From: Mario Limonciello <mario.limonciello@amd.com>
To: Lizhi Hou <lizhi.hou@amd.com>,
ogabbay@kernel.org, quic_jhugo@quicinc.com,
dri-devel@lists.freedesktop.org,
maciej.falkowski@linux.intel.com
Cc: Max Zhen <max.zhen@amd.com>,
linux-kernel@vger.kernel.org, sonal.santan@amd.com
Subject: Re: [PATCH V1] accel/amdxdna: Add per-process BO memory usage query support
Date: Tue, 24 Mar 2026 12:01:38 -0500 [thread overview]
Message-ID: <1e573513-b281-4ae3-9140-7a87ee9dab08@amd.com> (raw)
In-Reply-To: <20260324163159.2425461-1-lizhi.hou@amd.com>
On 3/24/26 11:31, Lizhi Hou wrote:
> From: Max Zhen <max.zhen@amd.com>
>
> Add support for querying per-process buffer object (BO) memory
> usage through the amdxdna GET_ARRAY UAPI.
>
> Introduce a new query type, DRM_AMDXDNA_BO_USAGE, along with
> struct amdxdna_drm_bo_usage to report BO memory usage statistics,
> including heap, total, and internal usage.
>
> Track BO memory usage on a per-client basis by maintaining counters
> in GEM open/close and heap allocation/free paths. This ensures the
> reported statistics reflect the current memory footprint of each
> process.
>
> Wire the new query into the GET_ARRAY implementation to expose
> the usage information to userspace.
>
> Signed-off-by: Max Zhen <max.zhen@amd.com>
> Reviewed-by: Lizhi Hou <lizhi.hou@amd.com>
> Signed-off-by: Lizhi Hou <lizhi.hou@amd.com>
I'm assuming you also have userspace side ready for this too right?
If you have a link handy can you please include it when committing.
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
> ---
> drivers/accel/amdxdna/aie2_pci.c | 4 +
> drivers/accel/amdxdna/amdxdna_gem.c | 134 ++++++++++++++++++++++--
> drivers/accel/amdxdna/amdxdna_gem.h | 7 +-
> drivers/accel/amdxdna/amdxdna_pci_drv.c | 6 +-
> drivers/accel/amdxdna/amdxdna_pci_drv.h | 4 +
> include/uapi/drm/amdxdna_accel.h | 35 +++++++
> 6 files changed, 177 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/accel/amdxdna/aie2_pci.c b/drivers/accel/amdxdna/aie2_pci.c
> index 9e39bfe75971..f1ac4e00bd9f 100644
> --- a/drivers/accel/amdxdna/aie2_pci.c
> +++ b/drivers/accel/amdxdna/aie2_pci.c
> @@ -865,6 +865,7 @@ static int aie2_hwctx_status_cb(struct amdxdna_hwctx *hwctx, void *arg)
> tmp->command_submissions = hwctx->priv->seq;
> tmp->command_completions = hwctx->priv->completed;
> tmp->pasid = hwctx->client->pasid;
> + tmp->heap_usage = hwctx->client->heap_usage;
> tmp->priority = hwctx->qos.priority;
> tmp->gops = hwctx->qos.gops;
> tmp->fps = hwctx->qos.fps;
> @@ -1148,6 +1149,9 @@ static int aie2_get_array(struct amdxdna_client *client,
> case DRM_AMDXDNA_HW_LAST_ASYNC_ERR:
> ret = aie2_get_array_async_error(xdna->dev_handle, args);
> break;
> + case DRM_AMDXDNA_BO_USAGE:
> + ret = amdxdna_drm_get_bo_usage(&xdna->ddev, args);
> + break;
> default:
> XDNA_ERR(xdna, "Not supported request parameter %u", args->param);
> ret = -EOPNOTSUPP;
> diff --git a/drivers/accel/amdxdna/amdxdna_gem.c b/drivers/accel/amdxdna/amdxdna_gem.c
> index 27712704e42d..238ee244d4a6 100644
> --- a/drivers/accel/amdxdna/amdxdna_gem.c
> +++ b/drivers/accel/amdxdna/amdxdna_gem.c
> @@ -63,6 +63,8 @@ amdxdna_gem_heap_alloc(struct amdxdna_gem_obj *abo)
> goto unlock_out;
> }
>
> + client->heap_usage += mem->size;
> +
> drm_gem_object_get(to_gobj(heap));
>
> unlock_out:
> @@ -74,16 +76,17 @@ amdxdna_gem_heap_alloc(struct amdxdna_gem_obj *abo)
> static void
> amdxdna_gem_heap_free(struct amdxdna_gem_obj *abo)
> {
> + struct amdxdna_client *client = abo->client;
> struct amdxdna_gem_obj *heap;
>
> - mutex_lock(&abo->client->mm_lock);
> + mutex_lock(&client->mm_lock);
>
> drm_mm_remove_node(&abo->mm_node);
> -
> - heap = abo->client->dev_heap;
> + client->heap_usage -= abo->mem.size;
> + heap = client->dev_heap;
> drm_gem_object_put(to_gobj(heap));
>
> - mutex_unlock(&abo->client->mm_lock);
> + mutex_unlock(&client->mm_lock);
> }
>
> static struct amdxdna_gem_obj *
> @@ -102,6 +105,8 @@ amdxdna_gem_create_obj(struct drm_device *dev, size_t size)
> abo->mem.dma_addr = AMDXDNA_INVALID_ADDR;
> abo->mem.uva = AMDXDNA_INVALID_ADDR;
> abo->mem.size = size;
> + abo->open_ref = 0;
> + abo->internal = false;
> INIT_LIST_HEAD(&abo->mem.umap_list);
>
> return abo;
> @@ -508,13 +513,55 @@ static void amdxdna_imported_obj_free(struct amdxdna_gem_obj *abo)
> kfree(abo);
> }
>
> +static inline bool
> +amdxdna_gem_skip_bo_usage(struct amdxdna_gem_obj *abo)
> +{
> + /* Do not count imported BOs since the buffer is not allocated by us. */
> + if (is_import_bo(abo))
> + return true;
> +
> + /* Already counted as part of HEAP BO */
> + if (abo->type == AMDXDNA_BO_DEV)
> + return true;
> +
> + return false;
> +}
> +
> +static void
> +amdxdna_gem_add_bo_usage(struct amdxdna_gem_obj *abo)
> +{
> + struct amdxdna_client *client = abo->client;
> +
> + if (amdxdna_gem_skip_bo_usage(abo))
> + return;
> +
> + guard(mutex)(&client->mm_lock);
> +
> + client->total_bo_usage += abo->mem.size;
> + if (abo->internal)
> + client->total_int_bo_usage += abo->mem.size;
> +}
> +
> +static void
> +amdxdna_gem_del_bo_usage(struct amdxdna_gem_obj *abo)
> +{
> + struct amdxdna_client *client = abo->client;
> +
> + if (amdxdna_gem_skip_bo_usage(abo))
> + return;
> +
> + guard(mutex)(&client->mm_lock);
> +
> + client->total_bo_usage -= abo->mem.size;
> + if (abo->internal)
> + client->total_int_bo_usage -= abo->mem.size;
> +}
> +
> 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);
>
> - XDNA_DBG(xdna, "BO type %d xdna_addr 0x%llx", abo->type, amdxdna_gem_dev_addr(abo));
> -
> amdxdna_hmm_unregister(abo, NULL);
> flush_workqueue(xdna->notifier_wq);
>
> @@ -543,9 +590,13 @@ static int amdxdna_gem_obj_open(struct drm_gem_object *gobj, struct drm_file *fi
> int ret;
>
> guard(mutex)(&abo->lock);
> + abo->open_ref++;
>
> - if (!abo->client)
> + if (abo->open_ref == 1) {
> + /* Attached to the client when first opened by it. */
> abo->client = filp->driver_priv;
> + amdxdna_gem_add_bo_usage(abo);
> + }
> if (amdxdna_iova_on(xdna)) {
> ret = amdxdna_iommu_map_bo(xdna, abo);
> if (ret)
> @@ -555,6 +606,20 @@ static int amdxdna_gem_obj_open(struct drm_gem_object *gobj, struct drm_file *fi
> return 0;
> }
>
> +static void amdxdna_gem_obj_close(struct drm_gem_object *gobj, struct drm_file *filp)
> +{
> + struct amdxdna_gem_obj *abo = to_xdna_obj(gobj);
> +
> + guard(mutex)(&abo->lock);
> + abo->open_ref--;
> +
> + if (abo->open_ref == 0) {
> + amdxdna_gem_del_bo_usage(abo);
> + /* Detach from the client when last closed by it. */
> + abo->client = NULL;
> + }
> +}
> +
> static int amdxdna_gem_dev_obj_vmap(struct drm_gem_object *obj, struct iosys_map *map)
> {
> struct amdxdna_gem_obj *abo = to_xdna_obj(obj);
> @@ -575,6 +640,7 @@ static const struct drm_gem_object_funcs amdxdna_gem_dev_obj_funcs = {
> static const struct drm_gem_object_funcs amdxdna_gem_shmem_funcs = {
> .free = amdxdna_gem_obj_free,
> .open = amdxdna_gem_obj_open,
> + .close = amdxdna_gem_obj_close,
> .print_info = drm_gem_shmem_object_print_info,
> .pin = drm_gem_shmem_object_pin,
> .unpin = drm_gem_shmem_object_unpin,
> @@ -708,10 +774,13 @@ amdxdna_drm_create_share_bo(struct drm_device *dev,
> if (IS_ERR(abo))
> return ERR_CAST(abo);
>
> - if (args->type == AMDXDNA_BO_DEV_HEAP)
> + if (args->type == AMDXDNA_BO_DEV_HEAP) {
> abo->type = AMDXDNA_BO_DEV_HEAP;
> - else
> + abo->internal = true;
> + } else {
> abo->type = AMDXDNA_BO_SHARE;
> + abo->internal = args->type == AMDXDNA_BO_CMD;
> + }
>
> return abo;
> }
> @@ -783,6 +852,11 @@ amdxdna_drm_create_dev_bo(struct drm_device *dev,
> gobj = to_gobj(abo);
> gobj->funcs = &amdxdna_gem_dev_obj_funcs;
> abo->type = AMDXDNA_BO_DEV;
> + abo->internal = true;
> + /*
> + * DEV BOs cannot be alive when client is gone, it's OK to
> + * always establish the connection.
> + */
> abo->client = client;
>
> ret = amdxdna_gem_heap_alloc(abo);
> @@ -826,7 +900,7 @@ int amdxdna_drm_create_bo_ioctl(struct drm_device *dev, void *data, struct drm_f
> if (IS_ERR(abo))
> return PTR_ERR(abo);
>
> - /* ready to publish object to userspace */
> + /* Ready to publish object to userspace and count for BO usage. */
> ret = drm_gem_handle_create(filp, to_gobj(abo), &args->handle);
> if (ret) {
> XDNA_ERR(xdna, "Create handle failed");
> @@ -986,3 +1060,43 @@ int amdxdna_drm_sync_bo_ioctl(struct drm_device *dev,
> drm_gem_object_put(gobj);
> return ret;
> }
> +
> +int amdxdna_drm_get_bo_usage(struct drm_device *dev, struct amdxdna_drm_get_array *args)
> +{
> + size_t min_sz = min(args->element_size, sizeof(struct amdxdna_drm_bo_usage));
> + char __user *buf = u64_to_user_ptr(args->buffer);
> + struct amdxdna_dev *xdna = to_xdna_dev(dev);
> + struct amdxdna_client *tmp_client;
> + struct amdxdna_drm_bo_usage tmp;
> +
> + drm_WARN_ON(dev, !mutex_is_locked(&xdna->dev_lock));
> +
> + if (args->num_element != 1)
> + return -EINVAL;
> +
> + if (copy_from_user(&tmp, buf, min_sz))
> + return -EFAULT;
> +
> + if (!tmp.pid)
> + return -EINVAL;
> +
> + tmp.total_usage = 0;
> + tmp.internal_usage = 0;
> + tmp.heap_usage = 0;
> +
> + list_for_each_entry(tmp_client, &xdna->client_list, node) {
> + if (tmp.pid != tmp_client->pid)
> + continue;
> +
> + mutex_lock(&tmp_client->mm_lock);
> + tmp.total_usage += tmp_client->total_bo_usage;
> + tmp.internal_usage += tmp_client->total_int_bo_usage;
> + tmp.heap_usage += tmp_client->heap_usage;
> + mutex_unlock(&tmp_client->mm_lock);
> + }
> +
> + if (copy_to_user(buf, &tmp, min_sz))
> + return -EFAULT;
> +
> + return 0;
> +}
> diff --git a/drivers/accel/amdxdna/amdxdna_gem.h b/drivers/accel/amdxdna/amdxdna_gem.h
> index a77d9344f8a4..4fc48a1189d2 100644
> --- a/drivers/accel/amdxdna/amdxdna_gem.h
> +++ b/drivers/accel/amdxdna/amdxdna_gem.h
> @@ -41,8 +41,9 @@ struct amdxdna_gem_obj {
> struct amdxdna_client *client;
> u8 type;
> bool pinned;
> - struct mutex lock; /* Protects: pinned, mem.kva */
> + struct mutex lock; /* Protects: pinned, mem.kva, open_ref */
> struct amdxdna_mem mem;
> + int open_ref;
>
> /* Below members are initialized when needed */
> struct drm_mm mm; /* For AMDXDNA_BO_DEV_HEAP */
> @@ -50,6 +51,9 @@ struct amdxdna_gem_obj {
> u32 assigned_hwctx;
> struct dma_buf *dma_buf;
> struct dma_buf_attachment *attach;
> +
> + /* True, if BO is managed by XRT, not application */
> + bool internal;
> };
>
> #define to_gobj(obj) (&(obj)->base.base)
> @@ -98,5 +102,6 @@ void amdxdna_gem_unpin(struct amdxdna_gem_obj *abo);
> int amdxdna_drm_create_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *filp);
> int amdxdna_drm_get_bo_info_ioctl(struct drm_device *dev, void *data, struct drm_file *filp);
> int amdxdna_drm_sync_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *filp);
> +int amdxdna_drm_get_bo_usage(struct drm_device *dev, struct amdxdna_drm_get_array *args);
>
> #endif /* _AMDXDNA_GEM_H_ */
> diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.c b/drivers/accel/amdxdna/amdxdna_pci_drv.c
> index d83be00daf2b..b50a7d1f8a11 100644
> --- a/drivers/accel/amdxdna/amdxdna_pci_drv.c
> +++ b/drivers/accel/amdxdna/amdxdna_pci_drv.c
> @@ -36,9 +36,10 @@ MODULE_FIRMWARE("amdnpu/17f0_11/npu_7.sbin");
> * 0.5: Support getting telemetry data
> * 0.6: Support preemption
> * 0.7: Support getting power and utilization data
> + * 0.8: Support BO usage query
> */
> #define AMDXDNA_DRIVER_MAJOR 0
> -#define AMDXDNA_DRIVER_MINOR 7
> +#define AMDXDNA_DRIVER_MINOR 8
>
> /*
> * Bind the driver base on (vendor_id, device_id) pair and later use the
> @@ -120,11 +121,12 @@ static void amdxdna_client_cleanup(struct amdxdna_client *client)
> amdxdna_hwctx_remove_all(client);
> xa_destroy(&client->hwctx_xa);
> cleanup_srcu_struct(&client->hwctx_srcu);
> - mutex_destroy(&client->mm_lock);
>
> if (client->dev_heap)
> drm_gem_object_put(to_gobj(client->dev_heap));
>
> + mutex_destroy(&client->mm_lock);
> +
> if (!IS_ERR_OR_NULL(client->sva))
> iommu_sva_unbind_device(client->sva);
> mmdrop(client->mm);
> diff --git a/drivers/accel/amdxdna/amdxdna_pci_drv.h b/drivers/accel/amdxdna/amdxdna_pci_drv.h
> index e91d14ae5190..0661749917d6 100644
> --- a/drivers/accel/amdxdna/amdxdna_pci_drv.h
> +++ b/drivers/accel/amdxdna/amdxdna_pci_drv.h
> @@ -138,6 +138,10 @@ struct amdxdna_client {
> struct iommu_sva *sva;
> int pasid;
> struct mm_struct *mm;
> +
> + size_t heap_usage;
> + size_t total_bo_usage;
> + size_t total_int_bo_usage;
> };
>
> #define amdxdna_for_each_hwctx(client, hwctx_id, entry) \
> diff --git a/include/uapi/drm/amdxdna_accel.h b/include/uapi/drm/amdxdna_accel.h
> index bddaaaf945cf..61d3686fa3b1 100644
> --- a/include/uapi/drm/amdxdna_accel.h
> +++ b/include/uapi/drm/amdxdna_accel.h
> @@ -591,8 +591,37 @@ struct amdxdna_async_error {
> __u64 ex_err_code;
> };
>
> +/**
> + * struct amdxdna_drm_bo_usage - all types of BO usage
> + * BOs managed by XRT/SHIM/driver is counted as internal.
> + * Others are counted as external which are managed by applications.
> + *
> + * Among all types of BOs:
> + * AMDXDNA_BO_DEV_HEAP - is counted for internal.
> + * AMDXDNA_BO_SHARE - is counted for external.
> + * AMDXDNA_BO_CMD - is counted for internal.
> + * AMDXDNA_BO_DEV - is counted by heap_usage only, not internal
> + * or external. It does not add to the total memory
> + * footprint since its mem comes from heap which is
> + * already counted as internal.
> + */
> +struct amdxdna_drm_bo_usage {
> + /** @pid: The ID of the process to query from. */
> + __s64 pid;
> + /** @total_usage: Total BO size used by process. */
> + __u64 total_usage;
> + /** @internal_usage: Total internal BO size used by process. */
> + __u64 internal_usage;
> + /** @heap_usage: Total device BO size used by process. */
> + __u64 heap_usage;
> +};
> +
> +/*
> + * Supported params in struct amdxdna_drm_get_array
> + */
> #define DRM_AMDXDNA_HW_CONTEXT_ALL 0
> #define DRM_AMDXDNA_HW_LAST_ASYNC_ERR 2
> +#define DRM_AMDXDNA_BO_USAGE 6
>
> /**
> * struct amdxdna_drm_get_array - Get information array.
> @@ -605,6 +634,12 @@ struct amdxdna_drm_get_array {
> *
> * %DRM_AMDXDNA_HW_CONTEXT_ALL:
> * Returns all created hardware contexts.
> + *
> + * %DRM_AMDXDNA_HW_LAST_ASYNC_ERR:
> + * Returns last async error.
> + *
> + * %DRM_AMDXDNA_BO_USAGE:
> + * Returns usage of heap/internal/external BOs.
> */
> __u32 param;
> /**
next prev parent reply other threads:[~2026-03-24 17:01 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-03-24 16:31 [PATCH V1] accel/amdxdna: Add per-process BO memory usage query support Lizhi Hou
2026-03-24 17:01 ` Mario Limonciello [this message]
2026-03-24 20:39 ` Claude review: " Claude Code Review Bot
2026-03-24 20:39 ` Claude Code Review Bot
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=1e573513-b281-4ae3-9140-7a87ee9dab08@amd.com \
--to=mario.limonciello@amd.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=linux-kernel@vger.kernel.org \
--cc=lizhi.hou@amd.com \
--cc=maciej.falkowski@linux.intel.com \
--cc=max.zhen@amd.com \
--cc=ogabbay@kernel.org \
--cc=quic_jhugo@quicinc.com \
--cc=sonal.santan@amd.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