Messages in this thread | | | Date | Sun, 11 Jun 2023 08:14:19 -0700 | From | Guenter Roeck <> | Subject | Re: [PATCH 5.15 000/159] 5.15.116-rc1 review |
| |
On 6/10/23 14:14, Guenter Roeck wrote: > Hi, > > On 6/10/23 12:23, Pavel Machek wrote: >> Hi! >> >>>> Build results: >>>> total: 155 pass: 155 fail: 0 >>>> Qemu test results: >>>> total: 499 pass: 498 fail: 1 >>>> Failed tests: >>>> arm:kudo-bmc:multi_v7_defconfig:npcm:usb0.1:nuvoton-npcm730-kudo:rootfs >>>> >>>> The test failure is spurious and not new. I observe it randomly on >>>> multi_v7_defconfig builds, primarily on npcm platforms. There is no error >>>> message, just a stalled boot. I have been trying to bisect for a while, >>>> but I have not been successful so far. No immediate concern; I just wanted >>>> to mention it in case someone else hits the same or a similar problem. >>>> >>> >>> I managed to revise my bisect script sufficiently enough to get reliable >>> results. It looks like the culprit is commit 503e554782c9 (" debugobject: >>> Ensure pool refill (again)"); see bisect log below. Bisect on four >>> different systems all have the same result. After reverting this patch, >>> I do not see the problem anymore (again, confirmed on four different >>> systems). If anyone has an idea how to debug this, please let me know. >>> I'll be happy to give it a try. >> >> You may want to comment out debug_objects_fill_pool() in >> debug_object_activate or debug_object_assert_init to see which one is >> causing the failure... >> >> CONFIG_PREEMPT_RT is disabled for you, right? (Should 5.15 even have >> that option?) >> > > CONFIG_PREEMPT_RT is disabled (it depends on ARCH_SUPPORTS_RT which is not > enabled by any architecture in v5.15.y). > > The added call in debug_object_activate() triggers the problem. > Any idea what to do about it or how to debug it further ? >
I did some more debugging. The call to debug_object_activate() from debug_hrtimer_activate() causes the immediate problem, and the call from debug_timer_activate() causes a second (less likely) problem, where the stall is seen during reboot.
In other words, the problem is (only) seen if DEBUG_OBJECTS_TIMERS is enabled.
Guenter
| |