Re: [PATCH V4 1/3] block: Allow mapping of vmalloc-ed buffers

[Christoph Hellwig <hch@lst.de>]

> +#ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE

That seems like an odd constructu, as you don't call
flush_kernel_dcache_page.  From looking whoe defines it it seems
to be about the right set of architectures, but that might be
by a mix of chance and similar requirements for cache flushing.

> +static void bio_invalidate_vmalloc_pages(struct bio *bio)
> +{
> +	if (bio->bi_private) {
> +		struct bvec_iter_all iter_all;
> +		struct bio_vec *bvec;
> +		unsigned long len = 0;
> +
> +		bio_for_each_segment_all(bvec, bio, iter_all)
> +			len += bvec->bv_len;
> +             invalidate_kernel_vmap_range(bio->bi_private, len);

We control the bio here, so we can directly iterate over the
segments instead of doing the fairly expensive bio_for_each_segment_all
call that goes to each page and builds a bvec for it.

> +	struct page *page;
>  	int offset, i;
>  	struct bio *bio;
>  
> @@ -1508,6 +1529,12 @@ struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
>  	if (!bio)
>  		return ERR_PTR(-ENOMEM);
>  
> +	if (is_vmalloc) {
> +		flush_kernel_vmap_range(data, len);
> +		if ((!op_is_write(bio_op(bio))))
> +			bio->bi_private = data;
> +	}

We've just allocate the bio, so bio->bi_opf is not actually set at
this point unfortunately.