Re: [BUG] drivers: staging: comedi: do not call blocking ops when !TASK_RUNNING; state=1

From: Ian Abbott
Date: Tue Jul 18 2017 - 07:17:18 EST


[Top-posting fixed. Please don't top-post!]
On 17/07/17 19:14, Piotr Gregor wrote:
On Mon, Jul 17, 2017 at 05:02:45PM +0100, Ian Abbott wrote:
On 17/07/17 15:48, Piotr Gregor wrote:
Calling blocking operations from wrong context.
Kernel:

Linux piotrpc 4.4.70-rt83 #1 SMP PREEMPT RT Thu Jul 13 08:42:02 BST 2017 x86_64 GNU/Linux

[ 80.542018] NOHZ: local_softirq_pending 80
[ 125.175471] ------------[ cut here ]------------
[ 125.175491] WARNING: CPU: 0 PID: 1497 at kernel/sched/core.c:7833 __might_sleep+0x9f/0xb0()
[ 125.175728] do not call blocking ops when !TASK_RUNNING; state=1 set at [<ffffffffc0575741>] comedi_read+0x1a1/0x610 [comedi]
[snip]
[ 125.175926] CPU: 0 PID: 1497 Comm: txrx Tainted: G WC 4.4.70-rt83 #1
[ 125.175928] Hardware name: NOVATECH LTD PC-XB04472/H110M-C, BIOS 3019 01/06/2017
[ 125.175935] 0000000000000000 ffff8802205b3cb0 ffffffff81387160 ffff8802205b3cf8
[ 125.175940] 0000000000000009 ffff8802205b3ce8 ffffffff81089766 ffffffffc057ca60
[ 125.175944] 00000000000002dc 0000000000000000 0000000000000001 0000000000000001
[ 125.175945] Call Trace:
[ 125.175955] [<ffffffff81387160>] dump_stack+0x85/0xc5
[ 125.175963] [<ffffffff81089766>] warn_slowpath_common+0x86/0xe0
[ 125.175971] [<ffffffff8108980c>] warn_slowpath_fmt+0x4c/0x50
[ 125.175980] [<ffffffff81185577>] ? trace_preempt_on+0x1a7/0x2b0
[ 125.175985] [<ffffffff8168ca95>] ? schedule+0x55/0xe0
[ 125.175994] [<ffffffffc0575741>] ? comedi_read+0x1a1/0x610 [comedi]
[ 125.176001] [<ffffffffc0575741>] ? comedi_read+0x1a1/0x610 [comedi]
[ 125.176005] [<ffffffff810b6d0f>] __might_sleep+0x9f/0xb0
[ 125.176012] [<ffffffff811f54cb>] __might_fault+0x3b/0xb0
[ 125.176020] [<ffffffffc057597f>] comedi_read+0x3df/0x610 [comedi]
[ 125.176034] [<ffffffff810bebf0>] ? wake_up_process+0x20/0x20
[ 125.176044] [<ffffffff812406d8>] __vfs_read+0x28/0xe0
[ 125.176053] [<ffffffff812f1f06>] ? security_file_permission+0xa6/0xc0
[ 125.176060] [<ffffffff81240c53>] ? rw_verify_area+0x53/0xf0
[ 125.176066] [<ffffffff81240d79>] vfs_read+0x89/0x130
[ 125.176074] [<ffffffff81241b79>] SyS_read+0x49/0xb0
[ 125.176084] [<ffffffff81691876>] entry_SYSCALL_64_fastpath+0x16/0x7a
[ 125.176095] ---[ end trace 0000000000000003 ]---


cheers,
Piotr

Thanks for the bug report. I think comedi_read() in
"drivers/staging/comedi/comedi_fops.c" needs a call to
set_current_state(TASK_RUNNING) just before the call to copy_to_user(...).

Similarly, comedi_write() needs a call to set_current_task(TASK_RUNNING)
just before the call to copy_from_user(...), and another call to
set_current_task(TASK_RUNNING) before the call to mutex_lock().

I'll work on a patch.

> Hi Ian,
>
> I am afraid I have more problems with running Comedi on rt-patched kernel than
> only that. I am still trying to figure out the specific reason - hopefully it is
> my fault and I scrued up my own code, but so far it seems pcie215
> support doesn't work at least in the same way on rt-patched kernel
> as it was on non-rt. I will dig it a little bit harder, likely coming with another
> bug reports shoud they be observed.
>
> Do you know about other experience other people may have using Comedi
> for pcie215 on rt-patched kernels, specifically 4.4.70?

Not in particular. People have used Comedi with RTAI in the past, but that's a different animal to the RT patch-set.

> Has Comedi been tested on rt kernels by yourself?

Not up until now. Initial testing with the "comedi_test" module suggests there may be some issues to be resolved (e.g. it locks up when running an asynchronous command), although "comedi_test" isn't a typical Comedi driver as it makes heavy use of kernel timers. I do have a pcie215, but haven't tried it on an RT kernel yet.

--
-=( Ian Abbott @ MEV Ltd. E-mail: <abbotti@xxxxxxxxx> )=-
-=( Web: http://www.mev.co.uk/ )=-