Re: [REGRESSION] 3.13-rc2: locks up hard on trying to transfer a file to mmc based internal SD card slot
Am Samstag, 30. November 2013, 14:53:51 schrieb Martin Steigerwald:
Just added linux-mmc. And I might git-bisect that at some time, but I do not intend to do it during my precious weekend. The chances of me bisecting it increase with workable suggestions on how to cut down the amount of iterations needed and avoid testing highly experimental between 3.12 and 3.13-rc1 kernels on a production laptop. I may be willing to test a patch or two. As I see there seem to have been quite some changes in MMC subsystem.
Hi!
Just does that on a ThinkPad T520 with:
merkaba:~> lspci -nn | grep MMC 0d:00.0 System peripheral [0880]: Ricoh Co Ltd PCIe SDXC/MMC Host Controller [1180:e823] (rev 08)
Mouse pointer freezes, no Ctrl-Alt-F1.
It just does that with
Linux version 3.13.0-rc6-tp520 (martin@merkaba) (gcc version 4.8.2 (Debian 4.8.2-10) ) #41 SMP PREEMPT Mon Dec 30 13:39:07 CET 2013
as well.
But, only when trying to write a file via desktop environment via dolphin from KDE in that case. When I am on tty1 it seems to be stable to write to the SD card. But with dolphin on writing a large few files vom /usr/bin mouse pointer froze again. But according to harddisk led from ThinkPad T520 there has been some write activity afterwards. The LED also lits up for MMC card accesses. Still after reboot there is none of the copied files visible on the FAT32 formatted SD card.
Thus adding Intel gfx and dri devel lists to CC.
This crashing only under GUI might still be a coindidence. I only tried once. But since the crash usually came almost immediately and it didn´t crash with reading or writing files on TTY1 and it somehow continued I/O according to harddisk led instead of seeming to be completely stopped… well I can try again to make sure. Would be good to make it crash on TTY1 since then I might see some kernel output.
Back to 3.12.6 for now. I just tried the same with that kernel and there the copying just works nice.
I can also report a bug at bugzilla.kernel.org if needed.
May comments about bisecting still applies. I do not feel comfortable with doing it on this production machine with production data on it… especially given the major block layer changes. There may be points in history were the kernel produces data corruption or so.
Thanks, Martin
merkaba:~> fdisk -l /dev/mmcblk0
Disk /dev/mmcblk0: 31.4 GB, 31439454208 bytes 255 heads, 63 sectors/track, 3822 cylinders, total 61405184 sectors Units = sectors of 1 * 512 = 512 bytes Sector size (logical/physical): 512 bytes / 512 bytes I/O size (minimum/optimal): 512 bytes / 512 bytes Disk identifier: 0x00000000
Device Boot Start End Blocks Id System/dev/mmcblk0p1 8192 61405183 30698496 c W95 FAT32 (LBA
merkaba:/sys/block/mmcblk0#2> grep . * 2>/dev/null alignment_offset:0 capability:10 dev:179:0 discard_alignment:0 ext_range:8 force_ro:0 inflight: 0 0 range:8 removable:0 ro:0 size:61405184 stat: 176 33 1672 102 0 0 0 0 0 102 102 uevent:MAJOR=179 uevent:MINOR=0 uevent:DEVNAME=mmcblk0 uevent:DEVTYPE=disk
I do not want to take the time to diagnose this further, especially as its one of those nasty "I just lock up and I don´t tell you what went wrong" kind of bugs. Thats just not a nice way to tell that there has been an error.
If there is any five or ten minute information gathering task, I am willing to provide more information, but right now there is no chance on Earth that I will be bisecting while having a long list of more interesting things to do than that.
Thus for now I just use 3.12 kernel again. Maybe I will try with some rc5 or so again.
Ciao,
Am Dienstag, 31. Dezember 2013, 13:41:22 schrieb Martin Steigerwald:
Am Samstag, 30. November 2013, 14:53:51 schrieb Martin Steigerwald:
Just added linux-mmc. And I might git-bisect that at some time, but I do not intend to do it during my precious weekend. The chances of me bisecting it increase with workable suggestions on how to cut down the amount of iterations needed and avoid testing highly experimental between 3.12 and 3.13-rc1 kernels on a production laptop. I may be willing to test a patch or two. As I see there seem to have been quite some changes in MMC subsystem.
Hi!
Just does that on a ThinkPad T520 with:
merkaba:~> lspci -nn | grep MMC 0d:00.0 System peripheral [0880]: Ricoh Co Ltd PCIe SDXC/MMC Host Controller [1180:e823] (rev 08)
Mouse pointer freezes, no Ctrl-Alt-F1.
It just does that with
Linux version 3.13.0-rc6-tp520 (martin@merkaba) (gcc version 4.8.2 (Debian 4.8.2-10) ) #41 SMP PREEMPT Mon Dec 30 13:39:07 CET 2013
as well.
I missed some important data. Kernel runs with threadirqs:
merkaba:~> cat /proc/cmdline BOOT_IMAGE=/vmlinuz-3.12.6-tp520 root=UUID=2f5c334d-249b-4c89-95cc-18572f750bd7 ro rootflags=subvol=root resume=/dev/mapper/merkaba-swap threadirqs i915_enable_rc6=7
Oh, and I see i915_enable_rc6=7. This always worked flawlessly. But maybe this changed? Cause according to powertop the GPU never entered deeper sleep states anyway. Maybe this now works (and thus may hang)?
These are values on 3.12.6:
| GPU | | | | Powered On 96,3% | | RC6 3,7% | | RC6p 0,0% | | RC6pp 0,0% | | | | |
I also attach kernel configs of non working 3.13-rc6 and working 3.12.6 kernels.
Its a ThinkPad T520 with Sandybridge:
merkaba:~> lspci -nn | grep VGA 00:02.0 VGA compatible controller [0300]: Intel Corporation 2nd Generation Core Processor Family Integrated Graphics Controller [8086:0126] (rev 09)
Debian kernel packages available. But well… optimized for ThinkPad T520, may not run nicely otherwere.
Please give suggestions on what to try next. Whats comes to mind is tryining without rc6_enable option and then without threadirq option.
Any other idea?
Ciao, Martin
But, only when trying to write a file via desktop environment via dolphin from KDE in that case. When I am on tty1 it seems to be stable to write to the SD card. But with dolphin on writing a large few files vom /usr/bin mouse pointer froze again. But according to harddisk led from ThinkPad T520 there has been some write activity afterwards. The LED also lits up for MMC card accesses. Still after reboot there is none of the copied files visible on the FAT32 formatted SD card.
Thus adding Intel gfx and dri devel lists to CC.
This crashing only under GUI might still be a coindidence. I only tried once. But since the crash usually came almost immediately and it didn´t crash with reading or writing files on TTY1 and it somehow continued I/O according to harddisk led instead of seeming to be completely stopped… well I can try again to make sure. Would be good to make it crash on TTY1 since then I might see some kernel output.
Back to 3.12.6 for now. I just tried the same with that kernel and there the copying just works nice.
I can also report a bug at bugzilla.kernel.org if needed.
May comments about bisecting still applies. I do not feel comfortable with doing it on this production machine with production data on it… especially given the major block layer changes. There may be points in history were the kernel produces data corruption or so.
Thanks, Martin
merkaba:~> fdisk -l /dev/mmcblk0
Disk /dev/mmcblk0: 31.4 GB, 31439454208 bytes 255 heads, 63 sectors/track, 3822 cylinders, total 61405184 sectors Units = sectors of 1 * 512 = 512 bytes Sector size (logical/physical): 512 bytes / 512 bytes I/O size (minimum/optimal): 512 bytes / 512 bytes Disk identifier: 0x00000000
Device Boot Start End Blocks Id System/dev/mmcblk0p1 8192 61405183 30698496 c W95 FAT32 (LBA
merkaba:/sys/block/mmcblk0#2> grep . * 2>/dev/null alignment_offset:0 capability:10 dev:179:0 discard_alignment:0 ext_range:8 force_ro:0 inflight: 0 0 range:8 removable:0 ro:0 size:61405184 stat: 176 33 1672 102 0 0 0 0 0 102 102 uevent:MAJOR=179 uevent:MINOR=0 uevent:DEVNAME=mmcblk0 uevent:DEVTYPE=disk
I do not want to take the time to diagnose this further, especially as its one of those nasty "I just lock up and I don´t tell you what went wrong" kind of bugs. Thats just not a nice way to tell that there has been an error.
If there is any five or ten minute information gathering task, I am willing to provide more information, but right now there is no chance on Earth that I will be bisecting while having a long list of more interesting things to do than that.
Thus for now I just use 3.12 kernel again. Maybe I will try with some rc5 or so again.
Ciao,
Am Dienstag, 31. Dezember 2013, 13:52:05 schrieb Martin Steigerwald:
Am Dienstag, 31. Dezember 2013, 13:41:22 schrieb Martin Steigerwald:
Am Samstag, 30. November 2013, 14:53:51 schrieb Martin Steigerwald:
Just added linux-mmc. And I might git-bisect that at some time, but I do not intend to do it during my precious weekend. The chances of me bisecting it increase with workable suggestions on how to cut down the amount of iterations needed and avoid testing highly experimental between 3.12 and 3.13-rc1 kernels on a production laptop. I may be willing to test a patch or two. As I see there seem to have been quite some changes in MMC subsystem.
Hi!
Just does that on a ThinkPad T520 with:
merkaba:~> lspci -nn | grep MMC 0d:00.0 System peripheral [0880]: Ricoh Co Ltd PCIe SDXC/MMC Host Controller [1180:e823] (rev 08)
Mouse pointer freezes, no Ctrl-Alt-F1.
It just does that with
Linux version 3.13.0-rc6-tp520 (martin@merkaba) (gcc version 4.8.2 (Debian 4.8.2-10) ) #41 SMP PREEMPT Mon Dec 30 13:39:07 CET 2013
as well.
I missed some important data. Kernel runs with threadirqs:
merkaba:~> cat /proc/cmdline BOOT_IMAGE=/vmlinuz-3.12.6-tp520 root=UUID=2f5c334d-249b-4c89-95cc-18572f750bd7 ro rootflags=subvol=root resume=/dev/mapper/merkaba-swap threadirqs i915_enable_rc6=7
Oh, and I see i915_enable_rc6=7. This always worked flawlessly. But maybe this changed? Cause according to powertop the GPU never entered deeper sleep states anyway. Maybe this now works (and thus may hang)?
These are values on 3.12.6: | GPU | | | Powered On 96,3% | | RC6 3,7% | | RC6p 0,0% | | RC6pp 0,0% |
I also attach kernel configs of non working 3.13-rc6 and working 3.12.6 kernels.
Its a ThinkPad T520 with Sandybridge:
merkaba:~> lspci -nn | grep VGA 00:02.0 VGA compatible controller [0300]: Intel Corporation 2nd Generation Core Processor Family Integrated Graphics Controller [8086:0126] (rev 09)
Debian kernel packages available. But well… optimized for ThinkPad T520, may not run nicely otherwere.
Please give suggestions on what to try next. Whats comes to mind is tryining without rc6_enable option and then without threadirq option.
Any other idea?
So there we go:
1) Without kernel options (except resume option) copying works in desktop.
2) But then otherwise as I expected: With just "threadirqs" kernel option it hangs. I would have suspected the "i915_enable_rc6" option instead.
3) But only when triggering copy via KDE desktop dolphin´s file manager. I tried copying on tty1 again and no hang there. Also again I/O seemed to go on after mouse pointer freeze (without Ctrl-Alt-F1) working. kwin ran in compositing mode.
4) With just "i915_enable_rc6=7" it works. But since according to powertop that option doesn´t give me the benefit of doing into deeper GPU sleep states anyway, I removed that one now as well.
So my laptop is on 3.13-rc6 now without any special options and I learned again:
Never try to tune the kernel :)
And if still tuning it: First remove any kernel option when facing any issue. Sorry for the noise that not adhering to this has caused.
This is second time that threadirqs option caused problems here. Thus CCing rt-users mailing list as well.
If wished I compile this all into a bugzilla.kernel.org bug report in a concise format as well. But thats for another day :)
Have a good shift into new year if not already in it… otherwise a happy new year, Martin
Ciao, Martin
But, only when trying to write a file via desktop environment via dolphin from KDE in that case. When I am on tty1 it seems to be stable to write to the SD card. But with dolphin on writing a large few files vom /usr/bin mouse pointer froze again. But according to harddisk led from ThinkPad T520 there has been some write activity afterwards. The LED also lits up for MMC card accesses. Still after reboot there is none of the copied files visible on the FAT32 formatted SD card.
Thus adding Intel gfx and dri devel lists to CC.
This crashing only under GUI might still be a coindidence. I only tried once. But since the crash usually came almost immediately and it didn´t crash with reading or writing files on TTY1 and it somehow continued I/O according to harddisk led instead of seeming to be completely stopped… well I can try again to make sure. Would be good to make it crash on TTY1 since then I might see some kernel output.
Back to 3.12.6 for now. I just tried the same with that kernel and there the copying just works nice.
I can also report a bug at bugzilla.kernel.org if needed.
May comments about bisecting still applies. I do not feel comfortable with doing it on this production machine with production data on it… especially given the major block layer changes. There may be points in history were the kernel produces data corruption or so.
Thanks, Martin
merkaba:~> fdisk -l /dev/mmcblk0
Disk /dev/mmcblk0: 31.4 GB, 31439454208 bytes 255 heads, 63 sectors/track, 3822 cylinders, total 61405184 sectors Units = sectors of 1 * 512 = 512 bytes Sector size (logical/physical): 512 bytes / 512 bytes I/O size (minimum/optimal): 512 bytes / 512 bytes Disk identifier: 0x00000000
Device Boot Start End Blocks Id System/dev/mmcblk0p1 8192 61405183 30698496 c W95 FAT32 (LBA
merkaba:/sys/block/mmcblk0#2> grep . * 2>/dev/null alignment_offset:0 capability:10 dev:179:0 discard_alignment:0 ext_range:8 force_ro:0 inflight: 0 0 range:8 removable:0 ro:0 size:61405184 stat: 176 33 1672 102 0 0 0 0 0 102 102 uevent:MAJOR=179 uevent:MINOR=0 uevent:DEVNAME=mmcblk0 uevent:DEVTYPE=disk
I do not want to take the time to diagnose this further, especially as its one of those nasty "I just lock up and I don´t tell you what went wrong" kind of bugs. Thats just not a nice way to tell that there has been an error.
If there is any five or ten minute information gathering task, I am willing to provide more information, but right now there is no chance on Earth that I will be bisecting while having a long list of more interesting things to do than that.
Thus for now I just use 3.12 kernel again. Maybe I will try with some rc5 or so again.
Ciao,
dri-devel@lists.freedesktop.org
-
Martin Steigerwald