drm/amdgpu: Add check to prevent IH overflow
[linux-2.6-microblaze.git] / drivers / gpu / drm / amd / amdgpu / gmc_v9_0.c
index 3ebbddb..aedef90 100644 (file)
@@ -31,8 +31,6 @@
 #include "amdgpu_atomfirmware.h"
 #include "amdgpu_gem.h"
 
-#include "hdp/hdp_4_0_offset.h"
-#include "hdp/hdp_4_0_sh_mask.h"
 #include "gc/gc_9_0_sh_mask.h"
 #include "dce/dce_12_0_offset.h"
 #include "dce/dce_12_0_sh_mask.h"
@@ -283,20 +281,6 @@ static const char *mmhub_client_ids_arcturus[][2] = {
        [224+15][1] = "SDMA7",
 };
 
-static const u32 golden_settings_vega10_hdp[] =
-{
-       0xf64, 0x0fffffff, 0x00000000,
-       0xf65, 0x0fffffff, 0x00000000,
-       0xf66, 0x0fffffff, 0x00000000,
-       0xf67, 0x0fffffff, 0x00000000,
-       0xf68, 0x0fffffff, 0x00000000,
-       0xf6a, 0x0fffffff, 0x00000000,
-       0xf6b, 0x0fffffff, 0x00000000,
-       0xf6c, 0x0fffffff, 0x00000000,
-       0xf6d, 0x0fffffff, 0x00000000,
-       0xf6e, 0x0fffffff, 0x00000000,
-};
-
 static const struct soc15_reg_golden golden_settings_mmhub_1_0_0[] =
 {
        SOC15_REG_GOLDEN_VALUE(MMHUB, 0, mmDAGB1_WRCLI2, 0x00000007, 0xfe5fe0fa),
@@ -379,41 +363,6 @@ static const uint32_t ecc_umc_mcumc_ctrl_mask_addrs[] = {
        (0x001d43e0 + 0x00001800),
 };
 
-static const uint32_t ecc_umc_mcumc_status_addrs[] = {
-       (0x000143c2 + 0x00000000),
-       (0x000143c2 + 0x00000800),
-       (0x000143c2 + 0x00001000),
-       (0x000143c2 + 0x00001800),
-       (0x000543c2 + 0x00000000),
-       (0x000543c2 + 0x00000800),
-       (0x000543c2 + 0x00001000),
-       (0x000543c2 + 0x00001800),
-       (0x000943c2 + 0x00000000),
-       (0x000943c2 + 0x00000800),
-       (0x000943c2 + 0x00001000),
-       (0x000943c2 + 0x00001800),
-       (0x000d43c2 + 0x00000000),
-       (0x000d43c2 + 0x00000800),
-       (0x000d43c2 + 0x00001000),
-       (0x000d43c2 + 0x00001800),
-       (0x001143c2 + 0x00000000),
-       (0x001143c2 + 0x00000800),
-       (0x001143c2 + 0x00001000),
-       (0x001143c2 + 0x00001800),
-       (0x001543c2 + 0x00000000),
-       (0x001543c2 + 0x00000800),
-       (0x001543c2 + 0x00001000),
-       (0x001543c2 + 0x00001800),
-       (0x001943c2 + 0x00000000),
-       (0x001943c2 + 0x00000800),
-       (0x001943c2 + 0x00001000),
-       (0x001943c2 + 0x00001800),
-       (0x001d43c2 + 0x00000000),
-       (0x001d43c2 + 0x00000800),
-       (0x001d43c2 + 0x00001000),
-       (0x001d43c2 + 0x00001800),
-};
-
 static int gmc_v9_0_ecc_interrupt_state(struct amdgpu_device *adev,
                struct amdgpu_irq_src *src,
                unsigned type,
@@ -502,6 +451,7 @@ static int gmc_v9_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
                                WREG32(reg, tmp);
                        }
                }
+               break;
        default:
                break;
        }
@@ -510,122 +460,136 @@ static int gmc_v9_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
 }
 
 static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
-                               struct amdgpu_irq_src *source,
-                               struct amdgpu_iv_entry *entry)
+                                     struct amdgpu_irq_src *source,
+                                     struct amdgpu_iv_entry *entry)
 {
-       struct amdgpu_vmhub *hub;
        bool retry_fault = !!(entry->src_data[1] & 0x80);
        uint32_t status = 0, cid = 0, rw = 0;
-       u64 addr;
-       char hub_name[10];
+       struct amdgpu_task_info task_info;
+       struct amdgpu_vmhub *hub;
        const char *mmhub_cid;
+       const char *hub_name;
+       u64 addr;
 
        addr = (u64)entry->src_data[0] << 12;
        addr |= ((u64)entry->src_data[1] & 0xf) << 44;
 
-       if (retry_fault && amdgpu_gmc_filter_faults(adev, addr, entry->pasid,
-                                                   entry->timestamp))
-               return 1; /* This also prevents sending it to KFD */
+       if (retry_fault) {
+               /* Returning 1 here also prevents sending the IV to the KFD */
+
+               /* Process it onyl if it's the first fault for this address */
+               if (entry->ih != &adev->irq.ih_soft &&
+                   amdgpu_gmc_filter_faults(adev, addr, entry->pasid,
+                                            entry->timestamp))
+                       return 1;
+
+               /* Delegate it to a different ring if the hardware hasn't
+                * already done it.
+                */
+               if (in_interrupt()) {
+                       amdgpu_irq_delegate(adev, entry, 8);
+                       return 1;
+               }
+
+               /* Try to handle the recoverable page faults by filling page
+                * tables
+                */
+               if (amdgpu_vm_handle_fault(adev, entry->pasid, addr))
+                       return 1;
+       }
+
+       if (!printk_ratelimit())
+               return 0;
 
        if (entry->client_id == SOC15_IH_CLIENTID_VMC) {
-               snprintf(hub_name, sizeof(hub_name), "mmhub0");
+               hub_name = "mmhub0";
                hub = &adev->vmhub[AMDGPU_MMHUB_0];
        } else if (entry->client_id == SOC15_IH_CLIENTID_VMC1) {
-               snprintf(hub_name, sizeof(hub_name), "mmhub1");
+               hub_name = "mmhub1";
                hub = &adev->vmhub[AMDGPU_MMHUB_1];
        } else {
-               snprintf(hub_name, sizeof(hub_name), "gfxhub0");
+               hub_name = "gfxhub0";
                hub = &adev->vmhub[AMDGPU_GFXHUB_0];
        }
 
-       /* If it's the first fault for this address, process it normally */
-       if (retry_fault && !in_interrupt() &&
-           amdgpu_vm_handle_fault(adev, entry->pasid, addr))
-               return 1; /* This also prevents sending it to KFD */
+       memset(&task_info, 0, sizeof(struct amdgpu_task_info));
+       amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
 
-       if (!amdgpu_sriov_vf(adev)) {
-               /*
-                * Issue a dummy read to wait for the status register to
-                * be updated to avoid reading an incorrect value due to
-                * the new fast GRBM interface.
-                */
-               if (entry->vmid_src == AMDGPU_GFXHUB_0)
-                       RREG32(hub->vm_l2_pro_fault_status);
-
-               status = RREG32(hub->vm_l2_pro_fault_status);
-               cid = REG_GET_FIELD(status,
-                                   VM_L2_PROTECTION_FAULT_STATUS, CID);
-               rw = REG_GET_FIELD(status,
-                                  VM_L2_PROTECTION_FAULT_STATUS, RW);
-               WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
-       }
+       dev_err(adev->dev,
+               "[%s] %s page fault (src_id:%u ring:%u vmid:%u "
+               "pasid:%u, for process %s pid %d thread %s pid %d)\n",
+               hub_name, retry_fault ? "retry" : "no-retry",
+               entry->src_id, entry->ring_id, entry->vmid,
+               entry->pasid, task_info.process_name, task_info.tgid,
+               task_info.task_name, task_info.pid);
+       dev_err(adev->dev, "  in page starting at address 0x%012llx from client %d\n",
+               addr, entry->client_id);
 
-       if (printk_ratelimit()) {
-               struct amdgpu_task_info task_info;
-
-               memset(&task_info, 0, sizeof(struct amdgpu_task_info));
-               amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
-
-               dev_err(adev->dev,
-                       "[%s] %s page fault (src_id:%u ring:%u vmid:%u "
-                       "pasid:%u, for process %s pid %d thread %s pid %d)\n",
-                       hub_name, retry_fault ? "retry" : "no-retry",
-                       entry->src_id, entry->ring_id, entry->vmid,
-                       entry->pasid, task_info.process_name, task_info.tgid,
-                       task_info.task_name, task_info.pid);
-               dev_err(adev->dev, "  in page starting at address 0x%016llx from client %d\n",
-                       addr, entry->client_id);
-               if (!amdgpu_sriov_vf(adev)) {
-                       dev_err(adev->dev,
-                               "VM_L2_PROTECTION_FAULT_STATUS:0x%08X\n",
-                               status);
-                       if (hub == &adev->vmhub[AMDGPU_GFXHUB_0]) {
-                               dev_err(adev->dev, "\t Faulty UTCL2 client ID: %s (0x%x)\n",
-                                       cid >= ARRAY_SIZE(gfxhub_client_ids) ? "unknown" : gfxhub_client_ids[cid],
-                                       cid);
-                       } else {
-                               switch (adev->asic_type) {
-                               case CHIP_VEGA10:
-                                       mmhub_cid = mmhub_client_ids_vega10[cid][rw];
-                                       break;
-                               case CHIP_VEGA12:
-                                       mmhub_cid = mmhub_client_ids_vega12[cid][rw];
-                                       break;
-                               case CHIP_VEGA20:
-                                       mmhub_cid = mmhub_client_ids_vega20[cid][rw];
-                                       break;
-                               case CHIP_ARCTURUS:
-                                       mmhub_cid = mmhub_client_ids_arcturus[cid][rw];
-                                       break;
-                               case CHIP_RAVEN:
-                                       mmhub_cid = mmhub_client_ids_raven[cid][rw];
-                                       break;
-                               case CHIP_RENOIR:
-                                       mmhub_cid = mmhub_client_ids_renoir[cid][rw];
-                                       break;
-                               default:
-                                       mmhub_cid = NULL;
-                                       break;
-                               }
-                               dev_err(adev->dev, "\t Faulty UTCL2 client ID: %s (0x%x)\n",
-                                       mmhub_cid ? mmhub_cid : "unknown", cid);
-                       }
-                       dev_err(adev->dev, "\t MORE_FAULTS: 0x%lx\n",
-                               REG_GET_FIELD(status,
-                               VM_L2_PROTECTION_FAULT_STATUS, MORE_FAULTS));
-                       dev_err(adev->dev, "\t WALKER_ERROR: 0x%lx\n",
-                               REG_GET_FIELD(status,
-                               VM_L2_PROTECTION_FAULT_STATUS, WALKER_ERROR));
-                       dev_err(adev->dev, "\t PERMISSION_FAULTS: 0x%lx\n",
-                               REG_GET_FIELD(status,
-                               VM_L2_PROTECTION_FAULT_STATUS, PERMISSION_FAULTS));
-                       dev_err(adev->dev, "\t MAPPING_ERROR: 0x%lx\n",
-                               REG_GET_FIELD(status,
-                               VM_L2_PROTECTION_FAULT_STATUS, MAPPING_ERROR));
-                       dev_err(adev->dev, "\t RW: 0x%x\n", rw);
+       if (amdgpu_sriov_vf(adev))
+               return 0;
+
+       /*
+        * Issue a dummy read to wait for the status register to
+        * be updated to avoid reading an incorrect value due to
+        * the new fast GRBM interface.
+        */
+       if (entry->vmid_src == AMDGPU_GFXHUB_0)
+               RREG32(hub->vm_l2_pro_fault_status);
+
+       status = RREG32(hub->vm_l2_pro_fault_status);
+       cid = REG_GET_FIELD(status, VM_L2_PROTECTION_FAULT_STATUS, CID);
+       rw = REG_GET_FIELD(status, VM_L2_PROTECTION_FAULT_STATUS, RW);
+       WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
+
+
+       dev_err(adev->dev,
+               "VM_L2_PROTECTION_FAULT_STATUS:0x%08X\n",
+               status);
+       if (hub == &adev->vmhub[AMDGPU_GFXHUB_0]) {
+               dev_err(adev->dev, "\t Faulty UTCL2 client ID: %s (0x%x)\n",
+                       cid >= ARRAY_SIZE(gfxhub_client_ids) ? "unknown" :
+                       gfxhub_client_ids[cid],
+                       cid);
+       } else {
+               switch (adev->asic_type) {
+               case CHIP_VEGA10:
+                       mmhub_cid = mmhub_client_ids_vega10[cid][rw];
+                       break;
+               case CHIP_VEGA12:
+                       mmhub_cid = mmhub_client_ids_vega12[cid][rw];
+                       break;
+               case CHIP_VEGA20:
+                       mmhub_cid = mmhub_client_ids_vega20[cid][rw];
+                       break;
+               case CHIP_ARCTURUS:
+                       mmhub_cid = mmhub_client_ids_arcturus[cid][rw];
+                       break;
+               case CHIP_RAVEN:
+                       mmhub_cid = mmhub_client_ids_raven[cid][rw];
+                       break;
+               case CHIP_RENOIR:
+                       mmhub_cid = mmhub_client_ids_renoir[cid][rw];
+                       break;
+               default:
+                       mmhub_cid = NULL;
+                       break;
                }
+               dev_err(adev->dev, "\t Faulty UTCL2 client ID: %s (0x%x)\n",
+                       mmhub_cid ? mmhub_cid : "unknown", cid);
        }
-
+       dev_err(adev->dev, "\t MORE_FAULTS: 0x%lx\n",
+               REG_GET_FIELD(status,
+               VM_L2_PROTECTION_FAULT_STATUS, MORE_FAULTS));
+       dev_err(adev->dev, "\t WALKER_ERROR: 0x%lx\n",
+               REG_GET_FIELD(status,
+               VM_L2_PROTECTION_FAULT_STATUS, WALKER_ERROR));
+       dev_err(adev->dev, "\t PERMISSION_FAULTS: 0x%lx\n",
+               REG_GET_FIELD(status,
+               VM_L2_PROTECTION_FAULT_STATUS, PERMISSION_FAULTS));
+       dev_err(adev->dev, "\t MAPPING_ERROR: 0x%lx\n",
+               REG_GET_FIELD(status,
+               VM_L2_PROTECTION_FAULT_STATUS, MAPPING_ERROR));
+       dev_err(adev->dev, "\t RW: 0x%x\n", rw);
        return 0;
 }
 
@@ -711,6 +675,7 @@ static bool gmc_v9_0_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
  *
  * @adev: amdgpu_device pointer
  * @vmid: vm instance to flush
+ * @vmhub: which hub to flush
  * @flush_type: the flush type
  *
  * Flush the TLB for the requested page table using certain type.
@@ -827,6 +792,8 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
  *
  * @adev: amdgpu_device pointer
  * @pasid: pasid to be flush
+ * @flush_type: the flush type
+ * @all_hub: flush all hubs
  *
  * Flush the TLB for the requested pasid.
  */
@@ -1166,15 +1133,7 @@ static void gmc_v9_0_set_mmhub_funcs(struct amdgpu_device *adev)
 
 static void gmc_v9_0_set_gfxhub_funcs(struct amdgpu_device *adev)
 {
-       switch (adev->asic_type) {
-       case CHIP_ARCTURUS:
-       case CHIP_VEGA20:
-               adev->gfxhub.funcs = &gfxhub_v1_1_funcs;
-               break;
-       default:
-               adev->gfxhub.funcs = &gfxhub_v1_0_funcs;
-               break;
-       }
+       adev->gfxhub.funcs = &gfxhub_v1_0_funcs;
 }
 
 static int gmc_v9_0_early_init(void *handle)
@@ -1202,8 +1161,6 @@ static int gmc_v9_0_late_init(void *handle)
        struct amdgpu_device *adev = (struct amdgpu_device *)handle;
        int r;
 
-       amdgpu_bo_late_init(adev);
-
        r = amdgpu_gmc_allocate_vm_inv_eng(adev);
        if (r)
                return r;
@@ -1598,46 +1555,24 @@ static int gmc_v9_0_hw_init(void *handle)
        struct amdgpu_device *adev = (struct amdgpu_device *)handle;
        bool value;
        int r, i;
-       u32 tmp;
 
        /* The sequence of these two function calls matters.*/
        gmc_v9_0_init_golden_registers(adev);
 
        if (adev->mode_info.num_crtc) {
-               if (adev->asic_type != CHIP_ARCTURUS) {
-                       /* Lockout access through VGA aperture*/
-                       WREG32_FIELD15(DCE, 0, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
-
-                       /* disable VGA render */
-                       WREG32_FIELD15(DCE, 0, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
-               }
+               /* Lockout access through VGA aperture*/
+               WREG32_FIELD15(DCE, 0, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
+               /* disable VGA render */
+               WREG32_FIELD15(DCE, 0, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
        }
 
-       amdgpu_device_program_register_sequence(adev,
-                                               golden_settings_vega10_hdp,
-                                               ARRAY_SIZE(golden_settings_vega10_hdp));
-
        if (adev->mmhub.funcs->update_power_gating)
                adev->mmhub.funcs->update_power_gating(adev, true);
 
-       switch (adev->asic_type) {
-       case CHIP_ARCTURUS:
-               WREG32_FIELD15(HDP, 0, HDP_MMHUB_CNTL, HDP_MMHUB_GCC, 1);
-               break;
-       default:
-               break;
-       }
-
-       WREG32_FIELD15(HDP, 0, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 1);
-
-       tmp = RREG32_SOC15(HDP, 0, mmHDP_HOST_PATH_CNTL);
-       WREG32_SOC15(HDP, 0, mmHDP_HOST_PATH_CNTL, tmp);
-
-       WREG32_SOC15(HDP, 0, mmHDP_NONSURFACE_BASE, (adev->gmc.vram_start >> 8));
-       WREG32_SOC15(HDP, 0, mmHDP_NONSURFACE_BASE_HI, (adev->gmc.vram_start >> 40));
+       adev->hdp.funcs->init_registers(adev);
 
        /* After HDP is initialized, flush HDP.*/
-       adev->nbio.funcs->hdp_flush(adev, NULL);
+       adev->hdp.funcs->flush_hdp(adev, NULL);
 
        if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
                value = false;