On Wed, Dec 18, 2024 at 05:24:32PM +0800, Ferry Meng wrote:
> This patchset implements io_uring passthrough surppot in virtio-blk
> driver, bypass vfs and part of block layer logic, resulting in lower
> submit latency and increased flexibility when utilizing virtio-blk.
>
> In this version, currently only supports READ/WRITE vec/no-vec operations,
> others like discard or zoned ops not considered in. So the userspace-related
If WRITE is supported then FLUSH is also required so that written data
can be persisted without falling back to another API.
> struct is not complicated.
>
> struct virtblk_uring_cmd {
> __u32 type;
> __u32 ioprio;
> __u64 sector;
> /* above is related to out_hdr */
> __u64 data; // user buffer addr or iovec base addr.
> __u32 data_len; // user buffer length or iovec count.
> __u32 flag; // only contains whether a vector rw or not.
> };
>
> To test this patch series, I changed fio's code:
> 1. Added virtio-blk support to engines/io_uring.c.
> 2. Added virtio-blk support to the t/io_uring.c testing tool.
> Link: https://github.com/jdmfr/fio
>
>
> ===========
> Performance
> ===========
>
> Using t/io_uring-vblk, the performance of virtio-blk based on uring-cmd
> scales better than block device access. (such as below, Virtio-Blk with QEMU,
> 1-depth fio)
> (passthru) read: IOPS=17.2k, BW=67.4MiB/s (70.6MB/s)
> slat (nsec): min=2907, max=43592, avg=3981.87, stdev=595.10
> clat (usec): min=38, max=285,avg=53.47, stdev= 8.28
> lat (usec): min=44, max=288, avg=57.45, stdev= 8.28
> (block) read: IOPS=15.3k, BW=59.8MiB/s (62.7MB/s)
> slat (nsec): min=3408, max=35366, avg=5102.17, stdev=790.79
> clat (usec): min=35, max=343, avg=59.63, stdev=10.26
> lat (usec): min=43, max=349, avg=64.73, stdev=10.21
>
> Testing the virtio-blk device with fio using 'engines=io_uring_cmd'
> and 'engines=io_uring' also demonstrates improvements in submit latency.
> (passthru) taskset -c 0 t/io_uring-vblk -b4096 -d8 -c4 -s4 -p0 -F1 -B0 -O0 -n1 -u1 /dev/vdcc0
> IOPS=189.80K, BW=741MiB/s, IOS/call=4/3
> IOPS=187.68K, BW=733MiB/s, IOS/call=4/3
> (block) taskset -c 0 t/io_uring-vblk -b4096 -d8 -c4 -s4 -p0 -F1 -B0 -O0 -n1 -u0 /dev/vdc
> IOPS=101.51K, BW=396MiB/s, IOS/call=4/3
> IOPS=100.01K, BW=390MiB/s, IOS/call=4/4
This iodepth=8 (submission/completion batching 4) result surprised me
because the io_uring calls are already batched but there is still a 4
microsecond improvement per request.
I was expecting to see less improvement when iodepth is increased
because the syscall, io_uring, and some block layer cost is amortized
thanks to batching and block plugging.
Is the virtio-blk driver submitting 4 requests at a time for both
passthru and block? I wonder if something else is going on here.
>
> =======
> Changes
> =======
>
> Changes in v1:
> --------------
> * remove virtblk_is_write() helper
> * fix rq_flags type definition (blk_opf_t), add REQ_ALLOC_CACHE flag.
> https://lore.kernel.org/io-uring/202412042324.uKQ5KdkE-lkp@xxxxxxxxx/
>
> RFC discussion:
> ---------------
> https://lore.kernel.org/io-uring/20241203121424.19887-1-mengferry@xxxxxxxxxxxxxxxxx/
>
> Ferry Meng (3):
> virtio-blk: add virtio-blk chardev support.
> virtio-blk: add uring_cmd support for I/O passthru on chardev.
> virtio-blk: add uring_cmd iopoll support.
>
> drivers/block/virtio_blk.c | 320 +++++++++++++++++++++++++++++++-
> include/uapi/linux/virtio_blk.h | 16 ++
> 2 files changed, 331 insertions(+), 5 deletions(-)
>
> --
> 2.43.5
>
Attachment:
signature.asc
Description: PGP signature
