[PATCH 2/2] drm/amdgpu: add gfx9 gpr EDC workaround when RAS is enabled
Zhu, James
James.Zhu at amd.com
Tue May 7 15:24:13 UTC 2019
When RAS is enabled, initializes the VGPRs/LDS/SGPRs and
resets EDC error counts. This is done in late_init, before
RAS TA GFX enable.
Signed-off-by: James Zhu <James.Zhu at amd.com>
Reviewed-by: Alex Deucher <alexander.deucher at amd.com>
---
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 245 ++++++++++++++++++++++++++++++++++
drivers/gpu/drm/amd/amdgpu/soc15.h | 10 ++
2 files changed, 255 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index ba67d10..14e671d 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -34,6 +34,7 @@
#include "vega10_enum.h"
#include "hdp/hdp_4_0_offset.h"
+#include "soc15.h"
#include "soc15_common.h"
#include "clearstate_gfx9.h"
#include "v9_structs.h"
@@ -3529,6 +3530,245 @@ static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
(1 << (oa_size + oa_base)) - (1 << oa_base));
}
+static const u32 vgpr_init_compute_shader[] =
+{
+ 0xb07c0000, 0xbe8000ff,
+ 0x000000f8, 0xbf110800,
+ 0x7e000280, 0x7e020280,
+ 0x7e040280, 0x7e060280,
+ 0x7e080280, 0x7e0a0280,
+ 0x7e0c0280, 0x7e0e0280,
+ 0x80808800, 0xbe803200,
+ 0xbf84fff5, 0xbf9c0000,
+ 0xd28c0001, 0x0001007f,
+ 0xd28d0001, 0x0002027e,
+ 0x10020288, 0xb8810904,
+ 0xb7814000, 0xd1196a01,
+ 0x00000301, 0xbe800087,
+ 0xbefc00c1, 0xd89c4000,
+ 0x00020201, 0xd89cc080,
+ 0x00040401, 0x320202ff,
+ 0x00000800, 0x80808100,
+ 0xbf84fff8, 0x7e020280,
+ 0xbf810000, 0x00000000,
+};
+
+static const u32 sgpr_init_compute_shader[] =
+{
+ 0xb07c0000, 0xbe8000ff,
+ 0x0000005f, 0xbee50080,
+ 0xbe812c65, 0xbe822c65,
+ 0xbe832c65, 0xbe842c65,
+ 0xbe852c65, 0xb77c0005,
+ 0x80808500, 0xbf84fff8,
+ 0xbe800080, 0xbf810000,
+};
+
+static const struct soc15_reg_entry vgpr_init_regs[] = {
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x1000000 }, /* CU_GROUP_COUNT=1 */
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 256*2 },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 1 },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x100007f }, /* VGPRS=15 (256 logical VGPRs, SGPRS=1 (16 SGPRs, BULKY=1 */
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
+};
+
+static const struct soc15_reg_entry sgpr_init_regs[] = {
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x1000000 }, /* CU_GROUP_COUNT=1 */
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 256*2 },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 1 },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x340 }, /* SGPRS=13 (112 GPRS) */
+ { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
+};
+
+static const struct soc15_reg_entry sec_ded_counter_registers[] = {
+ { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED) },
+ { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO) },
+ { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2) },
+ { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2) },
+ { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT) },
+ { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2) },
+ { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT) },
+};
+
+static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
+{
+ struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
+ struct amdgpu_ib ib;
+ struct dma_fence *f = NULL;
+ int r, i, j;
+ u32 tmp;
+ unsigned total_size, vgpr_offset, sgpr_offset;
+ u64 gpu_addr;
+
+ /* only support when RAS is enabled */
+ if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
+ return 0;
+
+ /* bail if the compute ring is not ready */
+ if (!ring->sched.ready)
+ return 0;
+
+ tmp = RREG32_SOC15(GC, 0, mmGB_EDC_MODE);
+ WREG32_SOC15(GC, 0, mmGB_EDC_MODE, 0);
+
+ total_size =
+ ((ARRAY_SIZE(vgpr_init_regs) * 3) + 4 + 5 + 2) * 4;
+ total_size +=
+ ((ARRAY_SIZE(sgpr_init_regs) * 3) + 4 + 5 + 2) * 4;
+ total_size = ALIGN(total_size, 256);
+ vgpr_offset = total_size;
+ total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
+ sgpr_offset = total_size;
+ total_size += sizeof(sgpr_init_compute_shader);
+
+ /* allocate an indirect buffer to put the commands in */
+ memset(&ib, 0, sizeof(ib));
+ r = amdgpu_ib_get(adev, NULL, total_size, &ib);
+ if (r) {
+ DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
+ return r;
+ }
+
+ /* load the compute shaders */
+ for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
+ ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
+
+ for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
+ ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
+
+ /* init the ib length to 0 */
+ ib.length_dw = 0;
+
+ /* VGPR */
+ /* write the register state for the compute dispatch */
+ for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i++) {
+ ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
+ ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs[i])
+ - PACKET3_SET_SH_REG_START;
+ ib.ptr[ib.length_dw++] = vgpr_init_regs[i].reg_value;
+ }
+ /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
+ gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
+ ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
+ ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
+ - PACKET3_SET_SH_REG_START;
+ ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
+ ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
+
+ /* write dispatch packet */
+ ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
+ ib.ptr[ib.length_dw++] = 128; /* x */
+ ib.ptr[ib.length_dw++] = 1; /* y */
+ ib.ptr[ib.length_dw++] = 1; /* z */
+ ib.ptr[ib.length_dw++] =
+ REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
+
+ /* write CS partial flush packet */
+ ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
+ ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
+
+ /* SGPR */
+ /* write the register state for the compute dispatch */
+ for (i = 0; i < ARRAY_SIZE(sgpr_init_regs); i++) {
+ ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
+ ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr_init_regs[i])
+ - PACKET3_SET_SH_REG_START;
+ ib.ptr[ib.length_dw++] = sgpr_init_regs[i].reg_value;
+ }
+ /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
+ gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
+ ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
+ ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
+ - PACKET3_SET_SH_REG_START;
+ ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
+ ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
+
+ /* write dispatch packet */
+ ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
+ ib.ptr[ib.length_dw++] = 128; /* x */
+ ib.ptr[ib.length_dw++] = 1; /* y */
+ ib.ptr[ib.length_dw++] = 1; /* z */
+ ib.ptr[ib.length_dw++] =
+ REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
+
+ /* write CS partial flush packet */
+ ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
+ ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
+
+ /* shedule the ib on the ring */
+ r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
+ if (r) {
+ DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
+ goto fail;
+ }
+
+ /* wait for the GPU to finish processing the IB */
+ r = dma_fence_wait(f, false);
+ if (r) {
+ DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
+ goto fail;
+ }
+
+ /* read back registers to clear the counters */
+ mutex_lock(&adev->grbm_idx_mutex);
+ for (j = 0; j < 16; j++) {
+ gfx_v9_0_select_se_sh(adev, 0x01, 0x0, j);
+ for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
+ RREG32(SOC15_REG_ENTRY_OFFSET(sec_ded_counter_registers[i]));
+ gfx_v9_0_select_se_sh(adev, 0x02, 0x0, j);
+ for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
+ RREG32(SOC15_REG_ENTRY_OFFSET(sec_ded_counter_registers[i]));
+ gfx_v9_0_select_se_sh(adev, 0x03, 0x0, j);
+ for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
+ RREG32(SOC15_REG_ENTRY_OFFSET(sec_ded_counter_registers[i]));
+ gfx_v9_0_select_se_sh(adev, 0x04, 0x0, j);
+ for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
+ RREG32(SOC15_REG_ENTRY_OFFSET(sec_ded_counter_registers[i]));
+ }
+ WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000);
+ mutex_unlock(&adev->grbm_idx_mutex);
+
+fail:
+ amdgpu_ib_free(adev, &ib, NULL);
+ dma_fence_put(f);
+
+ return r;
+}
+
static int gfx_v9_0_early_init(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -3577,6 +3817,11 @@ static int gfx_v9_0_ecc_late_init(void *handle)
if (!*ras_if)
return -ENOMEM;
+ /* requires IBs so do in late init after IB pool is initialized */
+ r = gfx_v9_0_do_edc_gpr_workarounds(adev);
+ if (r)
+ return r;
+
**ras_if = ras_block;
r = amdgpu_ras_feature_enable_on_boot(adev, *ras_if, 1);
diff --git a/drivers/gpu/drm/amd/amdgpu/soc15.h b/drivers/gpu/drm/amd/amdgpu/soc15.h
index a66c8bf..06f39f5 100644
--- a/drivers/gpu/drm/amd/amdgpu/soc15.h
+++ b/drivers/gpu/drm/amd/amdgpu/soc15.h
@@ -42,8 +42,18 @@ struct soc15_reg_golden {
u32 or_mask;
};
+struct soc15_reg_entry {
+ uint32_t hwip;
+ uint32_t inst;
+ uint32_t seg;
+ uint32_t reg_offset;
+ uint32_t reg_value;
+};
+
#define SOC15_REG_ENTRY(ip, inst, reg) ip##_HWIP, inst, reg##_BASE_IDX, reg
+#define SOC15_REG_ENTRY_OFFSET(entry) (adev->reg_offset[entry.hwip][entry.inst][entry.seg] + entry.reg_offset)
+
#define SOC15_REG_GOLDEN_VALUE(ip, inst, reg, and_mask, or_mask) \
{ ip##_HWIP, inst, reg##_BASE_IDX, reg, and_mask, or_mask }
--
2.7.4
More information about the amd-gfx
mailing list