Hi Tejun, JFYI. I found a workaround which seems to be working in 2.6.18-rc2. During warm plug/unplug, if I don't delete the device through sysfs or proc and just rescan the host, it recovered just fine. Thanks, Fajun On 8/24/06, Fajun Chen <fajunchen@xxxxxxxxx> wrote:
Hi Tejun, I tested 2.6.18-rc4-mm2 on i386 platform with and without your patch. The behavior is the same as rc2 with your patch. No kernel error and the only relevant dmesg traces are: [ 125.048966] ata5: soft resetting port [ 125.202410] ata5.00: configured for UDMA/100 [ 125.202492] ata5: EH complete [End of dmesg] It did not proceed to add/attach sg device! In rc4-mm2 release without your patch, I forgot to power on pata drive and add drive will cause kernel error. Please see attachment for detailed log. I can reproduce the problem easily, please feel free to let me know what kind of log/traces can be collected. Thanks, Fajun On 8/24/06, Tejun Heo <htejun@xxxxxxxxx> wrote: > [cc'ing linux-ide again] > > Fajun Chen wrote: > > Hi Tejun, > > > > Your patch is successful at fixing the kernel error when adding pata > > drive on 2.6.18-rc2 but there's some sg availability issue. I did the > > same sequence of test: power off drive -> remove drive -> power on > > drive -> add drive. From dmesg log, pata drive was added back > > successfully as sg0 but somehow it is unavailable for upper level. For > > instance, sg_scan didn't detect it. It is not shown in > > /sys/class/scsi_generic/. Sg utilities sg_inq and sg_dd complained > > about no such device for sg0. Is there any magic command I need to > > issue to populate sg information up to the user space? If not, then > > there is still bug somewhere. Complete log of the test sequence is > > attached. > > > > I'll hold off the testing on 2.6.18-rc4 for now. Please let me know if > > you think it still makes sense to test your patch in the new release. > > Thanks for your help! > > The thing is that the patch shouldn't really change behavior. With the > patch, libata might be more forgiving about SCSI refcnt bugs but the > code before and after are identical correctness-wise. ISTR there has > been SCSI ref bug fixed sometime inbetween. I can't pinpoint it though. > So, please try 2.5.18-rc4 w/ or w/o the patch. Pull > libata-dev#pata-drivers, that's basically current linus #master + > libata-dev#upstream updates + Alan's pata-drivers. > > Of course, I might be wrong, but I kind of have too little lead to > follow on at the moment, especially because I can't reproduce it here. > If testing new kernel on your target platform takes a lot of effort, > just perform preliminary tests on x86. As far as SCSI and libata are > concerned, there's little difference. > > -- > tejun >
- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
