| Message ID | Y66T1UCBTqBE6GQ4@gondor.apana.org.au |
|---|---|
| State | Accepted |
| Commit | 8e613cec25196b51601dfac50c5bf229acd72bc6 |
| Headers | show |
| Series | crypto: talitos - Remove GFP_DMA and add DMA alignment padding | expand |
Looks good:
Reviewed-by: Christoph Hellwig <hch@lst.de>
Le 30/12/2022 à 08:31, Herbert Xu a écrit : > GFP_DMA does not guarantee that the returned memory is aligned > for DMA. It should be removed where it is superfluous. Doesn't GFP_DMA guarantees that the provided memory is addressable for DMA ? Or do we assume that all memory returned by kmalloc can be used for DMA ? > > However, kmalloc may start returning DMA-unaligned memory in future > so fix this by adding the alignment by hand. kmalloc() already returns not DMA aligned memory, why does it becomes a problem now ? Ok, that may be suboptimal, but is that a problem at all ? By the way, I'm not sure I understand the solution, what's the added value of aligning allocation length to the cache alignment ? > > Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au> > > diff --git a/drivers/crypto/talitos.c b/drivers/crypto/talitos.c > index 71db6450b6aa..d62ec68e3183 100644 > --- a/drivers/crypto/talitos.c > +++ b/drivers/crypto/talitos.c > @@ -1393,7 +1393,7 @@ static struct talitos_edesc *talitos_edesc_alloc(struct device *dev, > alloc_len += sizeof(struct talitos_desc); > alloc_len += ivsize; > > - edesc = kmalloc(alloc_len, GFP_DMA | flags); > + edesc = kmalloc(ALIGN(alloc_len, dma_get_cache_alignment()), flags); > if (!edesc) > return ERR_PTR(-ENOMEM); > if (ivsize) {
On Mon, Jan 09, 2023 at 07:18:27AM +0000, Christophe Leroy wrote: > > Doesn't GFP_DMA guarantees that the provided memory is addressable for > DMA ? Or do we assume that all memory returned by kmalloc can be used > for DMA ? No it does not. Please refer to the DMI API documentation. > kmalloc() already returns not DMA aligned memory, why does it becomes a > problem now ? As it stands kmalloc on arm64 returns DMA-aligned memory. This will soon no longer be the case. Cheers,
diff --git a/drivers/crypto/talitos.c b/drivers/crypto/talitos.c index 71db6450b6aa..d62ec68e3183 100644 --- a/drivers/crypto/talitos.c +++ b/drivers/crypto/talitos.c @@ -1393,7 +1393,7 @@ static struct talitos_edesc *talitos_edesc_alloc(struct device *dev, alloc_len += sizeof(struct talitos_desc); alloc_len += ivsize; - edesc = kmalloc(alloc_len, GFP_DMA | flags); + edesc = kmalloc(ALIGN(alloc_len, dma_get_cache_alignment()), flags); if (!edesc) return ERR_PTR(-ENOMEM); if (ivsize) {
GFP_DMA does not guarantee that the returned memory is aligned for DMA. It should be removed where it is superfluous. However, kmalloc may start returning DMA-unaligned memory in future so fix this by adding the alignment by hand. Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>