From: Karol Wachowski <karol.wachowski@linux.intel.com>
To: Maciej Falkowski <maciej.falkowski@linux.intel.com>,
dri-devel@lists.freedesktop.org
Cc: oded.gabbay@gmail.com, jeff.hugo@oss.qualcomm.com,
lizhi.hou@amd.com,
Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Subject: Re: [PATCH] accel/ivpu: Rework bind/unbind of imported buffers
Date: Wed, 1 Oct 2025 09:49:55 +0200 [thread overview]
Message-ID: <375ebdd2-6548-4997-8b19-923321472ccb@linux.intel.com> (raw)
In-Reply-To: <20250925145059.1446243-1-maciej.falkowski@linux.intel.com>
Reviewed-by: Karol Wachowski <karol.wachowski@linux.intel.com>
On 9/25/2025 4:50 PM, Maciej Falkowski wrote:
> From: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
>
> Ensure that imported buffers are properly mapped and unmapped in
> the same way as regular buffers to properly handle buffers during
> device's bind and unbind operations to prevent resource leaks and
> inconsistent buffer states.
>
> Imported buffers are now dma_mapped before submission and
> dma_unmapped in ivpu_bo_unbind(), guaranteeing they are unmapped
> when the device is unbound.
>
> Add also imported buffers to vdev->bo_list for consistent unmapping
> on device unbind. The bo->ctx_id is set in open() so imported
> buffers have a valid context ID.
>
> Debug logs have been updated to match the new code structure.
> The function ivpu_bo_pin() has been renamed to ivpu_bo_bind()
> to better reflect its purpose, and unbind tests have been refactored
> for improved coverage and clarity.
>
> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
> Signed-off-by: Maciej Falkowski <maciej.falkowski@linux.intel.com>
> ---
> drivers/accel/ivpu/ivpu_gem.c | 90 ++++++++++++++++++++++-------------
> drivers/accel/ivpu/ivpu_gem.h | 2 +-
> drivers/accel/ivpu/ivpu_job.c | 2 +-
> 3 files changed, 60 insertions(+), 34 deletions(-)
>
> diff --git a/drivers/accel/ivpu/ivpu_gem.c b/drivers/accel/ivpu/ivpu_gem.c
> index cceb07232e12..0cb48aff396c 100644
> --- a/drivers/accel/ivpu/ivpu_gem.c
> +++ b/drivers/accel/ivpu/ivpu_gem.c
> @@ -28,8 +28,8 @@ static const struct drm_gem_object_funcs ivpu_gem_funcs;
> static inline void ivpu_dbg_bo(struct ivpu_device *vdev, struct ivpu_bo *bo, const char *action)
> {
> 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_id,
> + "%6s: bo %8p size %9zu ctx %d vpu_addr %9llx pages %d sgt %d mmu_mapped %d wc %d imported %d\n",
> + action, bo, ivpu_bo_size(bo), bo->ctx_id, bo->vpu_addr,
> (bool)bo->base.pages, (bool)bo->base.sgt, bo->mmu_mapped, bo->base.map_wc,
> (bool)drm_gem_is_imported(&bo->base.base));
> }
> @@ -44,22 +44,46 @@ static inline void ivpu_bo_unlock(struct ivpu_bo *bo)
> dma_resv_unlock(bo->base.base.resv);
> }
>
> +static struct sg_table *ivpu_bo_map_attachment(struct ivpu_device *vdev, struct ivpu_bo *bo)
> +{
> + struct sg_table *sgt = bo->base.sgt;
> +
> + drm_WARN_ON(&vdev->drm, !bo->base.base.import_attach);
> +
> + ivpu_bo_lock(bo);
> +
> + if (!sgt) {
> + sgt = dma_buf_map_attachment(bo->base.base.import_attach, DMA_BIDIRECTIONAL);
> + if (IS_ERR(sgt))
> + ivpu_err(vdev, "Failed to map BO in IOMMU: %ld\n", PTR_ERR(sgt));
> + else
> + bo->base.sgt = sgt;
> + }
> +
> + ivpu_bo_unlock(bo);
> +
> + return sgt;
> +}
> +
> /*
> - * ivpu_bo_pin() - pin the backing physical pages and map them to VPU.
> + * ivpu_bo_bind() - pin the backing physical pages and map them to VPU.
> *
> * This function pins physical memory pages, then maps the physical pages
> * to IOMMU address space and finally updates the VPU MMU page tables
> * to allow the VPU to translate VPU address to IOMMU address.
> */
> -int __must_check ivpu_bo_pin(struct ivpu_bo *bo)
> +int __must_check ivpu_bo_bind(struct ivpu_bo *bo)
> {
> struct ivpu_device *vdev = ivpu_bo_to_vdev(bo);
> struct sg_table *sgt;
> int ret = 0;
>
> - ivpu_dbg_bo(vdev, bo, "pin");
> + ivpu_dbg_bo(vdev, bo, "bind");
>
> - sgt = drm_gem_shmem_get_pages_sgt(&bo->base);
> + if (bo->base.base.import_attach)
> + sgt = ivpu_bo_map_attachment(vdev, bo);
> + else
> + 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);
> @@ -100,7 +124,9 @@ ivpu_bo_alloc_vpu_addr(struct ivpu_bo *bo, struct ivpu_mmu_context *ctx,
> ret = ivpu_mmu_context_insert_node(ctx, range, ivpu_bo_size(bo), &bo->mm_node);
> if (!ret) {
> bo->ctx = ctx;
> + bo->ctx_id = ctx->id;
> bo->vpu_addr = bo->mm_node.start;
> + ivpu_dbg_bo(vdev, bo, "vaddr");
> } else {
> ivpu_err(vdev, "Failed to add BO to context %u: %d\n", ctx->id, ret);
> }
> @@ -116,7 +142,7 @@ static void ivpu_bo_unbind_locked(struct ivpu_bo *bo)
> {
> struct ivpu_device *vdev = ivpu_bo_to_vdev(bo);
>
> - lockdep_assert(dma_resv_held(bo->base.base.resv) || !kref_read(&bo->base.base.refcount));
> + dma_resv_assert_held(bo->base.base.resv);
>
> if (bo->mmu_mapped) {
> drm_WARN_ON(&vdev->drm, !bo->ctx);
> @@ -135,9 +161,14 @@ static void ivpu_bo_unbind_locked(struct ivpu_bo *bo)
> return;
>
> 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);
> + if (bo->base.base.import_attach) {
> + dma_buf_unmap_attachment(bo->base.base.import_attach,
> + bo->base.sgt, DMA_BIDIRECTIONAL);
> + } else {
> + 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;
> }
> }
> @@ -163,6 +194,7 @@ void ivpu_bo_unbind_all_bos_from_context(struct ivpu_device *vdev, struct ivpu_m
>
> struct drm_gem_object *ivpu_gem_create_object(struct drm_device *dev, size_t size)
> {
> + struct ivpu_device *vdev = to_ivpu_device(dev);
> struct ivpu_bo *bo;
>
> if (size == 0 || !PAGE_ALIGNED(size))
> @@ -177,6 +209,11 @@ struct drm_gem_object *ivpu_gem_create_object(struct drm_device *dev, size_t siz
>
> INIT_LIST_HEAD(&bo->bo_list_node);
>
> + mutex_lock(&vdev->bo_list_lock);
> + list_add_tail(&bo->bo_list_node, &vdev->bo_list);
> + mutex_unlock(&vdev->bo_list_lock);
> +
> + ivpu_dbg(vdev, BO, " alloc: bo %8p size %9zu\n", bo, size);
> return &bo->base.base;
> }
>
> @@ -185,7 +222,6 @@ struct drm_gem_object *ivpu_gem_prime_import(struct drm_device *dev,
> {
> struct device *attach_dev = dev->dev;
> struct dma_buf_attachment *attach;
> - struct sg_table *sgt;
> struct drm_gem_object *obj;
> int ret;
>
> @@ -195,16 +231,10 @@ struct drm_gem_object *ivpu_gem_prime_import(struct drm_device *dev,
>
> get_dma_buf(dma_buf);
>
> - sgt = dma_buf_map_attachment_unlocked(attach, DMA_BIDIRECTIONAL);
> - if (IS_ERR(sgt)) {
> - ret = PTR_ERR(sgt);
> - goto fail_detach;
> - }
> -
> - obj = drm_gem_shmem_prime_import_sg_table(dev, attach, sgt);
> + obj = drm_gem_shmem_prime_import_sg_table(dev, attach, NULL);
> if (IS_ERR(obj)) {
> ret = PTR_ERR(obj);
> - goto fail_unmap;
> + goto fail_detach;
> }
>
> obj->import_attach = attach;
> @@ -212,8 +242,6 @@ struct drm_gem_object *ivpu_gem_prime_import(struct drm_device *dev,
>
> return obj;
>
> -fail_unmap:
> - dma_buf_unmap_attachment_unlocked(attach, sgt, DMA_BIDIRECTIONAL);
> fail_detach:
> dma_buf_detach(dma_buf, attach);
> dma_buf_put(dma_buf);
> @@ -221,7 +249,7 @@ struct drm_gem_object *ivpu_gem_prime_import(struct drm_device *dev,
> return ERR_PTR(ret);
> }
>
> -static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags, u32 ctx_id)
> +static struct ivpu_bo *ivpu_bo_alloc(struct ivpu_device *vdev, u64 size, u32 flags)
> {
> struct drm_gem_shmem_object *shmem;
> struct ivpu_bo *bo;
> @@ -239,16 +267,9 @@ 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;
>
> - mutex_lock(&vdev->bo_list_lock);
> - list_add_tail(&bo->bo_list_node, &vdev->bo_list);
> - mutex_unlock(&vdev->bo_list_lock);
> -
> - ivpu_dbg_bo(vdev, bo, "alloc");
> -
> return bo;
> }
>
> @@ -282,6 +303,8 @@ static void ivpu_gem_bo_free(struct drm_gem_object *obj)
>
> ivpu_dbg_bo(vdev, bo, "free");
>
> + drm_WARN_ON(&vdev->drm, list_empty(&bo->bo_list_node));
> +
> mutex_lock(&vdev->bo_list_lock);
> list_del(&bo->bo_list_node);
> mutex_unlock(&vdev->bo_list_lock);
> @@ -291,7 +314,10 @@ static void ivpu_gem_bo_free(struct drm_gem_object *obj)
> drm_WARN_ON(&vdev->drm, ivpu_bo_size(bo) == 0);
> drm_WARN_ON(&vdev->drm, bo->base.vaddr);
>
> + ivpu_bo_lock(bo);
> ivpu_bo_unbind_locked(bo);
> + ivpu_bo_unlock(bo);
> +
> drm_WARN_ON(&vdev->drm, bo->mmu_mapped);
> drm_WARN_ON(&vdev->drm, bo->ctx);
>
> @@ -327,7 +353,7 @@ 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, file_priv->ctx.id);
> + bo = ivpu_bo_alloc(vdev, size, args->flags);
> 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);
> @@ -361,7 +387,7 @@ 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, IVPU_GLOBAL_CONTEXT_MMU_SSID);
> + bo = ivpu_bo_alloc(vdev, size, flags);
> 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);
> @@ -372,7 +398,7 @@ ivpu_bo_create(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx,
> if (ret)
> goto err_put;
>
> - ret = ivpu_bo_pin(bo);
> + ret = ivpu_bo_bind(bo);
> if (ret)
> goto err_put;
>
> diff --git a/drivers/accel/ivpu/ivpu_gem.h b/drivers/accel/ivpu/ivpu_gem.h
> index 3a208f3bf0a6..54452eb8a41f 100644
> --- a/drivers/accel/ivpu/ivpu_gem.h
> +++ b/drivers/accel/ivpu/ivpu_gem.h
> @@ -24,7 +24,7 @@ struct ivpu_bo {
> bool mmu_mapped;
> };
>
> -int ivpu_bo_pin(struct ivpu_bo *bo);
> +int ivpu_bo_bind(struct ivpu_bo *bo);
> void ivpu_bo_unbind_all_bos_from_context(struct ivpu_device *vdev, struct ivpu_mmu_context *ctx);
>
> struct drm_gem_object *ivpu_gem_create_object(struct drm_device *dev, size_t size);
> diff --git a/drivers/accel/ivpu/ivpu_job.c b/drivers/accel/ivpu/ivpu_job.c
> index 044268d0fc87..17273c68f84c 100644
> --- a/drivers/accel/ivpu/ivpu_job.c
> +++ b/drivers/accel/ivpu/ivpu_job.c
> @@ -751,7 +751,7 @@ ivpu_job_prepare_bos_for_submit(struct drm_file *file, struct ivpu_job *job, u32
>
> job->bos[i] = to_ivpu_bo(obj);
>
> - ret = ivpu_bo_pin(job->bos[i]);
> + ret = ivpu_bo_bind(job->bos[i]);
> if (ret)
> return ret;
> }
next prev parent reply other threads:[~2025-10-01 7:50 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-09-25 14:50 [PATCH] accel/ivpu: Rework bind/unbind of imported buffers Maciej Falkowski
2025-10-01 7:49 ` Karol Wachowski [this message]
2025-10-01 8:02 ` Karol Wachowski
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=375ebdd2-6548-4997-8b19-923321472ccb@linux.intel.com \
--to=karol.wachowski@linux.intel.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=jacek.lawrynowicz@linux.intel.com \
--cc=jeff.hugo@oss.qualcomm.com \
--cc=lizhi.hou@amd.com \
--cc=maciej.falkowski@linux.intel.com \
--cc=oded.gabbay@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox