// SPDX-License-Identifier: GPL-2.0 /* * Copyright 2016-2021 HabanaLabs, Ltd. * All Rights Reserved. */ #include #include "habanalabs.h" #include #include #define HL_CS_FLAGS_TYPE_MASK (HL_CS_FLAGS_SIGNAL | HL_CS_FLAGS_WAIT | \ HL_CS_FLAGS_COLLECTIVE_WAIT | HL_CS_FLAGS_RESERVE_SIGNALS_ONLY | \ HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY | HL_CS_FLAGS_ENGINE_CORE_COMMAND | \ HL_CS_FLAGS_FLUSH_PCI_HBW_WRITES) #define MAX_TS_ITER_NUM 100 /** * enum hl_cs_wait_status - cs wait status * @CS_WAIT_STATUS_BUSY: cs was not completed yet * @CS_WAIT_STATUS_COMPLETED: cs completed * @CS_WAIT_STATUS_GONE: cs completed but fence is already gone */ enum hl_cs_wait_status { CS_WAIT_STATUS_BUSY, CS_WAIT_STATUS_COMPLETED, CS_WAIT_STATUS_GONE }; static void job_wq_completion(struct work_struct *work); static int _hl_cs_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx, u64 timeout_us, u64 seq, enum hl_cs_wait_status *status, s64 *timestamp); static void cs_do_release(struct kref *ref); static void hl_push_cs_outcome(struct hl_device *hdev, struct hl_cs_outcome_store *outcome_store, u64 seq, ktime_t ts, int error) { struct hl_cs_outcome *node; unsigned long flags; /* * CS outcome store supports the following operations: * push outcome - store a recent CS outcome in the store * pop outcome - retrieve a SPECIFIC (by seq) CS outcome from the store * It uses 2 lists: used list and free list. * It has a pre-allocated amount of nodes, each node stores * a single CS outcome. * Initially, all the nodes are in the free list. * On push outcome, a node (any) is taken from the free list, its * information is filled in, and the node is moved to the used list. * It is possible, that there are no nodes left in the free list. * In this case, we will lose some information about old outcomes. We * will pop the OLDEST node from the used list, and make it free. * On pop, the node is searched for in the used list (using a search * index). * If found, the node is then removed from the used list, and moved * back to the free list. The outcome data that the node contained is * returned back to the user. */ spin_lock_irqsave(&outcome_store->db_lock, flags); if (list_empty(&outcome_store->free_list)) { node = list_last_entry(&outcome_store->used_list, struct hl_cs_outcome, list_link); hash_del(&node->map_link); dev_dbg(hdev->dev, "CS %llu outcome was lost\n", node->seq); } else { node = list_last_entry(&outcome_store->free_list, struct hl_cs_outcome, list_link); } list_del_init(&node->list_link); node->seq = seq; node->ts = ts; node->error = error; list_add(&node->list_link, &outcome_store->used_list); hash_add(outcome_store->outcome_map, &node->map_link, node->seq); spin_unlock_irqrestore(&outcome_store->db_lock, flags); } static bool hl_pop_cs_outcome(struct hl_cs_outcome_store *outcome_store, u64 seq, ktime_t *ts, int *error) { struct hl_cs_outcome *node; unsigned long flags; spin_lock_irqsave(&outcome_store->db_lock, flags); hash_for_each_possible(outcome_store->outcome_map, node, map_link, seq) if (node->seq == seq) { *ts = node->ts; *error = node->error; hash_del(&node->map_link); list_del_init(&node->list_link); list_add(&node->list_link, &outcome_store->free_list); spin_unlock_irqrestore(&outcome_store->db_lock, flags); return true; } spin_unlock_irqrestore(&outcome_store->db_lock, flags); return false; } static void hl_sob_reset(struct kref *ref) { struct hl_hw_sob *hw_sob = container_of(ref, struct hl_hw_sob, kref); struct hl_device *hdev = hw_sob->hdev; dev_dbg(hdev->dev, "reset sob id %u\n", hw_sob->sob_id); hdev->asic_funcs->reset_sob(hdev, hw_sob); hw_sob->need_reset = false; } void hl_sob_reset_error(struct kref *ref) { struct hl_hw_sob *hw_sob = container_of(ref, struct hl_hw_sob, kref); struct hl_device *hdev = hw_sob->hdev; dev_crit(hdev->dev, "SOB release shouldn't be called here, q_idx: %d, sob_id: %d\n", hw_sob->q_idx, hw_sob->sob_id); } void hw_sob_put(struct hl_hw_sob *hw_sob) { if (hw_sob) kref_put(&hw_sob->kref, hl_sob_reset); } static void hw_sob_put_err(struct hl_hw_sob *hw_sob) { if (hw_sob) kref_put(&hw_sob->kref, hl_sob_reset_error); } void hw_sob_get(struct hl_hw_sob *hw_sob) { if (hw_sob) kref_get(&hw_sob->kref); } /** * hl_gen_sob_mask() - Generates a sob mask to be used in a monitor arm packet * @sob_base: sob base id * @sob_mask: sob user mask, each bit represents a sob offset from sob base * @mask: generated mask * * Return: 0 if given parameters are valid */ int hl_gen_sob_mask(u16 sob_base, u8 sob_mask, u8 *mask) { int i; if (sob_mask == 0) return -EINVAL; if (sob_mask == 0x1) { *mask = ~(1 << (sob_base & 0x7)); } else { /* find msb in order to verify sob range is valid */ for (i = BITS_PER_BYTE - 1 ; i >= 0 ; i--) if (BIT(i) & sob_mask) break; if (i > (HL_MAX_SOBS_PER_MONITOR - (sob_base & 0x7) - 1)) return -EINVAL; *mask = ~sob_mask; } return 0; } static void hl_fence_release(struct kref *kref) { struct hl_fence *fence = container_of(kref, struct hl_fence, refcount); struct hl_cs_compl *hl_cs_cmpl = container_of(fence, struct hl_cs_compl, base_fence); kfree(hl_cs_cmpl); } void hl_fence_put(struct hl_fence *fence) { if (IS_ERR_OR_NULL(fence)) return; kref_put(&fence->refcount, hl_fence_release); } void hl_fences_put(struct hl_fence **fence, int len) { int i; for (i = 0; i < len; i++, fence++) hl_fence_put(*fence); } void hl_fence_get(struct hl_fence *fence) { if (fence) kref_get(&fence->refcount); } static void hl_fence_init(struct hl_fence *fence, u64 sequence) { kref_init(&fence->refcount); fence->cs_sequence = sequence; fence->error = 0; fence->timestamp = ktime_set(0, 0); fence->mcs_handling_done = false; init_completion(&fence->completion); } void cs_get(struct hl_cs *cs) { kref_get(&cs->refcount); } static int cs_get_unless_zero(struct hl_cs *cs) { return kref_get_unless_zero(&cs->refcount); } static void cs_put(struct hl_cs *cs) { kref_put(&cs->refcount, cs_do_release); } static void cs_job_do_release(struct kref *ref) { struct hl_cs_job *job = container_of(ref, struct hl_cs_job, refcount); kfree(job); } static void hl_cs_job_put(struct hl_cs_job *job) { kref_put(&job->refcount, cs_job_do_release); } bool cs_needs_completion(struct hl_cs *cs) { /* In case this is a staged CS, only the last CS in sequence should * get a completion, any non staged CS will always get a completion */ if (cs->staged_cs && !cs->staged_last) return false; return true; } bool cs_needs_timeout(struct hl_cs *cs) { /* In case this is a staged CS, only the first CS in sequence should * get a timeout, any non staged CS will always get a timeout */ if (cs->staged_cs && !cs->staged_first) return false; return true; } static bool is_cb_patched(struct hl_device *hdev, struct hl_cs_job *job) { /* * Patched CB is created for external queues jobs, and for H/W queues * jobs if the user CB was allocated by driver and MMU is disabled. */ return (job->queue_type == QUEUE_TYPE_EXT || (job->queue_type == QUEUE_TYPE_HW && job->is_kernel_allocated_cb && !hdev->mmu_enable)); } /* * cs_parser - parse the user command submission * * @hpriv : pointer to the private data of the fd * @job : pointer to the job that holds the command submission info * * The function parses the command submission of the user. It calls the * ASIC specific parser, which returns a list of memory blocks to send * to the device as different command buffers * */ static int cs_parser(struct hl_fpriv *hpriv, struct hl_cs_job *job) { struct hl_device *hdev = hpriv->hdev; struct hl_cs_parser parser; int rc; parser.ctx_id = job->cs->ctx->asid; parser.cs_sequence = job->cs->sequence; parser.job_id = job->id; parser.hw_queue_id = job->hw_queue_id; parser.job_userptr_list = &job->userptr_list; parser.patched_cb = NULL; parser.user_cb = job->user_cb; parser.user_cb_size = job->user_cb_size; parser.queue_type = job->queue_type; parser.is_kernel_allocated_cb = job->is_kernel_allocated_cb; job->patched_cb = NULL; parser.completion = cs_needs_completion(job->cs); rc = hdev->asic_funcs->cs_parser(hdev, &parser); if (is_cb_patched(hdev, job)) { if (!rc) { job->patched_cb = parser.patched_cb; job->job_cb_size = parser.patched_cb_size; job->contains_dma_pkt = parser.contains_dma_pkt; atomic_inc(&job->patched_cb->cs_cnt); } /* * Whether the parsing worked or not, we don't need the * original CB anymore because it was already parsed and * won't be accessed again for this CS */ atomic_dec(&job->user_cb->cs_cnt); hl_cb_put(job->user_cb); job->user_cb = NULL; } else if (!rc) { job->job_cb_size = job->user_cb_size; } return rc; } static void hl_complete_job(struct hl_device *hdev, struct hl_cs_job *job) { struct hl_cs *cs = job->cs; if (is_cb_patched(hdev, job)) { hl_userptr_delete_list(hdev, &job->userptr_list); /* * We might arrive here from rollback and patched CB wasn't * created, so we need to check it's not NULL */ if (job->patched_cb) { atomic_dec(&job->patched_cb->cs_cnt); hl_cb_put(job->patched_cb); } } /* For H/W queue jobs, if a user CB was allocated by driver and MMU is * enabled, the user CB isn't released in cs_parser() and thus should be * released here. This is also true for INT queues jobs which were * allocated by driver. */ if ((job->is_kernel_allocated_cb && ((job->queue_type == QUEUE_TYPE_HW && hdev->mmu_enable) || job->queue_type == QUEUE_TYPE_INT))) { atomic_dec(&job->user_cb->cs_cnt); hl_cb_put(job->user_cb); } /* * This is the only place where there can be multiple threads * modifying the list at the same time */ spin_lock(&cs->job_lock); list_del(&job->cs_node); spin_unlock(&cs->job_lock); hl_debugfs_remove_job(hdev, job); /* We decrement reference only for a CS that gets completion * because the reference was incremented only for this kind of CS * right before it was scheduled. * * In staged submission, only the last CS marked as 'staged_last' * gets completion, hence its release function will be called from here. * As for all the rest CS's in the staged submission which do not get * completion, their CS reference will be decremented by the * 'staged_last' CS during the CS release flow. * All relevant PQ CI counters will be incremented during the CS release * flow by calling 'hl_hw_queue_update_ci'. */ if (cs_needs_completion(cs) && (job->queue_type == QUEUE_TYPE_EXT || job->queue_type == QUEUE_TYPE_HW)) { /* In CS based completions, the timestamp is already available, * so no need to extract it from job */ if (hdev->asic_prop.completion_mode == HL_COMPLETION_MODE_JOB) cs->completion_timestamp = job->timestamp; cs_put(cs); } hl_cs_job_put(job); } /* * hl_staged_cs_find_first - locate the first CS in this staged submission * * @hdev: pointer to device structure * @cs_seq: staged submission sequence number * * @note: This function must be called under 'hdev->cs_mirror_lock' * * Find and return a CS pointer with the given sequence */ struct hl_cs *hl_staged_cs_find_first(struct hl_device *hdev, u64 cs_seq) { struct hl_cs *cs; list_for_each_entry_reverse(cs, &hdev->cs_mirror_list, mirror_node) if (cs->staged_cs && cs->staged_first && cs->sequence == cs_seq) return cs; return NULL; } /* * is_staged_cs_last_exists - returns true if the last CS in sequence exists * * @hdev: pointer to device structure * @cs: staged submission member * */ bool is_staged_cs_last_exists(struct hl_device *hdev, struct hl_cs *cs) { struct hl_cs *last_entry; last_entry = list_last_entry(&cs->staged_cs_node, struct hl_cs, staged_cs_node); if (last_entry->staged_last) return true; return false; } /* * staged_cs_get - get CS reference if this CS is a part of a staged CS * * @hdev: pointer to device structure * @cs: current CS * @cs_seq: staged submission sequence number * * Increment CS reference for every CS in this staged submission except for * the CS which get completion. */ static void staged_cs_get(struct hl_device *hdev, struct hl_cs *cs) { /* Only the last CS in this staged submission will get a completion. * We must increment the reference for all other CS's in this * staged submission. * Once we get a completion we will release the whole staged submission. */ if (!cs->staged_last) cs_get(cs); } /* * staged_cs_put - put a CS in case it is part of staged submission * * @hdev: pointer to device structure * @cs: CS to put * * This function decrements a CS reference (for a non completion CS) */ static void staged_cs_put(struct hl_device *hdev, struct hl_cs *cs) { /* We release all CS's in a staged submission except the last * CS which we have never incremented its reference. */ if (!cs_needs_completion(cs)) cs_put(cs); } static void cs_handle_tdr(struct hl_device *hdev, struct hl_cs *cs) { struct hl_cs *next = NULL, *iter, *first_cs; if (!cs_needs_timeout(cs)) return; spin_lock(&hdev->cs_mirror_lock); /* We need to handle tdr only once for the complete staged submission. * Hence, we choose the CS that reaches this function first which is * the CS marked as 'staged_last'. * In case single staged cs was submitted which has both first and last * indications, then "cs_find_first" below will return NULL, since we * removed the cs node from the list before getting here, * in such cases just continue with the cs to cancel it's TDR work. */ if (cs->staged_cs && cs->staged_last) { first_cs = hl_staged_cs_find_first(hdev, cs->staged_sequence); if (first_cs) cs = first_cs; } spin_unlock(&hdev->cs_mirror_lock); /* Don't cancel TDR in case this CS was timedout because we might be * running from the TDR context */ if (cs->timedout || hdev->timeout_jiffies == MAX_SCHEDULE_TIMEOUT) return; if (cs->tdr_active) cancel_delayed_work_sync(&cs->work_tdr); spin_lock(&hdev->cs_mirror_lock); /* queue TDR for next CS */ list_for_each_entry(iter, &hdev->cs_mirror_list, mirror_node) if (cs_needs_timeout(iter)) { next = iter; break; } if (next && !next->tdr_active) { next->tdr_active = true; schedule_delayed_work(&next->work_tdr, next->timeout_jiffies); } spin_unlock(&hdev->cs_mirror_lock); } /* * force_complete_multi_cs - complete all contexts that wait on multi-CS * * @hdev: pointer to habanalabs device structure */ static void force_complete_multi_cs(struct hl_device *hdev) { int i; for (i = 0; i < MULTI_CS_MAX_USER_CTX; i++) { struct multi_cs_completion *mcs_compl; mcs_compl = &hdev->multi_cs_completion[i]; spin_lock(&mcs_compl->lock); if (!mcs_compl->used) { spin_unlock(&mcs_compl->lock); continue; } /* when calling force complete no context should be waiting on * multi-cS. * We are calling the function as a protection for such case * to free any pending context and print error message */ dev_err(hdev->dev, "multi-CS completion context %d still waiting when calling force completion\n", i); complete_all(&mcs_compl->completion); spin_unlock(&mcs_compl->lock); } } /* * complete_multi_cs - complete all waiting entities on multi-CS * * @hdev: pointer to habanalabs device structure * @cs: CS structure * The function signals a waiting entity that has an overlapping stream masters * with the completed CS. * For example: * - a completed CS worked on stream master QID 4, multi CS completion * is actively waiting on stream master QIDs 3, 5. don't send signal as no * common stream master QID * - a completed CS worked on stream master QID 4, multi CS completion * is actively waiting on stream master QIDs 3, 4. send signal as stream * master QID 4 is common */ static void complete_multi_cs(struct hl_device *hdev, struct hl_cs *cs) { struct hl_fence *fence = cs->fence; int i; /* in case of multi CS check for completion only for the first CS */ if (cs->staged_cs && !cs->staged_first) return; for (i = 0; i < MULTI_CS_MAX_USER_CTX; i++) { struct multi_cs_completion *mcs_compl; mcs_compl = &hdev->multi_cs_completion[i]; if (!mcs_compl->used) continue; spin_lock(&mcs_compl->lock); /* * complete if: * 1. still waiting for completion * 2. the completed CS has at least one overlapping stream * master with the stream masters in the completion */ if (mcs_compl->used && (fence->stream_master_qid_map & mcs_compl->stream_master_qid_map)) { /* extract the timestamp only of first completed CS */ if (!mcs_compl->timestamp) mcs_compl->timestamp = ktime_to_ns(fence->timestamp); complete_all(&mcs_compl->completion); /* * Setting mcs_handling_done inside the lock ensures * at least one fence have mcs_handling_done set to * true before wait for mcs finish. This ensures at * least one CS will be set as completed when polling * mcs fences. */ fence->mcs_handling_done = true; } spin_unlock(&mcs_compl->lock); } /* In case CS completed without mcs completion initialized */ fence->mcs_handling_done = true; } static inline void cs_release_sob_reset_handler(struct hl_device *hdev, struct hl_cs *cs, struct hl_cs_compl *hl_cs_cmpl) { /* Skip this handler if the cs wasn't submitted, to avoid putting * the hw_sob twice, since this case already handled at this point, * also skip if the hw_sob pointer wasn't set. */ if (!hl_cs_cmpl->hw_sob || !cs->submitted) return; spin_lock(&hl_cs_cmpl->lock); /* * we get refcount upon reservation of signals or signal/wait cs for the * hw_sob object, and need to put it when the first staged cs * (which contains the encaps signals) or cs signal/wait is completed. */ if ((hl_cs_cmpl->type == CS_TYPE_SIGNAL) || (hl_cs_cmpl->type == CS_TYPE_WAIT) || (hl_cs_cmpl->type == CS_TYPE_COLLECTIVE_WAIT) || (!!hl_cs_cmpl->encaps_signals)) { dev_dbg(hdev->dev, "CS 0x%llx type %d finished, sob_id: %d, sob_val: %u\n", hl_cs_cmpl->cs_seq, hl_cs_cmpl->type, hl_cs_cmpl->hw_sob->sob_id, hl_cs_cmpl->sob_val); hw_sob_put(hl_cs_cmpl->hw_sob); if (hl_cs_cmpl->type == CS_TYPE_COLLECTIVE_WAIT) hdev->asic_funcs->reset_sob_group(hdev, hl_cs_cmpl->sob_group); } spin_unlock(&hl_cs_cmpl->lock); } static void cs_do_release(struct kref *ref) { struct hl_cs *cs = container_of(ref, struct hl_cs, refcount); struct hl_device *hdev = cs->ctx->hdev; struct hl_cs_job *job, *tmp; struct hl_cs_compl *hl_cs_cmpl = container_of(cs->fence, struct hl_cs_compl, base_fence); cs->completed = true; /* * Although if we reached here it means that all external jobs have * finished, because each one of them took refcnt to CS, we still * need to go over the internal jobs and complete them. Otherwise, we * will have leaked memory and what's worse, the CS object (and * potentially the CTX object) could be released, while the JOB * still holds a pointer to them (but no reference). */ list_for_each_entry_safe(job, tmp, &cs->job_list, cs_node) hl_complete_job(hdev, job); if (!cs->submitted) { /* * In case the wait for signal CS was submitted, the fence put * occurs in init_signal_wait_cs() or collective_wait_init_cs() * right before hanging on the PQ. */ if (cs->type == CS_TYPE_WAIT || cs->type == CS_TYPE_COLLECTIVE_WAIT) hl_fence_put(cs->signal_fence); goto out; } /* Need to update CI for all queue jobs that does not get completion */ hl_hw_queue_update_ci(cs); /* remove CS from CS mirror list */ spin_lock(&hdev->cs_mirror_lock); list_del_init(&cs->mirror_node); spin_unlock(&hdev->cs_mirror_lock); cs_handle_tdr(hdev, cs); if (cs->staged_cs) { /* the completion CS decrements reference for the entire * staged submission */ if (cs->staged_last) { struct hl_cs *staged_cs, *tmp_cs; list_for_each_entry_safe(staged_cs, tmp_cs, &cs->staged_cs_node, staged_cs_node) staged_cs_put(hdev, staged_cs); } /* A staged CS will be a member in the list only after it * was submitted. We used 'cs_mirror_lock' when inserting * it to list so we will use it again when removing it */ if (cs->submitted) { spin_lock(&hdev->cs_mirror_lock); list_del(&cs->staged_cs_node); spin_unlock(&hdev->cs_mirror_lock); } /* decrement refcount to handle when first staged cs * with encaps signals is completed. */ if (hl_cs_cmpl->encaps_signals) kref_put(&hl_cs_cmpl->encaps_sig_hdl->refcount, hl_encaps_release_handle_and_put_ctx); } if ((cs->type == CS_TYPE_WAIT || cs->type == CS_TYPE_COLLECTIVE_WAIT) && cs->encaps_signals) kref_put(&cs->encaps_sig_hdl->refcount, hl_encaps_release_handle_and_put_ctx); out: /* Must be called before hl_ctx_put because inside we use ctx to get * the device */ hl_debugfs_remove_cs(cs); hdev->shadow_cs_queue[cs->sequence & (hdev->asic_prop.max_pending_cs - 1)] = NULL; /* We need to mark an error for not submitted because in that case * the hl fence release flow is different. Mainly, we don't need * to handle hw_sob for signal/wait */ if (cs->timedout) cs->fence->error = -ETIMEDOUT; else if (cs->aborted) cs->fence->error = -EIO; else if (!cs->submitted) cs->fence->error = -EBUSY; if (unlikely(cs->skip_reset_on_timeout)) { dev_err(hdev->dev, "Command submission %llu completed after %llu (s)\n", cs->sequence, div_u64(jiffies - cs->submission_time_jiffies, HZ)); } if (cs->timestamp) { cs->fence->timestamp = cs->completion_timestamp; hl_push_cs_outcome(hdev, &cs->ctx->outcome_store, cs->sequence, cs->fence->timestamp, cs->fence->error); } hl_ctx_put(cs->ctx); complete_all(&cs->fence->completion); complete_multi_cs(hdev, cs); cs_release_sob_reset_handler(hdev, cs, hl_cs_cmpl); hl_fence_put(cs->fence); kfree(cs->jobs_in_queue_cnt); kfree(cs); } static void cs_timedout(struct work_struct *work) { struct hl_device *hdev; u64 event_mask = 0x0; int rc; struct hl_cs *cs = container_of(work, struct hl_cs, work_tdr.work); bool skip_reset_on_timeout = cs->skip_reset_on_timeout, device_reset = false; rc = cs_get_unless_zero(cs); if (!rc) return; if ((!cs->submitted) || (cs->completed)) { cs_put(cs); return; } hdev = cs->ctx->hdev; if (likely(!skip_reset_on_timeout)) { if (hdev->reset_on_lockup) device_reset = true; else hdev->reset_info.needs_reset = true; /* Mark the CS is timed out so we won't try to cancel its TDR */ cs->timedout = true; } /* Save only the first CS timeout parameters */ rc = atomic_cmpxchg(&hdev->captured_err_info.cs_timeout.write_enable, 1, 0); if (rc) { hdev->captured_err_info.cs_timeout.timestamp = ktime_get(); hdev->captured_err_info.cs_timeout.seq = cs->sequence; event_mask |= HL_NOTIFIER_EVENT_CS_TIMEOUT; } switch (cs->type) { case CS_TYPE_SIGNAL: dev_err(hdev->dev, "Signal command submission %llu has not finished in time!\n", cs->sequence); break; case CS_TYPE_WAIT: dev_err(hdev->dev, "Wait command submission %llu has not finished in time!\n", cs->sequence); break; case CS_TYPE_COLLECTIVE_WAIT: dev_err(hdev->dev, "Collective Wait command submission %llu has not finished in time!\n", cs->sequence); break; default: dev_err(hdev->dev, "Command submission %llu has not finished in time!\n", cs->sequence); break; } rc = hl_state_dump(hdev); if (rc) dev_err(hdev->dev, "Error during system state dump %d\n", rc); cs_put(cs); if (device_reset) { event_mask |= HL_NOTIFIER_EVENT_DEVICE_RESET; hl_device_cond_reset(hdev, HL_DRV_RESET_TDR, event_mask); } else if (event_mask) { hl_notifier_event_send_all(hdev, event_mask); } } static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx, enum hl_cs_type cs_type, u64 user_sequence, struct hl_cs **cs_new, u32 flags, u32 timeout) { struct hl_cs_counters_atomic *cntr; struct hl_fence *other = NULL; struct hl_cs_compl *cs_cmpl; struct hl_cs *cs; int rc; cntr = &hdev->aggregated_cs_counters; cs = kzalloc(sizeof(*cs), GFP_ATOMIC); if (!cs) cs = kzalloc(sizeof(*cs), GFP_KERNEL); if (!cs) { atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); atomic64_inc(&cntr->out_of_mem_drop_cnt); return -ENOMEM; } /* increment refcnt for context */ hl_ctx_get(ctx); cs->ctx = ctx; cs->submitted = false; cs->completed = false; cs->type = cs_type; cs->timestamp = !!(flags & HL_CS_FLAGS_TIMESTAMP); cs->encaps_signals = !!(flags & HL_CS_FLAGS_ENCAP_SIGNALS); cs->timeout_jiffies = timeout; cs->skip_reset_on_timeout = hdev->reset_info.skip_reset_on_timeout || !!(flags & HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT); cs->submission_time_jiffies = jiffies; INIT_LIST_HEAD(&cs->job_list); INIT_DELAYED_WORK(&cs->work_tdr, cs_timedout); kref_init(&cs->refcount); spin_lock_init(&cs->job_lock); cs_cmpl = kzalloc(sizeof(*cs_cmpl), GFP_ATOMIC); if (!cs_cmpl) cs_cmpl = kzalloc(sizeof(*cs_cmpl), GFP_KERNEL); if (!cs_cmpl) { atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); atomic64_inc(&cntr->out_of_mem_drop_cnt); rc = -ENOMEM; goto free_cs; } cs->jobs_in_queue_cnt = kcalloc(hdev->asic_prop.max_queues, sizeof(*cs->jobs_in_queue_cnt), GFP_ATOMIC); if (!cs->jobs_in_queue_cnt) cs->jobs_in_queue_cnt = kcalloc(hdev->asic_prop.max_queues, sizeof(*cs->jobs_in_queue_cnt), GFP_KERNEL); if (!cs->jobs_in_queue_cnt) { atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); atomic64_inc(&cntr->out_of_mem_drop_cnt); rc = -ENOMEM; goto free_cs_cmpl; } cs_cmpl->hdev = hdev; cs_cmpl->type = cs->type; spin_lock_init(&cs_cmpl->lock); cs->fence = &cs_cmpl->base_fence; spin_lock(&ctx->cs_lock); cs_cmpl->cs_seq = ctx->cs_sequence; other = ctx->cs_pending[cs_cmpl->cs_seq & (hdev->asic_prop.max_pending_cs - 1)]; if (other && !completion_done(&other->completion)) { /* If the following statement is true, it means we have reached * a point in which only part of the staged submission was * submitted and we don't have enough room in the 'cs_pending' * array for the rest of the submission. * This causes a deadlock because this CS will never be * completed as it depends on future CS's for completion. */ if (other->cs_sequence == user_sequence) dev_crit_ratelimited(hdev->dev, "Staged CS %llu deadlock due to lack of resources", user_sequence); dev_dbg_ratelimited(hdev->dev, "Rejecting CS because of too many in-flights CS\n"); atomic64_inc(&ctx->cs_counters.max_cs_in_flight_drop_cnt); atomic64_inc(&cntr->max_cs_in_flight_drop_cnt); rc = -EAGAIN; goto free_fence; } /* init hl_fence */ hl_fence_init(&cs_cmpl->base_fence, cs_cmpl->cs_seq); cs->sequence = cs_cmpl->cs_seq; ctx->cs_pending[cs_cmpl->cs_seq & (hdev->asic_prop.max_pending_cs - 1)] = &cs_cmpl->base_fence; ctx->cs_sequence++; hl_fence_get(&cs_cmpl->base_fence); hl_fence_put(other); spin_unlock(&ctx->cs_lock); *cs_new = cs; return 0; free_fence: spin_unlock(&ctx->cs_lock); kfree(cs->jobs_in_queue_cnt); free_cs_cmpl: kfree(cs_cmpl); free_cs: kfree(cs); hl_ctx_put(ctx); return rc; } static void cs_rollback(struct hl_device *hdev, struct hl_cs *cs) { struct hl_cs_job *job, *tmp; staged_cs_put(hdev, cs); list_for_each_entry_safe(job, tmp, &cs->job_list, cs_node) hl_complete_job(hdev, job); } /* * release_reserved_encaps_signals() - release reserved encapsulated signals. * @hdev: pointer to habanalabs device structure * * Release reserved encapsulated signals which weren't un-reserved, or for which a CS with * encapsulated signals wasn't submitted and thus weren't released as part of CS roll-back. * For these signals need also to put the refcount of the H/W SOB which was taken at the * reservation. */ static void release_reserved_encaps_signals(struct hl_device *hdev) { struct hl_ctx *ctx = hl_get_compute_ctx(hdev); struct hl_cs_encaps_sig_handle *handle; struct hl_encaps_signals_mgr *mgr; u32 id; if (!ctx) return; mgr = &ctx->sig_mgr; idr_for_each_entry(&mgr->handles, handle, id) if (handle->cs_seq == ULLONG_MAX) kref_put(&handle->refcount, hl_encaps_release_handle_and_put_sob_ctx); hl_ctx_put(ctx); } void hl_cs_rollback_all(struct hl_device *hdev, bool skip_wq_flush) { int i; struct hl_cs *cs, *tmp; if (!skip_wq_flush) { flush_workqueue(hdev->ts_free_obj_wq); /* flush all completions before iterating over the CS mirror list in * order to avoid a race with the release functions */ for (i = 0 ; i < hdev->asic_prop.completion_queues_count ; i++) flush_workqueue(hdev->cq_wq[i]); flush_workqueue(hdev->cs_cmplt_wq); } /* Make sure we don't have leftovers in the CS mirror list */ list_for_each_entry_safe(cs, tmp, &hdev->cs_mirror_list, mirror_node) { cs_get(cs); cs->aborted = true; dev_warn_ratelimited(hdev->dev, "Killing CS %d.%llu\n", cs->ctx->asid, cs->sequence); cs_rollback(hdev, cs); cs_put(cs); } force_complete_multi_cs(hdev); release_reserved_encaps_signals(hdev); } static void wake_pending_user_interrupt_threads(struct hl_user_interrupt *interrupt) { struct hl_user_pending_interrupt *pend, *temp; spin_lock(&interrupt->wait_list_lock); list_for_each_entry_safe(pend, temp, &interrupt->wait_list_head, wait_list_node) { if (pend->ts_reg_info.buf) { list_del(&pend->wait_list_node); hl_mmap_mem_buf_put(pend->ts_reg_info.buf); hl_cb_put(pend->ts_reg_info.cq_cb); } else { pend->fence.error = -EIO; complete_all(&pend->fence.completion); } } spin_unlock(&interrupt->wait_list_lock); } void hl_release_pending_user_interrupts(struct hl_device *hdev) { struct asic_fixed_properties *prop = &hdev->asic_prop; struct hl_user_interrupt *interrupt; int i; if (!prop->user_interrupt_count) return; /* We iterate through the user interrupt requests and waking up all * user threads waiting for interrupt completion. We iterate the * list under a lock, this is why all user threads, once awake, * will wait on the same lock and will release the waiting object upon * unlock. */ for (i = 0 ; i < prop->user_interrupt_count ; i++) { interrupt = &hdev->user_interrupt[i]; wake_pending_user_interrupt_threads(interrupt); } interrupt = &hdev->common_user_cq_interrupt; wake_pending_user_interrupt_threads(interrupt); interrupt = &hdev->common_decoder_interrupt; wake_pending_user_interrupt_threads(interrupt); } static void force_complete_cs(struct hl_device *hdev) { struct hl_cs *cs; spin_lock(&hdev->cs_mirror_lock); list_for_each_entry(cs, &hdev->cs_mirror_list, mirror_node) { cs->fence->error = -EIO; complete_all(&cs->fence->completion); } spin_unlock(&hdev->cs_mirror_lock); } void hl_abort_waitings_for_completion(struct hl_device *hdev) { force_complete_cs(hdev); force_complete_multi_cs(hdev); hl_release_pending_user_interrupts(hdev); } static void job_wq_completion(struct work_struct *work) { struct hl_cs_job *job = container_of(work, struct hl_cs_job, finish_work); struct hl_cs *cs = job->cs; struct hl_device *hdev = cs->ctx->hdev; /* job is no longer needed */ hl_complete_job(hdev, job); } static void cs_completion(struct work_struct *work) { struct hl_cs *cs = container_of(work, struct hl_cs, finish_work); struct hl_device *hdev = cs->ctx->hdev; struct hl_cs_job *job, *tmp; list_for_each_entry_safe(job, tmp, &cs->job_list, cs_node) hl_complete_job(hdev, job); } u32 hl_get_active_cs_num(struct hl_device *hdev) { u32 active_cs_num = 0; struct hl_cs *cs; spin_lock(&hdev->cs_mirror_lock); list_for_each_entry(cs, &hdev->cs_mirror_list, mirror_node) if (!cs->completed) active_cs_num++; spin_unlock(&hdev->cs_mirror_lock); return active_cs_num; } static int validate_queue_index(struct hl_device *hdev, struct hl_cs_chunk *chunk, enum hl_queue_type *queue_type, bool *is_kernel_allocated_cb) { struct asic_fixed_properties *asic = &hdev->asic_prop; struct hw_queue_properties *hw_queue_prop; /* This must be checked here to prevent out-of-bounds access to * hw_queues_props array */ if (chunk->queue_index >= asic->max_queues) { dev_err(hdev->dev, "Queue index %d is invalid\n", chunk->queue_index); return -EINVAL; } hw_queue_prop = &asic->hw_queues_props[chunk->queue_index]; if (hw_queue_prop->type == QUEUE_TYPE_NA) { dev_err(hdev->dev, "Queue index %d is not applicable\n", chunk->queue_index); return -EINVAL; } if (hw_queue_prop->binned) { dev_err(hdev->dev, "Queue index %d is binned out\n", chunk->queue_index); return -EINVAL; } if (hw_queue_prop->driver_only) { dev_err(hdev->dev, "Queue index %d is restricted for the kernel driver\n", chunk->queue_index); return -EINVAL; } /* When hw queue type isn't QUEUE_TYPE_HW, * USER_ALLOC_CB flag shall be referred as "don't care". */ if (hw_queue_prop->type == QUEUE_TYPE_HW) { if (chunk->cs_chunk_flags & HL_CS_CHUNK_FLAGS_USER_ALLOC_CB) { if (!(hw_queue_prop->cb_alloc_flags & CB_ALLOC_USER)) { dev_err(hdev->dev, "Queue index %d doesn't support user CB\n", chunk->queue_index); return -EINVAL; } *is_kernel_allocated_cb = false; } else { if (!(hw_queue_prop->cb_alloc_flags & CB_ALLOC_KERNEL)) { dev_err(hdev->dev, "Queue index %d doesn't support kernel CB\n", chunk->queue_index); return -EINVAL; } *is_kernel_allocated_cb = true; } } else { *is_kernel_allocated_cb = !!(hw_queue_prop->cb_alloc_flags & CB_ALLOC_KERNEL); } *queue_type = hw_queue_prop->type; return 0; } static struct hl_cb *get_cb_from_cs_chunk(struct hl_device *hdev, struct hl_mem_mgr *mmg, struct hl_cs_chunk *chunk) { struct hl_cb *cb; cb = hl_cb_get(mmg, chunk->cb_handle); if (!cb) { dev_err(hdev->dev, "CB handle 0x%llx invalid\n", chunk->cb_handle); return NULL; } if ((chunk->cb_size < 8) || (chunk->cb_size > cb->size)) { dev_err(hdev->dev, "CB size %u invalid\n", chunk->cb_size); goto release_cb; } atomic_inc(&cb->cs_cnt); return cb; release_cb: hl_cb_put(cb); return NULL; } struct hl_cs_job *hl_cs_allocate_job(struct hl_device *hdev, enum hl_queue_type queue_type, bool is_kernel_allocated_cb) { struct hl_cs_job *job; job = kzalloc(sizeof(*job), GFP_ATOMIC); if (!job) job = kzalloc(sizeof(*job), GFP_KERNEL); if (!job) return NULL; kref_init(&job->refcount); job->queue_type = queue_type; job->is_kernel_allocated_cb = is_kernel_allocated_cb; if (is_cb_patched(hdev, job)) INIT_LIST_HEAD(&job->userptr_list); if (job->queue_type == QUEUE_TYPE_EXT) INIT_WORK(&job->finish_work, job_wq_completion); return job; } static enum hl_cs_type hl_cs_get_cs_type(u32 cs_type_flags) { if (cs_type_flags & HL_CS_FLAGS_SIGNAL) return CS_TYPE_SIGNAL; else if (cs_type_flags & HL_CS_FLAGS_WAIT) return CS_TYPE_WAIT; else if (cs_type_flags & HL_CS_FLAGS_COLLECTIVE_WAIT) return CS_TYPE_COLLECTIVE_WAIT; else if (cs_type_flags & HL_CS_FLAGS_RESERVE_SIGNALS_ONLY) return CS_RESERVE_SIGNALS; else if (cs_type_flags & HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY) return CS_UNRESERVE_SIGNALS; else if (cs_type_flags & HL_CS_FLAGS_ENGINE_CORE_COMMAND) return CS_TYPE_ENGINE_CORE; else if (cs_type_flags & HL_CS_FLAGS_FLUSH_PCI_HBW_WRITES) return CS_TYPE_FLUSH_PCI_HBW_WRITES; else return CS_TYPE_DEFAULT; } static int hl_cs_sanity_checks(struct hl_fpriv *hpriv, union hl_cs_args *args) { struct hl_device *hdev = hpriv->hdev; struct hl_ctx *ctx = hpriv->ctx; u32 cs_type_flags, num_chunks; enum hl_device_status status; enum hl_cs_type cs_type; bool is_sync_stream; int i; for (i = 0 ; i < sizeof(args->in.pad) ; i++) if (args->in.pad[i]) { dev_dbg(hdev->dev, "Padding bytes must be 0\n"); return -EINVAL; } if (!hl_device_operational(hdev, &status)) { return -EBUSY; } if ((args->in.cs_flags & HL_CS_FLAGS_STAGED_SUBMISSION) && !hdev->supports_staged_submission) { dev_err(hdev->dev, "staged submission not supported"); return -EPERM; } cs_type_flags = args->in.cs_flags & HL_CS_FLAGS_TYPE_MASK; if (unlikely(cs_type_flags && !is_power_of_2(cs_type_flags))) { dev_err(hdev->dev, "CS type flags are mutually exclusive, context %d\n", ctx->asid); return -EINVAL; } cs_type = hl_cs_get_cs_type(cs_type_flags); num_chunks = args->in.num_chunks_execute; is_sync_stream = (cs_type == CS_TYPE_SIGNAL || cs_type == CS_TYPE_WAIT || cs_type == CS_TYPE_COLLECTIVE_WAIT); if (unlikely(is_sync_stream && !hdev->supports_sync_stream)) { dev_err(hdev->dev, "Sync stream CS is not supported\n"); return -EINVAL; } if (cs_type == CS_TYPE_DEFAULT) { if (!num_chunks) { dev_err(hdev->dev, "Got execute CS with 0 chunks, context %d\n", ctx->asid); return -EINVAL; } } else if (is_sync_stream && num_chunks != 1) { dev_err(hdev->dev, "Sync stream CS mandates one chunk only, context %d\n", ctx->asid); return -EINVAL; } return 0; } static int hl_cs_copy_chunk_array(struct hl_device *hdev, struct hl_cs_chunk **cs_chunk_array, void __user *chunks, u32 num_chunks, struct hl_ctx *ctx) { u32 size_to_copy; if (num_chunks > HL_MAX_JOBS_PER_CS) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&hdev->aggregated_cs_counters.validation_drop_cnt); dev_err(hdev->dev, "Number of chunks can NOT be larger than %d\n", HL_MAX_JOBS_PER_CS); return -EINVAL; } *cs_chunk_array = kmalloc_array(num_chunks, sizeof(**cs_chunk_array), GFP_ATOMIC); if (!*cs_chunk_array) *cs_chunk_array = kmalloc_array(num_chunks, sizeof(**cs_chunk_array), GFP_KERNEL); if (!*cs_chunk_array) { atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); atomic64_inc(&hdev->aggregated_cs_counters.out_of_mem_drop_cnt); return -ENOMEM; } size_to_copy = num_chunks * sizeof(struct hl_cs_chunk); if (copy_from_user(*cs_chunk_array, chunks, size_to_copy)) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&hdev->aggregated_cs_counters.validation_drop_cnt); dev_err(hdev->dev, "Failed to copy cs chunk array from user\n"); kfree(*cs_chunk_array); return -EFAULT; } return 0; } static int cs_staged_submission(struct hl_device *hdev, struct hl_cs *cs, u64 sequence, u32 flags, u32 encaps_signal_handle) { if (!(flags & HL_CS_FLAGS_STAGED_SUBMISSION)) return 0; cs->staged_last = !!(flags & HL_CS_FLAGS_STAGED_SUBMISSION_LAST); cs->staged_first = !!(flags & HL_CS_FLAGS_STAGED_SUBMISSION_FIRST); if (cs->staged_first) { /* Staged CS sequence is the first CS sequence */ INIT_LIST_HEAD(&cs->staged_cs_node); cs->staged_sequence = cs->sequence; if (cs->encaps_signals) cs->encaps_sig_hdl_id = encaps_signal_handle; } else { /* User sequence will be validated in 'hl_hw_queue_schedule_cs' * under the cs_mirror_lock */ cs->staged_sequence = sequence; } /* Increment CS reference if needed */ staged_cs_get(hdev, cs); cs->staged_cs = true; return 0; } static u32 get_stream_master_qid_mask(struct hl_device *hdev, u32 qid) { int i; for (i = 0; i < hdev->stream_master_qid_arr_size; i++) if (qid == hdev->stream_master_qid_arr[i]) return BIT(i); return 0; } static int cs_ioctl_default(struct hl_fpriv *hpriv, void __user *chunks, u32 num_chunks, u64 *cs_seq, u32 flags, u32 encaps_signals_handle, u32 timeout, u16 *signal_initial_sob_count) { bool staged_mid, int_queues_only = true, using_hw_queues = false; struct hl_device *hdev = hpriv->hdev; struct hl_cs_chunk *cs_chunk_array; struct hl_cs_counters_atomic *cntr; struct hl_ctx *ctx = hpriv->ctx; struct hl_cs_job *job; struct hl_cs *cs; struct hl_cb *cb; u64 user_sequence; u8 stream_master_qid_map = 0; int rc, i; cntr = &hdev->aggregated_cs_counters; user_sequence = *cs_seq; *cs_seq = ULLONG_MAX; rc = hl_cs_copy_chunk_array(hdev, &cs_chunk_array, chunks, num_chunks, hpriv->ctx); if (rc) goto out; if ((flags & HL_CS_FLAGS_STAGED_SUBMISSION) && !(flags & HL_CS_FLAGS_STAGED_SUBMISSION_FIRST)) staged_mid = true; else staged_mid = false; rc = allocate_cs(hdev, hpriv->ctx, CS_TYPE_DEFAULT, staged_mid ? user_sequence : ULLONG_MAX, &cs, flags, timeout); if (rc) goto free_cs_chunk_array; *cs_seq = cs->sequence; hl_debugfs_add_cs(cs); rc = cs_staged_submission(hdev, cs, user_sequence, flags, encaps_signals_handle); if (rc) goto free_cs_object; /* If this is a staged submission we must return the staged sequence * rather than the internal CS sequence */ if (cs->staged_cs) *cs_seq = cs->staged_sequence; /* Validate ALL the CS chunks before submitting the CS */ for (i = 0 ; i < num_chunks ; i++) { struct hl_cs_chunk *chunk = &cs_chunk_array[i]; enum hl_queue_type queue_type; bool is_kernel_allocated_cb; rc = validate_queue_index(hdev, chunk, &queue_type, &is_kernel_allocated_cb); if (rc) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); goto free_cs_object; } if (is_kernel_allocated_cb) { cb = get_cb_from_cs_chunk(hdev, &hpriv->mem_mgr, chunk); if (!cb) { atomic64_inc( &ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); rc = -EINVAL; goto free_cs_object; } } else { cb = (struct hl_cb *) (uintptr_t) chunk->cb_handle; } if (queue_type == QUEUE_TYPE_EXT || queue_type == QUEUE_TYPE_HW) { int_queues_only = false; /* * store which stream are being used for external/HW * queues of this CS */ if (hdev->supports_wait_for_multi_cs) stream_master_qid_map |= get_stream_master_qid_mask(hdev, chunk->queue_index); } if (queue_type == QUEUE_TYPE_HW) using_hw_queues = true; job = hl_cs_allocate_job(hdev, queue_type, is_kernel_allocated_cb); if (!job) { atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); atomic64_inc(&cntr->out_of_mem_drop_cnt); dev_err(hdev->dev, "Failed to allocate a new job\n"); rc = -ENOMEM; if (is_kernel_allocated_cb) goto release_cb; goto free_cs_object; } job->id = i + 1; job->cs = cs; job->user_cb = cb; job->user_cb_size = chunk->cb_size; job->hw_queue_id = chunk->queue_index; cs->jobs_in_queue_cnt[job->hw_queue_id]++; cs->jobs_cnt++; list_add_tail(&job->cs_node, &cs->job_list); /* * Increment CS reference. When CS reference is 0, CS is * done and can be signaled to user and free all its resources * Only increment for JOB on external or H/W queues, because * only for those JOBs we get completion */ if (cs_needs_completion(cs) && (job->queue_type == QUEUE_TYPE_EXT || job->queue_type == QUEUE_TYPE_HW)) cs_get(cs); hl_debugfs_add_job(hdev, job); rc = cs_parser(hpriv, job); if (rc) { atomic64_inc(&ctx->cs_counters.parsing_drop_cnt); atomic64_inc(&cntr->parsing_drop_cnt); dev_err(hdev->dev, "Failed to parse JOB %d.%llu.%d, err %d, rejecting the CS\n", cs->ctx->asid, cs->sequence, job->id, rc); goto free_cs_object; } } /* We allow a CS with any queue type combination as long as it does * not get a completion */ if (int_queues_only && cs_needs_completion(cs)) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); dev_err(hdev->dev, "Reject CS %d.%llu since it contains only internal queues jobs and needs completion\n", cs->ctx->asid, cs->sequence); rc = -EINVAL; goto free_cs_object; } if (using_hw_queues) INIT_WORK(&cs->finish_work, cs_completion); /* * store the (external/HW queues) streams used by the CS in the * fence object for multi-CS completion */ if (hdev->supports_wait_for_multi_cs) cs->fence->stream_master_qid_map = stream_master_qid_map; rc = hl_hw_queue_schedule_cs(cs); if (rc) { if (rc != -EAGAIN) dev_err(hdev->dev, "Failed to submit CS %d.%llu to H/W queues, error %d\n", cs->ctx->asid, cs->sequence, rc); goto free_cs_object; } *signal_initial_sob_count = cs->initial_sob_count; rc = HL_CS_STATUS_SUCCESS; goto put_cs; release_cb: atomic_dec(&cb->cs_cnt); hl_cb_put(cb); free_cs_object: cs_rollback(hdev, cs); *cs_seq = ULLONG_MAX; /* The path below is both for good and erroneous exits */ put_cs: /* We finished with the CS in this function, so put the ref */ cs_put(cs); free_cs_chunk_array: kfree(cs_chunk_array); out: return rc; } static int hl_cs_ctx_switch(struct hl_fpriv *hpriv, union hl_cs_args *args, u64 *cs_seq) { struct hl_device *hdev = hpriv->hdev; struct hl_ctx *ctx = hpriv->ctx; bool need_soft_reset = false; int rc = 0, do_ctx_switch = 0; void __user *chunks; u32 num_chunks, tmp; u16 sob_count; int ret; if (hdev->supports_ctx_switch) do_ctx_switch = atomic_cmpxchg(&ctx->thread_ctx_switch_token, 1, 0); if (do_ctx_switch || (args->in.cs_flags & HL_CS_FLAGS_FORCE_RESTORE)) { mutex_lock(&hpriv->restore_phase_mutex); if (do_ctx_switch) { rc = hdev->asic_funcs->context_switch(hdev, ctx->asid); if (rc) { dev_err_ratelimited(hdev->dev, "Failed to switch to context %d, rejecting CS! %d\n", ctx->asid, rc); /* * If we timedout, or if the device is not IDLE * while we want to do context-switch (-EBUSY), * we need to soft-reset because QMAN is * probably stuck. However, we can't call to * reset here directly because of deadlock, so * need to do it at the very end of this * function */ if ((rc == -ETIMEDOUT) || (rc == -EBUSY)) need_soft_reset = true; mutex_unlock(&hpriv->restore_phase_mutex); goto out; } } hdev->asic_funcs->restore_phase_topology(hdev); chunks = (void __user *) (uintptr_t) args->in.chunks_restore; num_chunks = args->in.num_chunks_restore; if (!num_chunks) { dev_dbg(hdev->dev, "Need to run restore phase but restore CS is empty\n"); rc = 0; } else { rc = cs_ioctl_default(hpriv, chunks, num_chunks, cs_seq, 0, 0, hdev->timeout_jiffies, &sob_count); } mutex_unlock(&hpriv->restore_phase_mutex); if (rc) { dev_err(hdev->dev, "Failed to submit restore CS for context %d (%d)\n", ctx->asid, rc); goto out; } /* Need to wait for restore completion before execution phase */ if (num_chunks) { enum hl_cs_wait_status status; wait_again: ret = _hl_cs_wait_ioctl(hdev, ctx, jiffies_to_usecs(hdev->timeout_jiffies), *cs_seq, &status, NULL); if (ret) { if (ret == -ERESTARTSYS) { usleep_range(100, 200); goto wait_again; } dev_err(hdev->dev, "Restore CS for context %d failed to complete %d\n", ctx->asid, ret); rc = -ENOEXEC; goto out; } } if (hdev->supports_ctx_switch) ctx->thread_ctx_switch_wait_token = 1; } else if (hdev->supports_ctx_switch && !ctx->thread_ctx_switch_wait_token) { rc = hl_poll_timeout_memory(hdev, &ctx->thread_ctx_switch_wait_token, tmp, (tmp == 1), 100, jiffies_to_usecs(hdev->timeout_jiffies), false); if (rc == -ETIMEDOUT) { dev_err(hdev->dev, "context switch phase timeout (%d)\n", tmp); goto out; } } out: if ((rc == -ETIMEDOUT || rc == -EBUSY) && (need_soft_reset)) hl_device_reset(hdev, 0); return rc; } /* * hl_cs_signal_sob_wraparound_handler: handle SOB value wrapaound case. * if the SOB value reaches the max value move to the other SOB reserved * to the queue. * @hdev: pointer to device structure * @q_idx: stream queue index * @hw_sob: the H/W SOB used in this signal CS. * @count: signals count * @encaps_sig: tells whether it's reservation for encaps signals or not. * * Note that this function must be called while hw_queues_lock is taken. */ int hl_cs_signal_sob_wraparound_handler(struct hl_device *hdev, u32 q_idx, struct hl_hw_sob **hw_sob, u32 count, bool encaps_sig) { struct hl_sync_stream_properties *prop; struct hl_hw_sob *sob = *hw_sob, *other_sob; u8 other_sob_offset; prop = &hdev->kernel_queues[q_idx].sync_stream_prop; hw_sob_get(sob); /* check for wraparound */ if (prop->next_sob_val + count >= HL_MAX_SOB_VAL) { /* * Decrement as we reached the max value. * The release function won't be called here as we've * just incremented the refcount right before calling this * function. */ hw_sob_put_err(sob); /* * check the other sob value, if it still in use then fail * otherwise make the switch */ other_sob_offset = (prop->curr_sob_offset + 1) % HL_RSVD_SOBS; other_sob = &prop->hw_sob[other_sob_offset]; if (kref_read(&other_sob->kref) != 1) { dev_err(hdev->dev, "error: Cannot switch SOBs q_idx: %d\n", q_idx); return -EINVAL; } /* * next_sob_val always points to the next available signal * in the sob, so in encaps signals it will be the next one * after reserving the required amount. */ if (encaps_sig) prop->next_sob_val = count + 1; else prop->next_sob_val = count; /* only two SOBs are currently in use */ prop->curr_sob_offset = other_sob_offset; *hw_sob = other_sob; /* * check if other_sob needs reset, then do it before using it * for the reservation or the next signal cs. * we do it here, and for both encaps and regular signal cs * cases in order to avoid possible races of two kref_put * of the sob which can occur at the same time if we move the * sob reset(kref_put) to cs_do_release function. * in addition, if we have combination of cs signal and * encaps, and at the point we need to reset the sob there was * no more reservations and only signal cs keep coming, * in such case we need signal_cs to put the refcount and * reset the sob. */ if (other_sob->need_reset) hw_sob_put(other_sob); if (encaps_sig) { /* set reset indication for the sob */ sob->need_reset = true; hw_sob_get(other_sob); } dev_dbg(hdev->dev, "switched to SOB %d, q_idx: %d\n", prop->curr_sob_offset, q_idx); } else { prop->next_sob_val += count; } return 0; } static int cs_ioctl_extract_signal_seq(struct hl_device *hdev, struct hl_cs_chunk *chunk, u64 *signal_seq, struct hl_ctx *ctx, bool encaps_signals) { u64 *signal_seq_arr = NULL; u32 size_to_copy, signal_seq_arr_len; int rc = 0; if (encaps_signals) { *signal_seq = chunk->encaps_signal_seq; return 0; } signal_seq_arr_len = chunk->num_signal_seq_arr; /* currently only one signal seq is supported */ if (signal_seq_arr_len != 1) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&hdev->aggregated_cs_counters.validation_drop_cnt); dev_err(hdev->dev, "Wait for signal CS supports only one signal CS seq\n"); return -EINVAL; } signal_seq_arr = kmalloc_array(signal_seq_arr_len, sizeof(*signal_seq_arr), GFP_ATOMIC); if (!signal_seq_arr) signal_seq_arr = kmalloc_array(signal_seq_arr_len, sizeof(*signal_seq_arr), GFP_KERNEL); if (!signal_seq_arr) { atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); atomic64_inc(&hdev->aggregated_cs_counters.out_of_mem_drop_cnt); return -ENOMEM; } size_to_copy = signal_seq_arr_len * sizeof(*signal_seq_arr); if (copy_from_user(signal_seq_arr, u64_to_user_ptr(chunk->signal_seq_arr), size_to_copy)) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&hdev->aggregated_cs_counters.validation_drop_cnt); dev_err(hdev->dev, "Failed to copy signal seq array from user\n"); rc = -EFAULT; goto out; } /* currently it is guaranteed to have only one signal seq */ *signal_seq = signal_seq_arr[0]; out: kfree(signal_seq_arr); return rc; } static int cs_ioctl_signal_wait_create_jobs(struct hl_device *hdev, struct hl_ctx *ctx, struct hl_cs *cs, enum hl_queue_type q_type, u32 q_idx, u32 encaps_signal_offset) { struct hl_cs_counters_atomic *cntr; struct hl_cs_job *job; struct hl_cb *cb; u32 cb_size; cntr = &hdev->aggregated_cs_counters; job = hl_cs_allocate_job(hdev, q_type, true); if (!job) { atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); atomic64_inc(&cntr->out_of_mem_drop_cnt); dev_err(hdev->dev, "Failed to allocate a new job\n"); return -ENOMEM; } if (cs->type == CS_TYPE_WAIT) cb_size = hdev->asic_funcs->get_wait_cb_size(hdev); else cb_size = hdev->asic_funcs->get_signal_cb_size(hdev); cb = hl_cb_kernel_create(hdev, cb_size, q_type == QUEUE_TYPE_HW && hdev->mmu_enable); if (!cb) { atomic64_inc(&ctx->cs_counters.out_of_mem_drop_cnt); atomic64_inc(&cntr->out_of_mem_drop_cnt); kfree(job); return -EFAULT; } job->id = 0; job->cs = cs; job->user_cb = cb; atomic_inc(&job->user_cb->cs_cnt); job->user_cb_size = cb_size; job->hw_queue_id = q_idx; if ((cs->type == CS_TYPE_WAIT || cs->type == CS_TYPE_COLLECTIVE_WAIT) && cs->encaps_signals) job->encaps_sig_wait_offset = encaps_signal_offset; /* * No need in parsing, user CB is the patched CB. * We call hl_cb_destroy() out of two reasons - we don't need the CB in * the CB idr anymore and to decrement its refcount as it was * incremented inside hl_cb_kernel_create(). */ job->patched_cb = job->user_cb; job->job_cb_size = job->user_cb_size; hl_cb_destroy(&hdev->kernel_mem_mgr, cb->buf->handle); /* increment refcount as for external queues we get completion */ cs_get(cs); cs->jobs_in_queue_cnt[job->hw_queue_id]++; cs->jobs_cnt++; list_add_tail(&job->cs_node, &cs->job_list); hl_debugfs_add_job(hdev, job); return 0; } static int cs_ioctl_reserve_signals(struct hl_fpriv *hpriv, u32 q_idx, u32 count, u32 *handle_id, u32 *sob_addr, u32 *signals_count) { struct hw_queue_properties *hw_queue_prop; struct hl_sync_stream_properties *prop; struct hl_device *hdev = hpriv->hdev; struct hl_cs_encaps_sig_handle *handle; struct hl_encaps_signals_mgr *mgr; struct hl_hw_sob *hw_sob; int hdl_id; int rc = 0; if (count >= HL_MAX_SOB_VAL) { dev_err(hdev->dev, "signals count(%u) exceeds the max SOB value\n", count); rc = -EINVAL; goto out; } if (q_idx >= hdev->asic_prop.max_queues) { dev_err(hdev->dev, "Queue index %d is invalid\n", q_idx); rc = -EINVAL; goto out; } hw_queue_prop = &hdev->asic_prop.hw_queues_props[q_idx]; if (!hw_queue_prop->supports_sync_stream) { dev_err(hdev->dev, "Queue index %d does not support sync stream operations\n", q_idx); rc = -EINVAL; goto out; } prop = &hdev->kernel_queues[q_idx].sync_stream_prop; handle = kzalloc(sizeof(*handle), GFP_KERNEL); if (!handle) { rc = -ENOMEM; goto out; } handle->count = count; hl_ctx_get(hpriv->ctx); handle->ctx = hpriv->ctx; mgr = &hpriv->ctx->sig_mgr; spin_lock(&mgr->lock); hdl_id = idr_alloc(&mgr->handles, handle, 1, 0, GFP_ATOMIC); spin_unlock(&mgr->lock); if (hdl_id < 0) { dev_err(hdev->dev, "Failed to allocate IDR for a new signal reservation\n"); rc = -EINVAL; goto put_ctx; } handle->id = hdl_id; handle->q_idx = q_idx; handle->hdev = hdev; kref_init(&handle->refcount); hdev->asic_funcs->hw_queues_lock(hdev); hw_sob = &prop->hw_sob[prop->curr_sob_offset]; /* * Increment the SOB value by count by user request * to reserve those signals * check if the signals amount to reserve is not exceeding the max sob * value, if yes then switch sob. */ rc = hl_cs_signal_sob_wraparound_handler(hdev, q_idx, &hw_sob, count, true); if (rc) { dev_err(hdev->dev, "Failed to switch SOB\n"); hdev->asic_funcs->hw_queues_unlock(hdev); rc = -EINVAL; goto remove_idr; } /* set the hw_sob to the handle after calling the sob wraparound handler * since sob could have changed. */ handle->hw_sob = hw_sob; /* store the current sob value for unreserve validity check, and * signal offset support */ handle->pre_sob_val = prop->next_sob_val - handle->count; handle->cs_seq = ULLONG_MAX; *signals_count = prop->next_sob_val; hdev->asic_funcs->hw_queues_unlock(hdev); *sob_addr = handle->hw_sob->sob_addr; *handle_id = hdl_id; dev_dbg(hdev->dev, "Signals reserved, sob_id: %d, sob addr: 0x%x, last sob_val: %u, q_idx: %d, hdl_id: %d\n", hw_sob->sob_id, handle->hw_sob->sob_addr, prop->next_sob_val - 1, q_idx, hdl_id); goto out; remove_idr: spin_lock(&mgr->lock); idr_remove(&mgr->handles, hdl_id); spin_unlock(&mgr->lock); put_ctx: hl_ctx_put(handle->ctx); kfree(handle); out: return rc; } static int cs_ioctl_unreserve_signals(struct hl_fpriv *hpriv, u32 handle_id) { struct hl_cs_encaps_sig_handle *encaps_sig_hdl; struct hl_sync_stream_properties *prop; struct hl_device *hdev = hpriv->hdev; struct hl_encaps_signals_mgr *mgr; struct hl_hw_sob *hw_sob; u32 q_idx, sob_addr; int rc = 0; mgr = &hpriv->ctx->sig_mgr; spin_lock(&mgr->lock); encaps_sig_hdl = idr_find(&mgr->handles, handle_id); if (encaps_sig_hdl) { dev_dbg(hdev->dev, "unreserve signals, handle: %u, SOB:0x%x, count: %u\n", handle_id, encaps_sig_hdl->hw_sob->sob_addr, encaps_sig_hdl->count); hdev->asic_funcs->hw_queues_lock(hdev); q_idx = encaps_sig_hdl->q_idx; prop = &hdev->kernel_queues[q_idx].sync_stream_prop; hw_sob = &prop->hw_sob[prop->curr_sob_offset]; sob_addr = hdev->asic_funcs->get_sob_addr(hdev, hw_sob->sob_id); /* Check if sob_val got out of sync due to other * signal submission requests which were handled * between the reserve-unreserve calls or SOB switch * upon reaching SOB max value. */ if (encaps_sig_hdl->pre_sob_val + encaps_sig_hdl->count != prop->next_sob_val || sob_addr != encaps_sig_hdl->hw_sob->sob_addr) { dev_err(hdev->dev, "Cannot unreserve signals, SOB val ran out of sync, expected: %u, actual val: %u\n", encaps_sig_hdl->pre_sob_val, (prop->next_sob_val - encaps_sig_hdl->count)); hdev->asic_funcs->hw_queues_unlock(hdev); rc = -EINVAL; goto out; } /* * Decrement the SOB value by count by user request * to unreserve those signals */ prop->next_sob_val -= encaps_sig_hdl->count; hdev->asic_funcs->hw_queues_unlock(hdev); hw_sob_put(hw_sob); /* Release the id and free allocated memory of the handle */ idr_remove(&mgr->handles, handle_id); hl_ctx_put(encaps_sig_hdl->ctx); kfree(encaps_sig_hdl); } else { rc = -EINVAL; dev_err(hdev->dev, "failed to unreserve signals, cannot find handler\n"); } out: spin_unlock(&mgr->lock); return rc; } static int cs_ioctl_signal_wait(struct hl_fpriv *hpriv, enum hl_cs_type cs_type, void __user *chunks, u32 num_chunks, u64 *cs_seq, u32 flags, u32 timeout, u32 *signal_sob_addr_offset, u16 *signal_initial_sob_count) { struct hl_cs_encaps_sig_handle *encaps_sig_hdl = NULL; bool handle_found = false, is_wait_cs = false, wait_cs_submitted = false, cs_encaps_signals = false; struct hl_cs_chunk *cs_chunk_array, *chunk; bool staged_cs_with_encaps_signals = false; struct hw_queue_properties *hw_queue_prop; struct hl_device *hdev = hpriv->hdev; struct hl_cs_compl *sig_waitcs_cmpl; u32 q_idx, collective_engine_id = 0; struct hl_cs_counters_atomic *cntr; struct hl_fence *sig_fence = NULL; struct hl_ctx *ctx = hpriv->ctx; enum hl_queue_type q_type; struct hl_cs *cs; u64 signal_seq; int rc; cntr = &hdev->aggregated_cs_counters; *cs_seq = ULLONG_MAX; rc = hl_cs_copy_chunk_array(hdev, &cs_chunk_array, chunks, num_chunks, ctx); if (rc) goto out; /* currently it is guaranteed to have only one chunk */ chunk = &cs_chunk_array[0]; if (chunk->queue_index >= hdev->asic_prop.max_queues) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); dev_err(hdev->dev, "Queue index %d is invalid\n", chunk->queue_index); rc = -EINVAL; goto free_cs_chunk_array; } q_idx = chunk->queue_index; hw_queue_prop = &hdev->asic_prop.hw_queues_props[q_idx]; q_type = hw_queue_prop->type; if (!hw_queue_prop->supports_sync_stream) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); dev_err(hdev->dev, "Queue index %d does not support sync stream operations\n", q_idx); rc = -EINVAL; goto free_cs_chunk_array; } if (cs_type == CS_TYPE_COLLECTIVE_WAIT) { if (!(hw_queue_prop->collective_mode == HL_COLLECTIVE_MASTER)) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); dev_err(hdev->dev, "Queue index %d is invalid\n", q_idx); rc = -EINVAL; goto free_cs_chunk_array; } if (!hdev->nic_ports_mask) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); dev_err(hdev->dev, "Collective operations not supported when NIC ports are disabled"); rc = -EINVAL; goto free_cs_chunk_array; } collective_engine_id = chunk->collective_engine_id; } is_wait_cs = !!(cs_type == CS_TYPE_WAIT || cs_type == CS_TYPE_COLLECTIVE_WAIT); cs_encaps_signals = !!(flags & HL_CS_FLAGS_ENCAP_SIGNALS); if (is_wait_cs) { rc = cs_ioctl_extract_signal_seq(hdev, chunk, &signal_seq, ctx, cs_encaps_signals); if (rc) goto free_cs_chunk_array; if (cs_encaps_signals) { /* check if cs sequence has encapsulated * signals handle */ struct idr *idp; u32 id; spin_lock(&ctx->sig_mgr.lock); idp = &ctx->sig_mgr.handles; idr_for_each_entry(idp, encaps_sig_hdl, id) { if (encaps_sig_hdl->cs_seq == signal_seq) { /* get refcount to protect removing this handle from idr, * needed when multiple wait cs are used with offset * to wait on reserved encaps signals. * Since kref_put of this handle is executed outside the * current lock, it is possible that the handle refcount * is 0 but it yet to be removed from the list. In this * case need to consider the handle as not valid. */ if (kref_get_unless_zero(&encaps_sig_hdl->refcount)) handle_found = true; break; } } spin_unlock(&ctx->sig_mgr.lock); if (!handle_found) { /* treat as signal CS already finished */ dev_dbg(hdev->dev, "Cannot find encapsulated signals handle for seq 0x%llx\n", signal_seq); rc = 0; goto free_cs_chunk_array; } /* validate also the signal offset value */ if (chunk->encaps_signal_offset > encaps_sig_hdl->count) { dev_err(hdev->dev, "offset(%u) value exceed max reserved signals count(%u)!\n", chunk->encaps_signal_offset, encaps_sig_hdl->count); rc = -EINVAL; goto free_cs_chunk_array; } } sig_fence = hl_ctx_get_fence(ctx, signal_seq); if (IS_ERR(sig_fence)) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); dev_err(hdev->dev, "Failed to get signal CS with seq 0x%llx\n", signal_seq); rc = PTR_ERR(sig_fence); goto free_cs_chunk_array; } if (!sig_fence) { /* signal CS already finished */ rc = 0; goto free_cs_chunk_array; } sig_waitcs_cmpl = container_of(sig_fence, struct hl_cs_compl, base_fence); staged_cs_with_encaps_signals = !! (sig_waitcs_cmpl->type == CS_TYPE_DEFAULT && (flags & HL_CS_FLAGS_ENCAP_SIGNALS)); if (sig_waitcs_cmpl->type != CS_TYPE_SIGNAL && !staged_cs_with_encaps_signals) { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); dev_err(hdev->dev, "CS seq 0x%llx is not of a signal/encaps-signal CS\n", signal_seq); hl_fence_put(sig_fence); rc = -EINVAL; goto free_cs_chunk_array; } if (completion_done(&sig_fence->completion)) { /* signal CS already finished */ hl_fence_put(sig_fence); rc = 0; goto free_cs_chunk_array; } } rc = allocate_cs(hdev, ctx, cs_type, ULLONG_MAX, &cs, flags, timeout); if (rc) { if (is_wait_cs) hl_fence_put(sig_fence); goto free_cs_chunk_array; } /* * Save the signal CS fence for later initialization right before * hanging the wait CS on the queue. * for encaps signals case, we save the cs sequence and handle pointer * for later initialization. */ if (is_wait_cs) { cs->signal_fence = sig_fence; /* store the handle pointer, so we don't have to * look for it again, later on the flow * when we need to set SOB info in hw_queue. */ if (cs->encaps_signals) cs->encaps_sig_hdl = encaps_sig_hdl; } hl_debugfs_add_cs(cs); *cs_seq = cs->sequence; if (cs_type == CS_TYPE_WAIT || cs_type == CS_TYPE_SIGNAL) rc = cs_ioctl_signal_wait_create_jobs(hdev, ctx, cs, q_type, q_idx, chunk->encaps_signal_offset); else if (cs_type == CS_TYPE_COLLECTIVE_WAIT) rc = hdev->asic_funcs->collective_wait_create_jobs(hdev, ctx, cs, q_idx, collective_engine_id, chunk->encaps_signal_offset); else { atomic64_inc(&ctx->cs_counters.validation_drop_cnt); atomic64_inc(&cntr->validation_drop_cnt); rc = -EINVAL; } if (rc) goto free_cs_object; if (q_type == QUEUE_TYPE_HW) INIT_WORK(&cs->finish_work, cs_completion); rc = hl_hw_queue_schedule_cs(cs); if (rc) { /* In case wait cs failed here, it means the signal cs * already completed. we want to free all it's related objects * but we don't want to fail the ioctl. */ if (is_wait_cs) rc = 0; else if (rc != -EAGAIN) dev_err(hdev->dev, "Failed to submit CS %d.%llu to H/W queues, error %d\n", ctx->asid, cs->sequence, rc); goto free_cs_object; } *signal_sob_addr_offset = cs->sob_addr_offset; *signal_initial_sob_count = cs->initial_sob_count; rc = HL_CS_STATUS_SUCCESS; if (is_wait_cs) wait_cs_submitted = true; goto put_cs; free_cs_object: cs_rollback(hdev, cs); *cs_seq = ULLONG_MAX; /* The path below is both for good and erroneous exits */ put_cs: /* We finished with the CS in this function, so put the ref */ cs_put(cs); free_cs_chunk_array: if (!wait_cs_submitted && cs_encaps_signals && handle_found && is_wait_cs) kref_put(&encaps_sig_hdl->refcount, hl_encaps_release_handle_and_put_ctx); kfree(cs_chunk_array); out: return rc; } static int cs_ioctl_engine_cores(struct hl_fpriv *hpriv, u64 engine_cores, u32 num_engine_cores, u32 core_command) { int rc; struct hl_device *hdev = hpriv->hdev; void __user *engine_cores_arr; u32 *cores; if (!num_engine_cores || num_engine_cores > hdev->asic_prop.num_engine_cores) { dev_err(hdev->dev, "Number of engine cores %d is invalid\n", num_engine_cores); return -EINVAL; } if (core_command != HL_ENGINE_CORE_RUN && core_command != HL_ENGINE_CORE_HALT) { dev_err(hdev->dev, "Engine core command is invalid\n"); return -EINVAL; } engine_cores_arr = (void __user *) (uintptr_t) engine_cores; cores = kmalloc_array(num_engine_cores, sizeof(u32), GFP_KERNEL); if (!cores) return -ENOMEM; if (copy_from_user(cores, engine_cores_arr, num_engine_cores * sizeof(u32))) { dev_err(hdev->dev, "Failed to copy core-ids array from user\n"); kfree(cores); return -EFAULT; } rc = hdev->asic_funcs->set_engine_cores(hdev, cores, num_engine_cores, core_command); kfree(cores); return rc; } static int cs_ioctl_flush_pci_hbw_writes(struct hl_fpriv *hpriv) { struct hl_device *hdev = hpriv->hdev; struct asic_fixed_properties *prop = &hdev->asic_prop; if (!prop->hbw_flush_reg) { dev_dbg(hdev->dev, "HBW flush is not supported\n"); return -EOPNOTSUPP; } RREG32(prop->hbw_flush_reg); return 0; } int hl_cs_ioctl(struct hl_fpriv *hpriv, void *data) { union hl_cs_args *args = data; enum hl_cs_type cs_type = 0; u64 cs_seq = ULONG_MAX; void __user *chunks; u32 num_chunks, flags, timeout, signals_count = 0, sob_addr = 0, handle_id = 0; u16 sob_initial_count = 0; int rc; rc = hl_cs_sanity_checks(hpriv, args); if (rc) goto out; rc = hl_cs_ctx_switch(hpriv, args, &cs_seq); if (rc) goto out; cs_type = hl_cs_get_cs_type(args->in.cs_flags & ~HL_CS_FLAGS_FORCE_RESTORE); chunks = (void __user *) (uintptr_t) args->in.chunks_execute; num_chunks = args->in.num_chunks_execute; flags = args->in.cs_flags; /* In case this is a staged CS, user should supply the CS sequence */ if ((flags & HL_CS_FLAGS_STAGED_SUBMISSION) && !(flags & HL_CS_FLAGS_STAGED_SUBMISSION_FIRST)) cs_seq = args->in.seq; timeout = flags & HL_CS_FLAGS_CUSTOM_TIMEOUT ? msecs_to_jiffies(args->in.timeout * 1000) : hpriv->hdev->timeout_jiffies; switch (cs_type) { case CS_TYPE_SIGNAL: case CS_TYPE_WAIT: case CS_TYPE_COLLECTIVE_WAIT: rc = cs_ioctl_signal_wait(hpriv, cs_type, chunks, num_chunks, &cs_seq, args->in.cs_flags, timeout, &sob_addr, &sob_initial_count); break; case CS_RESERVE_SIGNALS: rc = cs_ioctl_reserve_signals(hpriv, args->in.encaps_signals_q_idx, args->in.encaps_signals_count, &handle_id, &sob_addr, &signals_count); break; case CS_UNRESERVE_SIGNALS: rc = cs_ioctl_unreserve_signals(hpriv, args->in.encaps_sig_handle_id); break; case CS_TYPE_ENGINE_CORE: rc = cs_ioctl_engine_cores(hpriv, args->in.engine_cores, args->in.num_engine_cores, args->in.core_command); break; case CS_TYPE_FLUSH_PCI_HBW_WRITES: rc = cs_ioctl_flush_pci_hbw_writes(hpriv); break; default: rc = cs_ioctl_default(hpriv, chunks, num_chunks, &cs_seq, args->in.cs_flags, args->in.encaps_sig_handle_id, timeout, &sob_initial_count); break; } out: if (rc != -EAGAIN) { memset(args, 0, sizeof(*args)); switch (cs_type) { case CS_RESERVE_SIGNALS: args->out.handle_id = handle_id; args->out.sob_base_addr_offset = sob_addr; args->out.count = signals_count; break; case CS_TYPE_SIGNAL: args->out.sob_base_addr_offset = sob_addr; args->out.sob_count_before_submission = sob_initial_count; args->out.seq = cs_seq; break; case CS_TYPE_DEFAULT: args->out.sob_count_before_submission = sob_initial_count; args->out.seq = cs_seq; break; default: args->out.seq = cs_seq; break; } args->out.status = rc; } return rc; } static int hl_wait_for_fence(struct hl_ctx *ctx, u64 seq, struct hl_fence *fence, enum hl_cs_wait_status *status, u64 timeout_us, s64 *timestamp) { struct hl_device *hdev = ctx->hdev; ktime_t timestamp_kt; long completion_rc; int rc = 0, error; if (IS_ERR(fence)) { rc = PTR_ERR(fence); if (rc == -EINVAL) dev_notice_ratelimited(hdev->dev, "Can't wait on CS %llu because current CS is at seq %llu\n", seq, ctx->cs_sequence); return rc; } if (!fence) { if (!hl_pop_cs_outcome(&ctx->outcome_store, seq, ×tamp_kt, &error)) { dev_dbg(hdev->dev, "Can't wait on seq %llu because current CS is at seq %llu (Fence is gone)\n", seq, ctx->cs_sequence); *status = CS_WAIT_STATUS_GONE; return 0; } completion_rc = 1; goto report_results; } if (!timeout_us) { completion_rc = completion_done(&fence->completion); } else { unsigned long timeout; timeout = (timeout_us == MAX_SCHEDULE_TIMEOUT) ? timeout_us : usecs_to_jiffies(timeout_us); completion_rc = wait_for_completion_interruptible_timeout( &fence->completion, timeout); } error = fence->error; timestamp_kt = fence->timestamp; report_results: if (completion_rc > 0) { *status = CS_WAIT_STATUS_COMPLETED; if (timestamp) *timestamp = ktime_to_ns(timestamp_kt); } else { *status = CS_WAIT_STATUS_BUSY; } if (completion_rc == -ERESTARTSYS) rc = completion_rc; else if (error == -ETIMEDOUT || error == -EIO) rc = error; return rc; } /* * hl_cs_poll_fences - iterate CS fences to check for CS completion * * @mcs_data: multi-CS internal data * @mcs_compl: multi-CS completion structure * * @return 0 on success, otherwise non 0 error code * * The function iterates on all CS sequence in the list and set bit in * completion_bitmap for each completed CS. * While iterating, the function sets the stream map of each fence in the fence * array in the completion QID stream map to be used by CSs to perform * completion to the multi-CS context. * This function shall be called after taking context ref */ static int hl_cs_poll_fences(struct multi_cs_data *mcs_data, struct multi_cs_completion *mcs_compl) { struct hl_fence **fence_ptr = mcs_data->fence_arr; struct hl_device *hdev = mcs_data->ctx->hdev; int i, rc, arr_len = mcs_data->arr_len; u64 *seq_arr = mcs_data->seq_arr; ktime_t max_ktime, first_cs_time; enum hl_cs_wait_status status; memset(fence_ptr, 0, arr_len * sizeof(struct hl_fence *)); /* get all fences under the same lock */ rc = hl_ctx_get_fences(mcs_data->ctx, seq_arr, fence_ptr, arr_len); if (rc) return rc; /* * re-initialize the completion here to handle 2 possible cases: * 1. CS will complete the multi-CS prior clearing the completion. in which * case the fence iteration is guaranteed to catch the CS completion. * 2. the completion will occur after re-init of the completion. * in which case we will wake up immediately in wait_for_completion. */ reinit_completion(&mcs_compl->completion); /* * set to maximum time to verify timestamp is valid: if at the end * this value is maintained- no timestamp was updated */ max_ktime = ktime_set(KTIME_SEC_MAX, 0); first_cs_time = max_ktime; for (i = 0; i < arr_len; i++, fence_ptr++) { struct hl_fence *fence = *fence_ptr; /* * In order to prevent case where we wait until timeout even though a CS associated * with the multi-CS actually completed we do things in the below order: * 1. for each fence set it's QID map in the multi-CS completion QID map. This way * any CS can, potentially, complete the multi CS for the specific QID (note * that once completion is initialized, calling complete* and then wait on the * completion will cause it to return at once) * 2. only after allowing multi-CS completion for the specific QID we check whether * the specific CS already completed (and thus the wait for completion part will * be skipped). if the CS not completed it is guaranteed that completing CS will * wake up the completion. */ if (fence) mcs_compl->stream_master_qid_map |= fence->stream_master_qid_map; /* * function won't sleep as it is called with timeout 0 (i.e. * poll the fence) */ rc = hl_wait_for_fence(mcs_data->ctx, seq_arr[i], fence, &status, 0, NULL); if (rc) { dev_err(hdev->dev, "wait_for_fence error :%d for CS seq %llu\n", rc, seq_arr[i]); break; } switch (status) { case CS_WAIT_STATUS_BUSY: /* CS did not finished, QID to wait on already stored */ break; case CS_WAIT_STATUS_COMPLETED: /* * Using mcs_handling_done to avoid possibility of mcs_data * returns to user indicating CS completed before it finished * all of its mcs handling, to avoid race the next time the * user waits for mcs. * note: when reaching this case fence is definitely not NULL * but NULL check was added to overcome static analysis */ if (fence && !fence->mcs_handling_done) { /* * in case multi CS is completed but MCS handling not done * we "complete" the multi CS to prevent it from waiting * until time-out and the "multi-CS handling done" will have * another chance at the next iteration */ complete_all(&mcs_compl->completion); break; } mcs_data->completion_bitmap |= BIT(i); /* * For all completed CSs we take the earliest timestamp. * For this we have to validate that the timestamp is * earliest of all timestamps so far. */ if (fence && mcs_data->update_ts && (ktime_compare(fence->timestamp, first_cs_time) < 0)) first_cs_time = fence->timestamp; break; case CS_WAIT_STATUS_GONE: mcs_data->update_ts = false; mcs_data->gone_cs = true; /* * It is possible to get an old sequence numbers from user * which related to already completed CSs and their fences * already gone. In this case, CS set as completed but * no need to consider its QID for mcs completion. */ mcs_data->completion_bitmap |= BIT(i); break; default: dev_err(hdev->dev, "Invalid fence status\n"); rc = -EINVAL; break; } } hl_fences_put(mcs_data->fence_arr, arr_len); if (mcs_data->update_ts && (ktime_compare(first_cs_time, max_ktime) != 0)) mcs_data->timestamp = ktime_to_ns(first_cs_time); return rc; } static int _hl_cs_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx, u64 timeout_us, u64 seq, enum hl_cs_wait_status *status, s64 *timestamp) { struct hl_fence *fence; int rc = 0; if (timestamp) *timestamp = 0; hl_ctx_get(ctx); fence = hl_ctx_get_fence(ctx, seq); rc = hl_wait_for_fence(ctx, seq, fence, status, timeout_us, timestamp); hl_fence_put(fence); hl_ctx_put(ctx); return rc; } static inline unsigned long hl_usecs64_to_jiffies(const u64 usecs) { if (usecs <= U32_MAX) return usecs_to_jiffies(usecs); /* * If the value in nanoseconds is larger than 64 bit, use the largest * 64 bit value. */ if (usecs >= ((u64)(U64_MAX / NSEC_PER_USEC))) return nsecs_to_jiffies(U64_MAX); return nsecs_to_jiffies(usecs * NSEC_PER_USEC); } /* * hl_wait_multi_cs_completion_init - init completion structure * * @hdev: pointer to habanalabs device structure * @stream_master_bitmap: stream master QIDs map, set bit indicates stream * master QID to wait on * * @return valid completion struct pointer on success, otherwise error pointer * * up to MULTI_CS_MAX_USER_CTX calls can be done concurrently to the driver. * the function gets the first available completion (by marking it "used") * and initialize its values. */ static struct multi_cs_completion *hl_wait_multi_cs_completion_init(struct hl_device *hdev) { struct multi_cs_completion *mcs_compl; int i; /* find free multi_cs completion structure */ for (i = 0; i < MULTI_CS_MAX_USER_CTX; i++) { mcs_compl = &hdev->multi_cs_completion[i]; spin_lock(&mcs_compl->lock); if (!mcs_compl->used) { mcs_compl->used = 1; mcs_compl->timestamp = 0; /* * init QID map to 0 to avoid completion by CSs. the actual QID map * to multi-CS CSs will be set incrementally at a later stage */ mcs_compl->stream_master_qid_map = 0; spin_unlock(&mcs_compl->lock); break; } spin_unlock(&mcs_compl->lock); } if (i == MULTI_CS_MAX_USER_CTX) { dev_err(hdev->dev, "no available multi-CS completion structure\n"); return ERR_PTR(-ENOMEM); } return mcs_compl; } /* * hl_wait_multi_cs_completion_fini - return completion structure and set as * unused * * @mcs_compl: pointer to the completion structure */ static void hl_wait_multi_cs_completion_fini( struct multi_cs_completion *mcs_compl) { /* * free completion structure, do it under lock to be in-sync with the * thread that signals completion */ spin_lock(&mcs_compl->lock); mcs_compl->used = 0; spin_unlock(&mcs_compl->lock); } /* * hl_wait_multi_cs_completion - wait for first CS to complete * * @mcs_data: multi-CS internal data * * @return 0 on success, otherwise non 0 error code */ static int hl_wait_multi_cs_completion(struct multi_cs_data *mcs_data, struct multi_cs_completion *mcs_compl) { long completion_rc; completion_rc = wait_for_completion_interruptible_timeout(&mcs_compl->completion, mcs_data->timeout_jiffies); /* update timestamp */ if (completion_rc > 0) mcs_data->timestamp = mcs_compl->timestamp; if (completion_rc == -ERESTARTSYS) return completion_rc; mcs_data->wait_status = completion_rc; return 0; } /* * hl_multi_cs_completion_init - init array of multi-CS completion structures * * @hdev: pointer to habanalabs device structure */ void hl_multi_cs_completion_init(struct hl_device *hdev) { struct multi_cs_completion *mcs_cmpl; int i; for (i = 0; i < MULTI_CS_MAX_USER_CTX; i++) { mcs_cmpl = &hdev->multi_cs_completion[i]; mcs_cmpl->used = 0; spin_lock_init(&mcs_cmpl->lock); init_completion(&mcs_cmpl->completion); } } /* * hl_multi_cs_wait_ioctl - implementation of the multi-CS wait ioctl * * @hpriv: pointer to the private data of the fd * @data: pointer to multi-CS wait ioctl in/out args * */ static int hl_multi_cs_wait_ioctl(struct hl_fpriv *hpriv, void *data) { struct multi_cs_completion *mcs_compl; struct hl_device *hdev = hpriv->hdev; struct multi_cs_data mcs_data = {}; union hl_wait_cs_args *args = data; struct hl_ctx *ctx = hpriv->ctx; struct hl_fence **fence_arr; void __user *seq_arr; u32 size_to_copy; u64 *cs_seq_arr; u8 seq_arr_len; int rc, i; for (i = 0 ; i < sizeof(args->in.pad) ; i++) if (args->in.pad[i]) { dev_dbg(hdev->dev, "Padding bytes must be 0\n"); return -EINVAL; } if (!hdev->supports_wait_for_multi_cs) { dev_err(hdev->dev, "Wait for multi CS is not supported\n"); return -EPERM; } seq_arr_len = args->in.seq_arr_len; if (seq_arr_len > HL_WAIT_MULTI_CS_LIST_MAX_LEN) { dev_err(hdev->dev, "Can wait only up to %d CSs, input sequence is of length %u\n", HL_WAIT_MULTI_CS_LIST_MAX_LEN, seq_arr_len); return -EINVAL; } /* allocate memory for sequence array */ cs_seq_arr = kmalloc_array(seq_arr_len, sizeof(*cs_seq_arr), GFP_KERNEL); if (!cs_seq_arr) return -ENOMEM; /* copy CS sequence array from user */ seq_arr = (void __user *) (uintptr_t) args->in.seq; size_to_copy = seq_arr_len * sizeof(*cs_seq_arr); if (copy_from_user(cs_seq_arr, seq_arr, size_to_copy)) { dev_err(hdev->dev, "Failed to copy multi-cs sequence array from user\n"); rc = -EFAULT; goto free_seq_arr; } /* allocate array for the fences */ fence_arr = kmalloc_array(seq_arr_len, sizeof(struct hl_fence *), GFP_KERNEL); if (!fence_arr) { rc = -ENOMEM; goto free_seq_arr; } /* initialize the multi-CS internal data */ mcs_data.ctx = ctx; mcs_data.seq_arr = cs_seq_arr; mcs_data.fence_arr = fence_arr; mcs_data.arr_len = seq_arr_len; hl_ctx_get(ctx); /* wait (with timeout) for the first CS to be completed */ mcs_data.timeout_jiffies = hl_usecs64_to_jiffies(args->in.timeout_us); mcs_compl = hl_wait_multi_cs_completion_init(hdev); if (IS_ERR(mcs_compl)) { rc = PTR_ERR(mcs_compl); goto put_ctx; } /* poll all CS fences, extract timestamp */ mcs_data.update_ts = true; rc = hl_cs_poll_fences(&mcs_data, mcs_compl); /* * skip wait for CS completion when one of the below is true: * - an error on the poll function * - one or more CS in the list completed * - the user called ioctl with timeout 0 */ if (rc || mcs_data.completion_bitmap || !args->in.timeout_us) goto completion_fini; while (true) { rc = hl_wait_multi_cs_completion(&mcs_data, mcs_compl); if (rc || (mcs_data.wait_status == 0)) break; /* * poll fences once again to update the CS map. * no timestamp should be updated this time. */ mcs_data.update_ts = false; rc = hl_cs_poll_fences(&mcs_data, mcs_compl); if (rc || mcs_data.completion_bitmap) break; /* * if hl_wait_multi_cs_completion returned before timeout (i.e. * it got a completion) it either got completed by CS in the multi CS list * (in which case the indication will be non empty completion_bitmap) or it * got completed by CS submitted to one of the shared stream master but * not in the multi CS list (in which case we should wait again but modify * the timeout and set timestamp as zero to let a CS related to the current * multi-CS set a new, relevant, timestamp) */ mcs_data.timeout_jiffies = mcs_data.wait_status; mcs_compl->timestamp = 0; } completion_fini: hl_wait_multi_cs_completion_fini(mcs_compl); put_ctx: hl_ctx_put(ctx); kfree(fence_arr); free_seq_arr: kfree(cs_seq_arr); if (rc == -ERESTARTSYS) { dev_err_ratelimited(hdev->dev, "user process got signal while waiting for Multi-CS\n"); rc = -EINTR; } if (rc) return rc; /* update output args */ memset(args, 0, sizeof(*args)); if (mcs_data.completion_bitmap) { args->out.status = HL_WAIT_CS_STATUS_COMPLETED; args->out.cs_completion_map = mcs_data.completion_bitmap; /* if timestamp not 0- it's valid */ if (mcs_data.timestamp) { args->out.timestamp_nsec = mcs_data.timestamp; args->out.flags |= HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD; } /* update if some CS was gone */ if (!mcs_data.timestamp) args->out.flags |= HL_WAIT_CS_STATUS_FLAG_GONE; } else { args->out.status = HL_WAIT_CS_STATUS_BUSY; } return 0; } static int hl_cs_wait_ioctl(struct hl_fpriv *hpriv, void *data) { struct hl_device *hdev = hpriv->hdev; union hl_wait_cs_args *args = data; enum hl_cs_wait_status status; u64 seq = args->in.seq; s64 timestamp; int rc; rc = _hl_cs_wait_ioctl(hdev, hpriv->ctx, args->in.timeout_us, seq, &status, ×tamp); if (rc == -ERESTARTSYS) { dev_err_ratelimited(hdev->dev, "user process got signal while waiting for CS handle %llu\n", seq); return -EINTR; } memset(args, 0, sizeof(*args)); if (rc) { if (rc == -ETIMEDOUT) { dev_err_ratelimited(hdev->dev, "CS %llu has timed-out while user process is waiting for it\n", seq); args->out.status = HL_WAIT_CS_STATUS_TIMEDOUT; } else if (rc == -EIO) { dev_err_ratelimited(hdev->dev, "CS %llu has been aborted while user process is waiting for it\n", seq); args->out.status = HL_WAIT_CS_STATUS_ABORTED; } return rc; } if (timestamp) { args->out.flags |= HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD; args->out.timestamp_nsec = timestamp; } switch (status) { case CS_WAIT_STATUS_GONE: args->out.flags |= HL_WAIT_CS_STATUS_FLAG_GONE; fallthrough; case CS_WAIT_STATUS_COMPLETED: args->out.status = HL_WAIT_CS_STATUS_COMPLETED; break; case CS_WAIT_STATUS_BUSY: default: args->out.status = HL_WAIT_CS_STATUS_BUSY; break; } return 0; } static int ts_buff_get_kernel_ts_record(struct hl_mmap_mem_buf *buf, struct hl_cb *cq_cb, u64 ts_offset, u64 cq_offset, u64 target_value, spinlock_t *wait_list_lock, struct hl_user_pending_interrupt **pend) { struct hl_ts_buff *ts_buff = buf->private; struct hl_user_pending_interrupt *requested_offset_record = (struct hl_user_pending_interrupt *)ts_buff->kernel_buff_address + ts_offset; struct hl_user_pending_interrupt *cb_last = (struct hl_user_pending_interrupt *)ts_buff->kernel_buff_address + (ts_buff->kernel_buff_size / sizeof(struct hl_user_pending_interrupt)); unsigned long iter_counter = 0; u64 current_cq_counter; ktime_t timestamp; /* Validate ts_offset not exceeding last max */ if (requested_offset_record >= cb_last) { dev_err(buf->mmg->dev, "Ts offset exceeds max CB offset(0x%llx)\n", (u64)(uintptr_t)cb_last); return -EINVAL; } timestamp = ktime_get(); start_over: spin_lock(wait_list_lock); /* Unregister only if we didn't reach the target value * since in this case there will be no handling in irq context * and then it's safe to delete the node out of the interrupt list * then re-use it on other interrupt */ if (requested_offset_record->ts_reg_info.in_use) { current_cq_counter = *requested_offset_record->cq_kernel_addr; if (current_cq_counter < requested_offset_record->cq_target_value) { list_del(&requested_offset_record->wait_list_node); spin_unlock(wait_list_lock); hl_mmap_mem_buf_put(requested_offset_record->ts_reg_info.buf); hl_cb_put(requested_offset_record->ts_reg_info.cq_cb); dev_dbg(buf->mmg->dev, "ts node removed from interrupt list now can re-use\n"); } else { dev_dbg(buf->mmg->dev, "ts node in middle of irq handling\n"); /* irq thread handling in the middle give it time to finish */ spin_unlock(wait_list_lock); usleep_range(100, 1000); if (++iter_counter == MAX_TS_ITER_NUM) { dev_err(buf->mmg->dev, "Timestamp offset processing reached timeout of %lld ms\n", ktime_ms_delta(ktime_get(), timestamp)); return -EAGAIN; } goto start_over; } } else { /* Fill up the new registration node info */ requested_offset_record->ts_reg_info.buf = buf; requested_offset_record->ts_reg_info.cq_cb = cq_cb; requested_offset_record->ts_reg_info.timestamp_kernel_addr = (u64 *) ts_buff->user_buff_address + ts_offset; requested_offset_record->cq_kernel_addr = (u64 *) cq_cb->kernel_address + cq_offset; requested_offset_record->cq_target_value = target_value; spin_unlock(wait_list_lock); } *pend = requested_offset_record; dev_dbg(buf->mmg->dev, "Found available node in TS kernel CB %p\n", requested_offset_record); return 0; } static int _hl_interrupt_wait_ioctl(struct hl_device *hdev, struct hl_ctx *ctx, struct hl_mem_mgr *cb_mmg, struct hl_mem_mgr *mmg, u64 timeout_us, u64 cq_counters_handle, u64 cq_counters_offset, u64 target_value, struct hl_user_interrupt *interrupt, bool register_ts_record, u64 ts_handle, u64 ts_offset, u32 *status, u64 *timestamp) { struct hl_user_pending_interrupt *pend; struct hl_mmap_mem_buf *buf; struct hl_cb *cq_cb; unsigned long timeout; long completion_rc; int rc = 0; timeout = hl_usecs64_to_jiffies(timeout_us); hl_ctx_get(ctx); cq_cb = hl_cb_get(cb_mmg, cq_counters_handle); if (!cq_cb) { rc = -EINVAL; goto put_ctx; } /* Validate the cq offset */ if (((u64 *) cq_cb->kernel_address + cq_counters_offset) >= ((u64 *) cq_cb->kernel_address + (cq_cb->size / sizeof(u64)))) { rc = -EINVAL; goto put_cq_cb; } if (register_ts_record) { dev_dbg(hdev->dev, "Timestamp registration: interrupt id: %u, ts offset: %llu, cq_offset: %llu\n", interrupt->interrupt_id, ts_offset, cq_counters_offset); buf = hl_mmap_mem_buf_get(mmg, ts_handle); if (!buf) { rc = -EINVAL; goto put_cq_cb; } /* get ts buffer record */ rc = ts_buff_get_kernel_ts_record(buf, cq_cb, ts_offset, cq_counters_offset, target_value, &interrupt->wait_list_lock, &pend); if (rc) goto put_ts_buff; } else { pend = kzalloc(sizeof(*pend), GFP_KERNEL); if (!pend) { rc = -ENOMEM; goto put_cq_cb; } hl_fence_init(&pend->fence, ULONG_MAX); pend->cq_kernel_addr = (u64 *) cq_cb->kernel_address + cq_counters_offset; pend->cq_target_value = target_value; } spin_lock(&interrupt->wait_list_lock); /* We check for completion value as interrupt could have been received * before we added the node to the wait list */ if (*pend->cq_kernel_addr >= target_value) { if (register_ts_record) pend->ts_reg_info.in_use = 0; spin_unlock(&interrupt->wait_list_lock); *status = HL_WAIT_CS_STATUS_COMPLETED; if (register_ts_record) { *pend->ts_reg_info.timestamp_kernel_addr = ktime_get_ns(); goto put_ts_buff; } else { pend->fence.timestamp = ktime_get(); goto set_timestamp; } } else if (!timeout_us) { spin_unlock(&interrupt->wait_list_lock); *status = HL_WAIT_CS_STATUS_BUSY; pend->fence.timestamp = ktime_get(); goto set_timestamp; } /* Add pending user interrupt to relevant list for the interrupt * handler to monitor. * Note that we cannot have sorted list by target value, * in order to shorten the list pass loop, since * same list could have nodes for different cq counter handle. * Note: * Mark ts buff offset as in use here in the spinlock protection area * to avoid getting in the re-use section in ts_buff_get_kernel_ts_record * before adding the node to the list. this scenario might happen when * multiple threads are racing on same offset and one thread could * set the ts buff in ts_buff_get_kernel_ts_record then the other thread * takes over and get to ts_buff_get_kernel_ts_record and then we will try * to re-use the same ts buff offset, and will try to delete a non existing * node from the list. */ if (register_ts_record) pend->ts_reg_info.in_use = 1; list_add_tail(&pend->wait_list_node, &interrupt->wait_list_head); spin_unlock(&interrupt->wait_list_lock); if (register_ts_record) { rc = *status = HL_WAIT_CS_STATUS_COMPLETED; goto ts_registration_exit; } /* Wait for interrupt handler to signal completion */ completion_rc = wait_for_completion_interruptible_timeout(&pend->fence.completion, timeout); if (completion_rc > 0) { *status = HL_WAIT_CS_STATUS_COMPLETED; } else { if (completion_rc == -ERESTARTSYS) { dev_err_ratelimited(hdev->dev, "user process got signal while waiting for interrupt ID %d\n", interrupt->interrupt_id); rc = -EINTR; *status = HL_WAIT_CS_STATUS_ABORTED; } else { if (pend->fence.error == -EIO) { dev_err_ratelimited(hdev->dev, "interrupt based wait ioctl aborted(error:%d) due to a reset cycle initiated\n", pend->fence.error); rc = -EIO; *status = HL_WAIT_CS_STATUS_ABORTED; } else { /* The wait has timed-out. We don't know anything beyond that * because the workload wasn't submitted through the driver. * Therefore, from driver's perspective, the workload is still * executing. */ rc = 0; *status = HL_WAIT_CS_STATUS_BUSY; } } } /* * We keep removing the node from list here, and not at the irq handler * for completion timeout case. and if it's a registration * for ts record, the node will be deleted in the irq handler after * we reach the target value. */ spin_lock(&interrupt->wait_list_lock); list_del(&pend->wait_list_node); spin_unlock(&interrupt->wait_list_lock); set_timestamp: *timestamp = ktime_to_ns(pend->fence.timestamp); kfree(pend); hl_cb_put(cq_cb); ts_registration_exit: hl_ctx_put(ctx); return rc; put_ts_buff: hl_mmap_mem_buf_put(buf); put_cq_cb: hl_cb_put(cq_cb); put_ctx: hl_ctx_put(ctx); return rc; } static int _hl_interrupt_wait_ioctl_user_addr(struct hl_device *hdev, struct hl_ctx *ctx, u64 timeout_us, u64 user_address, u64 target_value, struct hl_user_interrupt *interrupt, u32 *status, u64 *timestamp) { struct hl_user_pending_interrupt *pend; unsigned long timeout; u64 completion_value; long completion_rc; int rc = 0; timeout = hl_usecs64_to_jiffies(timeout_us); hl_ctx_get(ctx); pend = kzalloc(sizeof(*pend), GFP_KERNEL); if (!pend) { hl_ctx_put(ctx); return -ENOMEM; } hl_fence_init(&pend->fence, ULONG_MAX); /* Add pending user interrupt to relevant list for the interrupt * handler to monitor */ spin_lock(&interrupt->wait_list_lock); list_add_tail(&pend->wait_list_node, &interrupt->wait_list_head); spin_unlock(&interrupt->wait_list_lock); /* We check for completion value as interrupt could have been received * before we added the node to the wait list */ if (copy_from_user(&completion_value, u64_to_user_ptr(user_address), 8)) { dev_err(hdev->dev, "Failed to copy completion value from user\n"); rc = -EFAULT; goto remove_pending_user_interrupt; } if (completion_value >= target_value) { *status = HL_WAIT_CS_STATUS_COMPLETED; /* There was no interrupt, we assume the completion is now. */ pend->fence.timestamp = ktime_get(); } else { *status = HL_WAIT_CS_STATUS_BUSY; } if (!timeout_us || (*status == HL_WAIT_CS_STATUS_COMPLETED)) goto remove_pending_user_interrupt; wait_again: /* Wait for interrupt handler to signal completion */ completion_rc = wait_for_completion_interruptible_timeout(&pend->fence.completion, timeout); /* If timeout did not expire we need to perform the comparison. * If comparison fails, keep waiting until timeout expires */ if (completion_rc > 0) { spin_lock(&interrupt->wait_list_lock); /* reinit_completion must be called before we check for user * completion value, otherwise, if interrupt is received after * the comparison and before the next wait_for_completion, * we will reach timeout and fail */ reinit_completion(&pend->fence.completion); spin_unlock(&interrupt->wait_list_lock); if (copy_from_user(&completion_value, u64_to_user_ptr(user_address), 8)) { dev_err(hdev->dev, "Failed to copy completion value from user\n"); rc = -EFAULT; goto remove_pending_user_interrupt; } if (completion_value >= target_value) { *status = HL_WAIT_CS_STATUS_COMPLETED; } else if (pend->fence.error) { dev_err_ratelimited(hdev->dev, "interrupt based wait ioctl aborted(error:%d) due to a reset cycle initiated\n", pend->fence.error); /* set the command completion status as ABORTED */ *status = HL_WAIT_CS_STATUS_ABORTED; } else { timeout = completion_rc; goto wait_again; } } else if (completion_rc == -ERESTARTSYS) { dev_err_ratelimited(hdev->dev, "user process got signal while waiting for interrupt ID %d\n", interrupt->interrupt_id); rc = -EINTR; } else { /* The wait has timed-out. We don't know anything beyond that * because the workload wasn't submitted through the driver. * Therefore, from driver's perspective, the workload is still * executing. */ rc = 0; *status = HL_WAIT_CS_STATUS_BUSY; } remove_pending_user_interrupt: spin_lock(&interrupt->wait_list_lock); list_del(&pend->wait_list_node); spin_unlock(&interrupt->wait_list_lock); *timestamp = ktime_to_ns(pend->fence.timestamp); kfree(pend); hl_ctx_put(ctx); return rc; } static int hl_interrupt_wait_ioctl(struct hl_fpriv *hpriv, void *data) { u16 interrupt_id, first_interrupt, last_interrupt; struct hl_device *hdev = hpriv->hdev; struct asic_fixed_properties *prop; struct hl_user_interrupt *interrupt; union hl_wait_cs_args *args = data; u32 status = HL_WAIT_CS_STATUS_BUSY; u64 timestamp = 0; int rc, int_idx; prop = &hdev->asic_prop; if (!(prop->user_interrupt_count + prop->user_dec_intr_count)) { dev_err(hdev->dev, "no user interrupts allowed"); return -EPERM; } interrupt_id = FIELD_GET(HL_WAIT_CS_FLAGS_INTERRUPT_MASK, args->in.flags); first_interrupt = prop->first_available_user_interrupt; last_interrupt = prop->first_available_user_interrupt + prop->user_interrupt_count - 1; if (interrupt_id < prop->user_dec_intr_count) { /* Check if the requested core is enabled */ if (!(prop->decoder_enabled_mask & BIT(interrupt_id))) { dev_err(hdev->dev, "interrupt on a disabled core(%u) not allowed", interrupt_id); return -EINVAL; } interrupt = &hdev->user_interrupt[interrupt_id]; } else if (interrupt_id >= first_interrupt && interrupt_id <= last_interrupt) { int_idx = interrupt_id - first_interrupt + prop->user_dec_intr_count; interrupt = &hdev->user_interrupt[int_idx]; } else if (interrupt_id == HL_COMMON_USER_CQ_INTERRUPT_ID) { interrupt = &hdev->common_user_cq_interrupt; } else if (interrupt_id == HL_COMMON_DEC_INTERRUPT_ID) { interrupt = &hdev->common_decoder_interrupt; } else { dev_err(hdev->dev, "invalid user interrupt %u", interrupt_id); return -EINVAL; } if (args->in.flags & HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ) rc = _hl_interrupt_wait_ioctl(hdev, hpriv->ctx, &hpriv->mem_mgr, &hpriv->mem_mgr, args->in.interrupt_timeout_us, args->in.cq_counters_handle, args->in.cq_counters_offset, args->in.target, interrupt, !!(args->in.flags & HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT), args->in.timestamp_handle, args->in.timestamp_offset, &status, ×tamp); else rc = _hl_interrupt_wait_ioctl_user_addr(hdev, hpriv->ctx, args->in.interrupt_timeout_us, args->in.addr, args->in.target, interrupt, &status, ×tamp); if (rc) return rc; memset(args, 0, sizeof(*args)); args->out.status = status; if (timestamp) { args->out.timestamp_nsec = timestamp; args->out.flags |= HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD; } return 0; } int hl_wait_ioctl(struct hl_fpriv *hpriv, void *data) { struct hl_device *hdev = hpriv->hdev; union hl_wait_cs_args *args = data; u32 flags = args->in.flags; int rc; /* If the device is not operational, or if an error has happened and user should release the * device, there is no point in waiting for any command submission or user interrupt. */ if (!hl_device_operational(hpriv->hdev, NULL) || hdev->reset_info.watchdog_active) return -EBUSY; if (flags & HL_WAIT_CS_FLAGS_INTERRUPT) rc = hl_interrupt_wait_ioctl(hpriv, data); else if (flags & HL_WAIT_CS_FLAGS_MULTI_CS) rc = hl_multi_cs_wait_ioctl(hpriv, data); else rc = hl_cs_wait_ioctl(hpriv, data); return rc; }