Bug 618862
Description
Jon Masters
2010-07-27 21:37:28 UTC
Can you add loglevel=debug, run through the driver disk loading code, get dumped back to the "No driver found" dialog, press ctrl-Z to get dropped to a shell, and attach all the log files out of /tmp to this bug report? Created attachment 435471 [details]
/tmp/ dir loglevel=debug
Created attachment 435480 [details]
/tmp/ tarball with loglevel=7
Created attachment 435481 [details]
/tmp/ dir without loglevel specified.
Created attachment 435490 [details]
/tmp/ tarball with loglevel=debug
Hi Chirs, I have attached /tmp/ tarball, with and without loglevel specified. I found something: Whether the loglevel parameter is added or not, the anaconda.log always have the DEBUG log output. Any idea on this? Best, Chao We always record to the files at DEBUG, but change what gets logged to the console. That way we can see everything in bug reports but users don't get potentially confusing debugging messages. In other words, don't worry about it. Can you reproduce this problem with a module other than sym53c8xx.ko? If you look at the syslog, you'll see the following: <4>WARNING: at /builddir/build/BUILD/kernel-2.6.32/linux-2.6.32.x86_64/arch/x86/include/asm/dma-mapping.h:154 ___free_dma_mem_cluster+0x102/0x110 [sym53c8xx]() (Not tainted) <4>Hardware name: KVM <4>Modules linked in: jbd mbcache cdrom crc_t10dif sym53c8xx(-) scsi_transport_spi virtio_pci virtio_ring virtio pata_acpi ata_generic ata_piix ipv6 iscsi_ibft pcspkr edd floppy iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi squashfs cramfs [last unloaded: sd_mod] <4>Pid: 359, comm: rmmod Not tainted 2.6.32-52.el6.x86_64 #1 <4>Call Trace: <4> [<ffffffff8106b473>] warn_slowpath_common+0x83/0xc0 <4> [<ffffffff8106b4c4>] warn_slowpath_null+0x14/0x20 <4> [<ffffffffa0138af2>] ___free_dma_mem_cluster+0x102/0x110 [sym53c8xx] <4> [<ffffffffa01388d2>] __sym_mfree+0xd2/0x100 [sym53c8xx] <4> [<ffffffffa0138969>] __sym_mfree_dma+0x69/0xf0 [sym53c8xx] <4> [<ffffffffa01310d5>] sym_hcb_free+0x65/0x1f0 [sym53c8xx] <4> [<ffffffffa012e936>] sym_free_resources+0x46/0x90 [sym53c8xx] <4> [<ffffffffa012ea33>] sym_detach+0xb3/0xe0 [sym53c8xx] <4> [<ffffffffa012ea9b>] sym2_remove+0x3b/0x60 [sym53c8xx] <4> [<ffffffff81272be7>] pci_device_remove+0x37/0x70 <4> [<ffffffff81332c1f>] __device_release_driver+0x6f/0xe0 <4> [<ffffffff81332d58>] driver_detach+0xc8/0xd0 <4> [<ffffffff81331abe>] bus_remove_driver+0x8e/0x110 <4> [<ffffffff81333522>] driver_unregister+0x62/0xa0 <4> [<ffffffff81272ef4>] pci_unregister_driver+0x44/0xb0 <4> [<ffffffffa0139975>] sym2_exit+0x15/0x47 [sym53c8xx] <4> [<ffffffff810acbe6>] sys_delete_module+0x1a6/0x280 <4> [<ffffffff81013172>] system_call_fastpath+0x16/0x1b <4>---[ end trace 9955f74768a9ca36 ]--- I wonder if the end result of that trace is that the rmmod of the original sym53c8xx.ko from the initrd fails, so the module doesn't get unloaded, so we don't even try to load the new one. Chao Yang, have you had a chance to try this problem with a module other than sym53c8xx.ko? Hi Denise, Sure. I will try another driver, and let you know ASAP. Best, Chao Let's get this tested today. I'll spin another Driver Disk shortly with a SATA or virtIO or other storage driver in it. Jon. Created attachment 437759 [details]
console output from test session showing driver was loaded but not noticed
Created attachment 437762 [details]
2.6.32-59.1 x86_64 driver disk containing sym53c8xx version sym-2.2.3.rhtest60s10
Created attachment 437768 [details]
2.6.32-59.1 x86_64 driver disk containing virtio_blk from 2.6.32-61
Exactly the same behavior with virtio_blk. The updated driver is loaded (as witnessed by the obvious kernel debug message I inserted), but Anaconda behaves exactly as before. I have attached two driver disks for two different drivers - the virtio one will work out of the box with KVM, the SCSI one only requires you swap the virtio for a SCSI device - and think this is Anaconda's again issue now. I should have some more time in a couple of days, I have other stuff aswell. Jon. Created attachment 437774 [details]
console output from test session showing driver was loaded but not noticed (virtio_blk)
All of this is using 20100809.0 compose, so here you have two drivers - one for SCSI and one for virtio - and both of them load and work, but are not detected by Anaconda. Please keep me in the loop. Thanks! Comment 25 contains the wrong log file. I am attaching the right one. NOTE: A bug in virt-manager resulted in the deletion of the SCSI disk and swapping back to virtio still leaving also a SCSI device, which is why the sym53c8xx driver was still loaded and caused a backtrace in that example too. But in the corrected log file you will see there is no kernel warning when unloading the modules. The warning is not directly related to the issue here and the modules do function correctly in any case. Anyway, now we have proof that running with virtio_blk (totally different driver) behaves exactly the same way. Created attachment 437779 [details]
2.6.32-59.1 x86_64 driver disk containing virtio_blk from 2.6.32-61
The attached /tmp tarballs do not include /tmp/storage.log, which is critical. Dave: can you not trivially reproduce this for yourself now? I've attached two driver disks that work under KVM, the virtio_blk one works "out of the box" with any RHEL6 virtual machine you create in virt-manager... Jon. Additionally, in those examples, the kmod was not installed on the target. My guess is Anaconda decides it didn't load anything so shouldn't install it. The disk is detected and the driver is loaded during install, but the installation aborts with an error due to a python regex type issue: TypeError: unsupported operand type(s) for +: '_sre.SRE_Pattern' and 'str' </binding><binding fileName="/tmp/anaconda-tb-qvxy0m" name="pythonUnhandledException" type="text">anaconda 13.21.76 exception report Traceback (most recent call first): File "/usr/lib/anaconda/backend.py", line 77, in copyFirmware for f in glob.glob(DD_EXTRACTED+"/lib/firmware/*"): File "/usr/lib/anaconda/backend.py", line 98, in doPostInstall self.copyFirmware(anaconda) This likely should have come up in your testing. I am attaching the full log, along with updated copies of the Driver Update Disks built against kernel 2.6.32-63.el6.x86_64. The test installation in this case was performed using the virtio_blk driver on a stock KVM guest with virtio (default) and 20100815 x86_64, minimal server install using text mode. i.e. a completely default KVM guest. To reproduce, create a KVM guest (F13 used here), use the driver disk. Boom. Jon. Created attachment 438898 [details]
2.6.32-63 x86_64 driver disk containing sym53c8xx version sym-2.2.3.rhtest60s10
Created attachment 438899 [details]
2.6.32-63 x86_64 driver disk containing virtio_blk from 2.6.32-63
Created attachment 438900 [details]
anaconda log file from aborted test installation with 20100815 x86_64 Server nightly compose
Please let me know when there's another anaconda build to test. Thanks. During a default beaker install of RHEL6.0-20100816.n.0_nfs-Server-x86_64, I hit the traceback reported in comment #35. Not sure if this should be opened up in a new BZ or not. Performing post-installation configuration anaconda 13.21.76 exception report Traceback (most recent call first): File "/usr/lib/anaconda/backend.py", line 77, in copyFirmware for f in glob.glob(DD_EXTRACTED+"/lib/firmware/*"): File "/usr/lib/anaconda/backend.py", line 98, in doPostInstall self.copyFirmware(anaconda) File "/usr/lib/anaconda/yuminstall.py", line 1829, in doPostInstall AnacondaBackend.doPostInstall(self, anaconda) File "/usr/lib/anaconda/backend.py", line 299, in doPostInstall anaconda.backend.doPostInstall(anaconda) File "/usr/lib/anaconda/dispatch.py", line 208, in moveStep rc = stepFunc(self.anaconda) File "/usr/lib/anaconda/dispatch.py", line 126, in gotoNext self.moveStep() File "/usr/lib/anaconda/cmdline.py", line 187, in run anaconda.dispatch.gotoNext() File "/usr/bin/anaconda", line 1115, in <module> anaconda.intf.run(anaconda) TypeError: unsupported operand type(s) for +: '_sre.SRE_Pattern' and 'str' install exited abnormally [1/1] The system will be rebooted when you press Ctrl-C or Ctrl-Alt-Delete. *** Bug 623546 has been marked as a duplicate of this bug. *** Jon -- Looks like an updated anaconda just landed (anaconda-13.21.77-1). Yes, and it will be tested with the nightly from 20100817 once it's available. Using an "updates" image is nice, but it's not the same as an actual test. We need to know that actual media tests work, all the way through, out of the box :) TEST RESULTS ------------ The updates image supplied by Martin does get through the install, HOWEVER, the kmod is not installed on the finish system and the updated driver is not used: $ rpm -qa | grep kmod <nothing> This urgently requires fixing. Please investigate why the kmod RPM is not targeted for installation, and perform any necessary testing. Thanks for initial fix - we're half way there :) Jon. Created attachment 439101 [details]
Screenshot of the installed system with fixed anaconda
Ok, there was another problem with matching the module name to proper provide symbol during the install. That should be fixed now too.
My logs show, that the module got installed.
So expect this to hit anaconda-13.21.78-1 tonight. I can provide updates image before that.
If you're online, please do provide an updates image. Otherwise, I will test the moment a compose is available to do so. This testing is still blocking on a full 20100818 compose being available. Hopefully in a few hours, since that is being worked on. Provisional testing of 20100818 shows that driver disks are now working, under virtualized systems, with (in this example) a virtio_blk driver correctly installing, with the RPMs on the virtual disk and so forth. I am able to upgrade the kernel from -63 to -64 and the driver is picked up by the upgraded kernel, which boots correctly with the new driver. HOWEVER: We still need to do some bare metal testing, to be sure. *** Bug 625322 has been marked as a duplicate of this bug. *** Currently have a graphical install running too. CONFIRMED with graphical install. Bare metal tests will commence. My testing is done and I can verify that this is working on bare metal (laptop with replacement network drivers for wired and wireless in this test). moving to verified based on comment 59 Thanks Red Hat Enterprise Linux 6.0 is now available and should resolve the problem described in this bug report. This report is therefore being closed with a resolution of CURRENTRELEASE. You may reopen this bug report if the solution does not work for you. |