diff options
Diffstat (limited to 'drivers/accel')
27 files changed, 541 insertions, 211 deletions
diff --git a/drivers/accel/amdxdna/TODO b/drivers/accel/amdxdna/TODO index 5119bccd1917..ad8ac6e315b6 100644 --- a/drivers/accel/amdxdna/TODO +++ b/drivers/accel/amdxdna/TODO @@ -1,3 +1,2 @@ -- Add import and export BO support - Add debugfs support - Add debug BO support diff --git a/drivers/accel/amdxdna/aie2_ctx.c b/drivers/accel/amdxdna/aie2_ctx.c index 00d215ac866e..e04549f64d69 100644 --- a/drivers/accel/amdxdna/aie2_ctx.c +++ b/drivers/accel/amdxdna/aie2_ctx.c @@ -758,27 +758,42 @@ int aie2_hwctx_config(struct amdxdna_hwctx *hwctx, u32 type, u64 value, void *bu static int aie2_populate_range(struct amdxdna_gem_obj *abo) { struct amdxdna_dev *xdna = to_xdna_dev(to_gobj(abo)->dev); - struct mm_struct *mm = abo->mem.notifier.mm; - struct hmm_range range = { 0 }; + struct amdxdna_umap *mapp; unsigned long timeout; + struct mm_struct *mm; + bool found; int ret; - XDNA_INFO_ONCE(xdna, "populate memory range %llx size %lx", - abo->mem.userptr, abo->mem.size); - range.notifier = &abo->mem.notifier; - range.start = abo->mem.userptr; - range.end = abo->mem.userptr + abo->mem.size; - range.hmm_pfns = abo->mem.pfns; - range.default_flags = HMM_PFN_REQ_FAULT; + timeout = jiffies + msecs_to_jiffies(HMM_RANGE_DEFAULT_TIMEOUT); +again: + found = false; + down_write(&xdna->notifier_lock); + list_for_each_entry(mapp, &abo->mem.umap_list, node) { + if (mapp->invalid) { + found = true; + break; + } + } - if (!mmget_not_zero(mm)) + if (!found) { + abo->mem.map_invalid = false; + up_write(&xdna->notifier_lock); + return 0; + } + kref_get(&mapp->refcnt); + up_write(&xdna->notifier_lock); + + XDNA_DBG(xdna, "populate memory range %lx %lx", + mapp->vma->vm_start, mapp->vma->vm_end); + mm = mapp->notifier.mm; + if (!mmget_not_zero(mm)) { + amdxdna_umap_put(mapp); return -EFAULT; + } - timeout = jiffies + msecs_to_jiffies(HMM_RANGE_DEFAULT_TIMEOUT); -again: - range.notifier_seq = mmu_interval_read_begin(&abo->mem.notifier); + mapp->range.notifier_seq = mmu_interval_read_begin(&mapp->notifier); mmap_read_lock(mm); - ret = hmm_range_fault(&range); + ret = hmm_range_fault(&mapp->range); mmap_read_unlock(mm); if (ret) { if (time_after(jiffies, timeout)) { @@ -786,21 +801,27 @@ again: goto put_mm; } - if (ret == -EBUSY) + if (ret == -EBUSY) { + amdxdna_umap_put(mapp); goto again; + } goto put_mm; } - down_read(&xdna->notifier_lock); - if (mmu_interval_read_retry(&abo->mem.notifier, range.notifier_seq)) { - up_read(&xdna->notifier_lock); + down_write(&xdna->notifier_lock); + if (mmu_interval_read_retry(&mapp->notifier, mapp->range.notifier_seq)) { + up_write(&xdna->notifier_lock); + amdxdna_umap_put(mapp); goto again; } - abo->mem.map_invalid = false; - up_read(&xdna->notifier_lock); + mapp->invalid = false; + up_write(&xdna->notifier_lock); + amdxdna_umap_put(mapp); + goto again; put_mm: + amdxdna_umap_put(mapp); mmput(mm); return ret; } @@ -908,10 +929,6 @@ void aie2_hmm_invalidate(struct amdxdna_gem_obj *abo, struct drm_gem_object *gobj = to_gobj(abo); long ret; - down_write(&xdna->notifier_lock); - abo->mem.map_invalid = true; - mmu_interval_set_seq(&abo->mem.notifier, cur_seq); - up_write(&xdna->notifier_lock); ret = dma_resv_wait_timeout(gobj->resv, DMA_RESV_USAGE_BOOKKEEP, true, MAX_SCHEDULE_TIMEOUT); if (!ret || ret == -ERESTARTSYS) diff --git a/drivers/accel/amdxdna/aie2_message.c b/drivers/accel/amdxdna/aie2_message.c index bf4219e32cc1..82412eec9a4b 100644 --- a/drivers/accel/amdxdna/aie2_message.c +++ b/drivers/accel/amdxdna/aie2_message.c @@ -525,7 +525,7 @@ aie2_cmdlist_fill_one_slot_cf(void *cmd_buf, u32 offset, if (!payload) return -EINVAL; - if (!slot_cf_has_space(offset, payload_len)) + if (!slot_has_space(*buf, offset, payload_len)) return -ENOSPC; buf->cu_idx = cu_idx; @@ -558,7 +558,7 @@ aie2_cmdlist_fill_one_slot_dpu(void *cmd_buf, u32 offset, if (payload_len < sizeof(*sn) || arg_sz > MAX_DPU_ARGS_SIZE) return -EINVAL; - if (!slot_dpu_has_space(offset, arg_sz)) + if (!slot_has_space(*buf, offset, arg_sz)) return -ENOSPC; buf->inst_buf_addr = sn->buffer; @@ -569,7 +569,7 @@ aie2_cmdlist_fill_one_slot_dpu(void *cmd_buf, u32 offset, memcpy(buf->args, sn->prop_args, arg_sz); /* Accurate buf size to hint firmware to do necessary copy */ - *size += sizeof(*buf) + arg_sz; + *size = sizeof(*buf) + arg_sz; return 0; } diff --git a/drivers/accel/amdxdna/aie2_msg_priv.h b/drivers/accel/amdxdna/aie2_msg_priv.h index 4e02e744b470..6df9065b13f6 100644 --- a/drivers/accel/amdxdna/aie2_msg_priv.h +++ b/drivers/accel/amdxdna/aie2_msg_priv.h @@ -319,18 +319,16 @@ struct async_event_msg_resp { } __packed; #define MAX_CHAIN_CMDBUF_SIZE SZ_4K -#define slot_cf_has_space(offset, payload_size) \ - (MAX_CHAIN_CMDBUF_SIZE - ((offset) + (payload_size)) > \ - offsetof(struct cmd_chain_slot_execbuf_cf, args[0])) +#define slot_has_space(slot, offset, payload_size) \ + (MAX_CHAIN_CMDBUF_SIZE >= (offset) + (payload_size) + \ + sizeof(typeof(slot))) + struct cmd_chain_slot_execbuf_cf { __u32 cu_idx; __u32 arg_cnt; __u32 args[] __counted_by(arg_cnt); }; -#define slot_dpu_has_space(offset, payload_size) \ - (MAX_CHAIN_CMDBUF_SIZE - ((offset) + (payload_size)) > \ - offsetof(struct cmd_chain_slot_dpu, args[0])) struct cmd_chain_slot_dpu { __u64 inst_buf_addr; __u32 inst_size; diff --git a/drivers/accel/amdxdna/aie2_pci.c b/drivers/accel/amdxdna/aie2_pci.c index 5a058e565b01..c6cf7068d23c 100644 --- a/drivers/accel/amdxdna/aie2_pci.c +++ b/drivers/accel/amdxdna/aie2_pci.c @@ -512,12 +512,6 @@ static int aie2_init(struct amdxdna_dev *xdna) goto release_fw; } - ret = iommu_dev_enable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA); - if (ret) { - XDNA_ERR(xdna, "Enable PASID failed, ret %d", ret); - goto free_irq; - } - psp_conf.fw_size = fw->size; psp_conf.fw_buf = fw->data; for (i = 0; i < PSP_MAX_REGS; i++) @@ -526,14 +520,14 @@ static int aie2_init(struct amdxdna_dev *xdna) if (!ndev->psp_hdl) { XDNA_ERR(xdna, "failed to create psp"); ret = -ENOMEM; - goto disable_sva; + goto free_irq; } xdna->dev_handle = ndev; ret = aie2_hw_start(xdna); if (ret) { XDNA_ERR(xdna, "start npu failed, ret %d", ret); - goto disable_sva; + goto free_irq; } ret = aie2_mgmt_fw_query(ndev); @@ -584,8 +578,6 @@ async_event_free: aie2_error_async_events_free(ndev); stop_hw: aie2_hw_stop(xdna); -disable_sva: - iommu_dev_disable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA); free_irq: pci_free_irq_vectors(pdev); release_fw: @@ -601,7 +593,6 @@ static void aie2_fini(struct amdxdna_dev *xdna) aie2_hw_stop(xdna); aie2_error_async_events_free(ndev); - iommu_dev_disable_feature(&pdev->dev, IOMMU_DEV_FEAT_SVA); pci_free_irq_vectors(pdev); } diff --git a/drivers/accel/amdxdna/aie2_psp.c b/drivers/accel/amdxdna/aie2_psp.c index dc3a072ce3b6..f28a060a8810 100644 --- a/drivers/accel/amdxdna/aie2_psp.c +++ b/drivers/accel/amdxdna/aie2_psp.c @@ -126,8 +126,8 @@ struct psp_device *aie2m_psp_create(struct drm_device *ddev, struct psp_config * psp->ddev = ddev; memcpy(psp->psp_regs, conf->psp_regs, sizeof(psp->psp_regs)); - psp->fw_buf_sz = ALIGN(conf->fw_size, PSP_FW_ALIGN) + PSP_FW_ALIGN; - psp->fw_buffer = drmm_kmalloc(ddev, psp->fw_buf_sz, GFP_KERNEL); + psp->fw_buf_sz = ALIGN(conf->fw_size, PSP_FW_ALIGN); + psp->fw_buffer = drmm_kmalloc(ddev, psp->fw_buf_sz + PSP_FW_ALIGN, GFP_KERNEL); if (!psp->fw_buffer) { drm_err(ddev, "no memory for fw buffer"); return NULL; diff --git a/drivers/accel/amdxdna/amdxdna_ctx.c b/drivers/accel/amdxdna/amdxdna_ctx.c index 43442b9e273b..be073224bd69 100644 --- a/drivers/accel/amdxdna/amdxdna_ctx.c +++ b/drivers/accel/amdxdna/amdxdna_ctx.c @@ -496,11 +496,11 @@ static int amdxdna_drm_submit_execbuf(struct amdxdna_client *client, struct amdxdna_drm_exec_cmd *args) { struct amdxdna_dev *xdna = client->xdna; - u32 *arg_bo_hdls; + u32 *arg_bo_hdls = NULL; u32 cmd_bo_hdl; int ret; - if (!args->arg_count || args->arg_count > MAX_ARG_COUNT) { + if (args->arg_count > MAX_ARG_COUNT) { XDNA_ERR(xdna, "Invalid arg bo count %d", args->arg_count); return -EINVAL; } @@ -512,14 +512,16 @@ static int amdxdna_drm_submit_execbuf(struct amdxdna_client *client, } cmd_bo_hdl = (u32)args->cmd_handles; - arg_bo_hdls = kcalloc(args->arg_count, sizeof(u32), GFP_KERNEL); - if (!arg_bo_hdls) - return -ENOMEM; - ret = copy_from_user(arg_bo_hdls, u64_to_user_ptr(args->args), - args->arg_count * sizeof(u32)); - if (ret) { - ret = -EFAULT; - goto free_cmd_bo_hdls; + if (args->arg_count) { + arg_bo_hdls = kcalloc(args->arg_count, sizeof(u32), GFP_KERNEL); + if (!arg_bo_hdls) + return -ENOMEM; + ret = copy_from_user(arg_bo_hdls, u64_to_user_ptr(args->args), + args->arg_count * sizeof(u32)); + if (ret) { + ret = -EFAULT; + goto free_cmd_bo_hdls; + } } ret = amdxdna_cmd_submit(client, cmd_bo_hdl, arg_bo_hdls, diff --git a/drivers/accel/amdxdna/amdxdna_gem.c b/drivers/accel/amdxdna/amdxdna_gem.c index 606433d73236..26831ec69f89 100644 --- a/drivers/accel/amdxdna/amdxdna_gem.c +++ b/drivers/accel/amdxdna/amdxdna_gem.c @@ -9,7 +9,10 @@ #include <drm/drm_gem.h> #include <drm/drm_gem_shmem_helper.h> #include <drm/gpu_scheduler.h> +#include <linux/dma-buf.h> +#include <linux/dma-direct.h> #include <linux/iosys-map.h> +#include <linux/pagemap.h> #include <linux/vmalloc.h> #include "amdxdna_ctx.h" @@ -18,6 +21,8 @@ #define XDNA_MAX_CMD_BO_SIZE SZ_32K +MODULE_IMPORT_NS("DMA_BUF"); + static int amdxdna_gem_insert_node_locked(struct amdxdna_gem_obj *abo, bool use_vmap) { @@ -55,57 +60,38 @@ amdxdna_gem_insert_node_locked(struct amdxdna_gem_obj *abo, bool use_vmap) return 0; } -static void amdxdna_gem_obj_free(struct drm_gem_object *gobj) -{ - struct amdxdna_dev *xdna = to_xdna_dev(gobj->dev); - struct amdxdna_gem_obj *abo = to_xdna_obj(gobj); - struct iosys_map map = IOSYS_MAP_INIT_VADDR(abo->mem.kva); - - XDNA_DBG(xdna, "BO type %d xdna_addr 0x%llx", abo->type, abo->mem.dev_addr); - if (abo->pinned) - amdxdna_gem_unpin(abo); - - if (abo->type == AMDXDNA_BO_DEV) { - mutex_lock(&abo->client->mm_lock); - drm_mm_remove_node(&abo->mm_node); - mutex_unlock(&abo->client->mm_lock); - - vunmap(abo->mem.kva); - drm_gem_object_put(to_gobj(abo->dev_heap)); - drm_gem_object_release(gobj); - mutex_destroy(&abo->lock); - kfree(abo); - return; - } - - if (abo->type == AMDXDNA_BO_DEV_HEAP) - drm_mm_takedown(&abo->mm); - - drm_gem_vunmap_unlocked(gobj, &map); - mutex_destroy(&abo->lock); - drm_gem_shmem_free(&abo->base); -} - -static const struct drm_gem_object_funcs amdxdna_gem_dev_obj_funcs = { - .free = amdxdna_gem_obj_free, -}; - static bool amdxdna_hmm_invalidate(struct mmu_interval_notifier *mni, const struct mmu_notifier_range *range, unsigned long cur_seq) { - struct amdxdna_gem_obj *abo = container_of(mni, struct amdxdna_gem_obj, - mem.notifier); - struct amdxdna_dev *xdna = to_xdna_dev(to_gobj(abo)->dev); + struct amdxdna_umap *mapp = container_of(mni, struct amdxdna_umap, notifier); + struct amdxdna_gem_obj *abo = mapp->abo; + struct amdxdna_dev *xdna; - XDNA_DBG(xdna, "Invalid range 0x%llx, 0x%lx, type %d", - abo->mem.userptr, abo->mem.size, abo->type); + xdna = to_xdna_dev(to_gobj(abo)->dev); + XDNA_DBG(xdna, "Invalidating range 0x%lx, 0x%lx, type %d", + mapp->vma->vm_start, mapp->vma->vm_end, abo->type); if (!mmu_notifier_range_blockable(range)) return false; + down_write(&xdna->notifier_lock); + abo->mem.map_invalid = true; + mapp->invalid = true; + mmu_interval_set_seq(&mapp->notifier, cur_seq); + up_write(&xdna->notifier_lock); + xdna->dev_info->ops->hmm_invalidate(abo, cur_seq); + if (range->event == MMU_NOTIFY_UNMAP) { + down_write(&xdna->notifier_lock); + if (!mapp->unmapped) { + queue_work(xdna->notifier_wq, &mapp->hmm_unreg_work); + mapp->unmapped = true; + } + up_write(&xdna->notifier_lock); + } + return true; } @@ -113,102 +99,310 @@ static const struct mmu_interval_notifier_ops amdxdna_hmm_ops = { .invalidate = amdxdna_hmm_invalidate, }; -static void amdxdna_hmm_unregister(struct amdxdna_gem_obj *abo) +static void amdxdna_hmm_unregister(struct amdxdna_gem_obj *abo, + struct vm_area_struct *vma) { struct amdxdna_dev *xdna = to_xdna_dev(to_gobj(abo)->dev); + struct amdxdna_umap *mapp; + + down_read(&xdna->notifier_lock); + list_for_each_entry(mapp, &abo->mem.umap_list, node) { + if (!vma || mapp->vma == vma) { + if (!mapp->unmapped) { + queue_work(xdna->notifier_wq, &mapp->hmm_unreg_work); + mapp->unmapped = true; + } + if (vma) + break; + } + } + up_read(&xdna->notifier_lock); +} - if (!xdna->dev_info->ops->hmm_invalidate) - return; +static void amdxdna_umap_release(struct kref *ref) +{ + struct amdxdna_umap *mapp = container_of(ref, struct amdxdna_umap, refcnt); + struct vm_area_struct *vma = mapp->vma; + struct amdxdna_dev *xdna; + + mmu_interval_notifier_remove(&mapp->notifier); + if (is_import_bo(mapp->abo) && vma->vm_file && vma->vm_file->f_mapping) + mapping_clear_unevictable(vma->vm_file->f_mapping); + + xdna = to_xdna_dev(to_gobj(mapp->abo)->dev); + down_write(&xdna->notifier_lock); + list_del(&mapp->node); + up_write(&xdna->notifier_lock); + + kvfree(mapp->range.hmm_pfns); + kfree(mapp); +} + +void amdxdna_umap_put(struct amdxdna_umap *mapp) +{ + kref_put(&mapp->refcnt, amdxdna_umap_release); +} + +static void amdxdna_hmm_unreg_work(struct work_struct *work) +{ + struct amdxdna_umap *mapp = container_of(work, struct amdxdna_umap, + hmm_unreg_work); - mmu_interval_notifier_remove(&abo->mem.notifier); - kvfree(abo->mem.pfns); - abo->mem.pfns = NULL; + amdxdna_umap_put(mapp); } -static int amdxdna_hmm_register(struct amdxdna_gem_obj *abo, unsigned long addr, - size_t len) +static int amdxdna_hmm_register(struct amdxdna_gem_obj *abo, + struct vm_area_struct *vma) { struct amdxdna_dev *xdna = to_xdna_dev(to_gobj(abo)->dev); + unsigned long len = vma->vm_end - vma->vm_start; + unsigned long addr = vma->vm_start; + struct amdxdna_umap *mapp; u32 nr_pages; int ret; if (!xdna->dev_info->ops->hmm_invalidate) return 0; - if (abo->mem.pfns) - return -EEXIST; + mapp = kzalloc(sizeof(*mapp), GFP_KERNEL); + if (!mapp) + return -ENOMEM; nr_pages = (PAGE_ALIGN(addr + len) - (addr & PAGE_MASK)) >> PAGE_SHIFT; - abo->mem.pfns = kvcalloc(nr_pages, sizeof(*abo->mem.pfns), - GFP_KERNEL); - if (!abo->mem.pfns) - return -ENOMEM; + mapp->range.hmm_pfns = kvcalloc(nr_pages, sizeof(*mapp->range.hmm_pfns), + GFP_KERNEL); + if (!mapp->range.hmm_pfns) { + ret = -ENOMEM; + goto free_map; + } - ret = mmu_interval_notifier_insert_locked(&abo->mem.notifier, + ret = mmu_interval_notifier_insert_locked(&mapp->notifier, current->mm, addr, len, &amdxdna_hmm_ops); if (ret) { XDNA_ERR(xdna, "Insert mmu notifier failed, ret %d", ret); - kvfree(abo->mem.pfns); + goto free_pfns; } - abo->mem.userptr = addr; + mapp->range.notifier = &mapp->notifier; + mapp->range.start = vma->vm_start; + mapp->range.end = vma->vm_end; + mapp->range.default_flags = HMM_PFN_REQ_FAULT; + mapp->vma = vma; + mapp->abo = abo; + kref_init(&mapp->refcnt); + + if (abo->mem.userptr == AMDXDNA_INVALID_ADDR) + abo->mem.userptr = addr; + INIT_WORK(&mapp->hmm_unreg_work, amdxdna_hmm_unreg_work); + if (is_import_bo(abo) && vma->vm_file && vma->vm_file->f_mapping) + mapping_set_unevictable(vma->vm_file->f_mapping); + + down_write(&xdna->notifier_lock); + list_add_tail(&mapp->node, &abo->mem.umap_list); + up_write(&xdna->notifier_lock); + + return 0; + +free_pfns: + kvfree(mapp->range.hmm_pfns); +free_map: + kfree(mapp); return ret; } +static int amdxdna_insert_pages(struct amdxdna_gem_obj *abo, + struct vm_area_struct *vma) +{ + struct amdxdna_dev *xdna = to_xdna_dev(to_gobj(abo)->dev); + unsigned long num_pages = vma_pages(vma); + unsigned long offset = 0; + int ret; + + if (!is_import_bo(abo)) { + ret = drm_gem_shmem_mmap(&abo->base, vma); + if (ret) { + XDNA_ERR(xdna, "Failed shmem mmap %d", ret); + return ret; + } + + /* The buffer is based on memory pages. Fix the flag. */ + vm_flags_mod(vma, VM_MIXEDMAP, VM_PFNMAP); + ret = vm_insert_pages(vma, vma->vm_start, abo->base.pages, + &num_pages); + if (ret) { + XDNA_ERR(xdna, "Failed insert pages %d", ret); + vma->vm_ops->close(vma); + return ret; + } + + return 0; + } + + vma->vm_private_data = NULL; + vma->vm_ops = NULL; + ret = dma_buf_mmap(to_gobj(abo)->dma_buf, vma, 0); + if (ret) { + XDNA_ERR(xdna, "Failed to mmap dma buf %d", ret); + return ret; + } + + do { + vm_fault_t fault_ret; + + fault_ret = handle_mm_fault(vma, vma->vm_start + offset, + FAULT_FLAG_WRITE, NULL); + if (fault_ret & VM_FAULT_ERROR) { + vma->vm_ops->close(vma); + XDNA_ERR(xdna, "Fault in page failed"); + return -EFAULT; + } + + offset += PAGE_SIZE; + } while (--num_pages); + + /* Drop the reference drm_gem_mmap_obj() acquired.*/ + drm_gem_object_put(to_gobj(abo)); + + return 0; +} + static int amdxdna_gem_obj_mmap(struct drm_gem_object *gobj, struct vm_area_struct *vma) { + struct amdxdna_dev *xdna = to_xdna_dev(gobj->dev); struct amdxdna_gem_obj *abo = to_xdna_obj(gobj); - unsigned long num_pages; int ret; - ret = amdxdna_hmm_register(abo, vma->vm_start, gobj->size); + ret = amdxdna_hmm_register(abo, vma); if (ret) return ret; + ret = amdxdna_insert_pages(abo, vma); + if (ret) { + XDNA_ERR(xdna, "Failed insert pages, ret %d", ret); + goto hmm_unreg; + } + + XDNA_DBG(xdna, "BO map_offset 0x%llx type %d userptr 0x%lx size 0x%lx", + drm_vma_node_offset_addr(&gobj->vma_node), abo->type, + vma->vm_start, gobj->size); + return 0; + +hmm_unreg: + amdxdna_hmm_unregister(abo, vma); + return ret; +} + +static int amdxdna_gem_dmabuf_mmap(struct dma_buf *dma_buf, struct vm_area_struct *vma) +{ + struct drm_gem_object *gobj = dma_buf->priv; + struct amdxdna_gem_obj *abo = to_xdna_obj(gobj); + unsigned long num_pages = vma_pages(vma); + int ret; + + vma->vm_ops = &drm_gem_shmem_vm_ops; + vma->vm_private_data = gobj; + + drm_gem_object_get(gobj); ret = drm_gem_shmem_mmap(&abo->base, vma); if (ret) - goto hmm_unreg; + goto put_obj; - num_pages = gobj->size >> PAGE_SHIFT; - /* Try to insert the pages */ + /* The buffer is based on memory pages. Fix the flag. */ vm_flags_mod(vma, VM_MIXEDMAP, VM_PFNMAP); - ret = vm_insert_pages(vma, vma->vm_start, abo->base.pages, &num_pages); + ret = vm_insert_pages(vma, vma->vm_start, abo->base.pages, + &num_pages); if (ret) - XDNA_ERR(abo->client->xdna, "Failed insert pages, ret %d", ret); + goto close_vma; return 0; -hmm_unreg: - amdxdna_hmm_unregister(abo); +close_vma: + vma->vm_ops->close(vma); +put_obj: + drm_gem_object_put(gobj); return ret; } -static vm_fault_t amdxdna_gem_vm_fault(struct vm_fault *vmf) +static const struct dma_buf_ops amdxdna_dmabuf_ops = { + .attach = drm_gem_map_attach, + .detach = drm_gem_map_detach, + .map_dma_buf = drm_gem_map_dma_buf, + .unmap_dma_buf = drm_gem_unmap_dma_buf, + .release = drm_gem_dmabuf_release, + .mmap = amdxdna_gem_dmabuf_mmap, + .vmap = drm_gem_dmabuf_vmap, + .vunmap = drm_gem_dmabuf_vunmap, +}; + +static struct dma_buf *amdxdna_gem_prime_export(struct drm_gem_object *gobj, int flags) { - return drm_gem_shmem_vm_ops.fault(vmf); + DEFINE_DMA_BUF_EXPORT_INFO(exp_info); + + exp_info.ops = &amdxdna_dmabuf_ops; + exp_info.size = gobj->size; + exp_info.flags = flags; + exp_info.priv = gobj; + exp_info.resv = gobj->resv; + + return drm_gem_dmabuf_export(gobj->dev, &exp_info); } -static void amdxdna_gem_vm_open(struct vm_area_struct *vma) +static void amdxdna_imported_obj_free(struct amdxdna_gem_obj *abo) { - drm_gem_shmem_vm_ops.open(vma); + dma_buf_unmap_attachment_unlocked(abo->attach, abo->base.sgt, DMA_BIDIRECTIONAL); + dma_buf_detach(abo->dma_buf, abo->attach); + dma_buf_put(abo->dma_buf); + drm_gem_object_release(to_gobj(abo)); + kfree(abo); } -static void amdxdna_gem_vm_close(struct vm_area_struct *vma) +static void amdxdna_gem_obj_free(struct drm_gem_object *gobj) { - struct drm_gem_object *gobj = vma->vm_private_data; + struct amdxdna_dev *xdna = to_xdna_dev(gobj->dev); + struct amdxdna_gem_obj *abo = to_xdna_obj(gobj); + struct iosys_map map = IOSYS_MAP_INIT_VADDR(abo->mem.kva); + + XDNA_DBG(xdna, "BO type %d xdna_addr 0x%llx", abo->type, abo->mem.dev_addr); + + amdxdna_hmm_unregister(abo, NULL); + flush_workqueue(xdna->notifier_wq); + + if (abo->pinned) + amdxdna_gem_unpin(abo); - amdxdna_hmm_unregister(to_xdna_obj(gobj)); - drm_gem_shmem_vm_ops.close(vma); + if (abo->type == AMDXDNA_BO_DEV) { + mutex_lock(&abo->client->mm_lock); + drm_mm_remove_node(&abo->mm_node); + mutex_unlock(&abo->client->mm_lock); + + vunmap(abo->mem.kva); + drm_gem_object_put(to_gobj(abo->dev_heap)); + drm_gem_object_release(gobj); + mutex_destroy(&abo->lock); + kfree(abo); + return; + } + + if (abo->type == AMDXDNA_BO_DEV_HEAP) + drm_mm_takedown(&abo->mm); + + drm_gem_vunmap(gobj, &map); + mutex_destroy(&abo->lock); + + if (is_import_bo(abo)) { + amdxdna_imported_obj_free(abo); + return; + } + + drm_gem_shmem_free(&abo->base); } -static const struct vm_operations_struct amdxdna_gem_vm_ops = { - .fault = amdxdna_gem_vm_fault, - .open = amdxdna_gem_vm_open, - .close = amdxdna_gem_vm_close, +static const struct drm_gem_object_funcs amdxdna_gem_dev_obj_funcs = { + .free = amdxdna_gem_obj_free, }; static const struct drm_gem_object_funcs amdxdna_gem_shmem_funcs = { @@ -220,7 +414,8 @@ static const struct drm_gem_object_funcs amdxdna_gem_shmem_funcs = { .vmap = drm_gem_shmem_object_vmap, .vunmap = drm_gem_shmem_object_vunmap, .mmap = amdxdna_gem_obj_mmap, - .vm_ops = &amdxdna_gem_vm_ops, + .vm_ops = &drm_gem_shmem_vm_ops, + .export = amdxdna_gem_prime_export, }; static struct amdxdna_gem_obj * @@ -239,6 +434,7 @@ amdxdna_gem_create_obj(struct drm_device *dev, size_t size) abo->mem.userptr = AMDXDNA_INVALID_ADDR; abo->mem.dev_addr = AMDXDNA_INVALID_ADDR; abo->mem.size = size; + INIT_LIST_HEAD(&abo->mem.umap_list); return abo; } @@ -258,6 +454,51 @@ amdxdna_gem_create_object_cb(struct drm_device *dev, size_t size) return to_gobj(abo); } +struct drm_gem_object * +amdxdna_gem_prime_import(struct drm_device *dev, struct dma_buf *dma_buf) +{ + struct dma_buf_attachment *attach; + struct amdxdna_gem_obj *abo; + struct drm_gem_object *gobj; + struct sg_table *sgt; + int ret; + + get_dma_buf(dma_buf); + + attach = dma_buf_attach(dma_buf, dev->dev); + if (IS_ERR(attach)) { + ret = PTR_ERR(attach); + goto put_buf; + } + + sgt = dma_buf_map_attachment_unlocked(attach, DMA_BIDIRECTIONAL); + if (IS_ERR(sgt)) { + ret = PTR_ERR(sgt); + goto fail_detach; + } + + gobj = drm_gem_shmem_prime_import_sg_table(dev, attach, sgt); + if (IS_ERR(gobj)) { + ret = PTR_ERR(gobj); + goto fail_unmap; + } + + abo = to_xdna_obj(gobj); + abo->attach = attach; + abo->dma_buf = dma_buf; + + return gobj; + +fail_unmap: + dma_buf_unmap_attachment_unlocked(attach, sgt, DMA_BIDIRECTIONAL); +fail_detach: + dma_buf_detach(dma_buf, attach); +put_buf: + dma_buf_put(dma_buf); + + return ERR_PTR(ret); +} + static struct amdxdna_gem_obj * amdxdna_drm_alloc_shmem(struct drm_device *dev, struct amdxdna_drm_create_bo *args, @@ -417,7 +658,7 @@ amdxdna_drm_create_cmd_bo(struct drm_device *dev, abo->type = AMDXDNA_BO_CMD; abo->client = filp->driver_priv; - ret = drm_gem_vmap_unlocked(to_gobj(abo), &map); + ret = drm_gem_vmap(to_gobj(abo), &map); if (ret) { XDNA_ERR(xdna, "Vmap cmd bo failed, ret %d", ret); goto release_obj; @@ -483,6 +724,9 @@ int amdxdna_gem_pin_nolock(struct amdxdna_gem_obj *abo) struct amdxdna_dev *xdna = to_xdna_dev(to_gobj(abo)->dev); int ret; + if (is_import_bo(abo)) + return 0; + switch (abo->type) { case AMDXDNA_BO_SHMEM: case AMDXDNA_BO_DEV_HEAP: @@ -515,6 +759,9 @@ int amdxdna_gem_pin(struct amdxdna_gem_obj *abo) void amdxdna_gem_unpin(struct amdxdna_gem_obj *abo) { + if (is_import_bo(abo)) + return; + if (abo->type == AMDXDNA_BO_DEV) abo = abo->dev_heap; @@ -606,7 +853,9 @@ int amdxdna_drm_sync_bo_ioctl(struct drm_device *dev, goto put_obj; } - if (abo->type == AMDXDNA_BO_DEV) + if (is_import_bo(abo)) + drm_clflush_sg(abo->base.sgt); + else if (abo->type == AMDXDNA_BO_DEV) drm_clflush_pages(abo->mem.pages, abo->mem.nr_pages); else drm_clflush_pages(abo->base.pages, gobj->size >> PAGE_SHIFT); diff --git a/drivers/accel/amdxdna/amdxdna_gem.h b/drivers/accel/amdxdna/amdxdna_gem.h index 8ccc0375dd9d..aee97e971d6d 100644 --- a/drivers/accel/amdxdna/amdxdna_gem.h +++ b/drivers/accel/amdxdna/amdxdna_gem.h @@ -6,6 +6,20 @@ #ifndef _AMDXDNA_GEM_H_ #define _AMDXDNA_GEM_H_ +#include <linux/hmm.h> + +struct amdxdna_umap { + struct vm_area_struct *vma; + struct mmu_interval_notifier notifier; + struct hmm_range range; + struct work_struct hmm_unreg_work; + struct amdxdna_gem_obj *abo; + struct list_head node; + struct kref refcnt; + bool invalid; + bool unmapped; +}; + struct amdxdna_mem { u64 userptr; void *kva; @@ -13,8 +27,7 @@ struct amdxdna_mem { size_t size; struct page **pages; u32 nr_pages; - struct mmu_interval_notifier notifier; - unsigned long *pfns; + struct list_head umap_list; bool map_invalid; }; @@ -31,9 +44,12 @@ struct amdxdna_gem_obj { struct amdxdna_gem_obj *dev_heap; /* For AMDXDNA_BO_DEV */ struct drm_mm_node mm_node; /* For AMDXDNA_BO_DEV */ u32 assigned_hwctx; + struct dma_buf *dma_buf; + struct dma_buf_attachment *attach; }; #define to_gobj(obj) (&(obj)->base.base) +#define is_import_bo(obj) ((obj)->attach) static inline struct amdxdna_gem_obj *to_xdna_obj(struct drm_gem_object *gobj) { @@ -47,8 +63,12 @@ static inline void amdxdna_gem_put_obj(struct amdxdna_gem_obj *abo) drm_gem_object_put(to_gobj(abo)); } +void amdxdna_umap_put(struct amdxdna_umap *mapp); + struct drm_gem_object * amdxdna_gem_create_object_cb(struct drm_device *dev, size_t size); +struct drm_gem_object * +amdxdna_gem_prime_import(struct drm_device *dev, struct dma_buf *dma_buf); struct amdxdna_gem_obj * amdxdna_drm_alloc_dev_bo(struct drm_device *dev, struct amdxdna_drm_create_bo *args, diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.c b/drivers/accel/amdxdna/amdxdna_pci_drv.c index f5b8497cf5ad..f2bf1d374cc7 100644 --- a/drivers/accel/amdxdna/amdxdna_pci_drv.c +++ b/drivers/accel/amdxdna/amdxdna_pci_drv.c @@ -226,6 +226,7 @@ const struct drm_driver amdxdna_drm_drv = { .num_ioctls = ARRAY_SIZE(amdxdna_drm_ioctls), .gem_create_object = amdxdna_gem_create_object_cb, + .gem_prime_import = amdxdna_gem_prime_import, }; static const struct amdxdna_dev_info * @@ -266,12 +267,16 @@ static int amdxdna_probe(struct pci_dev *pdev, const struct pci_device_id *id) fs_reclaim_release(GFP_KERNEL); } + xdna->notifier_wq = alloc_ordered_workqueue("notifier_wq", 0); + if (!xdna->notifier_wq) + return -ENOMEM; + mutex_lock(&xdna->dev_lock); ret = xdna->dev_info->ops->init(xdna); mutex_unlock(&xdna->dev_lock); if (ret) { XDNA_ERR(xdna, "Hardware init failed, ret %d", ret); - return ret; + goto destroy_notifier_wq; } ret = amdxdna_sysfs_init(xdna); @@ -301,6 +306,8 @@ failed_dev_fini: mutex_lock(&xdna->dev_lock); xdna->dev_info->ops->fini(xdna); mutex_unlock(&xdna->dev_lock); +destroy_notifier_wq: + destroy_workqueue(xdna->notifier_wq); return ret; } @@ -310,6 +317,8 @@ static void amdxdna_remove(struct pci_dev *pdev) struct device *dev = &pdev->dev; struct amdxdna_client *client; + destroy_workqueue(xdna->notifier_wq); + pm_runtime_get_noresume(dev); pm_runtime_forbid(dev); diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.h b/drivers/accel/amdxdna/amdxdna_pci_drv.h index 37848a8d8031..ab79600911aa 100644 --- a/drivers/accel/amdxdna/amdxdna_pci_drv.h +++ b/drivers/accel/amdxdna/amdxdna_pci_drv.h @@ -6,6 +6,7 @@ #ifndef _AMDXDNA_PCI_DRV_H_ #define _AMDXDNA_PCI_DRV_H_ +#include <linux/workqueue.h> #include <linux/xarray.h> #define XDNA_INFO(xdna, fmt, args...) drm_info(&(xdna)->ddev, fmt, ##args) @@ -98,6 +99,7 @@ struct amdxdna_dev { struct list_head client_list; struct amdxdna_fw_ver fw_ver; struct rw_semaphore notifier_lock; /* for mmu notifier*/ + struct workqueue_struct *notifier_wq; }; /* diff --git a/drivers/accel/habanalabs/Kconfig b/drivers/accel/habanalabs/Kconfig index be85336107f9..1919fbb169c7 100644 --- a/drivers/accel/habanalabs/Kconfig +++ b/drivers/accel/habanalabs/Kconfig @@ -6,7 +6,7 @@ config DRM_ACCEL_HABANALABS tristate "HabanaLabs AI accelerators" depends on DRM_ACCEL - depends on X86_64 + depends on X86 && X86_64 depends on PCI && HAS_IOMEM select GENERIC_ALLOCATOR select HWMON diff --git a/drivers/accel/habanalabs/common/habanalabs_ioctl.c b/drivers/accel/habanalabs/common/habanalabs_ioctl.c index 8729a0c57d78..dc80ca921d90 100644 --- a/drivers/accel/habanalabs/common/habanalabs_ioctl.c +++ b/drivers/accel/habanalabs/common/habanalabs_ioctl.c @@ -17,8 +17,6 @@ #include <linux/uaccess.h> #include <linux/vmalloc.h> -#include <asm/msr.h> - /* make sure there is space for all the signed info */ static_assert(sizeof(struct cpucp_info) <= SEC_DEV_INFO_BUF_SZ); diff --git a/drivers/accel/ivpu/ivpu_debugfs.c b/drivers/accel/ivpu/ivpu_debugfs.c index f0dad0c9ce33..cd24ccd20ba6 100644 --- a/drivers/accel/ivpu/ivpu_debugfs.c +++ b/drivers/accel/ivpu/ivpu_debugfs.c @@ -455,7 +455,7 @@ priority_bands_fops_write(struct file *file, const char __user *user_buf, size_t if (ret < 0) return ret; - buf[size] = '\0'; + buf[ret] = '\0'; ret = sscanf(buf, "%u %u %u %u", &band, &grace_period, &process_grace_period, &process_quantum); if (ret != 4) diff --git a/drivers/accel/ivpu/ivpu_drv.c b/drivers/accel/ivpu/ivpu_drv.c index eff1d3ca075f..0e7748c5e117 100644 --- a/drivers/accel/ivpu/ivpu_drv.c +++ b/drivers/accel/ivpu/ivpu_drv.c @@ -374,6 +374,9 @@ int ivpu_boot(struct ivpu_device *vdev) { int ret; + drm_WARN_ON(&vdev->drm, atomic_read(&vdev->job_timeout_counter)); + drm_WARN_ON(&vdev->drm, !xa_empty(&vdev->submitted_jobs_xa)); + /* Update boot params located at first 4KB of FW memory */ ivpu_fw_boot_params_setup(vdev, ivpu_bo_vaddr(vdev->fw->mem)); @@ -573,6 +576,7 @@ static int ivpu_dev_init(struct ivpu_device *vdev) vdev->context_xa_limit.min = IVPU_USER_CONTEXT_MIN_SSID; vdev->context_xa_limit.max = IVPU_USER_CONTEXT_MAX_SSID; atomic64_set(&vdev->unique_id_counter, 0); + atomic_set(&vdev->job_timeout_counter, 0); xa_init_flags(&vdev->context_xa, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_IRQ); xa_init_flags(&vdev->submitted_jobs_xa, XA_FLAGS_ALLOC1); xa_init_flags(&vdev->db_xa, XA_FLAGS_ALLOC1); diff --git a/drivers/accel/ivpu/ivpu_drv.h b/drivers/accel/ivpu/ivpu_drv.h index 92753effb1c9..5497e7030e91 100644 --- a/drivers/accel/ivpu/ivpu_drv.h +++ b/drivers/accel/ivpu/ivpu_drv.h @@ -154,6 +154,7 @@ struct ivpu_device { struct mutex submitted_jobs_lock; /* Protects submitted_jobs */ struct xarray submitted_jobs_xa; struct ivpu_ipc_consumer job_done_consumer; + atomic_t job_timeout_counter; atomic64_t unique_id_counter; diff --git a/drivers/accel/ivpu/ivpu_fw.c b/drivers/accel/ivpu/ivpu_fw.c index ccaaf6c100c0..9db741695401 100644 --- a/drivers/accel/ivpu/ivpu_fw.c +++ b/drivers/accel/ivpu/ivpu_fw.c @@ -55,18 +55,18 @@ static struct { int gen; const char *name; } fw_names[] = { - { IVPU_HW_IP_37XX, "vpu_37xx.bin" }, + { IVPU_HW_IP_37XX, "intel/vpu/vpu_37xx_v1.bin" }, { IVPU_HW_IP_37XX, "intel/vpu/vpu_37xx_v0.0.bin" }, - { IVPU_HW_IP_40XX, "vpu_40xx.bin" }, + { IVPU_HW_IP_40XX, "intel/vpu/vpu_40xx_v1.bin" }, { IVPU_HW_IP_40XX, "intel/vpu/vpu_40xx_v0.0.bin" }, - { IVPU_HW_IP_50XX, "vpu_50xx.bin" }, + { IVPU_HW_IP_50XX, "intel/vpu/vpu_50xx_v1.bin" }, { IVPU_HW_IP_50XX, "intel/vpu/vpu_50xx_v0.0.bin" }, }; /* Production fw_names from the table above */ -MODULE_FIRMWARE("intel/vpu/vpu_37xx_v0.0.bin"); -MODULE_FIRMWARE("intel/vpu/vpu_40xx_v0.0.bin"); -MODULE_FIRMWARE("intel/vpu/vpu_50xx_v0.0.bin"); +MODULE_FIRMWARE("intel/vpu/vpu_37xx_v1.bin"); +MODULE_FIRMWARE("intel/vpu/vpu_40xx_v1.bin"); +MODULE_FIRMWARE("intel/vpu/vpu_50xx_v1.bin"); static int ivpu_fw_request(struct ivpu_device *vdev) { diff --git a/drivers/accel/ivpu/ivpu_fw.h b/drivers/accel/ivpu/ivpu_fw.h index 1d0b2bd9d65c..9a3935be1c05 100644 --- a/drivers/accel/ivpu/ivpu_fw.h +++ b/drivers/accel/ivpu/ivpu_fw.h @@ -39,6 +39,7 @@ struct ivpu_fw_info { u64 read_only_addr; u32 read_only_size; u32 sched_mode; + u64 last_heartbeat; }; int ivpu_fw_init(struct ivpu_device *vdev); diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c index 8741c73b92ce..59cfcf3eaded 100644 --- a/drivers/accel/ivpu/ivpu_gem.c +++ b/drivers/accel/ivpu/ivpu_gem.c @@ -28,9 +28,19 @@ static inline void ivpu_dbg_bo(struct ivpu_device *vdev, struct ivpu_bo *bo, con { ivpu_dbg(vdev, BO, "%6s: bo %8p vpu_addr %9llx size %8zu ctx %d has_pages %d dma_mapped %d mmu_mapped %d wc %d imported %d\n", - action, bo, bo->vpu_addr, ivpu_bo_size(bo), bo->ctx ? bo->ctx->id : 0, + action, bo, bo->vpu_addr, ivpu_bo_size(bo), bo->ctx_id, (bool)bo->base.pages, (bool)bo->base.sgt, bo->mmu_mapped, bo->base.map_wc, - (bool)bo->base.base.import_attach); + (bool)drm_gem_is_imported(&bo->base.base)); +} + +static inline int ivpu_bo_lock(struct ivpu_bo *bo) +{ + return dma_resv_lock(bo->base.base.resv, NULL); +} + +static inline void ivpu_bo_unlock(struct ivpu_bo *bo) +{ + dma_resv_unlock(bo->base.base.resv); } /* @@ -43,22 +53,22 @@ static inline void ivpu_dbg_bo(struct ivpu_device *vdev, struct ivpu_bo *bo, con int __must_check ivpu_bo_pin(struct ivpu_bo *bo) { struct ivpu_device *vdev = ivpu_bo_to_vdev(bo); + struct sg_table *sgt; int ret = 0; - mutex_lock(&bo->lock); - ivpu_dbg_bo(vdev, bo, "pin"); - drm_WARN_ON(&vdev->drm, !bo->ctx); - if (!bo->mmu_mapped) { - struct sg_table *sgt = drm_gem_shmem_get_pages_sgt(&bo->base); + sgt = drm_gem_shmem_get_pages_sgt(&bo->base); + if (IS_ERR(sgt)) { + ret = PTR_ERR(sgt); + ivpu_err(vdev, "Failed to map BO in IOMMU: %d\n", ret); + return ret; + } - if (IS_ERR(sgt)) { - ret = PTR_ERR(sgt); - ivpu_err(vdev, "Failed to map BO in IOMMU: %d\n", ret); - goto unlock; - } + ivpu_bo_lock(bo); + if (!bo->mmu_mapped) { + drm_WARN_ON(&vdev->drm, !bo->ctx); ret = ivpu_mmu_context_map_sgt(vdev, bo->ctx, bo->vpu_addr, sgt, ivpu_bo_is_snooped(bo)); if (ret) { @@ -69,7 +79,7 @@ int __must_check ivpu_bo_pin(struct ivpu_bo *bo) } unlock: - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); return ret; } @@ -84,7 +94,7 @@ ivpu_bo_alloc_vpu_addr(struct ivpu_bo *bo, struct ivpu_mmu_context *ctx, if (!drm_dev_enter(&vdev->drm, &idx)) return -ENODEV; - mutex_lock(&bo->lock); + ivpu_bo_lock(bo); ret = ivpu_mmu_context_insert_node(ctx, range, ivpu_bo_size(bo), &bo->mm_node); if (!ret) { @@ -94,9 +104,7 @@ ivpu_bo_alloc_vpu_addr(struct ivpu_bo *bo, struct ivpu_mmu_context *ctx, ivpu_err(vdev, "Failed to add BO to context %u: %d\n", ctx->id, ret); } - ivpu_dbg_bo(vdev, bo, "alloc"); - - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); drm_dev_exit(idx); @@ -107,7 +115,7 @@ static void ivpu_bo_unbind_locked(struct ivpu_bo *bo) { struct ivpu_device *vdev = ivpu_bo_to_vdev(bo); - lockdep_assert(lockdep_is_held(&bo->lock) || !kref_read(&bo->base.base.refcount)); + lockdep_assert(dma_resv_held(bo->base.base.resv) || !kref_read(&bo->base.base.refcount)); if (bo->mmu_mapped) { drm_WARN_ON(&vdev->drm, !bo->ctx); @@ -122,17 +130,15 @@ static void ivpu_bo_unbind_locked(struct ivpu_bo *bo) bo->ctx = NULL; } - if (bo->base.base.import_attach) + if (drm_gem_is_imported(&bo->base.base)) return; - dma_resv_lock(bo->base.base.resv, NULL); if (bo->base.sgt) { dma_unmap_sgtable(vdev->drm.dev, bo->base.sgt, DMA_BIDIRECTIONAL, 0); sg_free_table(bo->base.sgt); kfree(bo->base.sgt); bo->base.sgt = NULL; } - dma_resv_unlock(bo->base.base.resv); } void ivpu_bo_unbind_all_bos_from_context(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx) @@ -144,12 +150,12 @@ void ivpu_bo_unbind_all_bos_from_context(struct ivpu_device *vdev, struct ivpu_m mutex_lock(&vdev->bo_list_lock); list_for_each_entry(bo, &vdev->bo_list, bo_list_node) { - mutex_lock(&bo->lock); + ivpu_bo_lock(bo); if (bo->ctx == ctx) { ivpu_dbg_bo(vdev, bo, "unbind"); ivpu_bo_unbind_locked(bo); } - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); } mutex_unlock(&vdev->bo_list_lock); } @@ -169,7 +175,6 @@ struct drm_gem_object *ivpu_gem_create_object(struct drm_device *dev, size_t siz bo->base.pages_mark_dirty_on_put = true; /* VPU can dirty a BO anytime */ INIT_LIST_HEAD(&bo->bo_list_node); - mutex_init(&bo->lock); return &bo->base.base; } @@ -215,7 +220,7 @@ fail_detach: return ERR_PTR(ret); } -static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags) +static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags, u32 ctx_id) { struct drm_gem_shmem_object *shmem; struct ivpu_bo *bo; @@ -233,6 +238,7 @@ static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 fla return ERR_CAST(shmem); bo = to_ivpu_bo(&shmem->base); + bo->ctx_id = ctx_id; bo->base.map_wc = flags & DRM_IVPU_BO_WC; bo->flags = flags; @@ -240,6 +246,8 @@ static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 fla list_add_tail(&bo->bo_list_node, &vdev->bo_list); mutex_unlock(&vdev->bo_list_lock); + ivpu_dbg_bo(vdev, bo, "alloc"); + return bo; } @@ -277,12 +285,16 @@ static void ivpu_gem_bo_free(struct drm_gem_object *obj) list_del(&bo->bo_list_node); mutex_unlock(&vdev->bo_list_lock); - drm_WARN_ON(&vdev->drm, !dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_READ)); + drm_WARN_ON(&vdev->drm, !drm_gem_is_imported(&bo->base.base) && + !dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_READ)); + drm_WARN_ON(&vdev->drm, ivpu_bo_size(bo) == 0); + drm_WARN_ON(&vdev->drm, bo->base.vaddr); ivpu_bo_unbind_locked(bo); - mutex_destroy(&bo->lock); + drm_WARN_ON(&vdev->drm, bo->mmu_mapped); + drm_WARN_ON(&vdev->drm, bo->ctx); - drm_WARN_ON(obj->dev, bo->base.pages_use_count > 1); + drm_WARN_ON(obj->dev, refcount_read(&bo->base.pages_use_count) > 1); drm_gem_shmem_free(&bo->base); } @@ -314,7 +326,7 @@ int ivpu_bo_create_ioctl(struct drm_device *dev, void *data, struct drm_file *fi if (size == 0) return -EINVAL; - bo = ivpu_bo_alloc(vdev, size, args->flags); + bo = ivpu_bo_alloc(vdev, size, args->flags, file_priv->ctx.id); if (IS_ERR(bo)) { ivpu_err(vdev, "Failed to allocate BO: %pe (ctx %u size %llu flags 0x%x)", bo, file_priv->ctx.id, args->size, args->flags); @@ -322,7 +334,10 @@ int ivpu_bo_create_ioctl(struct drm_device *dev, void *data, struct drm_file *fi } ret = drm_gem_handle_create(file, &bo->base.base, &args->handle); - if (!ret) + if (ret) + ivpu_err(vdev, "Failed to create handle for BO: %pe (ctx %u size %llu flags 0x%x)", + bo, file_priv->ctx.id, args->size, args->flags); + else args->vpu_addr = bo->vpu_addr; drm_gem_object_put(&bo->base.base); @@ -345,7 +360,7 @@ ivpu_bo_create(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx, drm_WARN_ON(&vdev->drm, !PAGE_ALIGNED(range->end)); drm_WARN_ON(&vdev->drm, !PAGE_ALIGNED(size)); - bo = ivpu_bo_alloc(vdev, size, flags); + bo = ivpu_bo_alloc(vdev, size, flags, IVPU_GLOBAL_CONTEXT_MMU_SSID); if (IS_ERR(bo)) { ivpu_err(vdev, "Failed to allocate BO: %pe (vpu_addr 0x%llx size %llu flags 0x%x)", bo, range->start, size, flags); @@ -361,9 +376,9 @@ ivpu_bo_create(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx, goto err_put; if (flags & DRM_IVPU_BO_MAPPABLE) { - dma_resv_lock(bo->base.base.resv, NULL); - ret = drm_gem_shmem_vmap(&bo->base, &map); - dma_resv_unlock(bo->base.base.resv); + ivpu_bo_lock(bo); + ret = drm_gem_shmem_vmap_locked(&bo->base, &map); + ivpu_bo_unlock(bo); if (ret) goto err_put; @@ -386,9 +401,9 @@ void ivpu_bo_free(struct ivpu_bo *bo) struct iosys_map map = IOSYS_MAP_INIT_VADDR(bo->base.vaddr); if (bo->flags & DRM_IVPU_BO_MAPPABLE) { - dma_resv_lock(bo->base.base.resv, NULL); - drm_gem_shmem_vunmap(&bo->base, &map); - dma_resv_unlock(bo->base.base.resv); + ivpu_bo_lock(bo); + drm_gem_shmem_vunmap_locked(&bo->base, &map); + ivpu_bo_unlock(bo); } drm_gem_object_put(&bo->base.base); @@ -407,12 +422,12 @@ int ivpu_bo_info_ioctl(struct drm_device *dev, void *data, struct drm_file *file bo = to_ivpu_bo(obj); - mutex_lock(&bo->lock); + ivpu_bo_lock(bo); args->flags = bo->flags; args->mmap_offset = drm_vma_node_offset_addr(&obj->vma_node); args->vpu_addr = bo->vpu_addr; args->size = obj->size; - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); drm_gem_object_put(obj); return ret; @@ -449,10 +464,10 @@ int ivpu_bo_wait_ioctl(struct drm_device *dev, void *data, struct drm_file *file static void ivpu_bo_print_info(struct ivpu_bo *bo, struct drm_printer *p) { - mutex_lock(&bo->lock); + ivpu_bo_lock(bo); drm_printf(p, "%-9p %-3u 0x%-12llx %-10lu 0x%-8x %-4u", - bo, bo->ctx ? bo->ctx->id : 0, bo->vpu_addr, bo->base.base.size, + bo, bo->ctx_id, bo->vpu_addr, bo->base.base.size, bo->flags, kref_read(&bo->base.base.refcount)); if (bo->base.pages) @@ -461,12 +476,12 @@ static void ivpu_bo_print_info(struct ivpu_bo *bo, struct drm_printer *p) if (bo->mmu_mapped) drm_printf(p, " mmu_mapped"); - if (bo->base.base.import_attach) + if (drm_gem_is_imported(&bo->base.base)) drm_printf(p, " imported"); drm_printf(p, "\n"); - mutex_unlock(&bo->lock); + ivpu_bo_unlock(bo); } void ivpu_bo_list(struct drm_device *dev, struct drm_printer *p) diff --git a/drivers/accel/ivpu/ivpu_gem.h b/drivers/accel/ivpu/ivpu_gem.h index a222a9ec9d61..aa8ff14f7aae 100644 --- a/drivers/accel/ivpu/ivpu_gem.h +++ b/drivers/accel/ivpu/ivpu_gem.h @@ -17,10 +17,10 @@ struct ivpu_bo { struct list_head bo_list_node; struct drm_mm_node mm_node; - struct mutex lock; /* Protects: ctx, mmu_mapped, vpu_addr */ u64 vpu_addr; u32 flags; u32 job_status; /* Valid only for command buffer */ + u32 ctx_id; bool mmu_mapped; }; diff --git a/drivers/accel/ivpu/ivpu_job.c b/drivers/accel/ivpu/ivpu_job.c index b28da35c30b6..fae8351aa330 100644 --- a/drivers/accel/ivpu/ivpu_job.c +++ b/drivers/accel/ivpu/ivpu_job.c @@ -247,6 +247,10 @@ static int ivpu_cmdq_unregister(struct ivpu_file_priv *file_priv, struct ivpu_cm if (!cmdq->db_id) return 0; + ret = ivpu_jsm_unregister_db(vdev, cmdq->db_id); + if (!ret) + ivpu_dbg(vdev, JOB, "DB %d unregistered\n", cmdq->db_id); + if (vdev->fw->sched_mode == VPU_SCHEDULING_MODE_HW) { ret = ivpu_jsm_hws_destroy_cmdq(vdev, file_priv->ctx.id, cmdq->id); if (!ret) @@ -254,10 +258,6 @@ static int ivpu_cmdq_unregister(struct ivpu_file_priv *file_priv, struct ivpu_cm cmdq->id, file_priv->ctx.id); } - ret = ivpu_jsm_unregister_db(vdev, cmdq->db_id); - if (!ret) - ivpu_dbg(vdev, JOB, "DB %d unregistered\n", cmdq->db_id); - xa_erase(&file_priv->vdev->db_xa, cmdq->db_id); cmdq->db_id = 0; @@ -986,7 +986,8 @@ void ivpu_context_abort_work_fn(struct work_struct *work) return; if (vdev->fw->sched_mode == VPU_SCHEDULING_MODE_HW) - ivpu_jsm_reset_engine(vdev, 0); + if (ivpu_jsm_reset_engine(vdev, 0)) + return; mutex_lock(&vdev->context_list_lock); xa_for_each(&vdev->context_xa, ctx_id, file_priv) { @@ -1009,7 +1010,8 @@ void ivpu_context_abort_work_fn(struct work_struct *work) if (vdev->fw->sched_mode != VPU_SCHEDULING_MODE_HW) goto runtime_put; - ivpu_jsm_hws_resume_engine(vdev, 0); + if (ivpu_jsm_hws_resume_engine(vdev, 0)) + return; /* * In hardware scheduling mode NPU already has stopped processing jobs * and won't send us any further notifications, thus we have to free job related resources diff --git a/drivers/accel/ivpu/ivpu_jsm_msg.c b/drivers/accel/ivpu/ivpu_jsm_msg.c index 219ab8afefab..0256b2dfefc1 100644 --- a/drivers/accel/ivpu/ivpu_jsm_msg.c +++ b/drivers/accel/ivpu/ivpu_jsm_msg.c @@ -7,6 +7,7 @@ #include "ivpu_hw.h" #include "ivpu_ipc.h" #include "ivpu_jsm_msg.h" +#include "ivpu_pm.h" #include "vpu_jsm_api.h" const char *ivpu_jsm_msg_type_to_str(enum vpu_ipc_msg_type type) @@ -163,8 +164,10 @@ int ivpu_jsm_reset_engine(struct ivpu_device *vdev, u32 engine) ret = ivpu_ipc_send_receive(vdev, &req, VPU_JSM_MSG_ENGINE_RESET_DONE, &resp, VPU_IPC_CHAN_ASYNC_CMD, vdev->timeout.jsm); - if (ret) + if (ret) { ivpu_err_ratelimited(vdev, "Failed to reset engine %d: %d\n", engine, ret); + ivpu_pm_trigger_recovery(vdev, "Engine reset failed"); + } return ret; } @@ -354,8 +357,10 @@ int ivpu_jsm_hws_resume_engine(struct ivpu_device *vdev, u32 engine) ret = ivpu_ipc_send_receive(vdev, &req, VPU_JSM_MSG_HWS_RESUME_ENGINE_DONE, &resp, VPU_IPC_CHAN_ASYNC_CMD, vdev->timeout.jsm); - if (ret) + if (ret) { ivpu_err_ratelimited(vdev, "Failed to resume engine %d: %d\n", engine, ret); + ivpu_pm_trigger_recovery(vdev, "Engine resume failed"); + } return ret; } diff --git a/drivers/accel/ivpu/ivpu_pm.c b/drivers/accel/ivpu/ivpu_pm.c index ac0e22454596..ea30db181cd7 100644 --- a/drivers/accel/ivpu/ivpu_pm.c +++ b/drivers/accel/ivpu/ivpu_pm.c @@ -34,6 +34,7 @@ module_param_named(tdr_timeout_ms, ivpu_tdr_timeout_ms, ulong, 0644); MODULE_PARM_DESC(tdr_timeout_ms, "Timeout for device hang detection, in milliseconds, 0 - default"); #define PM_RESCHEDULE_LIMIT 5 +#define PM_TDR_HEARTBEAT_LIMIT 30 static void ivpu_pm_prepare_cold_boot(struct ivpu_device *vdev) { @@ -44,6 +45,7 @@ static void ivpu_pm_prepare_cold_boot(struct ivpu_device *vdev) ivpu_fw_log_reset(vdev); ivpu_fw_load(vdev); fw->entry_point = fw->cold_boot_entry_point; + fw->last_heartbeat = 0; } static void ivpu_pm_prepare_warm_boot(struct ivpu_device *vdev) @@ -189,7 +191,24 @@ static void ivpu_job_timeout_work(struct work_struct *work) { struct ivpu_pm_info *pm = container_of(work, struct ivpu_pm_info, job_timeout_work.work); struct ivpu_device *vdev = pm->vdev; + u64 heartbeat; + if (ivpu_jsm_get_heartbeat(vdev, 0, &heartbeat) || heartbeat <= vdev->fw->last_heartbeat) { + ivpu_err(vdev, "Job timeout detected, heartbeat not progressed\n"); + goto recovery; + } + + if (atomic_fetch_inc(&vdev->job_timeout_counter) > PM_TDR_HEARTBEAT_LIMIT) { + ivpu_err(vdev, "Job timeout detected, heartbeat limit exceeded\n"); + goto recovery; + } + + vdev->fw->last_heartbeat = heartbeat; + ivpu_start_job_timeout_detection(vdev); + return; + +recovery: + atomic_set(&vdev->job_timeout_counter, 0); ivpu_pm_trigger_recovery(vdev, "TDR"); } @@ -204,6 +223,7 @@ void ivpu_start_job_timeout_detection(struct ivpu_device *vdev) void ivpu_stop_job_timeout_detection(struct ivpu_device *vdev) { cancel_delayed_work_sync(&vdev->pm->job_timeout_work); + atomic_set(&vdev->job_timeout_counter, 0); } int ivpu_pm_suspend_cb(struct device *dev) diff --git a/drivers/accel/qaic/Kconfig b/drivers/accel/qaic/Kconfig index a9f866230058..5e405a19c157 100644 --- a/drivers/accel/qaic/Kconfig +++ b/drivers/accel/qaic/Kconfig @@ -8,7 +8,6 @@ config DRM_ACCEL_QAIC depends on DRM_ACCEL depends on PCI && HAS_IOMEM depends on MHI_BUS - depends on MMU select CRC32 help Enables driver for Qualcomm's Cloud AI accelerator PCIe cards that are diff --git a/drivers/accel/qaic/qaic_data.c b/drivers/accel/qaic/qaic_data.c index 43aba57b48f0..1bce1af7c72c 100644 --- a/drivers/accel/qaic/qaic_data.c +++ b/drivers/accel/qaic/qaic_data.c @@ -609,7 +609,7 @@ static int qaic_gem_object_mmap(struct drm_gem_object *obj, struct vm_area_struc struct scatterlist *sg; int ret = 0; - if (obj->import_attach) + if (drm_gem_is_imported(obj)) return -EINVAL; for (sg = bo->sgt->sgl; sg; sg = sg_next(sg)) { @@ -630,7 +630,7 @@ static void qaic_free_object(struct drm_gem_object *obj) { struct qaic_bo *bo = to_qaic_bo(obj); - if (obj->import_attach) { + if (drm_gem_is_imported(obj)) { /* DMABUF/PRIME Path */ drm_prime_gem_destroy(obj, NULL); } else { @@ -870,7 +870,7 @@ static int qaic_prepare_bo(struct qaic_device *qdev, struct qaic_bo *bo, { int ret; - if (bo->base.import_attach) + if (drm_gem_is_imported(&bo->base)) ret = qaic_prepare_import_bo(bo, hdr); else ret = qaic_prepare_export_bo(qdev, bo, hdr); @@ -894,7 +894,7 @@ static void qaic_unprepare_export_bo(struct qaic_device *qdev, struct qaic_bo *b static void qaic_unprepare_bo(struct qaic_device *qdev, struct qaic_bo *bo) { - if (bo->base.import_attach) + if (drm_gem_is_imported(&bo->base)) qaic_unprepare_import_bo(bo); else qaic_unprepare_export_bo(qdev, bo); diff --git a/drivers/accel/qaic/qaic_debugfs.c b/drivers/accel/qaic/qaic_debugfs.c index ba0cf2f94732..a991b8198dc4 100644 --- a/drivers/accel/qaic/qaic_debugfs.c +++ b/drivers/accel/qaic/qaic_debugfs.c @@ -240,7 +240,6 @@ static int qaic_bootlog_mhi_probe(struct mhi_device *mhi_dev, const struct mhi_d mhi_unprepare: mhi_unprepare_from_transfer(mhi_dev); destroy_workqueue: - flush_workqueue(qdev->bootlog_wq); destroy_workqueue(qdev->bootlog_wq); out: return ret; @@ -253,7 +252,6 @@ static void qaic_bootlog_mhi_remove(struct mhi_device *mhi_dev) qdev = dev_get_drvdata(&mhi_dev->dev); mhi_unprepare_from_transfer(qdev->bootlog_ch); - flush_workqueue(qdev->bootlog_wq); destroy_workqueue(qdev->bootlog_wq); qdev->bootlog_ch = NULL; } diff --git a/drivers/accel/qaic/qaic_timesync.c b/drivers/accel/qaic/qaic_timesync.c index 972833fabcfc..3fac540f8e03 100644 --- a/drivers/accel/qaic/qaic_timesync.c +++ b/drivers/accel/qaic/qaic_timesync.c @@ -129,7 +129,7 @@ static void qaic_timesync_dl_xfer_cb(struct mhi_device *mhi_dev, struct mhi_resu static void qaic_timesync_timer(struct timer_list *t) { - struct mqts_dev *mqtsdev = from_timer(mqtsdev, t, timer); + struct mqts_dev *mqtsdev = timer_container_of(mqtsdev, t, timer); struct qts_host_time_sync_msg_data *sync_msg; u64 device_qtimer_us; u64 device_qtimer; |