Slow path detected in RTCAP

Slow path detected in RTCAP

[François Legal]

Hello,

I'm trying to port an old application that used to be running on DSPBIOS to Xenomai, using RTNET for the network part.
I'm getting some weird results with semaphores, but before that I got that debug dump :

[   49.201310] WARNING: CPU: 0 PID: 958 at /home/dev/UCC/linux-4.4.189/kernel/ipipe/core.c:1912 __ipipe_post_work_root+0xe8/0xfc()
[   49.214809] CPU: 0 PID: 958 Comm: Maître de Rés Not tainted 4.4.189 #47
[   49.222636] Hardware name: Xilinx Zynq Platform
[   49.227791] I-pipe domain: Xenomai
[   49.231705] [<c00159e8>] (unwind_backtrace) from [<c0012e24>] (show_stack+0x10/0x14)
[   49.240511] [<c0012e24>] (show_stack) from [<c0222798>] (dump_stack+0x104/0x120)
[   49.248887] [<c0222798>] (dump_stack) from [<c0025568>] (warn_slowpath_common+0x80/0xb0)
[   49.258162] [<c0025568>] (warn_slowpath_common) from [<c0025660>] (warn_slowpath_null+0x1c/0x24)
[   49.268172] [<c0025660>] (warn_slowpath_null) from [<c008b7bc>] (__ipipe_post_work_root+0xe8/0xfc)
[   49.278344] [<c008b7bc>] (__ipipe_post_work_root) from [<c00a4df0>] (rtdm_nrtsig_pend+0x24/0x2c)
[   49.288219] [<c00a4df0>] (rtdm_nrtsig_pend) from [<c034e2c8>] (rtcap_xmit_hook+0x150/0x1bc)
[   49.297550] [<c034e2c8>] (rtcap_xmit_hook) from [<c033fc64>] (rtdev_locked_xmit+0x28/0x3c)
[   49.307010] [<c033fc64>] (rtdev_locked_xmit) from [<c0340be4>] (rtdev_xmit+0x38/0x84)
[   49.315871] [<c0340be4>] (rtdev_xmit) from [<c034935c>] (rt_packet_sendmsg+0x2d4/0x304)
[   49.324865] [<c034935c>] (rt_packet_sendmsg) from [<c00a6380>] (rtdm_fd_sendmsg+0x80/0x98)
[   49.334205] [<c00a6380>] (rtdm_fd_sendmsg) from [<c00aa0d8>] (CoBaLt_sendmsg+0x78/0xa4)
[   49.343275] [<c00aa0d8>] (CoBaLt_sendmsg) from [<c00b602c>] (handle_head_syscall+0x168/0x320)
[   49.352953] [<c00b602c>] (handle_head_syscall) from [<c000f9d0>] (local_restart+0x20/0x44)
[

The indicated thread "Maître de Rés" is a xenomai posix thread.

I see that the call stack includes RTCAP code, and I checked that this does not happen when RTCAP is disabled in kernel build. Anyway, Is this slow path to be expected in RTCAP ?

Thanks

François

Re: Slow path detected in RTCAP

[Jan Kiszka]

On 24.03.20 17:05, François Legal via Xenomai wrote:
> Hello,
> 
> I'm trying to port an old application that used to be running on DSPBIOS to Xenomai, using RTNET for the network part.
> I'm getting some weird results with semaphores, but before that I got that debug dump :
> 
> [   49.201310] WARNING: CPU: 0 PID: 958 at /home/dev/UCC/linux-4.4.189/kernel/ipipe/core.c:1912 __ipipe_post_work_root+0xe8/0xfc()
> [   49.214809] CPU: 0 PID: 958 Comm: Maître de Rés Not tainted 4.4.189 #47
> [   49.222636] Hardware name: Xilinx Zynq Platform
> [   49.227791] I-pipe domain: Xenomai
> [   49.231705] [<c00159e8>] (unwind_backtrace) from [<c0012e24>] (show_stack+0x10/0x14)
> [   49.240511] [<c0012e24>] (show_stack) from [<c0222798>] (dump_stack+0x104/0x120)
> [   49.248887] [<c0222798>] (dump_stack) from [<c0025568>] (warn_slowpath_common+0x80/0xb0)
> [   49.258162] [<c0025568>] (warn_slowpath_common) from [<c0025660>] (warn_slowpath_null+0x1c/0x24)
> [   49.268172] [<c0025660>] (warn_slowpath_null) from [<c008b7bc>] (__ipipe_post_work_root+0xe8/0xfc)
> [   49.278344] [<c008b7bc>] (__ipipe_post_work_root) from [<c00a4df0>] (rtdm_nrtsig_pend+0x24/0x2c)
> [   49.288219] [<c00a4df0>] (rtdm_nrtsig_pend) from [<c034e2c8>] (rtcap_xmit_hook+0x150/0x1bc)
> [   49.297550] [<c034e2c8>] (rtcap_xmit_hook) from [<c033fc64>] (rtdev_locked_xmit+0x28/0x3c)
> [   49.307010] [<c033fc64>] (rtdev_locked_xmit) from [<c0340be4>] (rtdev_xmit+0x38/0x84)
> [   49.315871] [<c0340be4>] (rtdev_xmit) from [<c034935c>] (rt_packet_sendmsg+0x2d4/0x304)
> [   49.324865] [<c034935c>] (rt_packet_sendmsg) from [<c00a6380>] (rtdm_fd_sendmsg+0x80/0x98)
> [   49.334205] [<c00a6380>] (rtdm_fd_sendmsg) from [<c00aa0d8>] (CoBaLt_sendmsg+0x78/0xa4)
> [   49.343275] [<c00aa0d8>] (CoBaLt_sendmsg) from [<c00b602c>] (handle_head_syscall+0x168/0x320)
> [   49.352953] [<c00b602c>] (handle_head_syscall) from [<c000f9d0>] (local_restart+0x20/0x44)
> [
> 
> The indicated thread "Maître de Rés" is a xenomai posix thread.
> 
> I see that the call stack includes RTCAP code, and I checked that this does not happen when RTCAP is disabled in kernel build. Anyway, Is this slow path to be expected in RTCAP ?
> 

This issue is supposed to be fixed in 3.1 via b2f253fd21b8. Are you on 
3.0.x? If you could try out a backport, we could merge that into stable 
as well.

Jan

-- 
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux

Re: Slow path detected in RTCAP

[François Legal]

Le Mardi, Mars 24, 2020 18:01 CET, Jan Kiszka <jan.kiszka@siemens.com> a écrit:

> On 24.03.20 17:05, François Legal via Xenomai wrote:
> > Hello,
> >
> > I'm trying to port an old application that used to be running on DSPBIOS to Xenomai, using RTNET for the network part.
> > I'm getting some weird results with semaphores, but before that I got that debug dump :
> >
> > [   49.201310] WARNING: CPU: 0 PID: 958 at /home/dev/UCC/linux-4.4.189/kernel/ipipe/core.c:1912 __ipipe_post_work_root+0xe8/0xfc()
> > [   49.214809] CPU: 0 PID: 958 Comm: Maître de Rés Not tainted 4.4.189 #47
> > [   49.222636] Hardware name: Xilinx Zynq Platform
> > [   49.227791] I-pipe domain: Xenomai
> > [   49.231705] [<c00159e8>] (unwind_backtrace) from [<c0012e24>] (show_stack+0x10/0x14)
> > [   49.240511] [<c0012e24>] (show_stack) from [<c0222798>] (dump_stack+0x104/0x120)
> > [   49.248887] [<c0222798>] (dump_stack) from [<c0025568>] (warn_slowpath_common+0x80/0xb0)
> > [   49.258162] [<c0025568>] (warn_slowpath_common) from [<c0025660>] (warn_slowpath_null+0x1c/0x24)
> > [   49.268172] [<c0025660>] (warn_slowpath_null) from [<c008b7bc>] (__ipipe_post_work_root+0xe8/0xfc)
> > [   49.278344] [<c008b7bc>] (__ipipe_post_work_root) from [<c00a4df0>] (rtdm_nrtsig_pend+0x24/0x2c)
> > [   49.288219] [<c00a4df0>] (rtdm_nrtsig_pend) from [<c034e2c8>] (rtcap_xmit_hook+0x150/0x1bc)
> > [   49.297550] [<c034e2c8>] (rtcap_xmit_hook) from [<c033fc64>] (rtdev_locked_xmit+0x28/0x3c)
> > [   49.307010] [<c033fc64>] (rtdev_locked_xmit) from [<c0340be4>] (rtdev_xmit+0x38/0x84)
> > [   49.315871] [<c0340be4>] (rtdev_xmit) from [<c034935c>] (rt_packet_sendmsg+0x2d4/0x304)
> > [   49.324865] [<c034935c>] (rt_packet_sendmsg) from [<c00a6380>] (rtdm_fd_sendmsg+0x80/0x98)
> > [   49.334205] [<c00a6380>] (rtdm_fd_sendmsg) from [<c00aa0d8>] (CoBaLt_sendmsg+0x78/0xa4)
> > [   49.343275] [<c00aa0d8>] (CoBaLt_sendmsg) from [<c00b602c>] (handle_head_syscall+0x168/0x320)
> > [   49.352953] [<c00b602c>] (handle_head_syscall) from [<c000f9d0>] (local_restart+0x20/0x44)
> > [
> >
> > The indicated thread "Maître de Rés" is a xenomai posix thread.
> >
> > I see that the call stack includes RTCAP code, and I checked that this does not happen when RTCAP is disabled in kernel build. Anyway, Is this slow path to be expected in RTCAP ?
> >

Yes. On 3.0.9.

>
> This issue is supposed to be fixed in 3.1 via b2f253fd21b8. Are you on
> 3.0.x? If you could try out a backport, we could merge that into stable
> as well.

I'll check that then submit a patch

>
> Jan
>
> --
> Siemens AG, Corporate Technology, CT RDA IOT SES-DE
> Corporate Competence Center Embedded Linux