On Mon, Feb 20, 2023 at 08:59:32AM +0000, Krasnov Arseniy wrote:
On 16.02.2023 16:33, Stefano Garzarella wrote:
Hi Arseniy,
sorry for the delay, but I was offline.
Hello! Sure no problem!, i was also offline a little bit so i'm replying
just now
On Mon, Feb 06, 2023 at 06:51:55AM +0000, Arseniy Krasnov wrote:
Hello,
DESCRIPTION
this is MSG_ZEROCOPY feature support for virtio/vsock. I tried to follow
current implementation for TCP as much as possible:
1) Sender must enable SO_ZEROCOPY flag to use this feature. Without this
flag, data will be sent in "classic" copy manner and MSG_ZEROCOPY
flag will be ignored (e.g. without completion).
2) Kernel uses completions from socket's error queue. Single completion
for single tx syscall (or it can merge several completions to single
one). I used already implemented logic for MSG_ZEROCOPY support:
'msg_zerocopy_realloc()' etc.
I will review for the vsock point of view. Hope some net maintainers can
comment about SO_ZEROCOPY.
Anyway I think is a good idea to keep it as close as possible to the TCP
implementation.
Difference with copy way is not significant. During packet allocation,
non-linear skb is created, then I call 'get_user_pages()' for each page
from user's iov iterator (I think i don't need 'pin_user_pages()' as
Are these pages exposed to the host via virtqueues? If so, I think we
should pin them. What happens if the host accesses them but these pages
have been unmapped?
Yes, user pages with data will be used by the virtio device.
'pin' - You mean use 'pin_user_pages()'? Unmapped - You mean guest
will unmap it, while host must access it to copy packet? Such pages
have incremented refcount by 'get_user_pages()', so page is locked
and memory and will be locked until host finishes copying data.
Yep, I got it while reviewing patch 7 ;-)
I think it is better to discuss things related to 'get/pin_user_pages()'
in one place, for example in 07/12 where this function is called.
Agree!
there is no backing storage for these pages) and add each returned page
to the skb as fragment. There are also some updates for vhost and guest
parts of transport - in both cases i've added handling of non-linear skb
for virtio part. vhost copies data from such skb to the guest's rx virtio
buffers. In the guest, virtio transport fills virtio queue with pages
from skb.
I think doc in Documentation/networking/msg_zerocopy.rst could be also
updated in next versions.
Yep, good idea.
Ack, i'll do it in v2.
This version has several limits/problems:
1) As this feature totally depends on transport, there is no way (or it
is difficult) to check whether transport is able to handle it or not
during SO_ZEROCOPY setting. Seems I need to call AF_VSOCK specific
setsockopt callback from setsockopt callback for SOL_SOCKET, but this
leads to lock problem, because both AF_VSOCK and SOL_SOCKET callback
are not considered to be called from each other. So in current version
SO_ZEROCOPY is set successfully to any type (e.g. transport) of
AF_VSOCK socket, but if transport does not support MSG_ZEROCOPY,
tx routine will fail with EOPNOTSUPP.
I'll take a look, but if we have no alternative, I think it's okay to
make tx fail.>
Thanks
2) When MSG_ZEROCOPY is used, for each tx system call we need to enqueue
one completion. In each completion there is flag which shows how tx
was performed: zerocopy or copy. This leads that whole message must
be send in zerocopy or copy way - we can't send part of message with
copying and rest of message with zerocopy mode (or vice versa). Now,
we need to account vsock credit logic, e.g. we can't send whole data
once - only allowed number of bytes could sent at any moment. In case
of copying way there is no problem as in worst case we can send single
bytes, but zerocopy is more complex because smallest transmission
unit is single page. So if there is not enough space at peer's side
to send integer number of pages (at least one) - we will wait, thus
stalling tx side. To overcome this problem i've added simple rule -
zerocopy is possible only when there is enough space at another side
for whole message (to check, that current 'msghdr' was already used
in previous tx iterations i use 'iov_offset' field of it's iov iter).
I see the problem and I think your approach is the right one.
3) loopback transport is not supported, because it requires to implement
non-linear skb handling in dequeue logic (as we "send" fragged skb
and "receive" it from the same queue). I'm going to implement it in
next versions.
loopback is useful for testing and debugging, so it would be great to
have the support, but if it's too complicated, we can do it later.
Ok, i'll implement it in v2.
4) Current implementation sets max length of packet to 64KB. IIUC this
is due to 'kmalloc()' allocated data buffers. I think, in case of
Yep, I think so.
When I started touching this code, the limit was already there.
As you said, I think it was introduced to have a limit on (host/device
side?) allocation, but buf_alloc might be enough, so maybe we could
also remove it for copy mode.
The only problem I see is compatibility with old devices/drivers, so
maybe we need a feature in the spec to say the limit is gone, or have a
field in the virtio config space where the device specifies its limit
(for the driver, the limit is implicitly that of the buffer allocated
and put in the virtqueue).
This reminded me that Laura had proposed something similar before,
maybe we should take it up again:
https://markmail.org/message/3el4ckeakfilg5wo
MSG_ZEROCOPY this value could be increased, because 'kmalloc()' is
not touched for data - user space pages are used as buffers. Also
this limit trims every message which is > 64KB, thus such messages
will be send in copy mode due to 'iov_offset' check in 2).
The host still needs to allocate and copy, so maybe the limitation
could be to avoid large allocations in the host, but actually the host
can use vmalloc because it doesn't need them to be contiguous.
Hmmm, I think it is possible to solve this situation in the following
way - i can keep limitation for 64KB for copy mode, and remove it for
zero copy, but I'll limit each packet size to 64KB(of course technically
it is not exactly 64KB, it is min(max packet size, MAX_SKB_FRAGS * PAGE_SIZE)
where max packet size is 64Kb, but for simplicity let's call this limit 64Kb:) ).
E.g. when zerocopy transmission needs to send for example 129Kb(of course
peer's free space is big enough and this check is passed), I'll won't trim
129Kb to 64Kb + 64Kb + 1Kb in the current manner - by returning to af_vsock.c
after sending every skb. I'll alloc several skbs, (3 in this case - 64Kb + 64Kb + 1Kb)
in single call to the transport. Completion arg will be attached
only to the last one skb, and send these 3 skbs. Host still processes
64Kb(let it be 64Kb for simplicity again :) ) packets - no big allocations.
Make sense to me!
Moreover, i think that this logic could be a little optimization for
copy mode - why we allocate single skb and always return to af_vsock.c?
@Bobby, can you help us here?
May be we can iterate needed number of skbs in the loop and send them.
Yep, I think is doable.
Also about vmalloc(), IIUC there is already this idea which replaces 'kmalloc()'
to 'kvmalloc()'.
Yep, I think it is already merged:
0e3f72931fc4 ("vhost/vsock: Use kvmalloc/kvfree for larger packets.")
But this is in the vhost transport (device emulation running in the
host), where we don't need that the pages are pinned.
PERFORMANCE
Performance: it is a little bit tricky to compare performance between
copy and zerocopy transmissions. In zerocopy way we need to wait when
user buffers will be released by kernel, so it something like synchronous
path (wait until device driver will process it), while in copy way we
can feed data to kernel as many as we want, don't care about device
driver. So I compared only time which we spend in 'sendmsg()' syscall.
Also there is limit from 4) above so max buffer size is 64KB. I've
tested this patchset in the nested VM, but i think for V1 it is not a
big deal.
Sender:
./vsock_perf --sender <CID> --buf-size <buf size> --bytes 60M [--zc]
Receiver:
./vsock_perf --vsk-size 256M
Number in cell is seconds which senders spends inside tx syscall.
Guest to host transmission:
*-------------------------------*
| | | |
| buf size | copy | zerocopy |
| | | |
*-------------------------------*
| 4KB | 0.26 | 0.042 |
*-------------------------------*
| 16KB | 0.11 | 0.014 |
*-------------------------------*
| 32KB | 0.05 | 0.009 |
*-------------------------------*
| 64KB | 0.04 | 0.005 |
*-------------------------------*
Host to guest transmission:
*--------------------------------*
| | | |
| buf size | copy | zerocopy |
| | | |
*--------------------------------*
| 4KB | 0.049 | 0.034 |
*--------------------------------*
| 16KB | 0.03 | 0.024 |
*--------------------------------*
| 32KB | 0.025 | 0.01 |
*--------------------------------*
| 64KB | 0.028 | 0.01 |
*--------------------------------*
If host fails to send data with "Cannot allocate memory", check value
/proc/sys/net/core/optmem_max - it is accounted during completion skb
allocation.
Zerocopy is faster than classic copy mode, but of course it requires
specific architecture of application due to user pages pinning, buffer
size and alignment. In next versions i'm going to fix 64KB barrier to
perform tests with bigger buffer sizes.
Yep, I see.
Adjusting vsock_perf to compare also Gbps (can io_uring helps in this
case to have more requests in-flight?) would be great.
Yes, i'll add Gbps. Also I thought about adding io_uring test to
the current test suite because io_uring also have MSG_ZEROCOPY
type of request, and interesting thing is that io_uring uses it's
own already allocated uarg.
Cool!
TESTING
This patchset includes set of tests for MSG_ZEROCOPY feature. I tried to
cover new code as much as possible so there are different cases for
MSG_ZEROCOPY transmissions: with disabled SO_ZEROCOPY and several io
vector types (different sizes, alignments, with unmapped pages).
Great! Thanks for adding the tests!
I'll go through the patches between today and Monday.
Sorry again for taking so long!
Sure, Thanks for review! I think we are not hurry :)
Yep, thank you for this work!
Stefano
