Re: [PATCH v2] nvmet-tcp: Don't map pages which can't come from HIGHMEM

[Al Viro]

On Wed, Aug 31, 2022 at 12:05:33AM +0200, Fabio M. De Francesco wrote:
> kmap() is being deprecated in favor of kmap_local_page().[1]
> 
> There are two main problems with kmap(): (1) It comes with an overhead as
> mapping space is restricted and protected by a global lock for
> synchronization and (2) it also requires global TLB invalidation when the
> kmap’s pool wraps and it might block when the mapping space is fully
> utilized until a slot becomes available.
> 
> The pages which will be mapped are allocated in nvmet_tcp_map_data(),
> using the GFP_KERNEL flag. This assures that they cannot come from
> HIGHMEM. This imply that a straight page_address() can replace the kmap()
> of sg_page(sg) in nvmet_tcp_map_pdu_iovec(). As a side effect, we might
> also delete the field "nr_mapped" from struct "nvmet_tcp_cmd" because,
> after removing the kmap() calls, there would be no longer any need of it.
> 
> In addition, there is no reason to use a kvec for the command receive
> data buffers iovec, use a bio_vec instead and let iov_iter handle the
> buffer mapping and data copy.
> 
> Test with blktests on a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with
> HIGHMEM64GB enabled.

Looks sane...