AGX ORIN的gpu掉电问题

你好,我们目前有多台设备出现运行几个小时后会卡死分析的原因是GPU 断电流程中,alter_usermode_mappings 空转等内存映射锁,而且不让出 CPU;持有这把锁的进程正在缺页处理,等待 drain_all_pages;drain_all_pages 要在被占住的 CPU7 上执行任务。三方互相等待,形成死锁。请问是否是因为这个问题,其次能否提供一下解决方案?

meminfo.txt (1.4 KB)

dz.txt (6.3 KB)

blocked.txt (28.0 KB)

366.txt (29.3 KB)

Please clarify which Jetpack ersion you are using and how to reproduce this issue on NV devkit.

We won’t know what solution to provide with just these logs…

你好,我使用的是jp6.2,这个出现的分析的像是gpu掉电,我们的程序在获取页的时候导致进程没有被收回造成的死锁,如果能够取消掉gpu的掉电应该就会解决这个问题

As I already told, we don’t provide a solution just with such comment you shared. Please give out a method that could reproduce on JP6.2.2. or later BSP.

jp6.2 is too old to check.

那请问有办法可以防止gpu空闲时间掉电吗?

Hi,
If you don’t use Jetpack 6.2.2 r36.5 or 6.2.3 r36.5.2, we would suggest apply the patches:

Making sure you're not a bot!
Making sure you're not a bot!

These are critical issues and fix potential issues when using GPU hardware.

jp6.2可以使用吗

Hi,

It should be able to be applied to Jetpack 6.2 r36.4.3. Please download kernel source code and try.

And please note this known issue on r36.4.x:

Making sure you're not a bot!
[Orin] Display drivers install to the wrong path with official document instructions

Please manually copy the ko files.

Here is a post about re-building kernel:
No data from Joystick Logitech-f710 - #10 by DaneLLL

这些是否能够解决gpu空闲掉电的问题