Re: [PATCH] scatterlist: Allocate a contiguous array instead of chaining

[Sultan Alsawaf <sultan@kerneltoast.com>]

On Fri, Jul 12, 2019 at 09:06:40AM +0200, Thomas Gleixner wrote:
> On Fri, 12 Jul 2019, Ming Lei wrote:
> > vmalloc() may sleep, so it is impossible to be called in atomic context.
> 
> Allocations from atomic context should be avoided wherever possible and you
> really have to have a very convincing argument why an atomic allocation is
> absolutely necessary. I cleaned up quite some GFP_ATOMIC users over the
> last couple of years and all of them were doing it for the very wrong
> reasons and mostly just to silence the warning which is triggered with
> GFP_KERNEL when called from a non-sleepable context.
> 
> So I suggest to audit all call sites first and figure out whether they
> really must use GFP_ATOMIC and if possible clean them up, remove the GFP
> argument and then do the vmalloc thing on top.

Hello Thomas and Ming,

It looks like the following call sites are atomic:
drivers/crypto/qce/ablkcipher.c:92:	ret = sg_alloc_table(&rctx->dst_tbl, rctx->dst_nents, gfp);
drivers/crypto/ccp/ccp-crypto-aes-cmac.c:110:	ret = sg_alloc_table(&rctx->data_sg, sg_count, gfp);
drivers/crypto/ccp/ccp-crypto-sha.c:103:		ret = sg_alloc_table(&rctx->data_sg, sg_count, gfp);
drivers/spi/spi-pl022.c:1035:	ret = sg_alloc_table(&pl022->sgt_rx, pages, GFP_ATOMIC);
drivers/spi/spi-pl022.c:1039:	ret = sg_alloc_table(&pl022->sgt_tx, pages, GFP_ATOMIC);

The crypto ones are conditionally made atomic depending on the presence of
CRYPTO_TFM_REQ_MAY_SLEEP.

Additionally, the following allocation could be problematic with kvmalloc:
net/ceph/crypto.c:180:		ret = sg_alloc_table(sgt, chunk_cnt, GFP_NOFS);

This is a snippet from kvmalloc:
	/*
	 * vmalloc uses GFP_KERNEL for some internal allocations (e.g page tables)
	 * so the given set of flags has to be compatible.
	 */
	if ((flags & GFP_KERNEL) != GFP_KERNEL)
		return kmalloc_node(size, flags, node);

Use of GFP_NOFS in net/ceph/crypto.c would cause kvmalloc to fall back to
kmalloc_node, which could cause problems if the allocation size is too large for
kmalloc_node to reasonably accomodate.

Also, it looks like the vmalloc family doesn't have kvmalloc's GFP_KERNEL check.
Is this intentional, or does vmalloc really not require GFP_KERNEL context?

Thanks,
Sultan