Scheduled Maintenance: We are aware of an issue with Google, AOL, and Yahoo services as email providers which are blocking new registrations. We are trying to fix the issue and we have several internal and external support tickets in process to resolve the issue. Please see: viewtopic.php?t=158230

 

 

 

Wrong reading with radeontop

Need help with peripherals or devices?
Post Reply
Message
Author
mm3100
Posts: 336
Joined: 2020-10-21 21:39
Has thanked: 8 times
Been thanked: 13 times

Wrong reading with radeontop

#1 Post by mm3100 »

Hello there,
Since installing Bullseye I noticed there were wrong readings for radeontop utility for my radeon card
Screenshot_20220119_201233.png
Screenshot_20220119_201233.png (32.93 KiB) Viewed 4250 times
Card is
VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Seymour [Radeon HD 6400M/7400M Series]

All drivers are installed, card is working and seemingly without an issue. Attached picture is for when card is not in use, you would expect all variables to be 0, but they aren't. In face, VRAM usage only increases with use. At one point I managed to get it to 17 GB after weeks or so of use. It doesn't impact anything, just seems... wrong.

Now it is obviously that reading are wrong, so I am asking if rest have had similar issues and if it is kernel or driver bug of some kind. Or maybe radeontop it self, possibly unable to read states or something like it. I especially like how memory clock is infinite in use.

Similar to that, is when trying to use rovclock utility, for example when running rovclock -i I get

Code: Select all

Radeon overclock 0.6e by Hasw (hasw@hasw.net)

Error getting I/O permissions (root?).
I ensure to run it from root and with DRI_PRIME=1 variable, so I know it is trying to access it. Tried with and without it, same result. That would at least explain wrong clock reading.

CwF
Global Moderator
Global Moderator
Posts: 2625
Joined: 2018-06-20 15:16
Location: Colorado
Has thanked: 41 times
Been thanked: 190 times

Re: Wrong reading with radeontop

#2 Post by CwF »

Interesting! I take it this is a laptop?

I have not seen this issue on a desktop(server). I do use radeontop and can comment it does not depend on the gpu's 'driver', only the firmware and sysfs enumeration. With multiple cards radeontop can monitor a vfio passed card on the host while a guest chews on it.
Like this

Code: Select all

# radeontop --bus 130
with the high bus number on another socket and in another numa zone as a guest, quite convenient!

Truth is I have no need for radeontop within Bullseye so far. Even though I have a complete bullseye mammoth hypervisor built it is still in training and runs as a hosted guest host under Buster! I will keep an eye out.

Having tried to build some power monitoring things where there is a "roll over array" type issue when the stat has no zero and needs a time block reference to come up with a human readable number, this doesn't make sense for clock and memory stat to have an issue like this?
Interesting.

mm3100
Posts: 336
Joined: 2020-10-21 21:39
Has thanked: 8 times
Been thanked: 13 times

Re: Wrong reading with radeontop

#3 Post by mm3100 »

Yeah, it is a laptop, with integrated intel 4000 HD graphics and that radeon dGPU card. I use radeon card only for some specific uses, problem arises that I haven't seen completely wrong behaviour on Buster, at least memory wise. Clock was still being wrongly used.

Now here we come to interesting part. For some reason I didn't bother to look at dmesg before commenting here. And here is result

When using radeontop

Code: Select all

[32686.004396] [drm] enabling PCIE gen 2 link speeds, disable with radeon.pcie_gen2=0
[32686.016512] [drm] PCIE GART of 1024M enabled (table at 0x0000000000162000).
[32686.016637] radeon 0000:01:00.0: WB enabled
[32686.016642] radeon 0000:01:00.0: fence driver on ring 0 use gpu addr 0x0000000040000c00
[32686.016644] radeon 0000:01:00.0: fence driver on ring 3 use gpu addr 0x0000000040000c0c
[32686.017441] radeon 0000:01:00.0: fence driver on ring 5 use gpu addr 0x0000000000072118
[32686.033865] [drm] ring test on 0 succeeded in 2 usecs
[32686.033900] [drm] ring test on 3 succeeded in 20 usecs
[32686.209805] [drm] ring test on 5 succeeded in 2 usecs
[32686.209825] [drm] UVD initialized successfully.
[32686.209977] [drm] ib test on ring 0 succeeded in 0 usecs
[32686.210111] [drm] ib test on ring 3 succeeded in 0 usecs
[32686.868026] [drm] ib test on ring 5 succeeded
All seems fine, no warning or error messages or any odd messages. But here is the fun bit, when running rovclock -i

Code: Select all

[32730.551070] Lockdown: rovclock: raw io port access is restricted; see man kernel_lockdown.7
For what I can say, that is connected to secure boot? I could try to disable it and test utilities again, but didn't expect that to be restricted due to secure boot. Unless I am wrong about it?

mm3100
Posts: 336
Joined: 2020-10-21 21:39
Has thanked: 8 times
Been thanked: 13 times

Re: Wrong reading with radeontop

#4 Post by mm3100 »

Okay, tried with secure boot turned off. I still get wrong readings from radeontop, memory keeps increasing and clock is still not readable.

When running rovclock -i I get

Code: Select all

Radeon overclock 0.6e by Hasw (hasw@hasw.net)

Found ATI card on e00:00, device id: 0x6760
I/O space not enabled, aborting.
Again, not working, but different reason at least, looking it up a bit didn't give me anything. I am new to this issue.

Post Reply