diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 1255 |
1 files changed, 787 insertions, 468 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c index 5c51d4910650..088d09cc7a72 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c @@ -19,19 +19,8 @@ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR * OTHER DEALINGS IN THE SOFTWARE. */ - -#define pr_fmt(fmt) "kfd2kgd: " fmt - -#include <linux/module.h> -#include <linux/fdtable.h> -#include <linux/uaccess.h> -#include <linux/firmware.h> -#include <linux/mmu_context.h> -#include <drm/drmP.h> #include "amdgpu.h" #include "amdgpu_amdkfd.h" -#include "amdgpu_ucode.h" -#include "soc15_hw_ip.h" #include "gc/gc_9_0_offset.h" #include "gc/gc_9_0_sh_mask.h" #include "vega10_enum.h" @@ -47,200 +36,71 @@ #include "v9_structs.h" #include "soc15.h" #include "soc15d.h" -#include "mmhub_v1_0.h" -#include "gfxhub_v1_0.h" - - -#define V9_PIPE_PER_MEC (4) -#define V9_QUEUES_PER_PIPE_MEC (8) +#include "gfx_v9_0.h" +#include "amdgpu_amdkfd_gfx_v9.h" +#include <uapi/linux/kfd_ioctl.h> enum hqd_dequeue_request_type { NO_ACTION = 0, DRAIN_PIPE, - RESET_WAVES + RESET_WAVES, + SAVE_WAVES }; -/* - * Register access functions - */ - -static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid, - uint32_t sh_mem_config, - uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit, - uint32_t sh_mem_bases); -static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid, - unsigned int vmid); -static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id); -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_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); -static int kgd_hqd_destroy(struct kgd_dev *kgd, void *mqd, - enum kfd_preempt_type reset_type, - unsigned int utimeout, uint32_t pipe_id, - uint32_t queue_id); -static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, - unsigned int utimeout); -static int kgd_address_watch_disable(struct kgd_dev *kgd); -static int kgd_address_watch_execute(struct kgd_dev *kgd, - unsigned int watch_point_id, - uint32_t cntl_val, - uint32_t addr_hi, - uint32_t addr_lo); -static int kgd_wave_control_execute(struct kgd_dev *kgd, - uint32_t gfx_index_val, - uint32_t sq_cmd); -static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd, - unsigned int watch_point_id, - unsigned int reg_offset); - -static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, - uint8_t vmid); -static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, - uint8_t vmid); -static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid, - uint64_t page_table_base); -static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type); -static void set_scratch_backing_va(struct kgd_dev *kgd, - uint64_t va, uint32_t vmid); -static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid); -static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid); - -/* Because of REG_GET_FIELD() being used, we put this function in the - * asic specific file. - */ -static int amdgpu_amdkfd_get_tile_config(struct kgd_dev *kgd, - struct tile_config *config) +static void kgd_gfx_v9_lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe, + uint32_t queue, uint32_t vmid, uint32_t inst) { - struct amdgpu_device *adev = (struct amdgpu_device *)kgd; - - config->gb_addr_config = adev->gfx.config.gb_addr_config; - - config->tile_config_ptr = adev->gfx.config.tile_mode_array; - config->num_tile_configs = - ARRAY_SIZE(adev->gfx.config.tile_mode_array); - config->macro_tile_config_ptr = - adev->gfx.config.macrotile_mode_array; - config->num_macro_tile_configs = - ARRAY_SIZE(adev->gfx.config.macrotile_mode_array); - - return 0; -} - -static const struct kfd2kgd_calls kfd2kgd = { - .program_sh_mem_settings = kgd_program_sh_mem_settings, - .set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping, - .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, - .hqd_sdma_destroy = kgd_hqd_sdma_destroy, - .address_watch_disable = kgd_address_watch_disable, - .address_watch_execute = kgd_address_watch_execute, - .wave_control_execute = kgd_wave_control_execute, - .address_watch_get_offset = kgd_address_watch_get_offset, - .get_atc_vmid_pasid_mapping_pasid = - get_atc_vmid_pasid_mapping_pasid, - .get_atc_vmid_pasid_mapping_valid = - get_atc_vmid_pasid_mapping_valid, - .get_fw_version = get_fw_version, - .set_scratch_backing_va = set_scratch_backing_va, - .get_tile_config = amdgpu_amdkfd_get_tile_config, - .set_vm_context_page_table_base = set_vm_context_page_table_base, - .invalidate_tlbs = invalidate_tlbs, - .invalidate_tlbs_vmid = invalidate_tlbs_vmid, - .get_hive_id = amdgpu_amdkfd_get_hive_id, -}; - -struct kfd2kgd_calls *amdgpu_amdkfd_gfx_9_0_get_functions(void) -{ - return (struct kfd2kgd_calls *)&kfd2kgd; -} - -static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd) -{ - return (struct amdgpu_device *)kgd; -} - -static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe, - uint32_t queue, uint32_t vmid) -{ - struct amdgpu_device *adev = get_amdgpu_device(kgd); - mutex_lock(&adev->srbm_mutex); - soc15_grbm_select(adev, mec, pipe, queue, vmid); + soc15_grbm_select(adev, mec, pipe, queue, vmid, GET_INST(GC, inst)); } -static void unlock_srbm(struct kgd_dev *kgd) +static void kgd_gfx_v9_unlock_srbm(struct amdgpu_device *adev, uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); - - soc15_grbm_select(adev, 0, 0, 0, 0); + soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, inst)); mutex_unlock(&adev->srbm_mutex); } -static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id, - uint32_t queue_id) +void kgd_gfx_v9_acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id, + uint32_t queue_id, uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); - uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1; uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec); - lock_srbm(kgd, mec, pipe, queue_id, 0); + kgd_gfx_v9_lock_srbm(adev, mec, pipe, queue_id, 0, inst); } -static uint32_t get_queue_mask(struct amdgpu_device *adev, +uint64_t kgd_gfx_v9_get_queue_mask(struct amdgpu_device *adev, uint32_t pipe_id, uint32_t queue_id) { - unsigned int bit = (pipe_id * adev->gfx.mec.num_queue_per_pipe + - queue_id) & 31; + unsigned int bit = pipe_id * adev->gfx.mec.num_queue_per_pipe + + queue_id; - return ((uint32_t)1) << bit; + return 1ull << bit; } -static void release_queue(struct kgd_dev *kgd) +void kgd_gfx_v9_release_queue(struct amdgpu_device *adev, uint32_t inst) { - unlock_srbm(kgd); + kgd_gfx_v9_unlock_srbm(adev, inst); } -static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid, +void kgd_gfx_v9_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid, uint32_t sh_mem_config, uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit, - uint32_t sh_mem_bases) + uint32_t sh_mem_bases, uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); + kgd_gfx_v9_lock_srbm(adev, 0, 0, 0, vmid, inst); - lock_srbm(kgd, 0, 0, 0, vmid); - - WREG32(SOC15_REG_OFFSET(GC, 0, mmSH_MEM_CONFIG), sh_mem_config); - WREG32(SOC15_REG_OFFSET(GC, 0, mmSH_MEM_BASES), sh_mem_bases); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmSH_MEM_CONFIG, sh_mem_config); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmSH_MEM_BASES, sh_mem_bases); /* APE1 no longer exists on GFX9 */ - unlock_srbm(kgd); + kgd_gfx_v9_unlock_srbm(adev, inst); } -static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid, - unsigned int vmid) +int kgd_gfx_v9_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid, + unsigned int vmid, uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); - /* * We have to assume that there is no outstanding mapping. * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because @@ -297,44 +157,56 @@ static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid, * but still works */ -static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) +int kgd_gfx_v9_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id, + uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); uint32_t mec; uint32_t pipe; mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1; pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec); - lock_srbm(kgd, mec, pipe, 0, 0); + kgd_gfx_v9_lock_srbm(adev, mec, pipe, 0, 0, inst); - WREG32(SOC15_REG_OFFSET(GC, 0, mmCPC_INT_CNTL), + WREG32_SOC15(GC, GET_INST(GC, inst), mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK | CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK); - unlock_srbm(kgd); + kgd_gfx_v9_unlock_srbm(adev, inst); return 0; } -static uint32_t get_sdma_base_addr(struct amdgpu_device *adev, +static uint32_t get_sdma_rlc_reg_offset(struct amdgpu_device *adev, unsigned int engine_id, unsigned int queue_id) { - uint32_t base[2] = { - SOC15_REG_OFFSET(SDMA0, 0, - mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL, - SOC15_REG_OFFSET(SDMA1, 0, - mmSDMA1_RLC0_RB_CNTL) - mmSDMA1_RLC0_RB_CNTL - }; - uint32_t retval; + uint32_t sdma_engine_reg_base = 0; + uint32_t sdma_rlc_reg_offset; + + switch (engine_id) { + default: + dev_warn(adev->dev, + "Invalid sdma engine id (%d), using engine id 0\n", + engine_id); + fallthrough; + case 0: + sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA0, 0, + mmSDMA0_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL; + break; + case 1: + sdma_engine_reg_base = SOC15_REG_OFFSET(SDMA1, 0, + mmSDMA1_RLC0_RB_CNTL) - mmSDMA0_RLC0_RB_CNTL; + break; + } - retval = base[engine_id] + queue_id * (mmSDMA0_RLC1_RB_CNTL - - mmSDMA0_RLC0_RB_CNTL); + sdma_rlc_reg_offset = sdma_engine_reg_base + + queue_id * (mmSDMA0_RLC1_RB_CNTL - mmSDMA0_RLC0_RB_CNTL); - pr_debug("sdma base address: 0x%x\n", retval); + pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", engine_id, + queue_id, sdma_rlc_reg_offset); - return retval; + return sdma_rlc_reg_offset; } static inline struct v9_mqd *get_mqd(void *mqd) @@ -347,48 +219,33 @@ static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd) return (struct v9_sdma_mqd *)mqd; } -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) +int kgd_gfx_v9_hqd_load(struct amdgpu_device *adev, 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, + uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); struct v9_mqd *m; uint32_t *mqd_hqd; uint32_t reg, hqd_base, data; m = get_mqd(mqd); - acquire_queue(kgd, pipe_id, queue_id); - - /* HIQ is set during driver init period with vmid set to 0*/ - if (m->cp_hqd_vmid == 0) { - uint32_t value, mec, pipe; - - mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1; - pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec); - - pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n", - mec, pipe, queue_id); - value = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_CP_SCHEDULERS)); - value = REG_SET_FIELD(value, RLC_CP_SCHEDULERS, scheduler1, - ((mec << 5) | (pipe << 3) | queue_id | 0x80)); - WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_CP_SCHEDULERS), value); - } + kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst); /* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */ mqd_hqd = &m->cp_mqd_base_addr_lo; - hqd_base = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR); + hqd_base = SOC15_REG_OFFSET(GC, GET_INST(GC, inst), mmCP_MQD_BASE_ADDR); for (reg = hqd_base; - reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++) - WREG32(reg, mqd_hqd[reg - hqd_base]); + reg <= SOC15_REG_OFFSET(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_HI); reg++) + WREG32_XCC(reg, mqd_hqd[reg - hqd_base], inst); /* Activate doorbell logic before triggering WPTR poll. */ data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control, CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1); - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL), data); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_DOORBELL_CONTROL, data); if (wptr) { /* Don't read wptr with get_user because the user @@ -417,36 +274,86 @@ static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id, guessed_wptr += m->cp_hqd_pq_wptr_lo & ~(queue_size - 1); guessed_wptr += (uint64_t)m->cp_hqd_pq_wptr_hi << 32; - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_LO), - lower_32_bits(guessed_wptr)); - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI), - upper_32_bits(guessed_wptr)); - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR), - lower_32_bits((uintptr_t)wptr)); - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI), - upper_32_bits((uintptr_t)wptr)); - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_PQ_WPTR_POLL_CNTL1), - get_queue_mask(adev, pipe_id, queue_id)); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_LO, + lower_32_bits(guessed_wptr)); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_HI, + upper_32_bits(guessed_wptr)); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_POLL_ADDR, + lower_32_bits((uintptr_t)wptr)); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_POLL_ADDR_HI, + upper_32_bits((uintptr_t)wptr)); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_PQ_WPTR_POLL_CNTL1, + (uint32_t)kgd_gfx_v9_get_queue_mask(adev, pipe_id, queue_id)); } /* Start the EOP fetcher */ - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_EOP_RPTR), - REG_SET_FIELD(m->cp_hqd_eop_rptr, - CP_HQD_EOP_RPTR, INIT_FETCHER, 1)); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_EOP_RPTR, + REG_SET_FIELD(m->cp_hqd_eop_rptr, CP_HQD_EOP_RPTR, INIT_FETCHER, 1)); data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1); - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_ACTIVE), data); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE, data); - release_queue(kgd); + kgd_gfx_v9_release_queue(adev, inst); return 0; } -static int kgd_hqd_dump(struct kgd_dev *kgd, +int kgd_gfx_v9_hiq_mqd_load(struct amdgpu_device *adev, void *mqd, + uint32_t pipe_id, uint32_t queue_id, + uint32_t doorbell_off, uint32_t inst) +{ + struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[inst].ring; + struct v9_mqd *m; + uint32_t mec, pipe; + int r; + + m = get_mqd(mqd); + + kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst); + + mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1; + pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec); + + pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n", + mec, pipe, queue_id); + + spin_lock(&adev->gfx.kiq[inst].ring_lock); + r = amdgpu_ring_alloc(kiq_ring, 7); + if (r) { + pr_err("Failed to alloc KIQ (%d).\n", r); + goto out_unlock; + } + + amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5)); + amdgpu_ring_write(kiq_ring, + PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */ + PACKET3_MAP_QUEUES_VMID(m->cp_hqd_vmid) | /* VMID */ + PACKET3_MAP_QUEUES_QUEUE(queue_id) | + PACKET3_MAP_QUEUES_PIPE(pipe) | + PACKET3_MAP_QUEUES_ME((mec - 1)) | + PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */ + PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */ + PACKET3_MAP_QUEUES_ENGINE_SEL(1) | /* engine_sel: hiq */ + PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */ + amdgpu_ring_write(kiq_ring, + PACKET3_MAP_QUEUES_DOORBELL_OFFSET(doorbell_off)); + amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_lo); + amdgpu_ring_write(kiq_ring, m->cp_mqd_base_addr_hi); + amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_lo); + amdgpu_ring_write(kiq_ring, m->cp_hqd_pq_wptr_poll_addr_hi); + amdgpu_ring_commit(kiq_ring); + +out_unlock: + spin_unlock(&adev->gfx.kiq[inst].ring_lock); + kgd_gfx_v9_release_queue(adev, inst); + + return r; +} + +int kgd_gfx_v9_hqd_dump(struct amdgpu_device *adev, uint32_t pipe_id, uint32_t queue_id, - uint32_t (**dump)[2], uint32_t *n_regs) + uint32_t (**dump)[2], uint32_t *n_regs, uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); uint32_t i = 0, reg; #define HQD_N_REGS 56 #define DUMP_REG(addr) do { \ @@ -456,17 +363,17 @@ static int kgd_hqd_dump(struct kgd_dev *kgd, (*dump)[i++][1] = RREG32(addr); \ } while (0) - *dump = kmalloc_array(HQD_N_REGS * 2, sizeof(uint32_t), GFP_KERNEL); + *dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL); if (*dump == NULL) return -ENOMEM; - acquire_queue(kgd, pipe_id, queue_id); + kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst); - for (reg = SOC15_REG_OFFSET(GC, 0, mmCP_MQD_BASE_ADDR); - reg <= SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_WPTR_HI); reg++) + for (reg = SOC15_REG_OFFSET(GC, GET_INST(GC, inst), mmCP_MQD_BASE_ADDR); + reg <= SOC15_REG_OFFSET(GC, GET_INST(GC, inst), mmCP_HQD_PQ_WPTR_HI); reg++) DUMP_REG(reg); - release_queue(kgd); + kgd_gfx_v9_release_queue(adev, inst); WARN_ON_ONCE(i != HQD_N_REGS); *n_regs = i; @@ -474,104 +381,99 @@ static int kgd_hqd_dump(struct kgd_dev *kgd, return 0; } -static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, +static int kgd_hqd_sdma_load(struct amdgpu_device *adev, void *mqd, uint32_t __user *wptr, struct mm_struct *mm) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); struct v9_sdma_mqd *m; - uint32_t sdma_base_addr, sdmax_gfx_context_cntl; + uint32_t sdma_rlc_reg_offset; unsigned long end_jiffies; uint32_t data; uint64_t data64; uint64_t __user *wptr64 = (uint64_t __user *)wptr; m = get_sdma_mqd(mqd); - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, + sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id, m->sdma_queue_id); - sdmax_gfx_context_cntl = m->sdma_engine_id ? - SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_GFX_CONTEXT_CNTL) : - SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_GFX_CONTEXT_CNTL); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, + WREG32(sdma_rlc_reg_offset + 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); + data = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS); if (data & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK) break; - if (time_after(jiffies, end_jiffies)) + if (time_after(jiffies, end_jiffies)) { + pr_err("SDMA RLC not idle in %s\n", __func__); return -ETIME; + } usleep_range(500, 1000); } - data = RREG32(sdmax_gfx_context_cntl); - data = REG_SET_FIELD(data, SDMA0_GFX_CONTEXT_CNTL, - RESUME_CTX, 0); - WREG32(sdmax_gfx_context_cntl, data); - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL_OFFSET, + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL_OFFSET, m->sdmax_rlcx_doorbell_offset); 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); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI, + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR, + m->sdmax_rlcx_rb_rptr); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI, m->sdmax_rlcx_rb_rptr_hi); - WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 1); if (read_user_wptr(mm, wptr64, data64)) { - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, lower_32_bits(data64)); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI, + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI, upper_32_bits(data64)); } else { - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, m->sdmax_rlcx_rb_rptr); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR_HI, + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR_HI, m->sdmax_rlcx_rb_rptr_hi); } - WREG32(sdma_base_addr + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_MINOR_PTR_UPDATE, 0); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE_HI, + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE, m->sdmax_rlcx_rb_base); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_BASE_HI, m->sdmax_rlcx_rb_base_hi); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_ADDR_LO, m->sdmax_rlcx_rb_rptr_addr_lo); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_ADDR_HI, + WREG32(sdma_rlc_reg_offset + 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); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, data); return 0; } -static int kgd_hqd_sdma_dump(struct kgd_dev *kgd, +static int kgd_hqd_sdma_dump(struct amdgpu_device *adev, 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_base_addr = get_sdma_base_addr(adev, engine_id, queue_id); + uint32_t sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, + engine_id, queue_id); uint32_t i = 0, reg; #undef HQD_N_REGS #define HQD_N_REGS (19+6+7+10) - *dump = kmalloc_array(HQD_N_REGS * 2, sizeof(uint32_t), GFP_KERNEL); + *dump = kmalloc_array(HQD_N_REGS, sizeof(**dump), GFP_KERNEL); if (*dump == NULL) return -ENOMEM; for (reg = mmSDMA0_RLC0_RB_CNTL; reg <= mmSDMA0_RLC0_DOORBELL; reg++) - DUMP_REG(sdma_base_addr + reg); + DUMP_REG(sdma_rlc_reg_offset + reg); for (reg = mmSDMA0_RLC0_STATUS; reg <= mmSDMA0_RLC0_CSA_ADDR_HI; reg++) - DUMP_REG(sdma_base_addr + reg); + DUMP_REG(sdma_rlc_reg_offset + reg); for (reg = mmSDMA0_RLC0_IB_SUB_REMAIN; reg <= mmSDMA0_RLC0_MINOR_PTR_UPDATE; reg++) - DUMP_REG(sdma_base_addr + reg); + DUMP_REG(sdma_rlc_reg_offset + reg); for (reg = mmSDMA0_RLC0_MIDCMD_DATA0; reg <= mmSDMA0_RLC0_MIDCMD_CNTL; reg++) - DUMP_REG(sdma_base_addr + reg); + DUMP_REG(sdma_rlc_reg_offset + reg); WARN_ON_ONCE(i != HQD_N_REGS); *n_regs = i; @@ -579,40 +481,39 @@ static int kgd_hqd_sdma_dump(struct kgd_dev *kgd, return 0; } -static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address, - uint32_t pipe_id, uint32_t queue_id) +bool kgd_gfx_v9_hqd_is_occupied(struct amdgpu_device *adev, + uint64_t queue_address, uint32_t pipe_id, + uint32_t queue_id, uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); uint32_t act; bool retval = false; uint32_t low, high; - acquire_queue(kgd, pipe_id, queue_id); - act = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_ACTIVE)); + kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst); + act = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE); if (act) { low = lower_32_bits(queue_address >> 8); high = upper_32_bits(queue_address >> 8); - if (low == RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_BASE)) && - high == RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_PQ_BASE_HI))) + if (low == RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE) && + high == RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE_HI)) retval = true; } - release_queue(kgd); + kgd_gfx_v9_release_queue(adev, inst); return retval; } -static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) +static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); struct v9_sdma_mqd *m; - uint32_t sdma_base_addr; + uint32_t sdma_rlc_reg_offset; uint32_t sdma_rlc_rb_cntl; m = get_sdma_mqd(mqd); - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, + sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id, m->sdma_queue_id); - sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); + sdma_rlc_rb_cntl = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL); if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK) return true; @@ -620,24 +521,23 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) return false; } -static int kgd_hqd_destroy(struct kgd_dev *kgd, void *mqd, +int kgd_gfx_v9_hqd_destroy(struct amdgpu_device *adev, void *mqd, enum kfd_preempt_type reset_type, unsigned int utimeout, uint32_t pipe_id, - uint32_t queue_id) + uint32_t queue_id, uint32_t inst) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); enum hqd_dequeue_request_type type; unsigned long end_jiffies; uint32_t temp; struct v9_mqd *m = get_mqd(mqd); - if (adev->in_gpu_reset) + if (amdgpu_in_reset(adev)) return -EIO; - acquire_queue(kgd, pipe_id, queue_id); + kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst); if (m->cp_hqd_vmid == 0) - WREG32_FIELD15(GC, 0, RLC_CP_SCHEDULERS, scheduler1, 0); + WREG32_FIELD15_RLC(GC, GET_INST(GC, inst), RLC_CP_SCHEDULERS, scheduler1, 0); switch (reset_type) { case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN: @@ -646,300 +546,719 @@ static int kgd_hqd_destroy(struct kgd_dev *kgd, void *mqd, case KFD_PREEMPT_TYPE_WAVEFRONT_RESET: type = RESET_WAVES; break; + case KFD_PREEMPT_TYPE_WAVEFRONT_SAVE: + type = SAVE_WAVES; + break; default: type = DRAIN_PIPE; break; } - WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_DEQUEUE_REQUEST), type); + WREG32_SOC15_RLC(GC, GET_INST(GC, inst), mmCP_HQD_DEQUEUE_REQUEST, type); end_jiffies = (utimeout * HZ / 1000) + jiffies; while (true) { - temp = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_HQD_ACTIVE)); + temp = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE); if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK)) break; if (time_after(jiffies, end_jiffies)) { pr_err("cp queue preemption time out.\n"); - release_queue(kgd); + kgd_gfx_v9_release_queue(adev, inst); return -ETIME; } usleep_range(500, 1000); } - release_queue(kgd); + kgd_gfx_v9_release_queue(adev, inst); return 0; } -static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, +static int kgd_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd, unsigned int utimeout) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); struct v9_sdma_mqd *m; - uint32_t sdma_base_addr; + uint32_t sdma_rlc_reg_offset; uint32_t temp; unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies; m = get_sdma_mqd(mqd); - sdma_base_addr = get_sdma_base_addr(adev, m->sdma_engine_id, + sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(adev, m->sdma_engine_id, m->sdma_queue_id); - temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL); + temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL); temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK; - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, temp); while (true) { - temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS); + temp = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_CONTEXT_STATUS); if (temp & SDMA0_RLC0_CONTEXT_STATUS__IDLE_MASK) break; - if (time_after(jiffies, end_jiffies)) + if (time_after(jiffies, end_jiffies)) { + pr_err("SDMA RLC not idle in %s\n", __func__); return -ETIME; + } usleep_range(500, 1000); } - WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, - RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL) | + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, 0); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL, + RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_CNTL) | SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK); - m->sdmax_rlcx_rb_rptr = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR); + m->sdmax_rlcx_rb_rptr = RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR); m->sdmax_rlcx_rb_rptr_hi = - RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR_HI); + RREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_RPTR_HI); return 0; } -static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, - uint8_t vmid) +bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev, + uint8_t vmid, uint16_t *p_pasid) { - uint32_t reg; - struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + uint32_t value; - reg = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING) + value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING) + vmid); - return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK; + *p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK; + + return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK); } -static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, - uint8_t vmid) +int kgd_gfx_v9_wave_control_execute(struct amdgpu_device *adev, + uint32_t gfx_index_val, + uint32_t sq_cmd, uint32_t inst) { - uint32_t reg; - struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + uint32_t data = 0; - reg = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING) - + vmid); - return reg & ATC_VMID0_PASID_MAPPING__PASID_MASK; -} - -static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid) -{ - struct amdgpu_device *adev = (struct amdgpu_device *) kgd; - - /* Use legacy mode tlb invalidation. - * - * Currently on Raven the code below is broken for anything but - * legacy mode due to a MMHUB power gating problem. A workaround - * is for MMHUB to wait until the condition PER_VMID_INVALIDATE_REQ - * == PER_VMID_INVALIDATE_ACK instead of simply waiting for the ack - * bit. - * - * TODO 1: agree on the right set of invalidation registers for - * KFD use. Use the last one for now. Invalidate both GC and - * MMHUB. - * - * TODO 2: support range-based invalidation, requires kfg2kgd - * interface change - */ - amdgpu_gmc_flush_gpu_tlb(adev, vmid, 0); -} - -static int invalidate_tlbs_with_kiq(struct amdgpu_device *adev, uint16_t pasid) -{ - signed long r; - uint32_t seq; - struct amdgpu_ring *ring = &adev->gfx.kiq.ring; - - spin_lock(&adev->gfx.kiq.ring_lock); - amdgpu_ring_alloc(ring, 12); /* fence + invalidate_tlbs package*/ - amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0)); - amdgpu_ring_write(ring, - PACKET3_INVALIDATE_TLBS_DST_SEL(1) | - PACKET3_INVALIDATE_TLBS_ALL_HUB(1) | - PACKET3_INVALIDATE_TLBS_PASID(pasid) | - PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(0)); /* legacy */ - amdgpu_fence_emit_polling(ring, &seq); - amdgpu_ring_commit(ring); - spin_unlock(&adev->gfx.kiq.ring_lock); - - r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout); - if (r < 1) { - DRM_ERROR("wait for kiq fence error: %ld.\n", r); - return -ETIME; - } + mutex_lock(&adev->grbm_idx_mutex); + + WREG32_SOC15_RLC_SHADOW(GC, GET_INST(GC, inst), mmGRBM_GFX_INDEX, gfx_index_val); + WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_CMD, sq_cmd); + + data = REG_SET_FIELD(data, GRBM_GFX_INDEX, + INSTANCE_BROADCAST_WRITES, 1); + data = REG_SET_FIELD(data, GRBM_GFX_INDEX, + SH_BROADCAST_WRITES, 1); + data = REG_SET_FIELD(data, GRBM_GFX_INDEX, + SE_BROADCAST_WRITES, 1); + + WREG32_SOC15_RLC_SHADOW(GC, GET_INST(GC, inst), mmGRBM_GFX_INDEX, data); + mutex_unlock(&adev->grbm_idx_mutex); return 0; } -static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid) +/* + * GFX9 helper for wave launch stall requirements on debug trap setting. + * + * vmid: + * Target VMID to stall/unstall. + * + * stall: + * 0-unstall wave launch (enable), 1-stall wave launch (disable). + * After wavefront launch has been stalled, allocated waves must drain from + * SPI in order for debug trap settings to take effect on those waves. + * This is roughly a ~96 clock cycle wait on SPI where a read on + * SPI_GDBG_WAVE_CNTL translates to ~32 clock cycles. + * KGD_GFX_V9_WAVE_LAUNCH_SPI_DRAIN_LATENCY indicates the number of reads required. + * + * NOTE: We can afford to clear the entire STALL_VMID field on unstall + * because GFX9.4.1 cannot support multi-process debugging due to trap + * configuration and masking being limited to global scope. Always assume + * single process conditions. + */ +#define KGD_GFX_V9_WAVE_LAUNCH_SPI_DRAIN_LATENCY 3 +void kgd_gfx_v9_set_wave_launch_stall(struct amdgpu_device *adev, + uint32_t vmid, + bool stall) { - struct amdgpu_device *adev = (struct amdgpu_device *) kgd; - int vmid; - struct amdgpu_ring *ring = &adev->gfx.kiq.ring; + int i; + uint32_t data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL)); - if (adev->in_gpu_reset) - return -EIO; + if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 1)) + data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_VMID, + stall ? 1 << vmid : 0); + else + data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL, STALL_RA, + stall ? 1 : 0); - if (ring->sched.ready) - return invalidate_tlbs_with_kiq(adev, pasid); - - for (vmid = 0; vmid < 16; vmid++) { - if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) - continue; - if (get_atc_vmid_pasid_mapping_valid(kgd, vmid)) { - if (get_atc_vmid_pasid_mapping_pasid(kgd, vmid) - == pasid) { - write_vmid_invalidate_request(kgd, vmid); - break; - } - } - } + WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), data); + + if (!stall) + return; + + for (i = 0; i < KGD_GFX_V9_WAVE_LAUNCH_SPI_DRAIN_LATENCY; i++) + RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL)); +} + +/* + * restore_dbg_registers is ignored here but is a general interface requirement + * for devices that support GFXOFF and where the RLC save/restore list + * does not support hw registers for debugging i.e. the driver has to manually + * initialize the debug mode registers after it has disabled GFX off during the + * debug session. + */ +uint32_t kgd_gfx_v9_enable_debug_trap(struct amdgpu_device *adev, + bool restore_dbg_registers, + uint32_t vmid) +{ + mutex_lock(&adev->grbm_idx_mutex); + + kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true); + + WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0); + + kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false); + + mutex_unlock(&adev->grbm_idx_mutex); return 0; } -static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid) +/* + * keep_trap_enabled is ignored here but is a general interface requirement + * for devices that support multi-process debugging where the performance + * overhead from trap temporary setup needs to be bypassed when the debug + * session has ended. + */ +uint32_t kgd_gfx_v9_disable_debug_trap(struct amdgpu_device *adev, + bool keep_trap_enabled, + uint32_t vmid) { - struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + mutex_lock(&adev->grbm_idx_mutex); - if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) { - pr_err("non kfd vmid %d\n", vmid); - return 0; - } + kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true); + + WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0); + + kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false); + + mutex_unlock(&adev->grbm_idx_mutex); - write_vmid_invalidate_request(kgd, vmid); return 0; } -static int kgd_address_watch_disable(struct kgd_dev *kgd) +int kgd_gfx_v9_validate_trap_override_request(struct amdgpu_device *adev, + uint32_t trap_override, + uint32_t *trap_mask_supported) { + *trap_mask_supported &= KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH; + + /* The SPI_GDBG_TRAP_MASK register is global and affects all + * processes. Only allow OR-ing the address-watch bit, since + * this only affects processes under the debugger. Other bits + * should stay 0 to avoid the debugger interfering with other + * processes. + */ + if (trap_override != KFD_DBG_TRAP_OVERRIDE_OR) + return -EINVAL; + return 0; } -static int kgd_address_watch_execute(struct kgd_dev *kgd, - unsigned int watch_point_id, - uint32_t cntl_val, - uint32_t addr_hi, - uint32_t addr_lo) +uint32_t kgd_gfx_v9_set_wave_launch_trap_override(struct amdgpu_device *adev, + uint32_t vmid, + uint32_t trap_override, + uint32_t trap_mask_bits, + uint32_t trap_mask_request, + uint32_t *trap_mask_prev, + uint32_t kfd_dbg_cntl_prev) { + uint32_t data, wave_cntl_prev; + + mutex_lock(&adev->grbm_idx_mutex); + + wave_cntl_prev = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL)); + + kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true); + + data = RREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK)); + *trap_mask_prev = REG_GET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN); + + trap_mask_bits = (trap_mask_bits & trap_mask_request) | + (*trap_mask_prev & ~trap_mask_request); + + data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, EXCP_EN, trap_mask_bits); + data = REG_SET_FIELD(data, SPI_GDBG_TRAP_MASK, REPLACE, trap_override); + WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), data); + + /* We need to preserve wave launch mode stall settings. */ + WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL), wave_cntl_prev); + + mutex_unlock(&adev->grbm_idx_mutex); + return 0; } -static int kgd_wave_control_execute(struct kgd_dev *kgd, - uint32_t gfx_index_val, - uint32_t sq_cmd) +uint32_t kgd_gfx_v9_set_wave_launch_mode(struct amdgpu_device *adev, + uint8_t wave_launch_mode, + uint32_t vmid) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); uint32_t data = 0; + bool is_mode_set = !!wave_launch_mode; mutex_lock(&adev->grbm_idx_mutex); - WREG32(SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX), gfx_index_val); - WREG32(SOC15_REG_OFFSET(GC, 0, mmSQ_CMD), sq_cmd); + kgd_gfx_v9_set_wave_launch_stall(adev, vmid, true); - data = REG_SET_FIELD(data, GRBM_GFX_INDEX, - INSTANCE_BROADCAST_WRITES, 1); - data = REG_SET_FIELD(data, GRBM_GFX_INDEX, - SH_BROADCAST_WRITES, 1); - data = REG_SET_FIELD(data, GRBM_GFX_INDEX, - SE_BROADCAST_WRITES, 1); + data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2, + VMID_MASK, is_mode_set ? 1 << vmid : 0); + data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2, + MODE, is_mode_set ? wave_launch_mode : 0); + WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data); + + kgd_gfx_v9_set_wave_launch_stall(adev, vmid, false); - WREG32(SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX), data); mutex_unlock(&adev->grbm_idx_mutex); return 0; } -static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd, - unsigned int watch_point_id, - unsigned int reg_offset) +#define TCP_WATCH_STRIDE (mmTCP_WATCH1_ADDR_H - mmTCP_WATCH0_ADDR_H) +uint32_t kgd_gfx_v9_set_address_watch(struct amdgpu_device *adev, + uint64_t watch_address, + uint32_t watch_address_mask, + uint32_t watch_id, + uint32_t watch_mode, + uint32_t debug_vmid, + uint32_t inst) { + uint32_t watch_address_high; + uint32_t watch_address_low; + uint32_t watch_address_cntl; + + watch_address_cntl = 0; + + watch_address_low = lower_32_bits(watch_address); + watch_address_high = upper_32_bits(watch_address) & 0xffff; + + watch_address_cntl = REG_SET_FIELD(watch_address_cntl, + TCP_WATCH0_CNTL, + VMID, + debug_vmid); + watch_address_cntl = REG_SET_FIELD(watch_address_cntl, + TCP_WATCH0_CNTL, + MODE, + watch_mode); + watch_address_cntl = REG_SET_FIELD(watch_address_cntl, + TCP_WATCH0_CNTL, + MASK, + watch_address_mask >> 6); + + /* Turning off this watch point until we set all the registers */ + watch_address_cntl = REG_SET_FIELD(watch_address_cntl, + TCP_WATCH0_CNTL, + VALID, + 0); + + WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) + + (watch_id * TCP_WATCH_STRIDE)), + watch_address_cntl); + + WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_H) + + (watch_id * TCP_WATCH_STRIDE)), + watch_address_high); + + WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_ADDR_L) + + (watch_id * TCP_WATCH_STRIDE)), + watch_address_low); + + /* Enable the watch point */ + watch_address_cntl = REG_SET_FIELD(watch_address_cntl, + TCP_WATCH0_CNTL, + VALID, + 1); + + WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) + + (watch_id * TCP_WATCH_STRIDE)), + watch_address_cntl); + return 0; } -static void set_scratch_backing_va(struct kgd_dev *kgd, - uint64_t va, uint32_t vmid) +uint32_t kgd_gfx_v9_clear_address_watch(struct amdgpu_device *adev, + uint32_t watch_id) { - /* No longer needed on GFXv9. The scratch base address is - * passed to the shader by the CP. It's the user mode driver's - * responsibility. - */ + uint32_t watch_address_cntl; + + watch_address_cntl = 0; + + WREG32_RLC((SOC15_REG_OFFSET(GC, 0, mmTCP_WATCH0_CNTL) + + (watch_id * TCP_WATCH_STRIDE)), + watch_address_cntl); + + return 0; } -/* FIXME: Does this need to be ASIC-specific code? */ -static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type) +/* kgd_gfx_v9_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values + * The values read are: + * ib_offload_wait_time -- Wait Count for Indirect Buffer Offloads. + * atomic_offload_wait_time -- Wait Count for L2 and GDS Atomics Offloads. + * wrm_offload_wait_time -- Wait Count for WAIT_REG_MEM Offloads. + * gws_wait_time -- Wait Count for Global Wave Syncs. + * que_sleep_wait_time -- Wait Count for Dequeue Retry. + * sch_wave_wait_time -- Wait Count for Scheduling Wave Message. + * sem_rearm_wait_time -- Wait Count for Semaphore re-arm. + * deq_retry_wait_time -- Wait Count for Global Wave Syncs. + */ +void kgd_gfx_v9_get_iq_wait_times(struct amdgpu_device *adev, + uint32_t *wait_times, + uint32_t inst) + { - struct amdgpu_device *adev = (struct amdgpu_device *) kgd; - const union amdgpu_firmware_header *hdr; + *wait_times = RREG32_SOC15_RLC(GC, GET_INST(GC, inst), + mmCP_IQ_WAIT_TIME2); +} - switch (type) { - case KGD_ENGINE_PFP: - hdr = (const union amdgpu_firmware_header *)adev->gfx.pfp_fw->data; - break; +void kgd_gfx_v9_set_vm_context_page_table_base(struct amdgpu_device *adev, + uint32_t vmid, uint64_t page_table_base) +{ + if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) { + pr_err("trying to set page table base for wrong VMID %u\n", + vmid); + return; + } - case KGD_ENGINE_ME: - hdr = (const union amdgpu_firmware_header *)adev->gfx.me_fw->data; - break; + adev->mmhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base); - case KGD_ENGINE_CE: - hdr = (const union amdgpu_firmware_header *)adev->gfx.ce_fw->data; - break; + adev->gfxhub.funcs->setup_vm_pt_regs(adev, vmid, page_table_base); +} - case KGD_ENGINE_MEC1: - hdr = (const union amdgpu_firmware_header *)adev->gfx.mec_fw->data; - break; +static void lock_spi_csq_mutexes(struct amdgpu_device *adev) +{ + mutex_lock(&adev->srbm_mutex); + mutex_lock(&adev->grbm_idx_mutex); - case KGD_ENGINE_MEC2: - hdr = (const union amdgpu_firmware_header *)adev->gfx.mec2_fw->data; - break; +} - case KGD_ENGINE_RLC: - hdr = (const union amdgpu_firmware_header *)adev->gfx.rlc_fw->data; - break; +static void unlock_spi_csq_mutexes(struct amdgpu_device *adev) +{ + mutex_unlock(&adev->grbm_idx_mutex); + mutex_unlock(&adev->srbm_mutex); +} - case KGD_ENGINE_SDMA1: - hdr = (const union amdgpu_firmware_header *)adev->sdma.instance[0].fw->data; - break; +/** + * get_wave_count: Read device registers to get number of waves in flight for + * a particular queue. The method also returns the VMID associated with the + * queue. + * + * @adev: Handle of device whose registers are to be read + * @queue_idx: Index of queue in the queue-map bit-field + * @queue_cnt: Stores the wave count and doorbell offset for an active queue + * @inst: xcc's instance number on a multi-XCC setup + */ +static void get_wave_count(struct amdgpu_device *adev, int queue_idx, + struct kfd_cu_occupancy *queue_cnt, uint32_t inst) +{ + int pipe_idx; + int queue_slot; + unsigned int reg_val; + unsigned int wave_cnt; + /* + * Program GRBM with appropriate MEID, PIPEID, QUEUEID and VMID + * parameters to read out waves in flight. Get VMID if there are + * non-zero waves in flight. + */ + pipe_idx = queue_idx / adev->gfx.mec.num_queue_per_pipe; + queue_slot = queue_idx % adev->gfx.mec.num_queue_per_pipe; + soc15_grbm_select(adev, 1, pipe_idx, queue_slot, 0, GET_INST(GC, inst)); + reg_val = RREG32_SOC15_IP(GC, SOC15_REG_OFFSET(GC, GET_INST(GC, inst), + mmSPI_CSQ_WF_ACTIVE_COUNT_0) + queue_slot); + wave_cnt = reg_val & SPI_CSQ_WF_ACTIVE_COUNT_0__COUNT_MASK; + if (wave_cnt != 0) { + queue_cnt->wave_cnt += wave_cnt; + queue_cnt->doorbell_off = + (RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_DOORBELL_CONTROL) & + CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK) >> + CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT; + } +} - case KGD_ENGINE_SDMA2: - hdr = (const union amdgpu_firmware_header *)adev->sdma.instance[1].fw->data; - break; +/** + * kgd_gfx_v9_get_cu_occupancy: Reads relevant registers associated with each + * shader engine and aggregates the number of waves that are in flight for the + * process whose pasid is provided as a parameter. The process could have ZERO + * or more queues running and submitting waves to compute units. + * + * @adev: Handle of device from which to get number of waves in flight + * @cu_occupancy: Array that gets filled with wave_cnt and doorbell offset + * for comparison later. + * @max_waves_per_cu: Output parameter updated with maximum number of waves + * possible per Compute Unit + * @inst: xcc's instance number on a multi-XCC setup + * + * Note: It's possible that the device has too many queues (oversubscription) + * in which case a VMID could be remapped to a different PASID. This could lead + * to an inaccurate wave count. Following is a high-level sequence: + * Time T1: vmid = getVmid(); vmid is associated with Pasid P1 + * Time T2: passId = getPasId(vmid); vmid is associated with Pasid P2 + * In the sequence above wave count obtained from time T1 will be incorrectly + * lost or added to total wave count. + * + * The registers that provide the waves in flight are: + * + * SPI_CSQ_WF_ACTIVE_STATUS - bit-map of queues per pipe. The bit is ON if a + * queue is slotted, OFF if there is no queue. A process could have ZERO or + * more queues slotted and submitting waves to be run on compute units. Even + * when there is a queue it is possible there could be zero wave fronts, this + * can happen when queue is waiting on top-of-pipe events - e.g. waitRegMem + * command + * + * For each bit that is ON from above: + * + * Read (SPI_CSQ_WF_ACTIVE_COUNT_0 + queue_idx) register. It provides the + * number of waves that are in flight for the queue at specified index. The + * index ranges from 0 to 7. + * + * If non-zero waves are in flight, store the corresponding doorbell offset + * of the queue, along with the wave count. + * + * Determine if the queue belongs to the process by comparing the doorbell + * offset against the process's queues. If it matches, aggregate the wave + * count for the process. + * + * Reading registers referenced above involves programming GRBM appropriately + */ +void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, + struct kfd_cu_occupancy *cu_occupancy, + int *max_waves_per_cu, uint32_t inst) +{ + int qidx; + int se_idx; + int se_cnt; + int queue_map; + int max_queue_cnt; + DECLARE_BITMAP(cp_queue_bitmap, AMDGPU_MAX_QUEUES); - default: - return 0; + lock_spi_csq_mutexes(adev); + soc15_grbm_select(adev, 1, 0, 0, 0, GET_INST(GC, inst)); + + /* + * Iterate through the shader engines and arrays of the device + * to get number of waves in flight + */ + bitmap_complement(cp_queue_bitmap, adev->gfx.mec_bitmap[0].queue_bitmap, + AMDGPU_MAX_QUEUES); + max_queue_cnt = adev->gfx.mec.num_pipe_per_mec * + adev->gfx.mec.num_queue_per_pipe; + se_cnt = adev->gfx.config.max_shader_engines; + for (se_idx = 0; se_idx < se_cnt; se_idx++) { + amdgpu_gfx_select_se_sh(adev, se_idx, 0, 0xffffffff, inst); + queue_map = RREG32_SOC15(GC, GET_INST(GC, inst), mmSPI_CSQ_WF_ACTIVE_STATUS); + + /* + * Assumption: queue map encodes following schema: four + * pipes per each micro-engine, with each pipe mapping + * eight queues. This schema is true for GFX9 devices + * and must be verified for newer device families + */ + for (qidx = 0; qidx < max_queue_cnt; qidx++) { + /* Skip qeueus that are not associated with + * compute functions + */ + if (!test_bit(qidx, cp_queue_bitmap)) + continue; + + if (!(queue_map & (1 << qidx))) + continue; + + /* Get number of waves in flight and aggregate them */ + get_wave_count(adev, qidx, &cu_occupancy[qidx], + inst); + } } - if (hdr == NULL) - return 0; + amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, inst); + soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, inst)); + unlock_spi_csq_mutexes(adev); - /* Only 12 bit in use*/ - return hdr->common.ucode_version; + /* Update the output parameters and return */ + *max_waves_per_cu = adev->gfx.cu_info.simd_per_cu * + adev->gfx.cu_info.max_waves_per_simd; } -static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid, - uint64_t page_table_base) +void kgd_gfx_v9_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev, + uint32_t wait_times, + uint32_t sch_wave, + uint32_t que_sleep, + uint32_t *reg_offset, + uint32_t *reg_data) { - struct amdgpu_device *adev = get_amdgpu_device(kgd); + *reg_data = wait_times; + + if (sch_wave) + *reg_data = REG_SET_FIELD(*reg_data, + CP_IQ_WAIT_TIME2, + SCH_WAVE, + sch_wave); + if (que_sleep) + *reg_data = REG_SET_FIELD(*reg_data, + CP_IQ_WAIT_TIME2, + QUE_SLEEP, + que_sleep); + + *reg_offset = SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2); +} - if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) { - pr_err("trying to set page table base for wrong VMID %u\n", - vmid); - return; - } +void kgd_gfx_v9_program_trap_handler_settings(struct amdgpu_device *adev, + uint32_t vmid, uint64_t tba_addr, uint64_t tma_addr, uint32_t inst) +{ + kgd_gfx_v9_lock_srbm(adev, 0, 0, 0, vmid, inst); + + /* + * Program TBA registers + */ + WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_SHADER_TBA_LO, + lower_32_bits(tba_addr >> 8)); + WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_SHADER_TBA_HI, + upper_32_bits(tba_addr >> 8)); - /* TODO: take advantage of per-process address space size. For - * now, all processes share the same address space size, like - * on GFX8 and older. + /* + * Program TMA registers */ - mmhub_v1_0_setup_vm_pt_regs(adev, vmid, page_table_base); + WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_SHADER_TMA_LO, + lower_32_bits(tma_addr >> 8)); + WREG32_SOC15(GC, GET_INST(GC, inst), mmSQ_SHADER_TMA_HI, + upper_32_bits(tma_addr >> 8)); + + kgd_gfx_v9_unlock_srbm(adev, inst); +} + +uint64_t kgd_gfx_v9_hqd_get_pq_addr(struct amdgpu_device *adev, + uint32_t pipe_id, uint32_t queue_id, + uint32_t inst) +{ + uint32_t low, high; + uint64_t queue_addr = 0; + + kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst); + amdgpu_gfx_rlc_enter_safe_mode(adev, inst); + + if (!RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE)) + goto unlock_out; + + low = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE); + high = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE_HI); + + /* only concerned with user queues. */ + if (!high) + goto unlock_out; + + queue_addr = (((queue_addr | high) << 32) | low) << 8; + +unlock_out: + amdgpu_gfx_rlc_exit_safe_mode(adev, inst); + kgd_gfx_v9_release_queue(adev, inst); + + return queue_addr; +} + +/* assume queue acquired */ +static int kgd_gfx_v9_hqd_dequeue_wait(struct amdgpu_device *adev, uint32_t inst, + unsigned int utimeout) +{ + unsigned long end_jiffies = (utimeout * HZ / 1000) + jiffies; + + while (true) { + uint32_t temp = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE); + + if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK)) + return 0; + + if (time_after(jiffies, end_jiffies)) + return -ETIME; + + usleep_range(500, 1000); + } +} + +uint64_t kgd_gfx_v9_hqd_reset(struct amdgpu_device *adev, + uint32_t pipe_id, uint32_t queue_id, + uint32_t inst, unsigned int utimeout) +{ + uint32_t low, high, pipe_reset_data = 0; + uint64_t queue_addr = 0; + + kgd_gfx_v9_acquire_queue(adev, pipe_id, queue_id, inst); + amdgpu_gfx_rlc_enter_safe_mode(adev, inst); + + if (!RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_ACTIVE)) + goto unlock_out; + + low = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE); + high = RREG32_SOC15(GC, GET_INST(GC, inst), mmCP_HQD_PQ_BASE_HI); + + /* only concerned with user queues. */ + if (!high) + goto unlock_out; + + queue_addr = (((queue_addr | high) << 32) | low) << 8; + + pr_debug("Attempting queue reset on XCC %i pipe id %i queue id %i\n", + inst, pipe_id, queue_id); + + /* assume previous dequeue request issued will take affect after reset */ + WREG32_SOC15(GC, GET_INST(GC, inst), mmSPI_COMPUTE_QUEUE_RESET, 0x1); + + if (!kgd_gfx_v9_hqd_dequeue_wait(adev, inst, utimeout)) + goto unlock_out; - gfxhub_v1_0_setup_vm_pt_regs(adev, vmid, page_table_base); + pr_debug("Attempting pipe reset on XCC %i pipe id %i\n", inst, pipe_id); + + pipe_reset_data = REG_SET_FIELD(pipe_reset_data, CP_MEC_CNTL, MEC_ME1_PIPE0_RESET, 1); + pipe_reset_data = pipe_reset_data << pipe_id; + + WREG32_SOC15(GC, GET_INST(GC, inst), mmCP_MEC_CNTL, pipe_reset_data); + WREG32_SOC15(GC, GET_INST(GC, inst), mmCP_MEC_CNTL, 0); + + if (kgd_gfx_v9_hqd_dequeue_wait(adev, inst, utimeout)) + queue_addr = 0; + +unlock_out: + pr_debug("queue reset on XCC %i pipe id %i queue id %i %s\n", + inst, pipe_id, queue_id, !!queue_addr ? "succeeded!" : "failed!"); + amdgpu_gfx_rlc_exit_safe_mode(adev, inst); + kgd_gfx_v9_release_queue(adev, inst); + + return queue_addr; +} + +uint32_t kgd_gfx_v9_hqd_sdma_get_doorbell(struct amdgpu_device *adev, + int engine, int queue) + +{ + return 0; } + +const struct kfd2kgd_calls gfx_v9_kfd2kgd = { + .program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings, + .set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping, + .init_interrupts = kgd_gfx_v9_init_interrupts, + .hqd_load = kgd_gfx_v9_hqd_load, + .hiq_mqd_load = kgd_gfx_v9_hiq_mqd_load, + .hqd_sdma_load = kgd_hqd_sdma_load, + .hqd_dump = kgd_gfx_v9_hqd_dump, + .hqd_sdma_dump = kgd_hqd_sdma_dump, + .hqd_is_occupied = kgd_gfx_v9_hqd_is_occupied, + .hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied, + .hqd_destroy = kgd_gfx_v9_hqd_destroy, + .hqd_sdma_destroy = kgd_hqd_sdma_destroy, + .wave_control_execute = kgd_gfx_v9_wave_control_execute, + .get_atc_vmid_pasid_mapping_info = + kgd_gfx_v9_get_atc_vmid_pasid_mapping_info, + .set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base, + .enable_debug_trap = kgd_gfx_v9_enable_debug_trap, + .disable_debug_trap = kgd_gfx_v9_disable_debug_trap, + .validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request, + .set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override, + .set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode, + .set_address_watch = kgd_gfx_v9_set_address_watch, + .clear_address_watch = kgd_gfx_v9_clear_address_watch, + .get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times, + .build_dequeue_wait_counts_packet_info = kgd_gfx_v9_build_dequeue_wait_counts_packet_info, + .get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy, + .program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings, + .hqd_get_pq_addr = kgd_gfx_v9_hqd_get_pq_addr, + .hqd_reset = kgd_gfx_v9_hqd_reset, + .hqd_sdma_get_doorbell = kgd_gfx_v9_hqd_sdma_get_doorbell +}; |