[bug report] drm/ttm: add transparent huge page support for DMA allocations v2
Koenig, Christian
Christian.Koenig at amd.com
Thu Jun 27 16:19:14 UTC 2019
Am 27.06.19 um 16:06 schrieb Dan Carpenter:
> Hello Christian König,
>
> The patch 648bc3574716: "drm/ttm: add transparent huge page support
> for DMA allocations v2" from Jul 6, 2017, leads to the following
> static checker warning:
>
> drivers/gpu/drm/ttm/ttm_page_alloc_dma.c:314 __ttm_dma_alloc_page()
> error: 'vaddr' came from dma_alloc_coherent() so we can't do virt_to_phys()
>
> drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
> 295 static struct dma_page *__ttm_dma_alloc_page(struct dma_pool *pool)
> 296 {
> 297 struct dma_page *d_page;
> 298 unsigned long attrs = 0;
> 299 void *vaddr;
> 300
> 301 d_page = kmalloc(sizeof(struct dma_page), GFP_KERNEL);
> 302 if (!d_page)
> 303 return NULL;
> 304
> 305 if (pool->type & IS_HUGE)
> 306 attrs = DMA_ATTR_NO_WARN;
> 307
> 308 vaddr = dma_alloc_attrs(pool->dev, pool->size, &d_page->dma,
> 309 pool->gfp_flags, attrs);
> 310 if (vaddr) {
> 311 if (is_vmalloc_addr(vaddr))
> 312 d_page->p = vmalloc_to_page(vaddr);
> 313 else
> 314 d_page->p = virt_to_page(vaddr);
>
> Christoph was explaining this earlier and I don't think it's sufficient
> to just check is_vmalloc_addr(). See https://lkml.org/lkml/2019/6/17/155
> Apparently the vaddr might not have a page backing...
That stuff was copied from the older non-huge page code and I actually
think it is correct.
If not then the real question is how to correctly get the backing page
from dma_alloc_attrs().
Christian.
>
> (I am a newbie to this so I may be wrong or have misunderstood also).
>
> 315 d_page->vaddr = (unsigned long)vaddr;
> 316 if (pool->type & IS_HUGE)
> 317 d_page->vaddr |= VADDR_FLAG_HUGE_POOL;
> 318 } else {
> 319 kfree(d_page);
> 320 d_page = NULL;
> 321 }
> 322 return d_page;
> 323 }
>
> regards,
> dan carpenter
More information about the dri-devel
mailing list