Using identical kernel configuration for MMC/SD support, the upgrade to the 2.6.60 series kernels stops a listing appearing in /dev for the MMC/SD reader. Boot back into the 2.6.19 kernel, it works again. Reproducible: Always Steps to Reproduce: 1. Boot with 2.6.20 kernel. 2. Plug in MMC/SD card 2. Look in /dev Actual Results: No mention of MMC/SD device in /dev Expected Results: MMC/SD device should appear in /dev. Kernel config files: Working: http://www.smiffysplace.com/files/kernel_config_2.6.19_20070316 Non-working: http://www.smiffysplace.com/files/kernel_config_2.6.20_20070316 Dmesg output: http://www.smiffysplace.com/files/dmesg_20070316 The latter lines show me inserting and removing the card a couple of times, with 2.6.20-gentoo-r2, nothing appearing in /dev.
Please, attach all the files here; third party links render bugs useless in days.
Created attachment 113442 [details] Kernel config, 2.6.19
Created attachment 113443 [details] Kernel config, 2.6.20
Created attachment 113445 [details] Dmesg output
Files attached as requested.
Which device node are you looking for? You seem to have both sdhci and tifm hardware. Do you know which one you are trying to use? Can you boot 2.6.19, and run "dmesg". Then plug a card in, ensure the device node is created, run "dmesg" again and paste here which messages have been added to the end since you plugged the card in. Then do the same for 2.6.20.
(In reply to comment #6) > Which device node are you looking for? These are what appear with 2.6.19: brw-rw---- 1 root disk 254, 0 Mar 23 15:50 mmcblk0 brw-rw---- 1 root disk 254, 1 Mar 23 15:50 mmcblk0p1 > You seem to have both sdhci and tifm hardware. Do you know which one you are > trying to use? Sorry, I wouldn't know. I just enabled it in the kernel and it just worked in 2.6.19. > Can you boot 2.6.19, and run "dmesg". Then plug a card in, ensure the device > node is created, run "dmesg" again and paste here which messages have been > added to the end since you plugged the card in. Then do the same for 2.6.20. I have done this, diff'ing the results, and am submitting attachments dmesg_19.diff and dmesg_20.diff, respectively.
Created attachment 114308 [details] Booted to 2.6.19, difference between dmesg without card and dmesg with card. Note: Although this shows errors, card can be mounted and accessed without problems.
Created attachment 114309 [details] Booted to 2.6.20, difference between dmesg without card and dmesg with card.
Nothing obvious there. Can you reproduce this on the latest development kernel, currently 2.6.21-rc4?
(In reply to comment #10) > Nothing obvious there. Can you reproduce this on the latest development kernel, > currently 2.6.21-rc4? No gentoo-sources package available for amd64 post 2.6.20-r3. Looks like I'll have to pull down source from kernel.org to do this, so may take a while to get it configured and working on this laptop. I will report back when I have been able to do this.
gentoo-sources doesn't do -rc releases but switching to a -rc is generally easy as gentoo-sources doesn't include much. emerge -n ketchup cd /usr/src mkdir -p linux-2.6.21-rc4 cd linux-2.6.21-rc4 ketchup 2.6.21-rc4
Thanks for the guidance on ketchup. I have now built 2.6.21-rc4, the problem persists: no device nodes are created when the card is inserted. Full dmesg and post-insertion diff to be attached.
Created attachment 114318 [details] 2.6.21-rc4 dmesg, card not inserted
Created attachment 114320 [details] diff between dmesg before and after card insertion, kernel 2.6.21-rc4
Thanks for the quick testing. The next step is to file a bug upstream against the tifm driver at http://bugzilla.kernel.org. Mention that 2.6.19 works, 2.6.20 breaks, and 2.6.21-rc4 is still broken. As this is a regression you can track down the exact buggy commit which introduced the bug, however it is a fairly time consuming process. If you're keen, see http://www.reactivated.net/weblog/archives/2006/01/using-git-bisect-to-find-buggy-kernel-patches/ You might want to wait for 2-3 days after filing the bug before going down this route, in case it is not needed.
please post the new bug URL here when you have filed it upstream
I did a quick check on the kernel Bugzilla before opening a new one and found one open (but nothing happening) which I guess is the same. I have, therefore, posted against: http://bugzilla.kernel.org/show_bug.cgi?id=8052 Thanks for your help, Daniel, I just hope someone up there knows how to fix this otherwise I'm going to be trying to shoehorn 2.6.19 code back into the new kernel.
Your 2.6.21 logs show that mmcblk0 and mmcblk0p1 are being detected, although there are a fair number of errors too. Do those device nodes really not appear under 2.6.21 or has the problem changed with this kernel release?
(In reply to comment #19) > Your 2.6.21 logs show that mmcblk0 and mmcblk0p1 are being detected, although > there are a fair number of errors too. Do those device nodes really not appear > under 2.6.21 or has the problem changed with this kernel release? I have now re-emerged all necessary packages so that I can run 2.6.21-r4 as my default, to save all the re-booting. Whilst dmesg shows mmc stuff when card inserted (it did on 2.6.20, too), I can confirm that nodes are still NOT found in /dev.
Created attachment 117456 [details, diff] tifm v0.8d for 2.6.21 Please test gentoo-sources-2.6.21. If it's still broken there, apply this patch on top of gentoo-sources-2.6.21 and see if it helps.
Device node is not created with linux-2.6.21-gentoo either before or after patching. Eject and insert events can be seen on dmesg - this is very similar to the way it was on 2.6.20 prior to using the external module build from the bug on kernel.org
Strange, I thought that patch should bring it in line with the external version. I'll check this soon. In the mean time, could you double check that you applied the patch correctly and installed the new kernel/module successfully? If you compile the driver as a module, you can check with "modinfo /lib/modules/<VERSION>/kernel/drivers/mmc/tifm_sd.ko" -- the version field should read 0.8, not 0.6 If you compile the driver into the kernel, check the last modified time of drivers/mmc/tifm_sd.c in the kernel directory -- it should be equal to when you applied the patch. If so, while booted into the kernel you believe is patched, run "uname -v". The value there is the time/date that kernel was compiled -- mentally check this is when you recompiled the kernel, and also check that it is newer than the modified time for the tifm_sd.c file.
Yep, this is version 0.8. Just for reference, here's the modinfo for all in that directory: filename: /lib/modules/2.6.21-gentoo/kernel/drivers/mmc/mmc_block.ko license: GPL description: Multimedia Card (MMC) block device driver vermagic: 2.6.21-gentoo SMP mod_unload depends: mmc_core parm: major:specify the major device number for MMC block driver (int) filename: /lib/modules/2.6.21-gentoo/kernel/drivers/mmc/mmc_core.ko license: GPL vermagic: 2.6.21-gentoo SMP mod_unload depends: filename: /lib/modules/2.6.21-gentoo/kernel/drivers/mmc/tifm_sd.ko author: Alex Dubov description: TI FlashMedia SD driver license: GPL version: 0.8 vermagic: 2.6.21-gentoo SMP mod_unload depends: mmc_core,tifm_core srcversion: C11C3EBB8F4E8ECA33554AC parm: fixed_timeout:bool parm: no_dma:bool And this is the one that lurks in ../misc: filename: /lib/modules/2.6.21-gentoo/kernel/drivers/misc/tifm_7xx1.ko author: Alex Dubov description: TI FlashMedia host driver license: GPL version: 0.8 vermagic: 2.6.21-gentoo SMP mod_unload depends: tifm_core alias: pci:v0000104Cd00008033sv*sd*bc*sc*i* alias: pci:v0000104Cd0000803Bsv*sd*bc*sc*i* alias: pci:v0000104Cd0000AC8Fsv*sd*bc*sc*i* srcversion: D736152CA27D661613954A9 filename: /lib/modules/2.6.21-gentoo/kernel/drivers/misc/tifm_core.ko license: GPL author: Alex Dubov description: TI FlashMedia core driver license: GPL version: 0.8 vermagic: 2.6.21-gentoo SMP mod_unload depends: srcversion: BE1DF8B52CD61BF90C0A9A2
Created attachment 117883 [details] diff of comment 21 patch against upstream .08d patched 2.6.21 (with patch from comment #21) and then diff'ed them against tifm_7xx1.c tifm_core.c tifm_sd.c and tifm.h from external version 0.8d from upstream
Created attachment 117885 [details, diff] Patch for 2.6.21 to upstream version 08d Smiffy, Can you please apply this patch to your 2.6.21 source and test? Thanks, Mike
Created attachment 117890 [details, diff] Patch for 2.6.21 to upstream version 08d Smiffy, Can you please ignore the last patch and instead apply this patch to your 2.6.21 source and test? Thanks, Mike
Created attachment 117908 [details, diff] Patch for 2.6.21 to upstream version 08d This one can be properly applied with -p1
I applied the patch to a fresh emerge by entering /usr/src/linux-2.6.21-gentoo and running: patch -p1 tifm-update-to-0.8d.patch This reported that the files had been patched successfully. However, when I came to make the kernel, I got the following error: CC [M] drivers/misc/tifm_core.o CC [M] drivers/misc/tifm_7xx1.o CC [M] drivers/mmc/mmc.o CC [M] drivers/mmc/mmc_sysfs.o CC [M] drivers/mmc/mmc_queue.o LD [M] drivers/mmc/mmc_core.o CC [M] drivers/mmc/mmc_block.o CC [M] drivers/mmc/tifm_sd.o drivers/mmc/tifm_sd.c: In function 'tifm_sd_probe': drivers/mmc/tifm_sd.c:987: error: 'struct mmc_host' has no member named 'max_sectors' make[2]: *** [drivers/mmc/tifm_sd.o] Error 1 make[1]: *** [drivers/mmc] Error 2 make: *** [drivers] Error 2 I will attach my .config, in case this has any bearing on the matter, although I'm sure that I have seen this error before, when I tried to apply the 2.6.20 patch to 2.6.21. (I may be wrong, but I have seen it before somewhere.)
Created attachment 117931 [details] Kernel .config to go with report on latest compliation attempt. Please see Comment #29: http://bugs.gentoo.org/show_bug.cgi?id=171123#c29
thanks for the .config It also failed for me. I am going to try applying this patch to 2.6.21.1 and see what happens.
Created attachment 117948 [details, diff] Patch for 2.6.21.1 to upstream version 08e Here's the patch to upstream version 0.8e This does compile when applied to 2.6.21.1, but not 2.6.21
Will 2.6.21.1 be released as a 2.6.21-gentoo-rx ebuild? If so, would this patch be already applied?
Smitty, I was incorrect when I said the version 08e patch doesn't compile against linux-2.6.21-gentoo. It did compile for me. Can you apply the patch "Patch for 2.6.21.1 to upstream version 08e" to linux-2.6.21, and test the result? Thanks
gentoo-sources now re-emerged and patched with tifm-update-to-0.8e.patch The kernel build was successful, but the device node still does not appear. Modules appear to be loaded (from lsmod): tifm_sd 13320 0 mmc_core 27208 1 tifm_sd tifm_7xx1 8576 0 tifm_core 11144 2 tifm_sd,tifm_7xx1 However, if I run modinfo, version only comes back as 0.8 - is this correct or should it say 0.8e?
That's correct. The plot thickens... Can you revert that patch, remove tifm from your kernel config, remove the modules from /lib/modules, and then install and test the external tifm-0.8e against 2.6.21? Direct link to download is http://prdownload.berlios.de/tifmxx/tifm-0.8e.tar.bz2 Basically we need to try and figure out what's going on.. the patch I posted (against 2.6.20) should have brought things just about inline with the external 0.8d (which you said worked fine), and Mike's patch above did similar, so its a bit of a mystery why these patches aren't working if the external versions are...
External module for 08e works fine, just like in the case of 2.6.20 and its external module. Whether it is relevant or not, the Makefile is broken in that it does not create /lib/modules/2.6.21-gentoo/kernel/drivers/misc/ and /lib/modules/2.6.21-gentoo/kernel/drivers/mmc which breaks make install. I have to create these by hand first, then run make install.
I've also been having issues with tifm (mainly buffer and I/O errors in dmesg, corrupt files, etc). I tried applying the 2.6.21.1 patch posted by Mike to gentoo-sources-2.6.21 and it has fixed my issue. If this could get into gentoo-sources-2.6.21-r1, that would be great. (It does appear that they are planning to merge it into 2.6.22, and I think tifm 0.8 is in the kernel git tree, I'd have to check though).
(In reply to comment #38) > I've also been having issues with tifm (mainly buffer and I/O errors in dmesg, > corrupt files, etc). I tried applying the 2.6.21.1 patch posted by Mike to > gentoo-sources-2.6.21 and it has fixed my issue. If this could get into > gentoo-sources-2.6.21-r1, that would be great. (It does appear that they are > planning to merge it into 2.6.22, and I think tifm 0.8 is in the kernel git > tree, I'd have to check though). > Err, I should clarify which patch I'm talking about: Patch for 2.6.21.1 to upstream version 08e (attachment #117948 [details, diff])
It seems that tifm 0.8e has been merged into 2.6.22-rc1. I gave it a try here, and it seems to have resolved the issue.
Smiffy, any chance you could test 2.6.22-rc1?
Yes, vanilla kernel with the appropriate options selected
I should be able to test 2.6.22-rc1 in the next couple of days - will advise results.
I've started the configuration of the new kernel, but there appears to be a problem in make menuconfig: <*> MMC/SD card support ---> This does not expand to offer the driver options as it does in the previous kernels, so I am unable to fulfill: "If you want MMC support, you should say Y here and also to the specific driver for your MMC interface."
Ignore previous, this has been hidden under Misc Devices for some reason.
(In reply to comment #45) > Ignore previous, this has been hidden under Misc Devices for some reason. > I think it was in Misc devices in 2.6.19 as well ;-)
OK, firstly, 2.6.22-rc1 works. However, to get it to work, I had to modprobe something called mmc_block, which I had not had to do before. (I only found it whilst I was having a nose around /lib/modules/----/kernel/drivers/; I modprobed it to see what it did and all of a sudden it worked...) Having discovered this, I rebooted to my (patched) 2.6.21-gentoo, did the same modprobe and found it worked there too. Please note, I did not have to do this modprobe with the external module in 2.6.21 or 2.6.20. Could this be the missing link? NOTE: make menuconfig on 2.6.22-rc1 doesn't set CONFIG_MMC_TIFM_SD - don't know if this is another parameter hidden in some daft place or whether it has been missed. I had to go back and set this manually to build the module and get this working. @nick - there were three things I spotted that had been moved around menuconfig; made it a helluva job to copy my .21 to my .22 configuration. Something's also been broken in the SATA AHCI driver so that I have to change the BIOS if I want to swap between 2.6.22 and Solaris (2.6.21 and Solaris, no problems). When someone puts GNU+Solaris kernel+Portage, I'll be there like a shot ;-)
Just a quick question, what do you have to change in the BIOS when you switch between 2.6.22 and Solaris? (I'm just kind of curious)
Comment #48 answered at length, off-list.
OK, thanks for the testing. Time to backport the patches from 2.6.22. This won't make it into the gentoo-sources-2.6.21 release being generated right now, but hopefully will make it into the next one.
I have just upgraded to kernel 2.6.21-gentoo-r1 (x86_64) and have applied patch tifm-update-to-0.8e.patch. I can confirm that my card reader is WORKING.
Fixed in gentoo-sources-2.6.21-r2 (genpatches-2.6.21-3), thanks for testing.
