Re: [PATCH v2] async_pq: Remove VLA usage

[Dan Williams]

[ adding linux-raid ]

Apologies for the delay, the recent ping from Kees made me take a closer look.

On Sat, May 5, 2018 at 12:58 AM, Kyle Spiers <ksspiers@xxxxxxxxxx> wrote:
> In the quest to remove VLAs from the kernel[1], this moves the
> allocation of coefs and blocks from the stack to being kmalloc()ed.
>
> [1] https://lkml.org/lkml/2018/3/7/621
>
> Signed-off-by: Kyle Spiers <ksspiers@xxxxxxxxxx>
> ---
> Forgot to add slab.h
> ---
>  crypto/async_tx/async_pq.c  | 18 ++++++++++++++----
>  crypto/async_tx/raid6test.c |  9 ++++++++-
>  2 files changed, 22 insertions(+), 5 deletions(-)
>
> diff --git a/crypto/async_tx/async_pq.c b/crypto/async_tx/async_pq.c
> index 56bd612927ab..af1912313a23 100644
> --- a/crypto/async_tx/async_pq.c
> +++ b/crypto/async_tx/async_pq.c
> @@ -194,9 +194,9 @@ async_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
>             (src_cnt <= dma_maxpq(device, 0) ||
>              dma_maxpq(device, DMA_PREP_CONTINUE) > 0) &&
>             is_dma_pq_aligned(device, offset, 0, len)) {
> -               struct dma_async_tx_descriptor *tx;
> +               struct dma_async_tx_descriptor *tx = NULL;
>                 enum dma_ctrl_flags dma_flags = 0;
> -               unsigned char coefs[src_cnt];
> +               unsigned char *coefs;
>                 int i, j;
>
>                 /* run the p+q asynchronously */
> @@ -207,6 +207,9 @@ async_gen_syndrome(struct page **blocks, unsigned int offset, int disks,
>                  * sources and update the coefficients accordingly
>                  */
>                 unmap->len = len;
> +               coefs = kmalloc_array(src_cnt, sizeof(*coefs), GFP_KERNEL);

At a minimum this needs to be GFP_NOIO since raid may be in the
page-reclaim path.

However, I think it may be better to allocate this with the rest of
the stripe resources and pass it in as a scratch buffer. See the usage
/ implementation of scribble_alloc() in drivers/md/raid5.c.