lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <9eb1bdbb-6a0c-4771-ae61-4ba9bd7fbd14@suse.de>
Date: Thu, 8 May 2025 11:30:39 +0200
From: Thomas Zimmermann <tzimmermann@...e.de>
To: oushixiong1025@....com, Christian König
 <christian.koenig@....com>
Cc: Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
 Maxime Ripard <mripard@...nel.org>, David Airlie <airlied@...il.com>,
 Simona Vetter <simona@...ll.ch>, Sean Paul <sean@...rly.run>,
 Jocelyn Falempe <jfalempe@...hat.com>, dri-devel@...ts.freedesktop.org,
 linux-kernel@...r.kernel.org, Shixiong Ou <oushixiong@...inos.cn>
Subject: Re: [PATCH v3 1/3] drm/shmem-helper: Import dmabuf without mapping
 its sg_table

Hi

Am 07.05.25 um 11:47 schrieb oushixiong1025@....com:
> From: Shixiong Ou <oushixiong@...inos.cn>
>
> [WHY]
> 1. Drivers using DRM_GEM_SHADOW_PLANE_HELPER_FUNCS and
>     DRM_GEM_SHMEM_DRIVER_OPS (e.g., udl, ast) do not require
>     sg_table import.
>     They only need dma_buf_vmap() to access the shared buffer's
>     kernel virtual address.
>
> 2. On certain Aspeed-based boards, a dma_mask of 0xffff_ffff may
>     trigger SWIOTLB during dmabuf import. However, IO_TLB_SEGSIZE
>     restricts the maximum DMA streaming mapping memory, resulting in
>     errors like:
>
>     ast 0000:07:00.0: swiotlb buffer is full (sz: 3145728 bytes), total 32768 (slots), used 0 (slots)
>
> [HOW]
> Provide a gem_prime_import implementation without sg_table mapping
> to avoid issues (e.g., "swiotlb buffer is full"). Drivers that do not
> require sg_table can adopt this.
>
> Signed-off-by: Shixiong Ou <oushixiong@...inos.cn>
> ---
> v1->v2:
>          Patch rebase.
> v2->v3:
>          Rename the import callback function.
>          Remove drm_gem_shmem_prime_export() and separate some codes
>          to drm_gem_prime_import_self().
>
>   drivers/gpu/drm/drm_gem_shmem_helper.c | 51 ++++++++++++++++++++++++++
>   drivers/gpu/drm/drm_prime.c            | 46 +++++++++++++++++------
>   include/drm/drm_gem_shmem_helper.h     | 15 ++++++++
>   include/drm/drm_prime.h                |  3 ++
>   4 files changed, 104 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> index aa43265f4f4f..85900ec1954a 100644
> --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> @@ -800,6 +800,57 @@ drm_gem_shmem_prime_import_sg_table(struct drm_device *dev,
>   }
>   EXPORT_SYMBOL_GPL(drm_gem_shmem_prime_import_sg_table);
>   
> +/**
> + * drm_gem_shmem_prime_import_no_sgt - Import dmabuf without mapping its sg_table
> + * @dev: Device to import into
> + * @dma_buf: dma-buf object to import
> + *
> + * Drivers that use the shmem helpers but also wants to import dmabuf without
> + * mapping its sg_table can use this as their &drm_driver.gem_prime_import
> + * implementation.
> + */
> +struct drm_gem_object *drm_gem_shmem_prime_import_no_sgt(struct drm_device *dev,
> +							 struct dma_buf *dma_buf)
> +{
> +	struct dma_buf_attachment *attach;
> +	struct drm_gem_shmem_object *shmem;
> +	struct drm_gem_object *obj;
> +	size_t size;
> +	int ret;
> +
> +	obj = drm_gem_prime_import_self(dev, dma_buf);
> +	if (obj)
> +		return obj;
> +
> +	attach = dma_buf_attach(dma_buf, dev->dev);
> +	if (IS_ERR(attach))
> +		return ERR_CAST(attach);
> +
> +	get_dma_buf(dma_buf);
> +
> +	size = PAGE_ALIGN(attach->dmabuf->size);
> +
> +	shmem = __drm_gem_shmem_create(dev, size, true, NULL);
> +	if (IS_ERR(shmem)) {
> +		ret = PTR_ERR(shmem);
> +		goto fail_detach;
> +	}
> +
> +	drm_dbg_prime(dev, "size = %zu\n", size);
> +
> +	shmem->base.import_attach = attach;
> +	shmem->base.resv = dma_buf->resv;
> +
> +	return &shmem->base;
> +
> +fail_detach:
> +	dma_buf_detach(dma_buf, attach);
> +	dma_buf_put(dma_buf);
> +
> +	return ERR_PTR(ret);
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_prime_import_no_sgt);
> +
>   MODULE_DESCRIPTION("DRM SHMEM memory-management helpers");
>   MODULE_IMPORT_NS("DMA_BUF");
>   MODULE_LICENSE("GPL v2");
> diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
> index d828502268b8..1e43b6022329 100644
> --- a/drivers/gpu/drm/drm_prime.c
> +++ b/drivers/gpu/drm/drm_prime.c
> @@ -910,6 +910,38 @@ struct dma_buf *drm_gem_prime_export(struct drm_gem_object *obj,
>   }
>   EXPORT_SYMBOL(drm_gem_prime_export);
>   
> +
> +/**
> + * drm_gem_prime_import_self - Import a DMA-BUF exported from the same DRM device.
> + * @dev: drm_device to check against
> + * @dma_buf: dma-buf object to import
> + *
> + * This function checks if the DMA-BUF was exported from a GEM object belonging
> + * to @dev. If so, it increments the GEM object's refcount and returns it directly.
> + *
> + * Return: GEM object if it belongs to @dev, NULL otherwise.
> + */
> +struct drm_gem_object *drm_gem_prime_import_self(struct drm_device *dev,
> +						 struct dma_buf *dma_buf)
> +{
> +	struct drm_gem_object *obj;
> +
> +	if (dma_buf->ops == &drm_gem_prime_dmabuf_ops) {
> +		obj = dma_buf->priv;
> +		if (obj->dev == dev) {
> +			/*
> +			 * Importing dmabuf exported from our own gem increases
> +			 * refcount on gem itself instead of f_count of dmabuf.
> +			 */
> +			drm_gem_object_get(obj);
> +			return obj;
> +		}
> +	}
> +
> +	return NULL;
> +}
> +EXPORT_SYMBOL(drm_gem_prime_import_self);
> +

I feel this helper is doing too much. Rather separate the test from the 
policy. Do something like

bool drm_gem_prime_exported_dma_buf(dev, dma_buf)
{
     obj = dma_buf->priv;

     return (dma_buf->ops == drm_gem_prime_dmabuf_ops) && (obj->dev == dev)
}

and in the callers do

   if (drm_gem_prime_exported_dma_buf(dev, dma_buf)) {
    /* ... */
     obj = dma_buf->obj;
     get(obj);
     return obj;
   }

>   /**
>    * drm_gem_prime_import_dev - core implementation of the import callback
>    * @dev: drm_device to import into
> @@ -933,17 +965,9 @@ struct drm_gem_object *drm_gem_prime_import_dev(struct drm_device *dev,
>   	struct drm_gem_object *obj;
>   	int ret;
>   
> -	if (dma_buf->ops == &drm_gem_prime_dmabuf_ops) {
> -		obj = dma_buf->priv;
> -		if (obj->dev == dev) {
> -			/*
> -			 * Importing dmabuf exported from our own gem increases
> -			 * refcount on gem itself instead of f_count of dmabuf.
> -			 */
> -			drm_gem_object_get(obj);
> -			return obj;
> -		}
> -	}
> +	obj = drm_gem_prime_import_self(dev, dma_buf);
> +	if (obj)
> +		return obj;
>   
>   	if (!dev->driver->gem_prime_import_sg_table)
>   		return ERR_PTR(-EINVAL);
> diff --git a/include/drm/drm_gem_shmem_helper.h b/include/drm/drm_gem_shmem_helper.h
> index b4f993da3cae..6e8e907bc43e 100644
> --- a/include/drm/drm_gem_shmem_helper.h
> +++ b/include/drm/drm_gem_shmem_helper.h
> @@ -287,6 +287,8 @@ drm_gem_shmem_prime_import_sg_table(struct drm_device *dev,
>   				    struct sg_table *sgt);
>   int drm_gem_shmem_dumb_create(struct drm_file *file, struct drm_device *dev,
>   			      struct drm_mode_create_dumb *args);
> +struct drm_gem_object *drm_gem_shmem_prime_import_no_sgt(struct drm_device *dev,
> +							 struct dma_buf *buf);
>   
>   /**
>    * DRM_GEM_SHMEM_DRIVER_OPS - Default shmem GEM operations
> @@ -298,4 +300,17 @@ int drm_gem_shmem_dumb_create(struct drm_file *file, struct drm_device *dev,
>   	.gem_prime_import_sg_table = drm_gem_shmem_prime_import_sg_table, \
>   	.dumb_create		   = drm_gem_shmem_dumb_create
>   
> +/**
> + * DRM_GEM_SHMEM_DRIVER_OPS_FOR_VMAP - shmem GEM operations
> + * 				       without mapping sg_table on
> + * 				       imported buffer.
> + *
> + * This macro provides a shortcut for setting the shmem GEM operations in
> + * the &drm_driver structure for drivers that do not require a sg_table on
> + * imported buffers. Only vmap is supported on imported buffer.

That final sentence is misleading. It would still be possible to 
pin/unpin and mmap the buffer. Only direct access from DMA-capable 
hardware is not. (Right?)

> + */
> +#define DRM_GEM_SHMEM_DRIVER_OPS_FOR_VMAP \

Rather call this DRM_GEM_SHMEM_DRIVER_OPS_NO_SGT please. Calling it 
_FOR_VMAP seems misleading to me. Best regards Thomas
> +	.gem_prime_import       = drm_gem_shmem_prime_import_no_sgt, \
> +	.dumb_create            = drm_gem_shmem_dumb_create
> +
>   #endif /* __DRM_GEM_SHMEM_HELPER_H__ */
> diff --git a/include/drm/drm_prime.h b/include/drm/drm_prime.h
> index fa085c44d4ca..b5179a9cde93 100644
> --- a/include/drm/drm_prime.h
> +++ b/include/drm/drm_prime.h
> @@ -100,6 +100,9 @@ struct dma_buf *drm_gem_prime_export(struct drm_gem_object *obj,
>   unsigned long drm_prime_get_contiguous_size(struct sg_table *sgt);
>   
>   /* helper functions for importing */
> +struct drm_gem_object *drm_gem_prime_import_self(struct drm_device *dev,
> +						 struct dma_buf *dma_buf);
> +
>   struct drm_gem_object *drm_gem_prime_import_dev(struct drm_device *dev,
>   						struct dma_buf *dma_buf,
>   						struct device *attach_dev);

-- 
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstrasse 146, 90461 Nuernberg, Germany
GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman
HRB 36809 (AG Nuernberg)


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ