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
  ...

9 files changed, 344 insertions, 40 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index bbe06e04dcb6..e40c5df55f95 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/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;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 896b16db58aa..335e454e2ee1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/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 *)kgd;
+	struct amdgpu_device *adev = (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;
+
+	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;
 
-	BUG_ON(kgd == NULL);
+	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);
 
-	return adev->mc.real_vram_size;
+	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 */
-	return adev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100;
+	/* the sclk is in quantas of 10kHz */
+	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]);
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 8d689ab7e429..2a519f9062ee 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/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)				\
 	({								\
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
index 1e3e9be7d77e..3d60e1fd9b68 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+++ b/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,
-				m->sdma_rlc_doorbell);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+	data = REG_SET_FIELD(m->sdma_rlc_doorbell, SDMA0_RLC0_DOORBELL,
+			     ENABLE, 1);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, data);
+	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);
-		timeout -= 20;
+		usleep_range(500, 1000);
 	}
 
 	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;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
index 056929b8ccd0..66b513e96437 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+++ b/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);
-		timeout -= 20;
+		usleep_range(500, 1000);
 	}
 
 	WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
-	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+	WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL,
+		RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) |
+		SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK);
+
+	m->sdmax_rlcx_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/cikd.h b/drivers/gpu/drm/amd/amdgpu/cikd.h
index 6a9e38a3d2a0..cee6e8a3ad9c 100644
--- a/drivers/gpu/drm/amd/amdgpu/cikd.h
+++ b/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 {
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index 83d94c23aa78..9c62ebd5a19c 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/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 =
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 46550b588982..e18c2e62a20b 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/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 =
diff --git a/drivers/gpu/drm/amd/amdgpu/vid.h b/drivers/gpu/drm/amd/amdgpu/vid.h
index dbf3703cbd1b..19ddd2312e00 100644
--- a/drivers/gpu/drm/amd/amdgpu/vid.h
+++ b/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)