selftest io_uring_zerocopy_tx.sh failed on VM

selftest io_uring_zerocopy_tx.sh failed on VM

[Hangbin Liu]

Hi Pavel,

I tried to run test selftest io_uring_zerocopy_tx.sh on VM, but it failed
with error like

+ ip netns exec ns-45iLeE2 ./msg_zerocopy -4 -t 2 -C 2 -S 192.168.1.1 -D 192.168.1.2 -r udp
cpu: unable to pin, may increase variance.
+ ip netns exec ns-45iLeE1 ./io_uring_zerocopy_tx -4 -t 1 -D 192.168.1.2 -m 1 -t 1 -n 32 udp
./io_uring_zerocopy_tx: io_uring: queue init: Unknown error -13

Do you know what's the reason? Should we update the test script to return
SKIP if io_uring init failed?

Thanks
Hangbin

Re: selftest io_uring_zerocopy_tx.sh failed on VM

[Pavel Begunkov]

On 7/17/23 05:11, Hangbin Liu wrote:
> Hi Pavel,
> 
> I tried to run test selftest io_uring_zerocopy_tx.sh on VM, but it failed
> with error like
> 
> + ip netns exec ns-45iLeE2 ./msg_zerocopy -4 -t 2 -C 2 -S 192.168.1.1 -D 192.168.1.2 -r udp
> cpu: unable to pin, may increase variance.
> + ip netns exec ns-45iLeE1 ./io_uring_zerocopy_tx -4 -t 1 -D 192.168.1.2 -m 1 -t 1 -n 32 udp
> ./io_uring_zerocopy_tx: io_uring: queue init: Unknown error -13
> 
> Do you know what's the reason? Should we update the test script to return
> SKIP if io_uring init failed?

I don't recall anything that can fail ring init with EACCES, probably
sth in your system disables io_uring with some syscall filters or so.
I think skipping the trace on EACCES is the right approach. Do you want
to send a patch?
Apart from that, if you're curious you can try to trace what's going on,
ftrace should give an idea.

-- 
Pavel Begunkov

Re: selftest io_uring_zerocopy_tx.sh failed on VM

[Hangbin Liu]

On Mon, Jul 17, 2023 at 01:34:04PM +0100, Pavel Begunkov wrote:
> On 7/17/23 05:11, Hangbin Liu wrote:
> > Hi Pavel,
> > 
> > I tried to run test selftest io_uring_zerocopy_tx.sh on VM, but it failed
> > with error like
> > 
> > + ip netns exec ns-45iLeE2 ./msg_zerocopy -4 -t 2 -C 2 -S 192.168.1.1 -D 192.168.1.2 -r udp
> > cpu: unable to pin, may increase variance.
> > + ip netns exec ns-45iLeE1 ./io_uring_zerocopy_tx -4 -t 1 -D 192.168.1.2 -m 1 -t 1 -n 32 udp
> > ./io_uring_zerocopy_tx: io_uring: queue init: Unknown error -13
> > 
> > Do you know what's the reason? Should we update the test script to return
> > SKIP if io_uring init failed?
> 
> I don't recall anything that can fail ring init with EACCES, probably
> sth in your system disables io_uring with some syscall filters or so.
> I think skipping the trace on EACCES is the right approach. Do you want
> to send a patch?
> Apart from that, if you're curious you can try to trace what's going on,
> ftrace should give an idea.

Ah, thanks for your reply. I just recalled that I also asked our storage team
member for this issue, and he told me I need to specify "io_uring.enable=y"
in the kernel commandline....

Hangbin