORBELL_MASK 0x8000000000000
 #define KFD_MMAP_EVENTS_MASK 0x4000000000000
+#define KFD_MMAP_RESERVED_MEM_MASK 0x2000000000000
 
 /*
  * When working with cp scheduler we should assign the HIQ manually or via
@@ -62,6 +65,15 @@
 #define KFD_MAX_NUM_OF_PROCESSES 512
 #define KFD_MAX_NUM_OF_QUEUES_PER_PROCESS 1024
 
+/*
+ * Size of the per-process TBA+TMA buffer: 2 pages
+ *
+ * The first page is the TBA used for the CWSR ISA code. The second
+ * page is used as TMA for daisy changing a user-mode trap handler.
+ */
+#define KFD_CWSR_TBA_TMA_SIZE (PAGE_SIZE * 2)
+#define KFD_CWSR_TMA_OFFSET PAGE_SIZE
+
 /*
  * Kernel module parameter to specify maximum number of supported queues per
  * device
@@ -78,12 +90,26 @@ extern int max_num_of_queues_per_device;
 /* Kernel module parameter to specify the scheduling policy */
 extern int sched_policy;
 
+/*
+ * Kernel module parameter to specify the maximum process
+ * number per HW scheduler
+ */
+extern int hws_max_conc_proc;
+
+extern int cwsr_enable;
+
 /*
  * Kernel module parameter to specify whether to send sigterm to HSA process on
  * unhandled exception
  */
 extern int send_sigterm;
 
+/*
+ * Ignore CRAT table during KFD initialization, can be used to work around
+ * broken CRAT tables on some AMD systems
+ */
+extern int ignore_crat;
+
 /**
  * enum kfd_sched_policy
  *
@@ -131,6 +157,7 @@ struct kfd_device_info {
 	size_t ih_ring_entry_size;
 	uint8_t num_of_watch_points;
 	uint16_t mqd_size_aligned;
+	bool supports_cwsr;
 };
 
 struct kfd_mem_obj {
@@ -200,6 +227,14 @@ struct kfd_dev {
 
 	/* Debug manager */
 	struct kfd_dbgmgr           *dbgmgr;
+
+	/* Maximum process number mapped to HW scheduler */
+	unsigned int max_proc_per_quantum;
+
+	/* CWSR */
+	bool cwsr_enabled;
+	const void *cwsr_isa;
+	unsigned int cwsr_isa_size;
 };
 
 /* KGD2KFD callbacks */
@@ -332,6 +367,9 @@ struct queue_properties {
 	uint32_t eop_ring_buffer_size;
 	uint64_t ctx_save_restore_area_address;
 	uint32_t ctx_save_restore_area_size;
+	uint32_t ctl_stack_size;
+	uint64_t tba_addr;
+	uint64_t tma_addr;
 };
 
 /**
@@ -439,6 +477,11 @@ struct qcm_process_device {
 	uint32_t num_gws;
 	uint32_t num_oac;
 	uint32_t sh_hidden_private_base;
+
+	/* CWSR memory */
+	void *cwsr_kaddr;
+	uint64_t tba_addr;
+	uint64_t tma_addr;
 };
 
 
@@ -501,6 +544,9 @@ struct kfd_process {
 	 */
 	void *mm;
 
+	struct kref ref;
+	struct work_struct release_work;
+
 	struct mutex mutex;
 
 	/*
@@ -563,9 +609,10 @@ struct amdkfd_ioctl_desc {
 
 void kfd_process_create_wq(void);
 void kfd_process_destroy_wq(void);
-struct kfd_process *kfd_create_process(const struct task_struct *);
+struct kfd_process *kfd_create_process(struct file *filep);
 struct kfd_process *kfd_get_process(const struct task_struct *);
 struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid);
+void kfd_unref_process(struct kfd_process *p);
 
 struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev,
 						struct kfd_process *p);
@@ -577,6 +624,9 @@ struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
 struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
 							struct kfd_process *p);
 
+int kfd_reserved_mem_mmap(struct kfd_process *process,
+			  struct vm_area_struct *vma);
+
 /* Process device data iterator */
 struct kfd_process_device *kfd_get_first_process_device_data(
 							struct kfd_process *p);
@@ -624,9 +674,12 @@ int kfd_topology_init(void);
 void kfd_topology_shutdown(void);
 int kfd_topology_add_device(struct kfd_dev *gpu);
 int kfd_topology_remove_device(struct kfd_dev *gpu);
+struct kfd_topology_device *kfd_topology_device_by_proximity_domain(
+						uint32_t proximity_domain);
 struct kfd_dev *kfd_device_by_id(uint32_t gpu_id);
 struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev);
-struct kfd_dev *kfd_topology_enum_kfd_devices(uint8_t idx);
+int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev);
+int kfd_numa_node_to_apic_id(int numa_node_id);
 
 /* Interrupts */
 int kfd_interrupt_init(struct kfd_dev *dev);
@@ -643,8 +696,6 @@ int kgd2kfd_resume(struct kfd_dev *kfd);
 int kfd_init_apertures(struct kfd_process *process);
 
 /* Queue Context Management */
-struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd);
-
 int init_queue(struct queue **q, const struct queue_properties *properties);
 void uninit_queue(struct queue *q);
 void print_queue_properties(struct queue_properties *q);
@@ -699,6 +750,7 @@ struct packet_manager {
 	struct mutex lock;
 	bool allocated;
 	struct kfd_mem_obj *ib_buffer_obj;
+	unsigned int ib_size_bytes;
 };
 
 int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm);
@@ -745,4 +797,23 @@ int kfd_event_destroy(struct kfd_process *p, uint32_t event_id);
 
 int dbgdev_wave_reset_wavefronts(struct kfd_dev *dev, struct kfd_process *p);
 
+/* Debugfs */
+#if defined(CONFIG_DEBUG_FS)
+
+void kfd_debugfs_init(void);
+void kfd_debugfs_fini(void);
+int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data);
+int pqm_debugfs_mqds(struct seq_file *m, void *data);
+int kfd_debugfs_hqds_by_device(struct seq_file *m, void *data);
+int dqm_debugfs_hqds(struct seq_file *m, void *data);
+int kfd_debugfs_rls_by_device(struct seq_file *m, void *data);
+int pm_debugfs_runlist(struct seq_file *m, void *data);
+
+#else
+
+static inline void kfd_debugfs_init(void) {}
+static inline void kfd_debugfs_fini(void) {}
+
+#endif
+
 #endif

drivers/gpu/drm/amd/amdkfd/kfd_process.c

@@ -24,10 +24,12 @@
 #include <linux/log2.h>
 #include <linux/sched.h>
 #include <linux/sched/mm.h>
+#include <linux/sched/task.h>
 #include <linux/slab.h>
 #include <linux/amd-iommu.h>
 #include <linux/notifier.h>
 #include <linux/compat.h>
+#include <linux/mman.h>
 
 struct mm_struct;
 
@@ -46,13 +48,12 @@ DEFINE_STATIC_SRCU(kfd_processes_srcu);
 
 static struct workqueue_struct *kfd_process_wq;
 
-struct kfd_process_release_work {
-	struct work_struct kfd_work;
-	struct kfd_process *p;
-};
-
 static struct kfd_process *find_process(const struct task_struct *thread);
-static struct kfd_process *create_process(const struct task_struct *thread);
+static void kfd_process_ref_release(struct kref *ref);
+static struct kfd_process *create_process(const struct task_struct *thread,
+					struct file *filep);
+static int kfd_process_init_cwsr(struct kfd_process *p, struct file *filep);
+
 
 void kfd_process_create_wq(void)
 {
@@ -68,9 +69,10 @@ void kfd_process_destroy_wq(void)
 	}
 }
 
-struct kfd_process *kfd_create_process(const struct task_struct *thread)
+struct kfd_process *kfd_create_process(struct file *filep)
 {
 	struct kfd_process *process;
+	struct task_struct *thread = current;
 
 	if (!thread->mm)
 		return ERR_PTR(-EINVAL);
@@ -79,9 +81,6 @@ struct kfd_process *kfd_create_process(const struct task_struct *thread)
 	if (thread->group_leader->mm != thread->mm)
 		return ERR_PTR(-EINVAL);
 
-	/* Take mmap_sem because we call __mmu_notifier_register inside */
-	down_write(&thread->mm->mmap_sem);
-
 	/*
 	 * take kfd processes mutex before starting of process creation
 	 * so there won't be a case where two threads of the same process
@@ -93,14 +92,11 @@ struct kfd_process *kfd_create_process(const struct task_struct *thread)
 	process = find_process(thread);
 	if (process)
 		pr_debug("Process already found\n");
-
+	else
-	if (!process)
+		process = create_process(thread, filep);
-		process = create_process(thread);
 
 	mutex_unlock(&kfd_processes_mutex);
 
-	up_write(&thread->mm->mmap_sem);
-
 	return process;
 }
 
@@ -144,63 +140,75 @@ static struct kfd_process *find_process(const struct task_struct *thread)
 	return p;
 }
 
-static void kfd_process_wq_release(struct work_struct *work)
+void kfd_unref_process(struct kfd_process *p)
+{
+	kref_put(&p->ref, kfd_process_ref_release);
+}
+
+static void kfd_process_destroy_pdds(struct kfd_process *p)
 {
-	struct kfd_process_release_work *my_work;
 	struct kfd_process_device *pdd, *temp;
-	struct kfd_process *p;
-
-	my_work = (struct kfd_process_release_work *) work;
-
-	p = my_work->p;
-
-	pr_debug("Releasing process (pasid %d) in workqueue\n",
-			p->pasid);
-
-	mutex_lock(&p->mutex);
 
 	list_for_each_entry_safe(pdd, temp, &p->per_device_data,
-							per_device_list) {
+				 per_device_list) {
-		pr_debug("Releasing pdd (topology id %d) for process (pasid %d) in workqueue\n",
+		pr_debug("Releasing pdd (topology id %d) for process (pasid %d)\n",
 				pdd->dev->id, p->pasid);
 
-		if (pdd->bound == PDD_BOUND)
-			amd_iommu_unbind_pasid(pdd->dev->pdev, p->pasid);
-
 		list_del(&pdd->per_device_list);
+
+		if (pdd->qpd.cwsr_kaddr)
+			free_pages((unsigned long)pdd->qpd.cwsr_kaddr,
+				get_order(KFD_CWSR_TBA_TMA_SIZE));
+
 		kfree(pdd);
 	}
+}
+
+/* No process locking is needed in this function, because the process
+ * is not findable any more. We must assume that no other thread is
+ * using it any more, otherwise we couldn't safely free the process
+ * structure in the end.
+ */
+static void kfd_process_wq_release(struct work_struct *work)
+{
+	struct kfd_process *p = container_of(work, struct kfd_process,
+					     release_work);
+	struct kfd_process_device *pdd;
+
+	pr_debug("Releasing process (pasid %d) in workqueue\n", p->pasid);
+
+	list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
+		if (pdd->bound == PDD_BOUND)
+			amd_iommu_unbind_pasid(pdd->dev->pdev, p->pasid);
+	}
+
+	kfd_process_destroy_pdds(p);
 
 	kfd_event_free_process(p);
 
 	kfd_pasid_free(p->pasid);
 	kfd_free_process_doorbells(p);
 
-	mutex_unlock(&p->mutex);
-
 	mutex_destroy(&p->mutex);
 
-	kfree(p);
+	put_task_struct(p->lead_thread);
 
-	kfree(work);
+	kfree(p);
+}
+
+static void kfd_process_ref_release(struct kref *ref)
+{
+	struct kfd_process *p = container_of(ref, struct kfd_process, ref);
+
+	INIT_WORK(&p->release_work, kfd_process_wq_release);
+	queue_work(kfd_process_wq, &p->release_work);
 }
 
 static void kfd_process_destroy_delayed(struct rcu_head *rcu)
 {
-	struct kfd_process_release_work *work;
+	struct kfd_process *p = container_of(rcu, struct kfd_process, rcu);
-	struct kfd_process *p;
 
-	p = container_of(rcu, struct kfd_process, rcu);
+	kfd_unref_process(p);
-
-	mmdrop(p->mm);
-
-	work = kmalloc(sizeof(struct kfd_process_release_work), GFP_ATOMIC);
-
-	if (work) {
-		INIT_WORK((struct work_struct *) work, kfd_process_wq_release);
-		work->p = p;
-		queue_work(kfd_process_wq, (struct work_struct *) work);
-	}
 }
 
 static void kfd_process_notifier_release(struct mmu_notifier *mn,
@@ -244,15 +252,12 @@ static void kfd_process_notifier_release(struct mmu_notifier *mn,
 	kfd_process_dequeue_from_all_devices(p);
 	pqm_uninit(&p->pqm);
 
+	/* Indicate to other users that MM is no longer valid */
+	p->mm = NULL;
+
 	mutex_unlock(&p->mutex);
 
-	/*
+	mmu_notifier_unregister_no_release(&p->mmu_notifier, mm);
-	 * Because we drop mm_count inside kfd_process_destroy_delayed
-	 * and because the mmu_notifier_unregister function also drop
-	 * mm_count we need to take an extra count here.
-	 */
-	mmgrab(p->mm);
-	mmu_notifier_unregister_no_release(&p->mmu_notifier, p->mm);
 	mmu_notifier_call_srcu(&p->rcu, &kfd_process_destroy_delayed);
 }
 
@@ -260,7 +265,44 @@ static const struct mmu_notifier_ops kfd_process_mmu_notifier_ops = {
 	.release = kfd_process_notifier_release,
 };
 
-static struct kfd_process *create_process(const struct task_struct *thread)
+static int kfd_process_init_cwsr(struct kfd_process *p, struct file *filep)
+{
+	unsigned long  offset;
+	struct kfd_process_device *pdd = NULL;
+	struct kfd_dev *dev = NULL;
+	struct qcm_process_device *qpd = NULL;
+
+	list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
+		dev = pdd->dev;
+		qpd = &pdd->qpd;
+		if (!dev->cwsr_enabled || qpd->cwsr_kaddr)
+			continue;
+		offset = (dev->id | KFD_MMAP_RESERVED_MEM_MASK) << PAGE_SHIFT;
+		qpd->tba_addr = (int64_t)vm_mmap(filep, 0,
+			KFD_CWSR_TBA_TMA_SIZE, PROT_READ | PROT_EXEC,
+			MAP_SHARED, offset);
+
+		if (IS_ERR_VALUE(qpd->tba_addr)) {
+			int err = qpd->tba_addr;
+
+			pr_err("Failure to set tba address. error %d.\n", err);
+			qpd->tba_addr = 0;
+			qpd->cwsr_kaddr = NULL;
+			return err;
+		}
+
+		memcpy(qpd->cwsr_kaddr, dev->cwsr_isa, dev->cwsr_isa_size);
+
+		qpd->tma_addr = qpd->tba_addr + KFD_CWSR_TMA_OFFSET;
+		pr_debug("set tba :0x%llx, tma:0x%llx, cwsr_kaddr:%p for pqm.\n",
+			qpd->tba_addr, qpd->tma_addr, qpd->cwsr_kaddr);
+	}
+
+	return 0;
+}
+
+static struct kfd_process *create_process(const struct task_struct *thread,
+					struct file *filep)
 {
 	struct kfd_process *process;
 	int err = -ENOMEM;
@@ -277,13 +319,15 @@ static struct kfd_process *create_process(const struct task_struct *thread)
 	if (kfd_alloc_process_doorbells(process) < 0)
 		goto err_alloc_doorbells;
 
+	kref_init(&process->ref);
+
 	mutex_init(&process->mutex);
 
 	process->mm = thread->mm;
 
 	/* register notifier */
 	process->mmu_notifier.ops = &kfd_process_mmu_notifier_ops;
-	err = __mmu_notifier_register(&process->mmu_notifier, process->mm);
+	err = mmu_notifier_register(&process->mmu_notifier, process->mm);
 	if (err)
 		goto err_mmu_notifier;
 
@@ -291,6 +335,7 @@ static struct kfd_process *create_process(const struct task_struct *thread)
 			(uintptr_t)process->mm);
 
 	process->lead_thread = thread->group_leader;
+	get_task_struct(process->lead_thread);
 
 	INIT_LIST_HEAD(&process->per_device_data);
 
@@ -306,8 +351,14 @@ static struct kfd_process *create_process(const struct task_struct *thread)
 	if (err != 0)
 		goto err_init_apertures;
 
+	err = kfd_process_init_cwsr(process, filep);
+	if (err)
+		goto err_init_cwsr;
+
 	return process;
 
+err_init_cwsr:
+	kfd_process_destroy_pdds(process);
 err_init_apertures:
 	pqm_uninit(&process->pqm);
 err_process_pqm_init:
@@ -343,16 +394,18 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
 	struct kfd_process_device *pdd = NULL;
 
 	pdd = kzalloc(sizeof(*pdd), GFP_KERNEL);
-	if (pdd != NULL) {
+	if (!pdd)
-		pdd->dev = dev;
+		return NULL;
-		INIT_LIST_HEAD(&pdd->qpd.queues_list);
+
-		INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
+	pdd->dev = dev;
-		pdd->qpd.dqm = dev->dqm;
+	INIT_LIST_HEAD(&pdd->qpd.queues_list);
-		pdd->process = p;
+	INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
-		pdd->bound = PDD_UNBOUND;
+	pdd->qpd.dqm = dev->dqm;
-		pdd->already_dequeued = false;
+	pdd->qpd.pqm = &p->pqm;
-		list_add(&pdd->per_device_list, &p->per_device_data);
+	pdd->process = p;
-	}
+	pdd->bound = PDD_UNBOUND;
+	pdd->already_dequeued = false;
+	list_add(&pdd->per_device_list, &p->per_device_data);
 
 	return pdd;
 }
@@ -483,6 +536,8 @@ void kfd_process_iommu_unbind_callback(struct kfd_dev *dev, unsigned int pasid)
 
 	mutex_unlock(kfd_get_dbgmgr_mutex());
 
+	mutex_lock(&p->mutex);
+
 	pdd = kfd_get_process_device_data(dev, p);
 	if (pdd)
 		/* For GPU relying on IOMMU, we need to dequeue here
@@ -491,6 +546,8 @@ void kfd_process_iommu_unbind_callback(struct kfd_dev *dev, unsigned int pasid)
 		kfd_process_dequeue_from_device(pdd);
 
 	mutex_unlock(&p->mutex);
+
+	kfd_unref_process(p);
 }
 
 struct kfd_process_device *kfd_get_first_process_device_data(
@@ -515,22 +572,86 @@ bool kfd_has_process_device_data(struct kfd_process *p)
 	return !(list_empty(&p->per_device_data));
 }
 
-/* This returns with process->mutex locked. */
+/* This increments the process->ref counter. */
 struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid)
 {
-	struct kfd_process *p;
+	struct kfd_process *p, *ret_p = NULL;
 	unsigned int temp;
 
 	int idx = srcu_read_lock(&kfd_processes_srcu);
 
 	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
 		if (p->pasid == pasid) {
-			mutex_lock(&p->mutex);
+			kref_get(&p->ref);
+			ret_p = p;
 			break;
 		}
 	}
 
 	srcu_read_unlock(&kfd_processes_srcu, idx);
 
-	return p;
+	return ret_p;
 }
+
+int kfd_reserved_mem_mmap(struct kfd_process *process,
+			  struct vm_area_struct *vma)
+{
+	struct kfd_dev *dev = kfd_device_by_id(vma->vm_pgoff);
+	struct kfd_process_device *pdd;
+	struct qcm_process_device *qpd;
+
+	if (!dev)
+		return -EINVAL;
+	if ((vma->vm_end - vma->vm_start) != KFD_CWSR_TBA_TMA_SIZE) {
+		pr_err("Incorrect CWSR mapping size.\n");
+		return -EINVAL;
+	}
+
+	pdd = kfd_get_process_device_data(dev, process);
+	if (!pdd)
+		return -EINVAL;
+	qpd = &pdd->qpd;
+
+	qpd->cwsr_kaddr = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
+					get_order(KFD_CWSR_TBA_TMA_SIZE));
+	if (!qpd->cwsr_kaddr) {
+		pr_err("Error allocating per process CWSR buffer.\n");
+		return -ENOMEM;
+	}
+
+	vma->vm_flags |= VM_IO | VM_DONTCOPY | VM_DONTEXPAND
+		| VM_NORESERVE | VM_DONTDUMP | VM_PFNMAP;
+	/* Mapping pages to user process */
+	return remap_pfn_range(vma, vma->vm_start,
+			       PFN_DOWN(__pa(qpd->cwsr_kaddr)),
+			       KFD_CWSR_TBA_TMA_SIZE, vma->vm_page_prot);
+}
+
+#if defined(CONFIG_DEBUG_FS)
+
+int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data)
+{
+	struct kfd_process *p;
+	unsigned int temp;
+	int r = 0;
+
+	int idx = srcu_read_lock(&kfd_processes_srcu);
+
+	hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
+		seq_printf(m, "Process %d PASID %d:\n",
+			   p->lead_thread->tgid, p->pasid);
+
+		mutex_lock(&p->mutex);
+		r = pqm_debugfs_mqds(m, &p->pqm);
+		mutex_unlock(&p->mutex);
+
+		if (r)
+			break;
+	}
+
+	srcu_read_unlock(&kfd_processes_srcu, idx);
+
+	return r;
+}
+
+#endif

drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c

@@ -178,10 +178,8 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 		return retval;
 
 	if (list_empty(&pdd->qpd.queues_list) &&
-	    list_empty(&pdd->qpd.priv_queue_list)) {
+	    list_empty(&pdd->qpd.priv_queue_list))
-		pdd->qpd.pqm = pqm;
 		dev->dqm->ops.register_process(dev->dqm, &pdd->qpd);
-	}
 
 	pqn = kzalloc(sizeof(*pqn), GFP_KERNEL);
 	if (!pqn) {
@@ -203,8 +201,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 			goto err_create_queue;
 		pqn->q = q;
 		pqn->kq = NULL;
-		retval = dev->dqm->ops.create_queue(dev->dqm, q, &pdd->qpd,
+		retval = dev->dqm->ops.create_queue(dev->dqm, q, &pdd->qpd);
-						&q->properties.vmid);
 		pr_debug("DQM returned %d for create_queue\n", retval);
 		print_queue(q);
 		break;
@@ -224,8 +221,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 			goto err_create_queue;
 		pqn->q = q;
 		pqn->kq = NULL;
-		retval = dev->dqm->ops.create_queue(dev->dqm, q, &pdd->qpd,
+		retval = dev->dqm->ops.create_queue(dev->dqm, q, &pdd->qpd);
-						&q->properties.vmid);
 		pr_debug("DQM returned %d for create_queue\n", retval);
 		print_queue(q);
 		break;
@@ -315,6 +311,10 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
 	if (pqn->q) {
 		dqm = pqn->q->device->dqm;
 		retval = dqm->ops.destroy_queue(dqm, &pdd->qpd, pqn->q);
+		if (retval) {
+			pr_debug("Destroy queue failed, returned %d\n", retval);
+			goto err_destroy_queue;
+		}
 		uninit_queue(pqn->q);
 	}
 
@@ -326,6 +326,7 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
 	    list_empty(&pdd->qpd.priv_queue_list))
 		dqm->ops.unregister_process(dqm, &pdd->qpd);
 
+err_destroy_queue:
 	return retval;
 }
 
@@ -367,4 +368,67 @@ struct kernel_queue *pqm_get_kernel_queue(
 	return NULL;
 }
 
+#if defined(CONFIG_DEBUG_FS)
 
+int pqm_debugfs_mqds(struct seq_file *m, void *data)
+{
+	struct process_queue_manager *pqm = data;
+	struct process_queue_node *pqn;
+	struct queue *q;
+	enum KFD_MQD_TYPE mqd_type;
+	struct mqd_manager *mqd_manager;
+	int r = 0;
+
+	list_for_each_entry(pqn, &pqm->queues, process_queue_list) {
+		if (pqn->q) {
+			q = pqn->q;
+			switch (q->properties.type) {
+			case KFD_QUEUE_TYPE_SDMA:
+				seq_printf(m, "  SDMA queue on device %x\n",
+					   q->device->id);
+				mqd_type = KFD_MQD_TYPE_SDMA;
+				break;
+			case KFD_QUEUE_TYPE_COMPUTE:
+				seq_printf(m, "  Compute queue on device %x\n",
+					   q->device->id);
+				mqd_type = KFD_MQD_TYPE_CP;
+				break;
+			default:
+				seq_printf(m,
+				"  Bad user queue type %d on device %x\n",
+					   q->properties.type, q->device->id);
+				continue;
+			}
+			mqd_manager = q->device->dqm->ops.get_mqd_manager(
+				q->device->dqm, mqd_type);
+		} else if (pqn->kq) {
+			q = pqn->kq->queue;
+			mqd_manager = pqn->kq->mqd;
+			switch (q->properties.type) {
+			case KFD_QUEUE_TYPE_DIQ:
+				seq_printf(m, "  DIQ on device %x\n",
+					   pqn->kq->dev->id);
+				mqd_type = KFD_MQD_TYPE_HIQ;
+				break;
+			default:
+				seq_printf(m,
+				"  Bad kernel queue type %d on device %x\n",
+					   q->properties.type,
+					   pqn->kq->dev->id);
+				continue;
+			}
+		} else {
+			seq_printf(m,
+		"  Weird: Queue node with neither kernel nor user queue\n");
+			continue;
+		}
+
+		r = mqd_manager->debugfs_show_mqd(m, q->mqd);
+		if (r != 0)
+			break;
+	}
+
+	return r;
+}
+
+#endif

drivers/gpu/drm/amd/amdkfd/kfd_topology.h

@@ -39,8 +39,13 @@
 #define HSA_CAP_WATCH_POINTS_SUPPORTED		0x00000080
 #define HSA_CAP_WATCH_POINTS_TOTALBITS_MASK	0x00000f00
 #define HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT	8
-#define HSA_CAP_RESERVED			0xfffff000
+#define HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK	0x00003000
-#define HSA_CAP_DOORBELL_PACKET_TYPE		0x00001000
+#define HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT	12
+#define HSA_CAP_RESERVED			0xffffc000
+
+#define HSA_CAP_DOORBELL_TYPE_PRE_1_0		0x0
+#define HSA_CAP_DOORBELL_TYPE_1_0		0x1
+#define HSA_CAP_AQL_QUEUE_DOUBLE_MAP		0x00004000
 
 struct kfd_node_properties {
 	uint32_t cpu_cores_count;
@@ -91,8 +96,6 @@ struct kfd_mem_properties {
 	struct attribute	attr;
 };
 
-#define KFD_TOPOLOGY_CPU_SIBLINGS 256
-
 #define HSA_CACHE_TYPE_DATA		0x00000001
 #define HSA_CACHE_TYPE_INSTRUCTION	0x00000002
 #define HSA_CACHE_TYPE_CPU		0x00000004
@@ -109,7 +112,7 @@ struct kfd_cache_properties {
 	uint32_t		cache_assoc;
 	uint32_t		cache_latency;
 	uint32_t		cache_type;
-	uint8_t			sibling_map[KFD_TOPOLOGY_CPU_SIBLINGS];
+	uint8_t			sibling_map[CRAT_SIBLINGMAP_SIZE];
 	struct kobject		*kobj;
 	struct attribute	attr;
 };
@@ -132,24 +135,36 @@ struct kfd_iolink_properties {
 	struct attribute	attr;
 };
 
+struct kfd_perf_properties {
+	struct list_head	list;
+	char			block_name[16];
+	uint32_t		max_concurrent;
+	struct attribute_group	*attr_group;
+};
+
 struct kfd_topology_device {
 	struct list_head		list;
 	uint32_t			gpu_id;
+	uint32_t			proximity_domain;
 	struct kfd_node_properties	node_props;
-	uint32_t			mem_bank_count;
 	struct list_head		mem_props;
 	uint32_t			cache_count;
 	struct list_head		cache_props;
 	uint32_t			io_link_count;
 	struct list_head		io_link_props;
+	struct list_head		perf_props;
 	struct kfd_dev			*gpu;
 	struct kobject			*kobj_node;
 	struct kobject			*kobj_mem;
 	struct kobject			*kobj_cache;
 	struct kobject			*kobj_iolink;
+	struct kobject			*kobj_perf;
 	struct attribute		attr_gpuid;
 	struct attribute		attr_name;
 	struct attribute		attr_props;
+	uint8_t				oem_id[CRAT_OEMID_LENGTH];
+	uint8_t				oem_table_id[CRAT_OEMTABLEID_LENGTH];
+	uint32_t			oem_revision;
 };
 
 struct kfd_system_properties {
@@ -164,6 +179,12 @@ struct kfd_system_properties {
 	struct attribute	attr_props;
 };
 
+struct kfd_topology_device *kfd_create_topology_device(
+		struct list_head *device_list);
+void kfd_release_topology_device_list(struct list_head *device_list);
 
+extern bool amd_iommu_pc_supported(void);
+extern u8 amd_iommu_pc_get_max_banks(u16 devid);
+extern u8 amd_iommu_pc_get_max_counters(u16 devid);
 
 #endif /* __KFD_TOPOLOGY_H__ */

drivers/gpu/drm/amd/include/kgd_kfd_interface.h

@@ -46,6 +46,28 @@ enum kfd_preempt_type {
 	KFD_PREEMPT_TYPE_WAVEFRONT_RESET,
 };
 
+struct kfd_cu_info {
+	uint32_t num_shader_engines;
+	uint32_t num_shader_arrays_per_engine;
+	uint32_t num_cu_per_sh;
+	uint32_t cu_active_number;
+	uint32_t cu_ao_mask;
+	uint32_t simd_per_cu;
+	uint32_t max_waves_per_simd;
+	uint32_t wave_front_size;
+	uint32_t max_scratch_slots_per_cu;
+	uint32_t lds_size;
+	uint32_t cu_bitmap[4][4];
+};
+
+/* For getting GPU local memory information from KGD */
+struct kfd_local_mem_info {
+	uint64_t local_mem_size_private;
+	uint64_t local_mem_size_public;
+	uint32_t vram_width;
+	uint32_t mem_clk_max;
+};
+
 enum kgd_memory_pool {
 	KGD_POOL_SYSTEM_CACHEABLE = 1,
 	KGD_POOL_SYSTEM_WRITECOMBINE = 2,
@@ -106,7 +128,7 @@ struct tile_config {
  *
  * @free_gtt_mem: Frees a buffer that was allocated on the gart aperture
  *
- * @get_vmem_size: Retrieves (physical) size of VRAM
+ * @get_local_mem_info: Retrieves information about GPU local memory
  *
  * @get_gpu_clock_counter: Retrieves GPU clock counter
  *
@@ -131,6 +153,12 @@ struct tile_config {
  * @hqd_sdma_load: Loads the SDMA mqd structure to a H/W SDMA hqd slot.
  * used only for no HWS mode.
  *
+ * @hqd_dump: Dumps CPC HQD registers to an array of address-value pairs.
+ * Array is allocated with kmalloc, needs to be freed with kfree by caller.
+ *
+ * @hqd_sdma_dump: Dumps SDMA HQD registers to an array of address-value pairs.
+ * Array is allocated with kmalloc, needs to be freed with kfree by caller.
+ *
  * @hqd_is_occupies: Checks if a hqd slot is occupied.
  *
  * @hqd_destroy: Destructs and preempts the queue assigned to that hqd slot.
@@ -147,6 +175,10 @@ struct tile_config {
  *
  * @get_tile_config: Returns GPU-specific tiling mode information
  *
+ * @get_cu_info: Retrieves activated cu info
+ *
+ * @get_vram_usage: Returns current VRAM usage
+ *
  * This structure contains function pointers to services that the kgd driver
  * provides to amdkfd driver.
  *
@@ -158,7 +190,8 @@ struct kfd2kgd_calls {
 
 	void (*free_gtt_mem)(struct kgd_dev *kgd, void *mem_obj);
 
-	uint64_t (*get_vmem_size)(struct kgd_dev *kgd);
+	void (*get_local_mem_info)(struct kgd_dev *kgd,
+			struct kfd_local_mem_info *mem_info);
 	uint64_t (*get_gpu_clock_counter)(struct kgd_dev *kgd);
 
 	uint32_t (*get_max_engine_clock_in_mhz)(struct kgd_dev *kgd);
@@ -184,7 +217,16 @@ struct kfd2kgd_calls {
 			uint32_t wptr_shift, uint32_t wptr_mask,
 			struct mm_struct *mm);
 
-	int (*hqd_sdma_load)(struct kgd_dev *kgd, void *mqd);
+	int (*hqd_sdma_load)(struct kgd_dev *kgd, void *mqd,
+			     uint32_t __user *wptr, struct mm_struct *mm);
+
+	int (*hqd_dump)(struct kgd_dev *kgd,
+			uint32_t pipe_id, uint32_t queue_id,
+			uint32_t (**dump)[2], uint32_t *n_regs);
+
+	int (*hqd_sdma_dump)(struct kgd_dev *kgd,
+			     uint32_t engine_id, uint32_t queue_id,
+			     uint32_t (**dump)[2], uint32_t *n_regs);
 
 	bool (*hqd_is_occupied)(struct kgd_dev *kgd, uint64_t queue_address,
 				uint32_t pipe_id, uint32_t queue_id);
@@ -224,6 +266,10 @@ struct kfd2kgd_calls {
 	void (*set_scratch_backing_va)(struct kgd_dev *kgd,
 				uint64_t va, uint32_t vmid);
 	int (*get_tile_config)(struct kgd_dev *kgd, struct tile_config *config);
+
+	void (*get_cu_info)(struct kgd_dev *kgd,
+			struct kfd_cu_info *cu_info);
+	uint64_t (*get_vram_usage)(struct kgd_dev *kgd);
 };
 
 /**

drivers/gpu/drm/amd/include/vi_structs.h

@@ -153,6 +153,8 @@ struct vi_sdma_mqd {
 	uint32_t reserved_125;
 	uint32_t reserved_126;
 	uint32_t reserved_127;
+	uint32_t sdma_engine_id;
+	uint32_t sdma_queue_id;
 };
 
 struct vi_mqd {

include/uapi/linux/kfd_ioctl.h

@@ -58,7 +58,8 @@ struct kfd_ioctl_create_queue_args {
 	__u64 eop_buffer_address;	/* to KFD */
 	__u64 eop_buffer_size;	/* to KFD */
 	__u64 ctx_save_restore_address; /* to KFD */
-	__u64 ctx_save_restore_size;	/* to KFD */
+	__u32 ctx_save_restore_size;	/* to KFD */
+	__u32 ctl_stack_size;		/* to KFD */
 };
 
 struct kfd_ioctl_destroy_queue_args {
@@ -261,6 +262,13 @@ struct kfd_ioctl_get_tile_config_args {
 	 */
 };
 
+struct kfd_ioctl_set_trap_handler_args {
+	uint64_t tba_addr;		/* to KFD */
+	uint64_t tma_addr;		/* to KFD */
+	uint32_t gpu_id;		/* to KFD */
+	uint32_t pad;
+};
+
 #define AMDKFD_IOCTL_BASE 'K'
 #define AMDKFD_IO(nr)			_IO(AMDKFD_IOCTL_BASE, nr)
 #define AMDKFD_IOR(nr, type)		_IOR(AMDKFD_IOCTL_BASE, nr, type)
@@ -321,7 +329,10 @@ struct kfd_ioctl_get_tile_config_args {
 #define AMDKFD_IOC_GET_TILE_CONFIG                                      \
 		AMDKFD_IOWR(0x12, struct kfd_ioctl_get_tile_config_args)
 
+#define AMDKFD_IOC_SET_TRAP_HANDLER		\
+		AMDKFD_IOW(0x13, struct kfd_ioctl_set_trap_handler_args)
+
 #define AMDKFD_COMMAND_START		0x01
-#define AMDKFD_COMMAND_END		0x13
+#define AMDKFD_COMMAND_END		0x14
 
 #endif