diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c | 597 |
1 files changed, 339 insertions, 258 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c index d2ce6b6a7ff6..326ecc8d37d2 100644 --- a/drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c @@ -33,7 +33,7 @@ #include "gc/gc_12_0_0_offset.h" #include "gc/gc_12_0_0_sh_mask.h" #include "hdp/hdp_6_0_0_offset.h" -#include "ivsrcid/gfx/irqsrcs_gfx_11_0_0.h" +#include "ivsrcid/gfx/irqsrcs_gfx_12_0_0.h" #include "soc15_common.h" #include "soc15.h" @@ -42,6 +42,8 @@ #include "sdma_common.h" #include "sdma_v7_0.h" #include "v12_structs.h" +#include "mes_userqueue.h" +#include "amdgpu_userq_fence.h" MODULE_FIRMWARE("amdgpu/sdma_7_0_0.bin"); MODULE_FIRMWARE("amdgpu/sdma_7_0_1.bin"); @@ -204,66 +206,39 @@ static uint64_t sdma_v7_0_ring_get_wptr(struct amdgpu_ring *ring) static void sdma_v7_0_ring_set_wptr(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; - uint32_t *wptr_saved; - uint32_t *is_queue_unmap; - uint64_t aggregated_db_index; - uint32_t mqd_size = adev->mqds[AMDGPU_HW_IP_DMA].mqd_size; DRM_DEBUG("Setting write pointer\n"); - if (ring->is_mes_queue) { - wptr_saved = (uint32_t *)(ring->mqd_ptr + mqd_size); - is_queue_unmap = (uint32_t *)(ring->mqd_ptr + mqd_size + - sizeof(uint32_t)); - aggregated_db_index = - amdgpu_mes_get_aggregated_doorbell_index(adev, - ring->hw_prio); - + if (ring->use_doorbell) { + DRM_DEBUG("Using doorbell -- " + "wptr_offs == 0x%08x " + "lower_32_bits(ring->wptr) << 2 == 0x%08x " + "upper_32_bits(ring->wptr) << 2 == 0x%08x\n", + ring->wptr_offs, + lower_32_bits(ring->wptr << 2), + upper_32_bits(ring->wptr << 2)); + /* XXX check if swapping is necessary on BE */ atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr << 2); - *wptr_saved = ring->wptr << 2; - if (*is_queue_unmap) { - WDOORBELL64(aggregated_db_index, ring->wptr << 2); - DRM_DEBUG("calling WDOORBELL64(0x%08x, 0x%016llx)\n", - ring->doorbell_index, ring->wptr << 2); - WDOORBELL64(ring->doorbell_index, ring->wptr << 2); - } else { - DRM_DEBUG("calling WDOORBELL64(0x%08x, 0x%016llx)\n", - ring->doorbell_index, ring->wptr << 2); - WDOORBELL64(ring->doorbell_index, ring->wptr << 2); - } + DRM_DEBUG("calling WDOORBELL64(0x%08x, 0x%016llx)\n", + ring->doorbell_index, ring->wptr << 2); + WDOORBELL64(ring->doorbell_index, ring->wptr << 2); } else { - if (ring->use_doorbell) { - DRM_DEBUG("Using doorbell -- " - "wptr_offs == 0x%08x " - "lower_32_bits(ring->wptr) << 2 == 0x%08x " - "upper_32_bits(ring->wptr) << 2 == 0x%08x\n", - ring->wptr_offs, - lower_32_bits(ring->wptr << 2), - upper_32_bits(ring->wptr << 2)); - /* XXX check if swapping is necessary on BE */ - atomic64_set((atomic64_t *)ring->wptr_cpu_addr, - ring->wptr << 2); - DRM_DEBUG("calling WDOORBELL64(0x%08x, 0x%016llx)\n", - ring->doorbell_index, ring->wptr << 2); - WDOORBELL64(ring->doorbell_index, ring->wptr << 2); - } else { - DRM_DEBUG("Not using doorbell -- " - "regSDMA%i_GFX_RB_WPTR == 0x%08x " - "regSDMA%i_GFX_RB_WPTR_HI == 0x%08x\n", - ring->me, - lower_32_bits(ring->wptr << 2), - ring->me, - upper_32_bits(ring->wptr << 2)); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, - ring->me, - regSDMA0_QUEUE0_RB_WPTR), - lower_32_bits(ring->wptr << 2)); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, - ring->me, - regSDMA0_QUEUE0_RB_WPTR_HI), - upper_32_bits(ring->wptr << 2)); - } + DRM_DEBUG("Not using doorbell -- " + "regSDMA%i_GFX_RB_WPTR == 0x%08x " + "regSDMA%i_GFX_RB_WPTR_HI == 0x%08x\n", + ring->me, + lower_32_bits(ring->wptr << 2), + ring->me, + upper_32_bits(ring->wptr << 2)); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, + ring->me, + regSDMA0_QUEUE0_RB_WPTR), + lower_32_bits(ring->wptr << 2)); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, + ring->me, + regSDMA0_QUEUE0_RB_WPTR_HI), + upper_32_bits(ring->wptr << 2)); } } @@ -407,11 +382,9 @@ static void sdma_v7_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 se } if (flags & AMDGPU_FENCE_FLAG_INT) { - uint32_t ctx = ring->is_mes_queue ? - (ring->hw_queue_id | AMDGPU_FENCE_MES_QUEUE_FLAG) : 0; /* generate an interrupt */ amdgpu_ring_write(ring, SDMA_PKT_COPY_LINEAR_HEADER_OP(SDMA_OP_TRAP)); - amdgpu_ring_write(ring, SDMA_PKT_TRAP_INT_CONTEXT_INT_CONTEXT(ctx)); + amdgpu_ring_write(ring, SDMA_PKT_TRAP_INT_CONTEXT_INT_CONTEXT(0)); } } @@ -490,162 +463,185 @@ static void sdma_v7_0_enable(struct amdgpu_device *adev, bool enable) } /** - * sdma_v7_0_gfx_resume - setup and start the async dma engines + * sdma_v7_0_gfx_resume_instance - start/restart a certain sdma engine * * @adev: amdgpu_device pointer + * @i: instance + * @restore: used to restore wptr when restart * - * Set up the gfx DMA ring buffers and enable them. - * Returns 0 for success, error for failure. + * Set up the gfx DMA ring buffers and enable them. On restart, we will restore wptr and rptr. + * Return 0 for success. */ -static int sdma_v7_0_gfx_resume(struct amdgpu_device *adev) +static int sdma_v7_0_gfx_resume_instance(struct amdgpu_device *adev, int i, bool restore) { struct amdgpu_ring *ring; u32 rb_cntl, ib_cntl; u32 rb_bufsz; u32 doorbell; u32 doorbell_offset; - u32 tmp; + u32 temp; u64 wptr_gpu_addr; - int i, r; - - for (i = 0; i < adev->sdma.num_instances; i++) { - ring = &adev->sdma.instance[i].ring; + int r; - //if (!amdgpu_sriov_vf(adev)) - // WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_SEM_WAIT_FAIL_TIMER_CNTL), 0); + ring = &adev->sdma.instance[i].ring; - /* Set ring buffer size in dwords */ - rb_bufsz = order_base_2(ring->ring_size / 4); - rb_cntl = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_CNTL)); - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RB_SIZE, rb_bufsz); + /* Set ring buffer size in dwords */ + rb_bufsz = order_base_2(ring->ring_size / 4); + rb_cntl = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_CNTL)); + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RB_SIZE, rb_bufsz); #ifdef __BIG_ENDIAN - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RB_SWAP_ENABLE, 1); - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, - RPTR_WRITEBACK_SWAP_ENABLE, 1); + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RB_SWAP_ENABLE, 1); + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, + RPTR_WRITEBACK_SWAP_ENABLE, 1); #endif - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RB_PRIV, 1); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_CNTL), rb_cntl); - - /* Initialize the ring buffer's read and write pointers */ + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RB_PRIV, 1); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_CNTL), rb_cntl); + + /* Initialize the ring buffer's read and write pointers */ + if (restore) { + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_RPTR), lower_32_bits(ring->wptr << 2)); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_RPTR_HI), upper_32_bits(ring->wptr << 2)); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR), lower_32_bits(ring->wptr << 2)); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR_HI), upper_32_bits(ring->wptr << 2)); + } else { WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_RPTR), 0); WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_RPTR_HI), 0); WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR), 0); WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR_HI), 0); + } + /* setup the wptr shadow polling */ + wptr_gpu_addr = ring->wptr_gpu_addr; + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR_POLL_ADDR_LO), + lower_32_bits(wptr_gpu_addr)); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR_POLL_ADDR_HI), + upper_32_bits(wptr_gpu_addr)); + + /* set the wb address whether it's enabled or not */ + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_RPTR_ADDR_HI), + upper_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFF); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_RPTR_ADDR_LO), + lower_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFC); + + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RPTR_WRITEBACK_ENABLE, 1); + if (amdgpu_sriov_vf(adev)) + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, WPTR_POLL_ENABLE, 1); + else + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, WPTR_POLL_ENABLE, 0); - /* setup the wptr shadow polling */ - wptr_gpu_addr = ring->wptr_gpu_addr; - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR_POLL_ADDR_LO), - lower_32_bits(wptr_gpu_addr)); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR_POLL_ADDR_HI), - upper_32_bits(wptr_gpu_addr)); - - /* set the wb address whether it's enabled or not */ - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_RPTR_ADDR_HI), - upper_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFF); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_RPTR_ADDR_LO), - lower_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFC); - - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RPTR_WRITEBACK_ENABLE, 1); - if (amdgpu_sriov_vf(adev)) - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, WPTR_POLL_ENABLE, 1); - else - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, WPTR_POLL_ENABLE, 0); - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, MCU_WPTR_POLL_ENABLE, 1); + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, MCU_WPTR_POLL_ENABLE, 1); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_BASE), ring->gpu_addr >> 8); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_BASE_HI), ring->gpu_addr >> 40); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_BASE), ring->gpu_addr >> 8); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_BASE_HI), ring->gpu_addr >> 40); + if (!restore) ring->wptr = 0; - /* before programing wptr to a less value, need set minor_ptr_update first */ - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_MINOR_PTR_UPDATE), 1); + /* before programing wptr to a less value, need set minor_ptr_update first */ + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_MINOR_PTR_UPDATE), 1); - if (!amdgpu_sriov_vf(adev)) { /* only bare-metal use register write for wptr */ - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR), lower_32_bits(ring->wptr) << 2); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR_HI), upper_32_bits(ring->wptr) << 2); - } + if (!amdgpu_sriov_vf(adev)) { /* only bare-metal use register write for wptr */ + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR), lower_32_bits(ring->wptr) << 2); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_WPTR_HI), upper_32_bits(ring->wptr) << 2); + } - doorbell = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_DOORBELL)); - doorbell_offset = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_DOORBELL_OFFSET)); + doorbell = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_DOORBELL)); + doorbell_offset = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_DOORBELL_OFFSET)); - if (ring->use_doorbell) { - doorbell = REG_SET_FIELD(doorbell, SDMA0_QUEUE0_DOORBELL, ENABLE, 1); - doorbell_offset = REG_SET_FIELD(doorbell_offset, SDMA0_QUEUE0_DOORBELL_OFFSET, - OFFSET, ring->doorbell_index); - } else { - doorbell = REG_SET_FIELD(doorbell, SDMA0_QUEUE0_DOORBELL, ENABLE, 0); - } - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_DOORBELL), doorbell); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_DOORBELL_OFFSET), doorbell_offset); - - if (i == 0) - adev->nbio.funcs->sdma_doorbell_range(adev, i, ring->use_doorbell, - ring->doorbell_index, - adev->doorbell_index.sdma_doorbell_range * adev->sdma.num_instances); - - if (amdgpu_sriov_vf(adev)) - sdma_v7_0_ring_set_wptr(ring); - - /* set minor_ptr_update to 0 after wptr programed */ - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_MINOR_PTR_UPDATE), 0); - - /* Set up sdma hang watchdog */ - tmp = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_WATCHDOG_CNTL)); - /* 100ms per unit */ - tmp = REG_SET_FIELD(tmp, SDMA0_WATCHDOG_CNTL, QUEUE_HANG_COUNT, - max(adev->usec_timeout/100000, 1)); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_WATCHDOG_CNTL), tmp); - - /* Set up RESP_MODE to non-copy addresses */ - tmp = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_UTCL1_CNTL)); - tmp = REG_SET_FIELD(tmp, SDMA0_UTCL1_CNTL, RESP_MODE, 3); - tmp = REG_SET_FIELD(tmp, SDMA0_UTCL1_CNTL, REDO_DELAY, 9); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_UTCL1_CNTL), tmp); - - /* program default cache read and write policy */ - tmp = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_UTCL1_PAGE)); - /* clean read policy and write policy bits */ - tmp &= 0xFF0FFF; - tmp |= ((CACHE_READ_POLICY_L2__DEFAULT << 12) | - (CACHE_WRITE_POLICY_L2__DEFAULT << 14)); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_UTCL1_PAGE), tmp); - - if (!amdgpu_sriov_vf(adev)) { - /* unhalt engine */ - tmp = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_MCU_CNTL)); - tmp = REG_SET_FIELD(tmp, SDMA0_MCU_CNTL, HALT, 0); - tmp = REG_SET_FIELD(tmp, SDMA0_MCU_CNTL, RESET, 0); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_MCU_CNTL), tmp); - } + if (ring->use_doorbell) { + doorbell = REG_SET_FIELD(doorbell, SDMA0_QUEUE0_DOORBELL, ENABLE, 1); + doorbell_offset = REG_SET_FIELD(doorbell_offset, SDMA0_QUEUE0_DOORBELL_OFFSET, + OFFSET, ring->doorbell_index); + } else { + doorbell = REG_SET_FIELD(doorbell, SDMA0_QUEUE0_DOORBELL, ENABLE, 0); + } + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_DOORBELL), doorbell); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_DOORBELL_OFFSET), doorbell_offset); - /* enable DMA RB */ - rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RB_ENABLE, 1); - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_CNTL), rb_cntl); + if (i == 0) + adev->nbio.funcs->sdma_doorbell_range(adev, i, ring->use_doorbell, + ring->doorbell_index, + adev->doorbell_index.sdma_doorbell_range * adev->sdma.num_instances); - ib_cntl = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_IB_CNTL)); - ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_QUEUE0_IB_CNTL, IB_ENABLE, 1); + if (amdgpu_sriov_vf(adev)) + sdma_v7_0_ring_set_wptr(ring); + + /* set minor_ptr_update to 0 after wptr programed */ + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_MINOR_PTR_UPDATE), 0); + + /* Set up sdma hang watchdog */ + temp = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_WATCHDOG_CNTL)); + /* 100ms per unit */ + temp = REG_SET_FIELD(temp, SDMA0_WATCHDOG_CNTL, QUEUE_HANG_COUNT, + max(adev->usec_timeout/100000, 1)); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_WATCHDOG_CNTL), temp); + + /* Set up RESP_MODE to non-copy addresses */ + temp = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_UTCL1_CNTL)); + temp = REG_SET_FIELD(temp, SDMA0_UTCL1_CNTL, RESP_MODE, 3); + temp = REG_SET_FIELD(temp, SDMA0_UTCL1_CNTL, REDO_DELAY, 9); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_UTCL1_CNTL), temp); + + /* program default cache read and write policy */ + temp = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_UTCL1_PAGE)); + /* clean read policy and write policy bits */ + temp &= 0xFF0FFF; + temp |= ((CACHE_READ_POLICY_L2__DEFAULT << 12) | + (CACHE_WRITE_POLICY_L2__DEFAULT << 14)); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_UTCL1_PAGE), temp); + + if (!amdgpu_sriov_vf(adev)) { + /* unhalt engine */ + temp = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_MCU_CNTL)); + temp = REG_SET_FIELD(temp, SDMA0_MCU_CNTL, HALT, 0); + temp = REG_SET_FIELD(temp, SDMA0_MCU_CNTL, RESET, 0); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_MCU_CNTL), temp); + } + + /* enable DMA RB */ + rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_QUEUE0_RB_CNTL, RB_ENABLE, 1); + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_RB_CNTL), rb_cntl); + + ib_cntl = RREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_IB_CNTL)); + ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_QUEUE0_IB_CNTL, IB_ENABLE, 1); #ifdef __BIG_ENDIAN - ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_QUEUE0_IB_CNTL, IB_SWAP_ENABLE, 1); + ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_QUEUE0_IB_CNTL, IB_SWAP_ENABLE, 1); #endif - /* enable DMA IBs */ - WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_IB_CNTL), ib_cntl); + /* enable DMA IBs */ + WREG32_SOC15_IP(GC, sdma_v7_0_get_reg_offset(adev, i, regSDMA0_QUEUE0_IB_CNTL), ib_cntl); + ring->sched.ready = true; + + if (amdgpu_sriov_vf(adev)) { /* bare-metal sequence doesn't need below to lines */ + sdma_v7_0_ctx_switch_enable(adev, true); + sdma_v7_0_enable(adev, true); + } - ring->sched.ready = true; + r = amdgpu_ring_test_helper(ring); + if (r) + ring->sched.ready = false; - if (amdgpu_sriov_vf(adev)) { /* bare-metal sequence doesn't need below to lines */ - sdma_v7_0_ctx_switch_enable(adev, true); - sdma_v7_0_enable(adev, true); - } + return r; +} - r = amdgpu_ring_test_helper(ring); - if (r) { - ring->sched.ready = false; - return r; - } +/** + * sdma_v7_0_gfx_resume - setup and start the async dma engines + * + * @adev: amdgpu_device pointer + * + * Set up the gfx DMA ring buffers and enable them. + * Returns 0 for success, error for failure. + */ +static int sdma_v7_0_gfx_resume(struct amdgpu_device *adev) +{ + int i, r; + for (i = 0; i < adev->sdma.num_instances; i++) { + r = sdma_v7_0_gfx_resume_instance(adev, i, false); + if (r) + return r; } return 0; + } /** @@ -806,6 +802,31 @@ static bool sdma_v7_0_check_soft_reset(struct amdgpu_ip_block *ip_block) return false; } +static int sdma_v7_0_reset_queue(struct amdgpu_ring *ring, + unsigned int vmid, + struct amdgpu_fence *timedout_fence) +{ + struct amdgpu_device *adev = ring->adev; + int r; + + if (ring->me >= adev->sdma.num_instances) { + dev_err(adev->dev, "sdma instance not found\n"); + return -EINVAL; + } + + amdgpu_ring_reset_helper_begin(ring, timedout_fence); + + r = amdgpu_mes_reset_legacy_queue(adev, ring, vmid, true); + if (r) + return r; + + r = sdma_v7_0_gfx_resume_instance(adev, ring->me, true); + if (r) + return r; + + return amdgpu_ring_reset_helper_end(ring, timedout_fence); +} + /** * sdma_v7_0_start - setup and start the async dma engines * @@ -887,6 +908,12 @@ static int sdma_v7_0_mqd_init(struct amdgpu_device *adev, void *mqd, m->sdmax_rlcx_rb_aql_cntl = 0x4000; //regSDMA0_QUEUE0_RB_AQL_CNTL_DEFAULT; m->sdmax_rlcx_dummy_reg = 0xf; //regSDMA0_QUEUE0_DUMMY_REG_DEFAULT; + m->sdmax_rlcx_csa_addr_lo = lower_32_bits(prop->csa_addr); + m->sdmax_rlcx_csa_addr_hi = upper_32_bits(prop->csa_addr); + + m->sdmax_rlcx_mcu_dbg0 = lower_32_bits(prop->fence_address); + m->sdmax_rlcx_mcu_dbg1 = upper_32_bits(prop->fence_address); + return 0; } @@ -913,33 +940,22 @@ static int sdma_v7_0_ring_test_ring(struct amdgpu_ring *ring) int r; u32 tmp; u64 gpu_addr; - volatile uint32_t *cpu_ptr = NULL; tmp = 0xCAFEDEAD; - if (ring->is_mes_queue) { - uint32_t offset = 0; - offset = amdgpu_mes_ctx_get_offs(ring, - AMDGPU_MES_CTX_PADDING_OFFS); - gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset); - cpu_ptr = amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset); - *cpu_ptr = tmp; - } else { - r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); - return r; - } - - gpu_addr = adev->wb.gpu_addr + (index * 4); - adev->wb.wb[index] = cpu_to_le32(tmp); + r = amdgpu_device_wb_get(adev, &index); + if (r) { + dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r); + return r; } + gpu_addr = adev->wb.gpu_addr + (index * 4); + adev->wb.wb[index] = cpu_to_le32(tmp); + r = amdgpu_ring_alloc(ring, 5); if (r) { DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r); - if (!ring->is_mes_queue) - amdgpu_device_wb_free(adev, index); + amdgpu_device_wb_free(adev, index); return r; } @@ -952,10 +968,7 @@ static int sdma_v7_0_ring_test_ring(struct amdgpu_ring *ring) amdgpu_ring_commit(ring); for (i = 0; i < adev->usec_timeout; i++) { - if (ring->is_mes_queue) - tmp = le32_to_cpu(*cpu_ptr); - else - tmp = le32_to_cpu(adev->wb.wb[index]); + tmp = le32_to_cpu(adev->wb.wb[index]); if (tmp == 0xDEADBEEF) break; if (amdgpu_emu_mode == 1) @@ -967,8 +980,7 @@ static int sdma_v7_0_ring_test_ring(struct amdgpu_ring *ring) if (i >= adev->usec_timeout) r = -ETIMEDOUT; - if (!ring->is_mes_queue) - amdgpu_device_wb_free(adev, index); + amdgpu_device_wb_free(adev, index); return r; } @@ -991,37 +1003,23 @@ static int sdma_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) long r; u32 tmp = 0; u64 gpu_addr; - volatile uint32_t *cpu_ptr = NULL; tmp = 0xCAFEDEAD; memset(&ib, 0, sizeof(ib)); - if (ring->is_mes_queue) { - uint32_t offset = 0; - offset = amdgpu_mes_ctx_get_offs(ring, AMDGPU_MES_CTX_IB_OFFS); - ib.gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset); - ib.ptr = (void *)amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset); - - offset = amdgpu_mes_ctx_get_offs(ring, - AMDGPU_MES_CTX_PADDING_OFFS); - gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset); - cpu_ptr = amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset); - *cpu_ptr = tmp; - } else { - r = amdgpu_device_wb_get(adev, &index); - if (r) { - dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); - return r; - } + r = amdgpu_device_wb_get(adev, &index); + if (r) { + dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r); + return r; + } - gpu_addr = adev->wb.gpu_addr + (index * 4); - adev->wb.wb[index] = cpu_to_le32(tmp); + gpu_addr = adev->wb.gpu_addr + (index * 4); + adev->wb.wb[index] = cpu_to_le32(tmp); - r = amdgpu_ib_get(adev, NULL, 256, AMDGPU_IB_POOL_DIRECT, &ib); - if (r) { - DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); - goto err0; - } + r = amdgpu_ib_get(adev, NULL, 256, AMDGPU_IB_POOL_DIRECT, &ib); + if (r) { + DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); + goto err0; } ib.ptr[0] = SDMA_PKT_COPY_LINEAR_HEADER_OP(SDMA_OP_WRITE) | @@ -1049,10 +1047,7 @@ static int sdma_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) goto err1; } - if (ring->is_mes_queue) - tmp = le32_to_cpu(*cpu_ptr); - else - tmp = le32_to_cpu(adev->wb.wb[index]); + tmp = le32_to_cpu(adev->wb.wb[index]); if (tmp == 0xDEADBEEF) r = 0; @@ -1060,11 +1055,10 @@ static int sdma_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) r = -EINVAL; err1: - amdgpu_ib_free(adev, &ib, NULL); + amdgpu_ib_free(&ib, NULL); dma_fence_put(f); err0: - if (!ring->is_mes_queue) - amdgpu_device_wb_free(adev, index); + amdgpu_device_wb_free(adev, index); return r; } @@ -1264,6 +1258,23 @@ static int sdma_v7_0_early_init(struct amdgpu_ip_block *ip_block) struct amdgpu_device *adev = ip_block->adev; int r; + switch (amdgpu_user_queue) { + case -1: + case 0: + default: + adev->sdma.no_user_submission = false; + adev->sdma.disable_uq = true; + break; + case 1: + adev->sdma.no_user_submission = false; + adev->sdma.disable_uq = false; + break; + case 2: + adev->sdma.no_user_submission = true; + adev->sdma.disable_uq = false; + break; + } + r = amdgpu_sdma_init_microcode(adev, 0, true); if (r) { DRM_ERROR("Failed to init sdma firmware!\n"); @@ -1289,16 +1300,24 @@ static int sdma_v7_0_sw_init(struct amdgpu_ip_block *ip_block) /* SDMA trap event */ r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GFX, - GFX_11_0_0__SRCID__SDMA_TRAP, + GFX_12_0_0__SRCID__SDMA_TRAP, &adev->sdma.trap_irq); if (r) return r; + /* SDMA user fence event */ + r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GFX, + GFX_12_0_0__SRCID__SDMA_FENCE, + &adev->sdma.fence_irq); + if (r) + return r; + for (i = 0; i < adev->sdma.num_instances; i++) { ring = &adev->sdma.instance[i].ring; ring->ring_obj = NULL; ring->use_doorbell = true; ring->me = i; + ring->no_user_submission = adev->sdma.no_user_submission; DRM_DEBUG("SDMA %d use_doorbell being set to: [%s]\n", i, ring->use_doorbell?"true":"false"); @@ -1316,6 +1335,14 @@ static int sdma_v7_0_sw_init(struct amdgpu_ip_block *ip_block) return r; } + adev->sdma.supported_reset = + amdgpu_get_soft_full_reset_mask(&adev->sdma.instance[0].ring); + if (!amdgpu_sriov_vf(adev)) + adev->sdma.supported_reset |= AMDGPU_RESET_TYPE_PER_QUEUE; + + r = amdgpu_sdma_sysfs_reset_mask_init(adev); + if (r) + return r; /* Allocate memory for SDMA IP Dump buffer */ ptr = kcalloc(adev->sdma.num_instances * reg_count, sizeof(uint32_t), GFP_KERNEL); if (ptr) @@ -1323,6 +1350,16 @@ static int sdma_v7_0_sw_init(struct amdgpu_ip_block *ip_block) else DRM_ERROR("Failed to allocated memory for SDMA IP Dump\n"); + switch (amdgpu_ip_version(adev, SDMA0_HWIP, 0)) { + case IP_VERSION(7, 0, 0): + case IP_VERSION(7, 0, 1): + if ((adev->sdma.instance[0].fw_version >= 7966358) && !adev->sdma.disable_uq) + adev->userq_funcs[AMDGPU_HW_IP_DMA] = &userq_mes_funcs; + break; + default: + break; + } + return r; } @@ -1334,6 +1371,7 @@ static int sdma_v7_0_sw_fini(struct amdgpu_ip_block *ip_block) for (i = 0; i < adev->sdma.num_instances; i++) amdgpu_ring_fini(&adev->sdma.instance[i].ring); + amdgpu_sdma_sysfs_reset_mask_fini(adev); amdgpu_sdma_destroy_inst_ctx(adev, true); if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) @@ -1344,11 +1382,39 @@ static int sdma_v7_0_sw_fini(struct amdgpu_ip_block *ip_block) return 0; } +static int sdma_v7_0_set_userq_trap_interrupts(struct amdgpu_device *adev, + bool enable) +{ + unsigned int irq_type; + int i, r; + + if (adev->userq_funcs[AMDGPU_HW_IP_DMA]) { + for (i = 0; i < adev->sdma.num_instances; i++) { + irq_type = AMDGPU_SDMA_IRQ_INSTANCE0 + i; + if (enable) + r = amdgpu_irq_get(adev, &adev->sdma.trap_irq, + irq_type); + else + r = amdgpu_irq_put(adev, &adev->sdma.trap_irq, + irq_type); + if (r) + return r; + } + } + + return 0; +} + static int sdma_v7_0_hw_init(struct amdgpu_ip_block *ip_block) { struct amdgpu_device *adev = ip_block->adev; + int r; - return sdma_v7_0_start(adev); + r = sdma_v7_0_start(adev); + if (r) + return r; + + return sdma_v7_0_set_userq_trap_interrupts(adev, true); } static int sdma_v7_0_hw_fini(struct amdgpu_ip_block *ip_block) @@ -1360,6 +1426,7 @@ static int sdma_v7_0_hw_fini(struct amdgpu_ip_block *ip_block) sdma_v7_0_ctx_switch_enable(adev, false); sdma_v7_0_enable(adev, false); + sdma_v7_0_set_userq_trap_interrupts(adev, false); return 0; } @@ -1374,9 +1441,9 @@ static int sdma_v7_0_resume(struct amdgpu_ip_block *ip_block) return sdma_v7_0_hw_init(ip_block); } -static bool sdma_v7_0_is_idle(void *handle) +static bool sdma_v7_0_is_idle(struct amdgpu_ip_block *ip_block) { - struct amdgpu_device *adev = (struct amdgpu_device *)handle; + struct amdgpu_device *adev = ip_block->adev; u32 i; for (i = 0; i < adev->sdma.num_instances; i++) { @@ -1477,25 +1544,9 @@ static int sdma_v7_0_process_trap_irq(struct amdgpu_device *adev, struct amdgpu_iv_entry *entry) { int instances, queue; - uint32_t mes_queue_id = entry->src_data[0]; DRM_DEBUG("IH: SDMA trap\n"); - if (adev->enable_mes && (mes_queue_id & AMDGPU_FENCE_MES_QUEUE_FLAG)) { - struct amdgpu_mes_queue *queue; - - mes_queue_id &= AMDGPU_FENCE_MES_QUEUE_ID_MASK; - - spin_lock(&adev->mes.queue_id_lock); - queue = idr_find(&adev->mes.queue_id_idr, mes_queue_id); - if (queue) { - DRM_DEBUG("process smda queue id = %d\n", mes_queue_id); - amdgpu_fence_process(queue->ring); - } - spin_unlock(&adev->mes.queue_id_lock); - return 0; - } - queue = entry->ring_id & 0xf; instances = (entry->ring_id & 0xf0) >> 4; if (instances > 1) { @@ -1517,6 +1568,29 @@ static int sdma_v7_0_process_trap_irq(struct amdgpu_device *adev, return 0; } +static int sdma_v7_0_process_fence_irq(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + struct amdgpu_iv_entry *entry) +{ + u32 doorbell_offset = entry->src_data[0]; + + if (adev->enable_mes && doorbell_offset) { + struct amdgpu_userq_fence_driver *fence_drv = NULL; + struct xarray *xa = &adev->userq_xa; + unsigned long flags; + + doorbell_offset >>= SDMA0_QUEUE0_DOORBELL_OFFSET__OFFSET__SHIFT; + + xa_lock_irqsave(xa, flags); + fence_drv = xa_load(xa, doorbell_offset); + if (fence_drv) + amdgpu_userq_fence_driver_process(fence_drv); + xa_unlock_irqrestore(xa, flags); + } + + return 0; +} + static int sdma_v7_0_process_illegal_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) @@ -1524,19 +1598,19 @@ static int sdma_v7_0_process_illegal_inst_irq(struct amdgpu_device *adev, return 0; } -static int sdma_v7_0_set_clockgating_state(void *handle, +static int sdma_v7_0_set_clockgating_state(struct amdgpu_ip_block *ip_block, enum amd_clockgating_state state) { return 0; } -static int sdma_v7_0_set_powergating_state(void *handle, +static int sdma_v7_0_set_powergating_state(struct amdgpu_ip_block *ip_block, enum amd_powergating_state state) { return 0; } -static void sdma_v7_0_get_clockgating_state(void *handle, u64 *flags) +static void sdma_v7_0_get_clockgating_state(struct amdgpu_ip_block *ip_block, u64 *flags) { } @@ -1636,6 +1710,7 @@ static const struct amdgpu_ring_funcs sdma_v7_0_ring_funcs = { .emit_reg_write_reg_wait = sdma_v7_0_ring_emit_reg_write_reg_wait, .init_cond_exec = sdma_v7_0_ring_init_cond_exec, .preempt_ib = sdma_v7_0_ring_preempt_ib, + .reset = sdma_v7_0_reset_queue, }; static void sdma_v7_0_set_ring_funcs(struct amdgpu_device *adev) @@ -1653,6 +1728,10 @@ static const struct amdgpu_irq_src_funcs sdma_v7_0_trap_irq_funcs = { .process = sdma_v7_0_process_trap_irq, }; +static const struct amdgpu_irq_src_funcs sdma_v7_0_fence_irq_funcs = { + .process = sdma_v7_0_process_fence_irq, +}; + static const struct amdgpu_irq_src_funcs sdma_v7_0_illegal_inst_irq_funcs = { .process = sdma_v7_0_process_illegal_inst_irq, }; @@ -1662,6 +1741,7 @@ static void sdma_v7_0_set_irq_funcs(struct amdgpu_device *adev) adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE0 + adev->sdma.num_instances; adev->sdma.trap_irq.funcs = &sdma_v7_0_trap_irq_funcs; + adev->sdma.fence_irq.funcs = &sdma_v7_0_fence_irq_funcs; adev->sdma.illegal_inst_irq.funcs = &sdma_v7_0_illegal_inst_irq_funcs; } @@ -1684,11 +1764,12 @@ static void sdma_v7_0_emit_copy_buffer(struct amdgpu_ib *ib, uint32_t byte_count, uint32_t copy_flags) { - uint32_t num_type, data_format, max_com; + uint32_t num_type, data_format, max_com, write_cm; max_com = AMDGPU_COPY_FLAGS_GET(copy_flags, MAX_COMPRESSED); data_format = AMDGPU_COPY_FLAGS_GET(copy_flags, DATA_FORMAT); num_type = AMDGPU_COPY_FLAGS_GET(copy_flags, NUMBER_TYPE); + write_cm = AMDGPU_COPY_FLAGS_GET(copy_flags, WRITE_COMPRESS_DISABLE) ? 2 : 1; ib->ptr[ib->length_dw++] = SDMA_PKT_COPY_LINEAR_HEADER_OP(SDMA_OP_COPY) | SDMA_PKT_COPY_LINEAR_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR) | @@ -1705,7 +1786,7 @@ static void sdma_v7_0_emit_copy_buffer(struct amdgpu_ib *ib, if ((copy_flags & (AMDGPU_COPY_FLAGS_READ_DECOMPRESSED | AMDGPU_COPY_FLAGS_WRITE_COMPRESSED))) ib->ptr[ib->length_dw++] = SDMA_DCC_DATA_FORMAT(data_format) | SDMA_DCC_NUM_TYPE(num_type) | ((copy_flags & AMDGPU_COPY_FLAGS_READ_DECOMPRESSED) ? SDMA_DCC_READ_CM(2) : 0) | - ((copy_flags & AMDGPU_COPY_FLAGS_WRITE_COMPRESSED) ? SDMA_DCC_WRITE_CM(1) : 0) | + ((copy_flags & AMDGPU_COPY_FLAGS_WRITE_COMPRESSED) ? SDMA_DCC_WRITE_CM(write_cm) : 0) | SDMA_DCC_MAX_COM(max_com) | SDMA_DCC_MAX_UCOM(1); else ib->ptr[ib->length_dw++] = 0; |