Re: Linux 6.10-rc1
Hi! > > Ok, so machine is ready to be thrown out of window, again. Trying to > > play 29C3 video should not make machine completely unusable ... as in > > keyboard looses keystrokes in terminal. > > Well, that at least sounds like you can bisect it with a very clear test-case? > > Even if you don't bisect all the way, just doing a handful of > bisections tends to narrow things down enough that we can at least > guess at what general kind of area it is... So... I guess I now know what went on. We got summer here, and I was running notebook closed. Appartently that affects cooling a _lot_. Open lid means more dust, but also better cooling... Best regards (and sorry for the noise), Pavel -- People of Russia, stop Putin before his war on Ukraine escalates. signature.asc Description: PGP signature
Re: Linux 6.10-rc1
Hello! Ideas welcome, especially some way to see what graphics is doing. I'm unsure about the distro you are using but try package intel_gpu_top. It displays irqs/s and a bunch of other utilization statistics. Hope this helps. -- BR, Gerhard
Re: Linux 6.10-rc1
On Wed, 19 Jun 2024 at 03:44, Pavel Machek wrote: > > Ok, so machine is ready to be thrown out of window, again. Trying to > play 29C3 video should not make machine completely unusable ... as in > keyboard looses keystrokes in terminal. Well, that at least sounds like you can bisect it with a very clear test-case? Even if you don't bisect all the way, just doing a handful of bisections tends to narrow things down enough that we can at least guess at what general kind of area it is... Linus
Re: Linux 6.10-rc1
Hi! > > > Let's bring in the actual gpu people.. Dave/Jani/others - does any of > > > this sound familiar? Pavel says things have gotten much slower in > > > 6.10: "something was very wrong with the performance, likely to do > > > with graphics" > > > > Actually, maybe it's not graphics at all. Rafael just sent me a pull > > request that fixes a "turbo is disabled at boot, but magically enabled > > at runtime by firmware" issue. > > > > The 6.10-rc1 kernel would notice that turbo was disabled, and stopped > > noticing that it magically got re-enabled. > > > > Pavel, that was with a very different laptop, but who knows... That > > would match the "laptop is much slower" thing. > > > > So current -git might be worth checking. > > So... I went to (then) current -git and I don't want to replace my > machine any more. So the problem should not exist in current mainline. > > (I did not have good objective data, so I'm not 100% sure problem was > real in the first place. More like 90% sure.) Ok, so machine is ready to be thrown out of window, again. Trying to play 29C3 video should not make machine completely unusable ... as in keyboard looses keystrokes in terminal. https://media.ccc.de/v/29c3-5333-en-gsm_cell_phone_network_review_h264#t=340 dmesg is kind-of unhappy: [130729.891961] usb 2-1.2.3: reset low-speed USB device number 13 using ehci-pci [130733.311644] usb 2-1.2.3: reset low-speed USB device number 13 using ehci-pci [130736.534601] i915 :00:02.0: [drm] *ERROR* Atomic update failure on pipe A (start=617818 end=617819) time 159 us, min 1017, max 1023, scanline start 1012, end 1024 [130738.625131] usb 2-1.2.3: reset low-speed USB device number 13 using ehci-pci [130745.451785] usb 2-1.2.3: reset low-speed USB device number 13 using ehci-pci ... [131631.941091] usb 2-1.2.3: reset low-speed USB device number 13 using ehci-pci [131634.817628] usb 2-1.2.3: reset low-speed USB device number 13 using ehci-pci [131639.536918] usb 2-1.2.3: reset low-speed USB device number 13 using ehci-pci [131790.153952] i915 :00:02.0: [drm] GPU HANG: ecode 6:1:95bc, in Xorg [3043] [131790.154245] i915 :00:02.0: [drm] GT0: Resetting chip for stopped heartbeat on rcs0 [131790.255994] i915 :00:02.0: [drm] Xorg[3043] context reset due to GPU hang Wifi is a bit too active, even on fairly idle system: 430 root -51 0 0 0 0 S 0.3 0.0 8:48.74 irq/17-iwlwifi Ideas welcome, especially some way to see what graphics is doing. Best regards, Pavel -- People of Russia, stop Putin before his war on Ukraine escalates. signature.asc Description: PGP signature
Re: Linux 6.10-rc1
Hi! > > Let's bring in the actual gpu people.. Dave/Jani/others - does any of > > this sound familiar? Pavel says things have gotten much slower in > > 6.10: "something was very wrong with the performance, likely to do > > with graphics" > > Actually, maybe it's not graphics at all. Rafael just sent me a pull > request that fixes a "turbo is disabled at boot, but magically enabled > at runtime by firmware" issue. > > The 6.10-rc1 kernel would notice that turbo was disabled, and stopped > noticing that it magically got re-enabled. > > Pavel, that was with a very different laptop, but who knows... That > would match the "laptop is much slower" thing. > > So current -git might be worth checking. So... I went to (then) current -git and I don't want to replace my machine any more. So the problem should not exist in current mainline. (I did not have good objective data, so I'm not 100% sure problem was real in the first place. More like 90% sure.) Best regards, Pavel -- People of Russia, stop Putin before his war on Ukraine escalates. signature.asc Description: PGP signature
Re: Linux 6.10-rc1
Hi! > > Let's bring in the actual gpu people.. Dave/Jani/others - does any of > > this sound familiar? Pavel says things have gotten much slower in > > 6.10: "something was very wrong with the performance, likely to do > > with graphics" > > Actually, maybe it's not graphics at all. Rafael just sent me a pull > request that fixes a "turbo is disabled at boot, but magically enabled > at runtime by firmware" issue. > > The 6.10-rc1 kernel would notice that turbo was disabled, and stopped > noticing that it magically got re-enabled. > > Pavel, that was with a very different laptop, but who knows... That > would match the "laptop is much slower" thing. > > So current -git might be worth checking. Is that: commit 0cac73eb3875f6ecb6105e533218dba1868d04c9 Merge: 94df82fe5bfd 350cbb5d2f67 Author: Linus Torvalds Date: Fri Jun 14 09:52:51 2024 -0700 Merge tag 'pm-6.10-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm Pull power management fix from Rafael Wysocki: "Restore the behavior of the no_turbo sysfs attribute in the intel_pstate driver which allowed users to make the driver start using turbo P-states if they have been enabled on the fly by the firmware after OS initialization (Rafael Wysocki)" * tag 'pm-6.10-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: cpufreq: intel_pstate: Check turbo_is_disabled() in store_no_turbo() ? I don't think I tweaking no_turbo in the sysfs. But the thermal stuff looks important: commit cee84c0b003f2e0f486f200a72eca2bcdb3a49a7 Merge: d20f6b3d747c b6846826982b Author: Linus Torvalds Date: Fri Jun 14 09:28:56 2024 -0700 Merge tag 'thermal-6.10-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm So I guess I'll have to try again. Thanks and best regards, Pavel -- People of Russia, stop Putin before his war on Ukraine escalates. signature.asc Description: PGP signature
Re: Linux 6.10-rc1
Hi! > > Let's bring in the actual gpu people.. Dave/Jani/others - does any of > > this sound familiar? Pavel says things have gotten much slower in > > 6.10: "something was very wrong with the performance, likely to do > > with graphics" > > Actually, maybe it's not graphics at all. Rafael just sent me a pull > request that fixes a "turbo is disabled at boot, but magically enabled > at runtime by firmware" issue. > > The 6.10-rc1 kernel would notice that turbo was disabled, and stopped > noticing that it magically got re-enabled. > > Pavel, that was with a very different laptop, but who knows... That > would match the "laptop is much slower" thing. > > So current -git might be worth checking. Ok, let me check. That sounds like something that could make machine hotter. My problem seems to be that machine seems to run way hotter with 6.10, and when it hovers around the 97C limit, it is unusable with all the throttling. It gets unusable with 6.9 at 97C, too, it is just that it is harder to make it so hot with 6.9. (And yes, I'm running Chromium, and yes, that means websites influence this. Media playback also does, 1080p video pushes thermals close to the limits even on good kernels.) Thanks and best regards, Pavel -- People of Russia, stop Putin before his war on Ukraine escalates. signature.asc Description: PGP signature
Re: Linux 6.10-rc1
On Fri, 14 Jun 2024 at 09:21, Linus Torvalds wrote: > > Let's bring in the actual gpu people.. Dave/Jani/others - does any of > this sound familiar? Pavel says things have gotten much slower in > 6.10: "something was very wrong with the performance, likely to do > with graphics" Actually, maybe it's not graphics at all. Rafael just sent me a pull request that fixes a "turbo is disabled at boot, but magically enabled at runtime by firmware" issue. The 6.10-rc1 kernel would notice that turbo was disabled, and stopped noticing that it magically got re-enabled. Pavel, that was with a very different laptop, but who knows... That would match the "laptop is much slower" thing. So current -git might be worth checking. Linus
Re: Linux 6.10-rc1
On Fri, 14 Jun 2024 at 02:02, Pavel Machek wrote: > > If I can get at least basic metric on the gpu (%idle? which process > use how much time?), it might be feasible. Is there tool similar for > top? Let's bring in the actual gpu people.. Dave/Jani/others - does any of this sound familiar? Pavel says things have gotten much slower in 6.10: "something was very wrong with the performance, likely to do with graphics" To bisect it, he'd need some way to judge it reasonably well and without too much of a bias. See https://lore.kernel.org/all/zmrtzozoi0t%2ft...@duo.ucw.cz/ for the original report. Thinkpad X220 - which I assume means old intel integrated GPU - at least one listing I found for that thing is i5-2430M, with "Intel® HD Graphics 3000". Linus
