[PATCH v3] drm/i915: Implement dma-buf mmap() for userspace access

[PATCH v3] drm/i915: Implement dma-buf mmap() for userspace access

[Ander Conselvan de Oliveira]

From: Rob Bradford <rob@linux.intel.com>

The aim is that this is functionally equivalent to using the
I915_GEM_MMAP ioctl.

v2: Add aperture check limit, limit to SNB+ and improve error handling.

v3 (anderco): Map without using the GTT.

Signed-off-by: Rob Bradford <rob@linux.intel.com>
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
---
 drivers/gpu/drm/i915/i915_gem_dmabuf.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
index 580aa42..2d31fca 100644
--- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c
+++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
@@ -193,7 +193,24 @@ static void i915_gem_dmabuf_kunmap(struct dma_buf *dma_buf, unsigned long page_n
 
 static int i915_gem_dmabuf_mmap(struct dma_buf *dma_buf, struct vm_area_struct *vma)
 {
-	return -EINVAL;
+	struct drm_i915_gem_object *obj = dma_buf_to_obj(dma_buf);
+	struct drm_device *dev = obj->base.dev;
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	int ret;
+
+	if (INTEL_INFO(dev)->gen < 6)
+		return -EINVAL;
+
+	ret = i915_mutex_lock_interruptible(dev);
+	if (ret)
+		return ret;
+
+	ret = obj->base.filp->f_op->mmap(obj->base.filp, vma);
+	fput(vma->vm_file);
+	vma->vm_file = get_file(obj->base.filp);
+
+	mutex_unlock(&dev->struct_mutex);
+	return ret;
 }
 
 static int i915_gem_begin_cpu_access(struct dma_buf *dma_buf, size_t start, size_t length, enum dma_data_direction direction)
-- 
1.8.3.2

Re: [PATCH v3] drm/i915: Implement dma-buf mmap() for userspace access

[Daniel Vetter]

On Thu, Jun 19, 2014 at 03:34:34PM +0300, Ander Conselvan de Oliveira wrote:
> From: Rob Bradford <rob@linux.intel.com>
> 
> The aim is that this is functionally equivalent to using the
> I915_GEM_MMAP ioctl.
> 
> v2: Add aperture check limit, limit to SNB+ and improve error handling.
> 
> v3 (anderco): Map without using the GTT.

Why that? This means you'll have a non-coherent mapping and atm dma-buf
doesn't expose and invalidate/flushing ioctls. Which really you want for
this.
-Daniel

> 
> Signed-off-by: Rob Bradford <rob@linux.intel.com>
> Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
> ---
>  drivers/gpu/drm/i915/i915_gem_dmabuf.c | 19 ++++++++++++++++++-
>  1 file changed, 18 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
> index 580aa42..2d31fca 100644
> --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c
> +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
> @@ -193,7 +193,24 @@ static void i915_gem_dmabuf_kunmap(struct dma_buf *dma_buf, unsigned long page_n
>  
>  static int i915_gem_dmabuf_mmap(struct dma_buf *dma_buf, struct vm_area_struct *vma)
>  {
> -	return -EINVAL;
> +	struct drm_i915_gem_object *obj = dma_buf_to_obj(dma_buf);
> +	struct drm_device *dev = obj->base.dev;
> +	struct drm_i915_private *dev_priv = dev->dev_private;
> +	int ret;
> +
> +	if (INTEL_INFO(dev)->gen < 6)
> +		return -EINVAL;
> +
> +	ret = i915_mutex_lock_interruptible(dev);
> +	if (ret)
> +		return ret;
> +
> +	ret = obj->base.filp->f_op->mmap(obj->base.filp, vma);
> +	fput(vma->vm_file);
> +	vma->vm_file = get_file(obj->base.filp);
> +
> +	mutex_unlock(&dev->struct_mutex);
> +	return ret;
>  }
>  
>  static int i915_gem_begin_cpu_access(struct dma_buf *dma_buf, size_t start, size_t length, enum dma_data_direction direction)
> -- 
> 1.8.3.2
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch

Re: [PATCH v3] drm/i915: Implement dma-buf mmap() for userspace access

[Ander Conselvan de Oliveira]

On 06/19/2014 03:37 PM, Daniel Vetter wrote:
> On Thu, Jun 19, 2014 at 03:34:34PM +0300, Ander Conselvan de Oliveira wrote:
>> From: Rob Bradford <rob@linux.intel.com>
>>
>> The aim is that this is functionally equivalent to using the
>> I915_GEM_MMAP ioctl.
>>
>> v2: Add aperture check limit, limit to SNB+ and improve error handling.
>>
>> v3 (anderco): Map without using the GTT.
>
> Why that? This means you'll have a non-coherent mapping and atm dma-buf
> doesn't expose and invalidate/flushing ioctls. Which really you want for
> this.

Hmm. I completely overlooked that important detail. In a discussion with 
Rob, we came to the conclusion we didn't want to use the GTT because of 
the performance penalty. Now I see that just modifying the mmap() 
support is not enough.

Thanks,
Ander

> -Daniel
>
>>
>> Signed-off-by: Rob Bradford <rob@linux.intel.com>
>> Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
>> ---
>>   drivers/gpu/drm/i915/i915_gem_dmabuf.c | 19 ++++++++++++++++++-
>>   1 file changed, 18 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
>> index 580aa42..2d31fca 100644
>> --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c
>> +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
>> @@ -193,7 +193,24 @@ static void i915_gem_dmabuf_kunmap(struct dma_buf *dma_buf, unsigned long page_n
>>
>>   static int i915_gem_dmabuf_mmap(struct dma_buf *dma_buf, struct vm_area_struct *vma)
>>   {
>> -	return -EINVAL;
>> +	struct drm_i915_gem_object *obj = dma_buf_to_obj(dma_buf);
>> +	struct drm_device *dev = obj->base.dev;
>> +	struct drm_i915_private *dev_priv = dev->dev_private;
>> +	int ret;
>> +
>> +	if (INTEL_INFO(dev)->gen < 6)
>> +		return -EINVAL;
>> +
>> +	ret = i915_mutex_lock_interruptible(dev);
>> +	if (ret)
>> +		return ret;
>> +
>> +	ret = obj->base.filp->f_op->mmap(obj->base.filp, vma);
>> +	fput(vma->vm_file);
>> +	vma->vm_file = get_file(obj->base.filp);
>> +
>> +	mutex_unlock(&dev->struct_mutex);
>> +	return ret;
>>   }
>>
>>   static int i915_gem_begin_cpu_access(struct dma_buf *dma_buf, size_t start, size_t length, enum dma_data_direction direction)
>> --
>> 1.8.3.2
>>
>> _______________________________________________
>> Intel-gfx mailing list
>> Intel-gfx@lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/intel-gfx
>

Re: [PATCH v3] drm/i915: Implement dma-buf mmap() for userspace access

[Chris Wilson]

On Thu, Jun 19, 2014 at 03:34:34PM +0300, Ander Conselvan de Oliveira wrote:
> From: Rob Bradford <rob@linux.intel.com>
> 
> The aim is that this is functionally equivalent to using the
> I915_GEM_MMAP ioctl.
> 
> v2: Add aperture check limit, limit to SNB+ and improve error handling.
> 
> v3 (anderco): Map without using the GTT.
> 
> Signed-off-by: Rob Bradford <rob@linux.intel.com>
> Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
> ---
>  drivers/gpu/drm/i915/i915_gem_dmabuf.c | 19 ++++++++++++++++++-
>  1 file changed, 18 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
> index 580aa42..2d31fca 100644
> --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c
> +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c
> @@ -193,7 +193,24 @@ static void i915_gem_dmabuf_kunmap(struct dma_buf *dma_buf, unsigned long page_n
>  
>  static int i915_gem_dmabuf_mmap(struct dma_buf *dma_buf, struct vm_area_struct *vma)
>  {
> -	return -EINVAL;
> +	struct drm_i915_gem_object *obj = dma_buf_to_obj(dma_buf);
> +	struct drm_device *dev = obj->base.dev;
> +	struct drm_i915_private *dev_priv = dev->dev_private;
> +	int ret;
> +
> +	if (INTEL_INFO(dev)->gen < 6)
> +		return -EINVAL;

For cpu mmapings, the check can die.
> +
> +	ret = i915_mutex_lock_interruptible(dev);
> +	if (ret)
> +		return ret;

This lock doesn't actually do anything useful.

obj->base.filp can be NULL.

> +
> +	ret = obj->base.filp->f_op->mmap(obj->base.filp, vma);
> +	fput(vma->vm_file);
> +	vma->vm_file = get_file(obj->base.filp);

Talk to me about how the user is expected to control coherency since
that appears to be outside of the scope of this function.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre