diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h | 62 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm | 37 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 12 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 85 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_events.c | 11 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 7 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 8 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.h | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process.c | 12 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_queue.c | 13 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 68 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.h | 1 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 8 |
13 files changed, 189 insertions, 136 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h index 0320163b6e74..f98c735b2905 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler.h @@ -3644,14 +3644,18 @@ static const uint32_t cwsr_trap_gfx9_4_3_hex[] = { }; static const uint32_t cwsr_trap_gfx12_hex[] = { - 0xbfa00001, 0xbfa002a2, - 0xb0804009, 0xb8f8f804, + 0xbfa00001, 0xbfa002b2, + 0xb0804009, 0xb8eef81a, + 0xbf880000, 0xb980081a, + 0x00000000, 0xb8f8f804, + 0x9177ff77, 0x0c000000, + 0x846e9a6e, 0x8c776e77, 0x9178ff78, 0x00008c00, 0xb8fbf811, 0x8b6eff78, 0x00004000, 0xbfa10008, 0x8b6eff7b, 0x00000080, 0xbfa20018, 0x8b6ea07b, - 0xbfa20042, 0xbf830010, + 0xbfa2004a, 0xbf830010, 0xb8fbf811, 0xbfa0fffb, 0x8b6eff7b, 0x00000bd0, 0xbfa20010, 0xb8eef812, @@ -3662,28 +3666,32 @@ static const uint32_t cwsr_trap_gfx12_hex[] = { 0xf0000000, 0xbfa20005, 0x8b6fff6f, 0x00000200, 0xbfa20002, 0x8b6ea07b, - 0xbfa2002c, 0xbefa4d82, + 0xbfa20034, 0xbefa4d82, 0xbf8a0000, 0x84fa887a, 0xbf0d8f7b, 0xbfa10002, 0x8c7bff7b, 0xffff0000, - 0xf4601bbd, 0xf8000010, - 0xbf8a0000, 0x846e976e, - 0x9177ff77, 0x00800000, - 0x8c776e77, 0xf4603bbd, - 0xf8000000, 0xbf8a0000, - 0xf4603ebd, 0xf8000008, - 0xbf8a0000, 0x8bee6e6e, - 0xbfa10001, 0xbe80486e, - 0x8b6eff6d, 0xf0000000, - 0xbfa20009, 0xb8eef811, - 0x8b6eff6e, 0x00000080, - 0xbfa20007, 0x8c78ff78, - 0x00004000, 0x80ec886c, - 0x82ed806d, 0xbfa00002, - 0x806c846c, 0x826d806d, - 0x8b6dff6d, 0x0000ffff, - 0x8bfe7e7e, 0x8bea6a6a, - 0x85788978, 0xb9783244, + 0x8b6eff77, 0x0c000000, + 0x916dff6d, 0x0c000000, + 0x8c6d6e6d, 0xf4601bbd, + 0xf8000010, 0xbf8a0000, + 0x846e976e, 0x9177ff77, + 0x00800000, 0x8c776e77, + 0xf4603bbd, 0xf8000000, + 0xbf8a0000, 0xf4603ebd, + 0xf8000008, 0xbf8a0000, + 0x8bee6e6e, 0xbfa10001, + 0xbe80486e, 0x8b6eff6d, + 0xf0000000, 0xbfa20009, + 0xb8eef811, 0x8b6eff6e, + 0x00000080, 0xbfa20007, + 0x8c78ff78, 0x00004000, + 0x80ec886c, 0x82ed806d, + 0xbfa00002, 0x806c846c, + 0x826d806d, 0x8b6dff6d, + 0x0000ffff, 0x8bfe7e7e, + 0x8bea6a6a, 0x85788978, + 0x936eff77, 0x0002001a, + 0xb96ef81a, 0xb9783244, 0xbe804a6c, 0xb8faf802, 0xbf0d987a, 0xbfa10001, 0xbfb00000, 0x8b6dff6d, @@ -3981,7 +3989,7 @@ static const uint32_t cwsr_trap_gfx12_hex[] = { 0x008ce800, 0x00000000, 0x807d817d, 0x8070ff70, 0x00000080, 0xbf0a7b7d, - 0xbfa2fff7, 0xbfa0016e, + 0xbfa2fff7, 0xbfa00171, 0xbef4007e, 0x8b75ff7f, 0x0000ffff, 0x8c75ff75, 0x00040000, 0xbef60080, @@ -4163,12 +4171,14 @@ static const uint32_t cwsr_trap_gfx12_hex[] = { 0xf8000074, 0xbf8a0000, 0x8b6dff6d, 0x0000ffff, 0x8bfe7e7e, 0x8bea6a6a, - 0xb97af804, 0xbe804ec2, - 0xbf94fffe, 0xbe804a6c, + 0x936eff77, 0x0002001a, + 0xb96ef81a, 0xb97af804, 0xbe804ec2, 0xbf94fffe, - 0xbfb10000, 0xbf9f0000, + 0xbe804a6c, 0xbe804ec2, + 0xbf94fffe, 0xbfb10000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, 0xbf9f0000, + 0xbf9f0000, 0x00000000, }; static const uint32_t cwsr_trap_gfx9_5_0_hex[] = { diff --git a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm index 5a1a1b1f897f..07999b4649de 100644 --- a/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm +++ b/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx12.asm @@ -78,9 +78,16 @@ var SQ_WAVE_EXCP_FLAG_PRIV_RESTORE_PART_2_SHIFT = SQ_WAVE_EXCP_FLAG_PRIV_ILLEGAL var SQ_WAVE_EXCP_FLAG_PRIV_RESTORE_PART_2_SIZE = SQ_WAVE_EXCP_FLAG_PRIV_HOST_TRAP_SHIFT - SQ_WAVE_EXCP_FLAG_PRIV_ILLEGAL_INST_SHIFT var SQ_WAVE_EXCP_FLAG_PRIV_RESTORE_PART_3_SHIFT = SQ_WAVE_EXCP_FLAG_PRIV_WAVE_START_SHIFT var SQ_WAVE_EXCP_FLAG_PRIV_RESTORE_PART_3_SIZE = 32 - SQ_WAVE_EXCP_FLAG_PRIV_RESTORE_PART_3_SHIFT + +var SQ_WAVE_SCHED_MODE_DEP_MODE_SHIFT = 0 +var SQ_WAVE_SCHED_MODE_DEP_MODE_SIZE = 2 + var BARRIER_STATE_SIGNAL_OFFSET = 16 var BARRIER_STATE_VALID_OFFSET = 0 +var TTMP11_SCHED_MODE_SHIFT = 26 +var TTMP11_SCHED_MODE_SIZE = 2 +var TTMP11_SCHED_MODE_MASK = 0xC000000 var TTMP11_DEBUG_TRAP_ENABLED_SHIFT = 23 var TTMP11_DEBUG_TRAP_ENABLED_MASK = 0x800000 @@ -160,8 +167,19 @@ L_JUMP_TO_RESTORE: s_branch L_RESTORE L_SKIP_RESTORE: + // Assume most relaxed scheduling mode is set. Save and revert to normal mode. + s_getreg_b32 ttmp2, hwreg(HW_REG_WAVE_SCHED_MODE) + s_wait_alu 0 + s_setreg_imm32_b32 hwreg(HW_REG_WAVE_SCHED_MODE, \ + SQ_WAVE_SCHED_MODE_DEP_MODE_SHIFT, SQ_WAVE_SCHED_MODE_DEP_MODE_SIZE), 0 + s_getreg_b32 s_save_state_priv, hwreg(HW_REG_WAVE_STATE_PRIV) //save STATUS since we will change SCC + // Save SCHED_MODE[1:0] into ttmp11[27:26]. + s_andn2_b32 ttmp11, ttmp11, TTMP11_SCHED_MODE_MASK + s_lshl_b32 ttmp2, ttmp2, TTMP11_SCHED_MODE_SHIFT + s_or_b32 ttmp11, ttmp11, ttmp2 + // Clear SPI_PRIO: do not save with elevated priority. // Clear ECC_ERR: prevents SQC store and triggers FATAL_HALT if setreg'd. s_andn2_b32 s_save_state_priv, s_save_state_priv, SQ_WAVE_STATE_PRIV_ALWAYS_CLEAR_MASK @@ -238,6 +256,13 @@ L_FETCH_2ND_TRAP: s_cbranch_scc0 L_NO_SIGN_EXTEND_TMA s_or_b32 ttmp15, ttmp15, 0xFFFF0000 L_NO_SIGN_EXTEND_TMA: +#if ASIC_FAMILY == CHIP_GFX12 + // Move SCHED_MODE[1:0] from ttmp11 to unused bits in ttmp1[27:26] (return PC_HI). + // The second-level trap will restore from ttmp1 for backwards compatibility. + s_and_b32 ttmp2, ttmp11, TTMP11_SCHED_MODE_MASK + s_andn2_b32 ttmp1, ttmp1, TTMP11_SCHED_MODE_MASK + s_or_b32 ttmp1, ttmp1, ttmp2 +#endif s_load_dword ttmp2, [ttmp14, ttmp15], 0x10 scope:SCOPE_SYS // debug trap enabled flag s_wait_idle @@ -287,6 +312,10 @@ L_EXIT_TRAP: // STATE_PRIV.BARRIER_COMPLETE may have changed since we read it. // Only restore fields which the trap handler changes. s_lshr_b32 s_save_state_priv, s_save_state_priv, SQ_WAVE_STATE_PRIV_SCC_SHIFT + + // Assume relaxed scheduling mode after this point. + restore_sched_mode(ttmp2) + s_setreg_b32 hwreg(HW_REG_WAVE_STATE_PRIV, SQ_WAVE_STATE_PRIV_SCC_SHIFT, \ SQ_WAVE_STATE_PRIV_POISON_ERR_SHIFT - SQ_WAVE_STATE_PRIV_SCC_SHIFT + 1), s_save_state_priv @@ -1043,6 +1072,9 @@ L_SKIP_BARRIER_RESTORE: s_and_b64 exec, exec, exec // Restore STATUS.EXECZ, not writable by s_setreg_b32 s_and_b64 vcc, vcc, vcc // Restore STATUS.VCCZ, not writable by s_setreg_b32 + // Assume relaxed scheduling mode after this point. + restore_sched_mode(s_restore_tmp) + s_setreg_b32 hwreg(HW_REG_WAVE_STATE_PRIV), s_restore_state_priv // SCC is included, which is changed by previous salu // Make barrier and LDS state visible to all waves in the group. @@ -1134,3 +1166,8 @@ function valu_sgpr_hazard end #endif end + +function restore_sched_mode(s_tmp) + s_bfe_u32 s_tmp, ttmp11, (TTMP11_SCHED_MODE_SHIFT | (TTMP11_SCHED_MODE_SIZE << 0x10)) + s_setreg_b32 hwreg(HW_REG_WAVE_SCHED_MODE), s_tmp +end diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 0f0719528bcc..22925df6a791 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -2826,7 +2826,7 @@ retry: static int runtime_disable(struct kfd_process *p) { - int i = 0, ret; + int i = 0, ret = 0; bool was_enabled = p->runtime_info.runtime_state == DEBUG_RUNTIME_STATE_ENABLED; p->runtime_info.runtime_state = DEBUG_RUNTIME_STATE_DISABLED; @@ -2863,6 +2863,7 @@ static int runtime_disable(struct kfd_process *p) /* disable ttmp setup */ for (i = 0; i < p->n_pdds; i++) { struct kfd_process_device *pdd = p->pdds[i]; + int last_err = 0; if (kfd_dbg_is_per_vmid_supported(pdd->dev)) { pdd->spi_dbg_override = @@ -2872,14 +2873,17 @@ static int runtime_disable(struct kfd_process *p) pdd->dev->vm_info.last_vmid_kfd); if (!pdd->dev->kfd->shared_resources.enable_mes) - debug_refresh_runlist(pdd->dev->dqm); + last_err = debug_refresh_runlist(pdd->dev->dqm); else - kfd_dbg_set_mes_debug_mode(pdd, + last_err = kfd_dbg_set_mes_debug_mode(pdd, !kfd_dbg_has_cwsr_workaround(pdd->dev)); + + if (last_err) + ret = last_err; } } - return 0; + return ret; } static int kfd_ioctl_runtime_enable(struct file *filep, struct kfd_process *p, void *data) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 6c5c7c1bf5ed..d7a2e7178ea9 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -1209,6 +1209,15 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm, pr_debug_ratelimited("Evicting process pid %d queues\n", pdd->process->lead_thread->pid); + if (dqm->dev->kfd->shared_resources.enable_mes) { + pdd->last_evict_timestamp = get_jiffies_64(); + retval = suspend_all_queues_mes(dqm); + if (retval) { + dev_err(dev, "Suspending all queues failed"); + goto out; + } + } + /* Mark all queues as evicted. Deactivate all active queues on * the qpd. */ @@ -1221,23 +1230,27 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm, decrement_queue_count(dqm, qpd, q); if (dqm->dev->kfd->shared_resources.enable_mes) { - int err; - - err = remove_queue_mes(dqm, q, qpd); - if (err) { + retval = remove_queue_mes(dqm, q, qpd); + if (retval) { dev_err(dev, "Failed to evict queue %d\n", q->properties.queue_id); - retval = err; + goto out; } } } - pdd->last_evict_timestamp = get_jiffies_64(); - if (!dqm->dev->kfd->shared_resources.enable_mes) + + if (!dqm->dev->kfd->shared_resources.enable_mes) { + pdd->last_evict_timestamp = get_jiffies_64(); retval = execute_queues_cpsch(dqm, qpd->is_debug ? KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES : KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0, USE_DEFAULT_GRACE_PERIOD); + } else { + retval = resume_all_queues_mes(dqm); + if (retval) + dev_err(dev, "Resuming all queues failed"); + } out: dqm_unlock(dqm); @@ -1884,6 +1897,8 @@ fail_packet_manager_init: static int stop_cpsch(struct device_queue_manager *dqm) { + int ret = 0; + dqm_lock(dqm); if (!dqm->sched_running) { dqm_unlock(dqm); @@ -1891,9 +1906,10 @@ static int stop_cpsch(struct device_queue_manager *dqm) } if (!dqm->dev->kfd->shared_resources.enable_mes) - unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0, USE_DEFAULT_GRACE_PERIOD, false); + ret = unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, + 0, USE_DEFAULT_GRACE_PERIOD, false); else - remove_all_kfd_queues_mes(dqm); + ret = remove_all_kfd_queues_mes(dqm); dqm->sched_running = false; @@ -1907,7 +1923,7 @@ static int stop_cpsch(struct device_queue_manager *dqm) dqm->detect_hang_info = NULL; dqm_unlock(dqm); - return 0; + return ret; } static int create_kernel_queue_cpsch(struct device_queue_manager *dqm, @@ -2078,7 +2094,8 @@ int amdkfd_fence_wait_timeout(struct device_queue_manager *dqm, while (*fence_addr != fence_value) { /* Fatal err detected, this response won't come */ - if (amdgpu_amdkfd_is_fed(dqm->dev->adev)) + if (amdgpu_amdkfd_is_fed(dqm->dev->adev) || + amdgpu_in_reset(dqm->dev->adev)) return -EIO; if (time_after(jiffies, end_jiffies)) { @@ -3098,61 +3115,17 @@ out: return ret; } -static int kfd_dqm_evict_pasid_mes(struct device_queue_manager *dqm, - struct qcm_process_device *qpd) -{ - struct device *dev = dqm->dev->adev->dev; - int ret = 0; - - /* Check if process is already evicted */ - dqm_lock(dqm); - if (qpd->evicted) { - /* Increment the evicted count to make sure the - * process stays evicted before its terminated. - */ - qpd->evicted++; - dqm_unlock(dqm); - goto out; - } - dqm_unlock(dqm); - - ret = suspend_all_queues_mes(dqm); - if (ret) { - dev_err(dev, "Suspending all queues failed"); - goto out; - } - - ret = dqm->ops.evict_process_queues(dqm, qpd); - if (ret) { - dev_err(dev, "Evicting process queues failed"); - goto out; - } - - ret = resume_all_queues_mes(dqm); - if (ret) - dev_err(dev, "Resuming all queues failed"); - -out: - return ret; -} - int kfd_evict_process_device(struct kfd_process_device *pdd) { struct device_queue_manager *dqm; struct kfd_process *p; - int ret = 0; p = pdd->process; dqm = pdd->dev->dqm; WARN(debug_evictions, "Evicting pid %d", p->lead_thread->pid); - if (dqm->dev->kfd->shared_resources.enable_mes) - ret = kfd_dqm_evict_pasid_mes(dqm, &pdd->qpd); - else - ret = dqm->ops.evict_process_queues(dqm, &pdd->qpd); - - return ret; + return dqm->ops.evict_process_queues(dqm, &pdd->qpd); } int reserve_debug_trap_vmid(struct device_queue_manager *dqm, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c index 82905f3e54dd..5a190dd6be4e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c @@ -748,16 +748,6 @@ void kfd_signal_event_interrupt(u32 pasid, uint32_t partial_id, uint64_t *slots = page_slots(p->signal_page); uint32_t id; - /* - * If id is valid but slot is not signaled, GPU may signal the same event twice - * before driver have chance to process the first interrupt, then signal slot is - * auto-reset after set_event wakeup the user space, just drop the second event as - * the application only need wakeup once. - */ - if ((valid_id_bits > 31 || (1U << valid_id_bits) >= KFD_SIGNAL_EVENT_LIMIT) && - partial_id < KFD_SIGNAL_EVENT_LIMIT && slots[partial_id] == UNSIGNALED_EVENT_SLOT) - goto out_unlock; - if (valid_id_bits) pr_debug_ratelimited("Partial ID invalid: %u (%u valid bits)\n", partial_id, valid_id_bits); @@ -786,7 +776,6 @@ void kfd_signal_event_interrupt(u32 pasid, uint32_t partial_id, } } -out_unlock: rcu_read_unlock(); kfd_unref_process(p); } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c index 4ceb251312a6..d76fb61869c7 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -28,6 +28,7 @@ #include "kfd_device_queue_manager.h" #include "kfd_smi_events.h" #include "amdgpu_ras.h" +#include "amdgpu_ras_mgr.h" /* * GFX9 SQ Interrupts @@ -228,7 +229,11 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev, kfd_signal_poison_consumed_event(dev, pasid); - event_id = amdgpu_ras_acquire_event_id(dev->adev, type); + if (amdgpu_uniras_enabled(dev->adev)) + event_id = amdgpu_ras_mgr_gen_ras_event_seqno(dev->adev, + RAS_SEQNO_TYPE_POISON_CONSUMPTION); + else + event_id = amdgpu_ras_acquire_event_id(dev->adev, type); RAS_EVENT_LOG(dev->adev, event_id, "poison is consumed by client %d, kick off gpu reset flow\n", client_id); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 59a5a3fea65d..af53e796ea1b 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -21,7 +21,6 @@ * OTHER DEALINGS IN THE SOFTWARE. */ #include <linux/types.h> -#include <linux/hmm.h> #include <linux/dma-direction.h> #include <linux/dma-mapping.h> #include <linux/migrate.h> @@ -218,7 +217,7 @@ svm_migrate_get_vram_page(struct svm_range *prange, unsigned long pfn) page = pfn_to_page(pfn); svm_range_bo_ref(prange->svm_bo); page->zone_device_data = prange->svm_bo; - zone_device_page_init(page); + zone_device_page_init(page, 0); } static void @@ -568,8 +567,9 @@ out: return r < 0 ? r : 0; } -static void svm_migrate_page_free(struct page *page) +static void svm_migrate_folio_free(struct folio *folio) { + struct page *page = &folio->page; struct svm_range_bo *svm_bo = page->zone_device_data; if (svm_bo) { @@ -1009,7 +1009,7 @@ out_mmput: } static const struct dev_pagemap_ops svm_migrate_pgmap_ops = { - .page_free = svm_migrate_page_free, + .folio_free = svm_migrate_folio_free, .migrate_to_ram = svm_migrate_to_ram, }; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h index 2eebf67f9c2c..2b7fd442d29c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h @@ -31,7 +31,6 @@ #include <linux/list.h> #include <linux/mutex.h> #include <linux/sched/mm.h> -#include <linux/hmm.h> #include "kfd_priv.h" #include "kfd_svm.h" diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index ddfe30c13e9d..a085faac9fe1 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -1083,7 +1083,6 @@ static void kfd_process_destroy_pdds(struct kfd_process *p) * for auto suspend */ if (pdd->runtime_inuse) { - pm_runtime_mark_last_busy(adev_to_drm(pdd->dev->adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(pdd->dev->adev)->dev); pdd->runtime_inuse = false; } @@ -1162,9 +1161,6 @@ static void kfd_process_wq_release(struct work_struct *work) release_work); struct dma_fence *ef; - kfd_process_dequeue_from_all_devices(p); - pqm_uninit(&p->pqm); - /* * If GPU in reset, user queues may still running, wait for reset complete. */ @@ -1226,6 +1222,14 @@ static void kfd_process_notifier_release_internal(struct kfd_process *p) cancel_delayed_work_sync(&p->eviction_work); cancel_delayed_work_sync(&p->restore_work); + /* + * Dequeue and destroy user queues, it is not safe for GPU to access + * system memory after mmu release notifier callback returns because + * exit_mmap free process memory afterwards. + */ + kfd_process_dequeue_from_all_devices(p); + pqm_uninit(&p->pqm); + for (i = 0; i < p->n_pdds; i++) { struct kfd_process_device *pdd = p->pdds[i]; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_queue.c index a65c67cf56ff..80c4fa2b0975 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_queue.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_queue.c @@ -297,16 +297,16 @@ int kfd_queue_acquire_buffers(struct kfd_process_device *pdd, struct queue_prope goto out_err_unreserve; } - if (properties->ctx_save_restore_area_size != topo_dev->node_props.cwsr_size) { - pr_debug("queue cwsr size 0x%x not equal to node cwsr size 0x%x\n", + if (properties->ctx_save_restore_area_size < topo_dev->node_props.cwsr_size) { + pr_debug("queue cwsr size 0x%x not sufficient for node cwsr size 0x%x\n", properties->ctx_save_restore_area_size, topo_dev->node_props.cwsr_size); err = -EINVAL; goto out_err_unreserve; } - total_cwsr_size = (topo_dev->node_props.cwsr_size + topo_dev->node_props.debug_memory_size) - * NUM_XCC(pdd->dev->xcc_mask); + total_cwsr_size = (properties->ctx_save_restore_area_size + + topo_dev->node_props.debug_memory_size) * NUM_XCC(pdd->dev->xcc_mask); total_cwsr_size = ALIGN(total_cwsr_size, PAGE_SIZE); err = kfd_queue_buffer_get(vm, (void *)properties->ctx_save_restore_area_address, @@ -352,8 +352,8 @@ int kfd_queue_release_buffers(struct kfd_process_device *pdd, struct queue_prope topo_dev = kfd_topology_device_by_id(pdd->dev->id); if (!topo_dev) return -EINVAL; - total_cwsr_size = (topo_dev->node_props.cwsr_size + topo_dev->node_props.debug_memory_size) - * NUM_XCC(pdd->dev->xcc_mask); + total_cwsr_size = (properties->ctx_save_restore_area_size + + topo_dev->node_props.debug_memory_size) * NUM_XCC(pdd->dev->xcc_mask); total_cwsr_size = ALIGN(total_cwsr_size, PAGE_SIZE); kfd_queue_buffer_svm_put(pdd, properties->ctx_save_restore_area_address, total_cwsr_size); @@ -409,6 +409,7 @@ static u32 kfd_get_vgpr_size_per_cu(u32 gfxv) vgpr_size = 0x80000; else if (gfxv == 110000 || /* GFX_VERSION_PLUM_BONITO */ gfxv == 110001 || /* GFX_VERSION_WHEAT_NAS */ + gfxv == 110501 || /* GFX_VERSION_GFX1151 */ gfxv == 120000 || /* GFX_VERSION_GFX1200 */ gfxv == 120001) /* GFX_VERSION_GFX1201 */ vgpr_size = 0x60000; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 9d72411c3379..79ea138897fc 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -1144,30 +1144,48 @@ static int svm_range_split_tail(struct svm_range *prange, uint64_t new_last, struct list_head *insert_list, struct list_head *remap_list) { + unsigned long last_align_down = ALIGN_DOWN(prange->last, 512); + unsigned long start_align = ALIGN(prange->start, 512); + bool huge_page_mapping = last_align_down > start_align; struct svm_range *tail = NULL; - int r = svm_range_split(prange, prange->start, new_last, &tail); + int r; - if (!r) { - list_add(&tail->list, insert_list); - if (!IS_ALIGNED(new_last + 1, 1UL << prange->granularity)) - list_add(&tail->update_list, remap_list); - } - return r; + r = svm_range_split(prange, prange->start, new_last, &tail); + + if (r) + return r; + + list_add(&tail->list, insert_list); + + if (huge_page_mapping && tail->start > start_align && + tail->start < last_align_down && (!IS_ALIGNED(tail->start, 512))) + list_add(&tail->update_list, remap_list); + + return 0; } static int svm_range_split_head(struct svm_range *prange, uint64_t new_start, struct list_head *insert_list, struct list_head *remap_list) { + unsigned long last_align_down = ALIGN_DOWN(prange->last, 512); + unsigned long start_align = ALIGN(prange->start, 512); + bool huge_page_mapping = last_align_down > start_align; struct svm_range *head = NULL; - int r = svm_range_split(prange, new_start, prange->last, &head); + int r; - if (!r) { - list_add(&head->list, insert_list); - if (!IS_ALIGNED(new_start, 1UL << prange->granularity)) - list_add(&head->update_list, remap_list); - } - return r; + r = svm_range_split(prange, new_start, prange->last, &head); + + if (r) + return r; + + list_add(&head->list, insert_list); + + if (huge_page_mapping && head->last + 1 > start_align && + head->last + 1 < last_align_down && (!IS_ALIGNED(head->last, 512))) + list_add(&head->update_list, remap_list); + + return 0; } static void @@ -1698,7 +1716,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm, start = map_start << PAGE_SHIFT; end = (map_last + 1) << PAGE_SHIFT; for (addr = start; !r && addr < end; ) { - struct hmm_range *hmm_range = NULL; + struct amdgpu_hmm_range *range = NULL; unsigned long map_start_vma; unsigned long map_last_vma; struct vm_area_struct *vma; @@ -1737,9 +1755,12 @@ static int svm_range_validate_and_map(struct mm_struct *mm, } WRITE_ONCE(p->svms.faulting_task, current); - r = amdgpu_hmm_range_get_pages(&prange->notifier, addr, npages, - readonly, owner, - &hmm_range); + range = amdgpu_hmm_range_alloc(NULL); + if (likely(range)) + r = amdgpu_hmm_range_get_pages(&prange->notifier, addr, npages, + readonly, owner, range); + else + r = -ENOMEM; WRITE_ONCE(p->svms.faulting_task, NULL); if (r) pr_debug("failed %d to get svm range pages\n", r); @@ -1750,7 +1771,7 @@ static int svm_range_validate_and_map(struct mm_struct *mm, if (!r) { offset = (addr >> PAGE_SHIFT) - prange->start; r = svm_range_dma_map(prange, ctx->bitmap, offset, npages, - hmm_range->hmm_pfns); + range->hmm_range.hmm_pfns); if (r) pr_debug("failed %d to dma map range\n", r); } @@ -1758,14 +1779,17 @@ static int svm_range_validate_and_map(struct mm_struct *mm, svm_range_lock(prange); /* Free backing memory of hmm_range if it was initialized - * Overrride return value to TRY AGAIN only if prior returns + * Override return value to TRY AGAIN only if prior returns * were successful */ - if (hmm_range && amdgpu_hmm_range_get_pages_done(hmm_range) && !r) { + if (range && !amdgpu_hmm_range_valid(range) && !r) { pr_debug("hmm update the range, need validate again\n"); r = -EAGAIN; } + /* Free the hmm range */ + amdgpu_hmm_range_free(range); + if (!r && !list_empty(&prange->child_list)) { pr_debug("range split by unmap in parallel, validate again\n"); r = -EAGAIN; @@ -3687,6 +3711,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, svm_range_apply_attrs(p, prange, nattr, attrs, &update_mapping); /* TODO: unmap ranges from GPU that lost access */ } + update_mapping |= !p->xnack_enabled && !list_empty(&remap_list); + list_for_each_entry_safe(prange, next, &remove_list, update_list) { pr_debug("unlink old 0x%p prange 0x%p [0x%lx 0x%lx]\n", prange->svms, prange, prange->start, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h index 01c7a4877904..a63dfc95b602 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h @@ -31,7 +31,6 @@ #include <linux/list.h> #include <linux/mutex.h> #include <linux/sched/mm.h> -#include <linux/hmm.h> #include "amdgpu.h" #include "kfd_priv.h" diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 5c98746eb72d..3eb32d58a120 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -491,6 +491,10 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr, dev->node_props.num_sdma_queues_per_engine); sysfs_show_32bit_prop(buffer, offs, "num_cp_queues", dev->node_props.num_cp_queues); + sysfs_show_32bit_prop(buffer, offs, "cwsr_size", + dev->node_props.cwsr_size); + sysfs_show_32bit_prop(buffer, offs, "ctl_stack_size", + dev->node_props.ctl_stack_size); if (dev->gpu) { log_max_watch_addr = @@ -530,7 +534,9 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr, sysfs_show_32bit_prop(buffer, offs, "sdma_fw_version", dev->gpu->kfd->sdma_fw_version); sysfs_show_64bit_prop(buffer, offs, "unique_id", - dev->gpu->xcp ? + dev->gpu->xcp && + (dev->gpu->xcp->xcp_mgr->mode != + AMDGPU_SPX_PARTITION_MODE) ? dev->gpu->xcp->unique_id : dev->gpu->adev->unique_id); sysfs_show_32bit_prop(buffer, offs, "num_xcc", |