diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 59 |
1 files changed, 31 insertions, 28 deletions
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 4894108c5d58..240441e15d52 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c @@ -260,13 +260,15 @@ static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id) return 0; } -static inline uint32_t get_sdma_base_addr(struct vi_sdma_mqd *m) +static inline uint32_t get_sdma_rlc_reg_offset(struct vi_sdma_mqd *m) { uint32_t retval; retval = m->sdma_engine_id * SDMA1_REGISTER_OFFSET + m->sdma_queue_id * KFD_VI_SDMA_QUEUE_OFFSET; - pr_debug("sdma base address: 0x%x\n", retval); + + pr_debug("RLC register offset for SDMA%d RLC%d: 0x%x\n", + m->sdma_engine_id, m->sdma_queue_id, retval); return retval; } @@ -398,17 +400,17 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, struct amdgpu_device *adev = get_amdgpu_device(kgd); struct vi_sdma_mqd *m; unsigned long end_jiffies; - uint32_t sdma_base_addr; + uint32_t sdma_rlc_reg_offset; uint32_t data; m = get_sdma_mqd(mqd); - sdma_base_addr = get_sdma_base_addr(m); - WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, + sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m); + 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)) { @@ -420,28 +422,29 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd, 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_rlc_reg_offset + mmSDMA0_RLC0_DOORBELL, data); + WREG32(sdma_rlc_reg_offset + 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); + WREG32(sdma_rlc_reg_offset + mmSDMA0_RLC0_RB_WPTR, data); 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_VIRTUAL_ADDR, + WREG32(sdma_rlc_reg_offset + 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, + 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; } @@ -508,13 +511,13 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd) { struct amdgpu_device *adev = get_amdgpu_device(kgd); struct vi_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(m); + sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m); - 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; @@ -632,19 +635,19 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, { struct amdgpu_device *adev = get_amdgpu_device(kgd); struct vi_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(m); + sdma_rlc_reg_offset = get_sdma_rlc_reg_offset(m); - 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)) { @@ -654,12 +657,12 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd, 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); return 0; } |