[PATCH 1/6] block: add support for REQ_OP_VERIFY

[Darrick J. Wong]

On Tue, Jul 05, 2022 at 04:50:33PM +0000, Chaitanya Kulkarni wrote:
> Darrik,
> 
> Thanks for the reply.
> 
> >> +
> >> +/**
> >> + * __blkdev_issue_verify - generate number of verify operations
> >> + * @bdev:	blockdev to issue
> >> + * @sector:	start sector
> >> + * @nr_sects:	number of sectors to verify
> >> + * @gfp_mask:	memory allocation flags (for bio_alloc())
> >> + * @biop:	pointer to anchor bio
> >> + *
> >> + * Description:
> >> + *  Verify a block range using hardware offload.
> >> + *
> >> + * The function will emulate verify operation if no explicit hardware
> >> + * offloading for verifying is provided.
> >> + */
> >> +int __blkdev_issue_verify(struct block_device *bdev, sector_t sector,
> >> +		sector_t nr_sects, gfp_t gfp_mask, struct bio **biop)
> >> +{
> >> +	unsigned int max_verify_sectors = bdev_verify_sectors(bdev);
> >> +	sector_t min_io_sect = (BIO_MAX_VECS << PAGE_SHIFT) >> 9;
> >> +	struct bio *bio = *biop;
> >> +	sector_t curr_sects;
> >> +	char *buf;
> >> +
> >> +	if (!max_verify_sectors) {
> >> +		int ret = 0;
> >> +
> >> +		buf = kzalloc(min_io_sect << 9, GFP_KERNEL);
> > 
> > k*z*alloc?  I don't think you need to zero a buffer that we're reading
> > into, right?
> > 
> > --D
> 
> we don't need to but I guess it is just a habit to make sure alloced
> buffer is zeored, should I remove it for any particular reason ?

What's the point in wasting CPU time zeroing a buffer if you're just
going to DMA into it?

--D

> -ck
> 
>