Bug 12530
Summary: | ACPI Exceptions and EC GPE storm - Apple MacBook Pro 1,1 | ||
---|---|---|---|
Product: | ACPI | Reporter: | Javier Marcet (jmarcet) |
Component: | EC | Assignee: | Alexey Starikovskiy (astarikovskiy) |
Status: | CLOSED UNREPRODUCIBLE | ||
Severity: | high | CC: | acpi-bugzilla, astarikovskiy, czoccolo, lenb, rui.zhang |
Priority: | P1 | ||
Hardware: | All | ||
OS: | Linux | ||
Kernel Version: | 2.6.28.1 | Subsystem: | |
Regression: | Yes | Bisected commit-id: | |
Attachments: | dmesg of my current system, still going on after a GPE storm |
Description
Javier Marcet
2009-01-23 16:49:12 UTC
Created attachment 19977 [details]
dmesg of my current system, still going on after a GPE storm
I found this report http://bugzilla.kernel.org/show_bug.cgi?id=12250 of which this could well be a dup. I can confirm that 2.6.24.7 works fine, without errors. 2.6.25 OTOH says it can't find the MMCFG space, although there aren't any ACPI exceptions thrown. However reverting b6ce068a1285a24185b01be8a49021827516b3e1 seems to fix it. So far I couldn't port it to any 2.6.26+ kernel successfully, hence I'm testing 2.6.25.20 with the above commit reverted and see how it behaves under load. So far so good. The only thing which has come up is: [ 560.582772] CE: hpet increasing min_delta_ns to 15000 nsec [ 563.807043] CE: hpet increasing min_delta_ns to 22500 nsec That and that ATM it hasn't survived a suspend/resume. Other than that it seems quite stable. I've already stress tested it a little (make -j3 the kernel while watching a 720p x264 mkv read over wifi from a nfs share) and haven't had any problem, which I could not say of any 2.6.26+ kernel, or 2.6.25 without reverting the mentioned commit. There are 4 issues mentioned in this bug report. 1. PCI config space warnings It seems backwards that building with Any yields: PCI: Using MMCONFIG for extended config space while building with MMConfig yields: PCI: Not using MMCONFIG. This may be a PCI bug. Perhaps you're the first person on an 2.6.25 MMCONFIG machine to not use "Any"? 2. CE: hpet increasing min_delta_ns to 15000 nsec Please file a separate bug against timers for this message. (I'm guessing that it is independent of #1 and #3) 3. GPE storm ACPI: EC: GPE storm detected, transactions will use polling mode ACPI: EC: missing confirmations, switch off interrupt mode. This may be a known issue, for I recall that there were some GPE issues on the Apple macbook a few releases back. 4. system freeze this is the most serious, but most mysterious thing. It is related in any way to #1-#3? Does it happen in graphics mode only? Sorry for taking so long to reply. I was trying to reproduce my results consistently. 1. I know what it looks like, but that's what happens. All the distributions I checked compile their kernels with Any as access mode. That could be why this has not been noticed any earlier. 2. I've realized that 2 & 3 are different and not involve the other one. Booting with nohz it's harder to get these messages, but they do happen, with or without X11. They don't seem to affect performance or stability, though. 3. This started to happen alongside the MMCONFIG issue. At 2.6.25. At first ( 2.6.25) I had a sharp performance decrease when the storm was detected, soon followed by a crash. Right now it is non deterministic at all, sometimes it takes several hours to happen, other times it does right after starting the X11 session. 4. I've had problems freezing my system the same way all this week. I tend to compile my kernels with voluntary preemption. I've tried with full preemption and I got several panics while starting X11; with no preemption the system had too much latency to be used as desktop. All in all, it seems the freezes are completely random. The only thing for certain is that before a GPE storm happens, the system works wonderfully, including suspend/resume. Once the storm kicks in, at the very least it can't survive a suspend anymore. Sadly I can't reproduce reliably one of those freezes, whatever I do. #1 nowadays (2.6.28) seems harmless. Whether I use Any or MMCONFIG with the subsequent warnings, the system behaves the same way. #2 happens in console mode too, without having had a #3. #3 only happens in X11. Or I might not have stressed enough the system without X11. If at all, #4 seems related to #3, not to #1 or #2. do you know if there is an ambient light sensor on your laptop? please attach the acpidump output. Javier, can you try the patches attached at: http://bugzilla.kernel.org/show_bug.cgi?id=1294 (comments 20 and 29), and see if they improve the GPE storm issue? Sorry, in previous comment, the correct bug url is the following: http://bugzilla.kernel.org/show_bug.cgi?id=12949 It's been a while since I saw a GPE storm. But a few minutes ago, I got one. I'm compiling 2.6.29.6 with it right now. Maybe it even fixes another issue that has come up recently... Anyway, since I don't use HPET I haven't had any of those problems anymore. I'll see what happens with the last patch on 12949, both with and without HPET. closing due to no activity in this bug report in a month. please re-open if this is still a problem in the latest stable kernel. |