On Wed, 2013-10-02 at 06:31 +0200, Thomas Glanzmann wrote: > Hello Nab, > > > Thomas, these are all fairly obvious fixes to me, and I've been able > > to confirm them on my setup. Please confirm on your end, and I'll > > plan to push them for v3.12-rc4 by the end of the week. > > I confirm that the fixes work. Thank you for fixing this. I did the > following: > > - Applied the patches on top of your for-next, recompiled, > installed modules and kernel got rid of > '/lib/modules/3.11.0-rc5+/extra/' which contained target > modules with an unknown symbol __dynamic_pr_debug (from the > debug build, I assume) > > - Started the target, discovered from two esx servers, created > VMFS (no error), deployed one VM, used XCOPY to clone it three > times, powered it on and did 4 svmotion in parallel. > > So the fix works. I also have the feeling that the xcopy code became > faster, but I first have to meassure it. >
Great, thanks for the confirmation here. > I'll today some more testing with rescans from 12 ESX and svmotion > (XCOPY) of 24 VMs in parallel. But everything is looking good from my > site. I'm also working on the discovery patch. > There is one other regression wrt to the struct iscsi_cmd pre-allocation changes in v3.12-rc1 code that I'm still tracking down, that you may end up encountering during stress testing. It's currently caused by tag starvation under heavy (remotely generated) I/O load, and manifests itself as percpu_ida_alloc() sleeping indefinitely in iscsit_allocate_cmd() waiting for new tag(s) to become available while StatSN acknowledgement is (AFAICT) being blocked elsewhere. FYI, the output of 'echo w > /proc/sysrq-trigger', and/or the hung task checker should produce a stack-trace to this effect. Please let me know if you encounter any issues, and a patch to address this bit should be along in the next 24-48 hours. --nab -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
