Re: [PATCH v2 1/2] drm/shmem: add support for per object caching attributes

[Daniel Vetter]

On Wed, Dec 11, 2019 at 11:07:25AM +0100, Thomas Zimmermann wrote:
> 
> 
> Am 11.12.19 um 10:58 schrieb Thomas Zimmermann:
> > Hi Gerd
> > 
> > Am 11.12.19 um 09:18 schrieb Gerd Hoffmann:
> >> Add caching field to drm_gem_shmem_object to specify the cachine
> >> attributes for mappings.  Add helper function to tweak pgprot
> >> accordingly.  Switch vmap and mmap functions to the new helper.
> >>
> >> Set caching to write-combine when creating the object so behavior
> >> doesn't change by default.  Drivers can override that later if
> >> needed.
> >>
> >> Signed-off-by: Gerd Hoffmann <kraxel@xxxxxxxxxx>
> > 
> > If you want to merge this patch, you have my
> > 
> > Reviewed-by: Thomas Zimmermann <tzimmermann@xxxxxxx>
> > 
> > Please see my comment below.
> > 
> >> ---
> >>  include/drm/drm_gem_shmem_helper.h     | 12 ++++++++++++
> >>  drivers/gpu/drm/drm_gem_shmem_helper.c | 24 +++++++++++++++++++++---
> >>  2 files changed, 33 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/include/drm/drm_gem_shmem_helper.h b/include/drm/drm_gem_shmem_helper.h
> >> index 6748379a0b44..9d6e02c6205f 100644
> >> --- a/include/drm/drm_gem_shmem_helper.h
> >> +++ b/include/drm/drm_gem_shmem_helper.h
> >> @@ -17,6 +17,11 @@ struct drm_mode_create_dumb;
> >>  struct drm_printer;
> >>  struct sg_table;
> >>  
> >> +enum drm_gem_shmem_caching {
> >> +	DRM_GEM_SHMEM_CACHED = 1,
> >> +	DRM_GEM_SHMEM_WC,
> >> +};
> >> +
> >>  /**
> >>   * struct drm_gem_shmem_object - GEM object backed by shmem
> >>   */
> >> @@ -83,6 +88,11 @@ struct drm_gem_shmem_object {
> >>  	 * The address are un-mapped when the count reaches zero.
> >>  	 */
> >>  	unsigned int vmap_use_count;
> >> +
> >> +	/**
> >> +	 * @caching: caching attributes for mappings.
> >> +	 */
> >> +	enum drm_gem_shmem_caching caching;
> >>  };
> >>  
> >>  #define to_drm_gem_shmem_obj(obj) \
> >> @@ -130,6 +140,8 @@ drm_gem_shmem_prime_import_sg_table(struct drm_device *dev,
> >>  
> >>  struct sg_table *drm_gem_shmem_get_pages_sgt(struct drm_gem_object *obj);
> >>  
> >> +pgprot_t drm_gem_shmem_caching(struct drm_gem_shmem_object *shmem, pgprot_t prot);
> >> +
> >>  /**
> >>   * DRM_GEM_SHMEM_DRIVER_OPS - Default shmem GEM operations
> >>   *
> >> diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> >> index a421a2eed48a..5bb94e130a50 100644
> >> --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> >> +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> >> @@ -76,6 +76,7 @@ struct drm_gem_shmem_object *drm_gem_shmem_create(struct drm_device *dev, size_t
> >>  	mutex_init(&shmem->pages_lock);
> >>  	mutex_init(&shmem->vmap_lock);
> >>  	INIT_LIST_HEAD(&shmem->madv_list);
> >> +	shmem->caching = DRM_GEM_SHMEM_WC;
> >>  
> >>  	/*
> >>  	 * Our buffers are kept pinned, so allocating them
> >> @@ -256,9 +257,11 @@ static void *drm_gem_shmem_vmap_locked(struct drm_gem_shmem_object *shmem)
> >>  
> >>  	if (obj->import_attach)
> >>  		shmem->vaddr = dma_buf_vmap(obj->import_attach->dmabuf);
> >> -	else
> >> +	else {
> >> +		pgprot_t prot = drm_gem_shmem_caching(shmem, PAGE_KERNEL);
> >>  		shmem->vaddr = vmap(shmem->pages, obj->size >> PAGE_SHIFT,
> >> -				    VM_MAP, pgprot_writecombine(PAGE_KERNEL));
> >> +				    VM_MAP, prot);
> >> +	}
> >>  
> >>  	if (!shmem->vaddr) {
> >>  		DRM_DEBUG_KMS("Failed to vmap pages\n");
> >> @@ -540,7 +543,8 @@ int drm_gem_shmem_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
> >>  	}
> >>  
> >>  	vma->vm_flags |= VM_MIXEDMAP | VM_DONTEXPAND;
> >> -	vma->vm_page_prot = pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
> >> +	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
> >> +	vma->vm_page_prot = drm_gem_shmem_caching(shmem, vma->vm_page_prot);
> >>  	vma->vm_page_prot = pgprot_decrypted(vma->vm_page_prot);
> >>  	vma->vm_ops = &drm_gem_shmem_vm_ops;
> >>  
> >> @@ -683,3 +687,17 @@ drm_gem_shmem_prime_import_sg_table(struct drm_device *dev,
> >>  	return ERR_PTR(ret);
> >>  }
> >>  EXPORT_SYMBOL_GPL(drm_gem_shmem_prime_import_sg_table);
> >> +
> >> +pgprot_t drm_gem_shmem_caching(struct drm_gem_shmem_object *shmem, pgprot_t prot)
> >> +{
> >> +	switch (shmem->caching) {
> >> +	case DRM_GEM_SHMEM_CACHED:
> >> +		return prot;
> >> +	case DRM_GEM_SHMEM_WC:
> >> +		return pgprot_writecombine(prot);
> >> +	default:
> >> +		WARN_ON_ONCE(1);
> >> +		return prot;
> >> +	}
> >> +}
> >> +EXPORT_SYMBOL_GPL(drm_gem_shmem_caching);
> > 
> > Two reason why I'd reconsider this design.
> > 
> > I don't like switch statements new the bottom of the call graph. The
> > code ends up with default warnings, such as this one.
> > 
> > Udl has different caching flags for imported and 'native' buffers. This
> > would require a new constant and additional code here.
> > 
> > What do you think about turning this function into a callback in struct
> > shmem_funcs? The default implementation would be for WC, virtio would
> > use CACHED. The individual implementations could still be located in the
> > shmem code. Udl would later provide its own code.
> 
> On a second thought, all this might be over-engineered and v1 of the
> patchset was the correct approach. You can add my

The udl use-case should be covered already, simply set the flag correctly
at create/import time? It's per-object ...

btw on why udl does this: Imported bo are usually rendered by real hw, and
reading it uncached/wc is the more defensive setting. It would be kinda
nice if dma-buf would expose this, but I fear dma-api maintainers would
murder us if we even just propose that ... so it's a mess right now.

btw the issue extends to dma access by devices too, e.g. both i915 and
amdgpu can select the coherency mode at runtime (using e.g. the pcie
no-snoop transaction mode), and we have similar uncoordinated hacks in
there too, like in udl.
-Daniel

> 
> Acked-by: Thomas Zimmermann <tzimmermann@xxxxxxx>
> 
> if you prefer to merge v1.
> 
> > 
> > Best regards
> > Thomas
> > 
> >>
> > 
> 
> -- 
> Thomas Zimmermann
> Graphics Driver Developer
> SUSE Software Solutions Germany GmbH
> Maxfeldstr. 5, 90409 Nürnberg, Germany
> (HRB 36809, AG Nürnberg)
> Geschäftsführer: Felix Imendörffer
> 




> _______________________________________________
> dri-devel mailing list
> dri-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.freedesktop.org/mailman/listinfo/dri-devel


-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch