Re: [Xen-ia64-devel] Re: [patch] ia64: kexec: add __va_efi
On Sat, 2008-03-15 at 09:08 +0900, Simon Horman wrote: Patch 13/15 seems nasty, so I think that ought to be left out. But the other patches that I suggested omitting seem harmless enough. I tested again with patches 1-12,14,15,__va_efi. This boots on all my systems! I guess 13 should indeed be dropped or fixed, but I'm not seeing the hang after freeing init mem issue it supposedly fixes. Thanks, Alex -- Alex Williamson HP Open Source Linux Org. ___ Xen-ia64-devel mailing list Xen-ia64-devel@lists.xensource.com http://lists.xensource.com/xen-ia64-devel
Re: [Xen-ia64-devel] Re: [patch] ia64: kexec: add __va_efi
On Mon, Mar 17, 2008 at 10:11:16AM -0600, Alex Williamson wrote: On Sat, 2008-03-15 at 09:08 +0900, Simon Horman wrote: Patch 13/15 seems nasty, so I think that ought to be left out. But the other patches that I suggested omitting seem harmless enough. I tested again with patches 1-12,14,15,__va_efi. This boots on all my systems! I guess 13 should indeed be dropped or fixed, but I'm not seeing the hang after freeing init mem issue it supposedly fixes. Hi Alex, that is great news! I think that patch 13 is an error (on my part) and it should be dropped. Can you confirm that 1-12,14,15,__va_efi also works on superdome? You mentioned some PAL problems in your previous email. Are these still a problem? I have spent a bit of time poking around to see extra places where __va_efi() may be needed. I have found a bunch of places were we pass in __va() addresses into pal/sal/efi calls, but I think that these are all safe as the appropriate fault handlers are in place. In any case, I will poke around a bit more. -- Horms ___ Xen-ia64-devel mailing list Xen-ia64-devel@lists.xensource.com http://lists.xensource.com/xen-ia64-devel
Re: [Xen-ia64-devel] Re: [patch] ia64: kexec: add __va_efi
Hi Simon, On Tue, 2008-03-18 at 11:32 +0900, Simon Horman wrote: that is great news! I think that patch 13 is an error (on my part) and it should be dropped. Can you confirm that 1-12,14,15,__va_efi also works on superdome? You mentioned some PAL problems in your previous email. Are these still a problem? Yes, I tested these on the superdome and it booted. I instrumented the PAL calls, not expecting dropping patch 13 to solve the problem, so I should probably go back and remove the instrumentation before declaring total victory. I can't speak to whether kexec actually works anywhere, but I think a kexec enabled xen is at least not a regression now. I have spent a bit of time poking around to see extra places where __va_efi() may be needed. I have found a bunch of places were we pass in __va() addresses into pal/sal/efi calls, but I think that these are all safe as the appropriate fault handlers are in place. In any case, I will poke around a bit more. Great. Thanks, Alex -- Alex Williamson HP Open Source Linux Org. ___ Xen-ia64-devel mailing list Xen-ia64-devel@lists.xensource.com http://lists.xensource.com/xen-ia64-devel
Re: [Xen-ia64-devel] Re: [patch] ia64: kexec: add __va_efi
On Mon, Mar 17, 2008 at 09:17:18PM -0600, Alex Williamson wrote: Hi Simon, On Tue, 2008-03-18 at 11:32 +0900, Simon Horman wrote: that is great news! I think that patch 13 is an error (on my part) and it should be dropped. Can you confirm that 1-12,14,15,__va_efi also works on superdome? You mentioned some PAL problems in your previous email. Are these still a problem? Yes, I tested these on the superdome and it booted. I instrumented the PAL calls, not expecting dropping patch 13 to solve the problem, so I should probably go back and remove the instrumentation before declaring total victory. I can't speak to whether kexec actually works anywhere, but I think a kexec enabled xen is at least not a regression now. Great. As you say, a working kexec is a slightly different issue. And one that should be easier to tackle in terms of test coverage if the code is merged. So non-regressing kexec enabled xen is certainly my goal for now. Though I will add that on my hw kexec is working rather well these days. -- Horms ___ Xen-ia64-devel mailing list Xen-ia64-devel@lists.xensource.com http://lists.xensource.com/xen-ia64-devel
Re: [Xen-ia64-devel] Re: [patch] ia64: kexec: add __va_efi
On Thu, 2008-03-13 at 21:03 -0600, Alex Williamson wrote: On Fri, 2008-03-14 at 10:51 +0900, Simon Horman wrote: On an HP rx2600 this eliminates the problem where the SAL rendezvous address can't be registered and subsequently the boot fails. I suspect it solves similar problems that have been seen on other HP machines too. Actually, its rather amazing that the HP rx2620 and Tiger2 that I have been using work without this fix. Wow, that is rather surprising. Congratulations on tracking this down. I'll try to verify if this also solves the problem on superdome tomorrow. Thanks, Hi Simon, Good news/bad news on the superdome. It was definitely affected by the issue you solved on the rx2600 and no longer hangs at boot. However, PAL calls are still broken. It does make it all the way up to dom0 userspace before it falls over completely though. If you have any test code I'd be happy to try it, otherwise I'll see if I can find a system I can share with you the exhibits the problem. FWIW, I didn't prune down the patch series for this test, I just added this on top of the previous 15 patches. Are you as concerned as I am about how many other machine specific issues might still be hidden in this code? Thanks, Alex -- Alex Williamson HP Open Source Linux Org. ___ Xen-ia64-devel mailing list Xen-ia64-devel@lists.xensource.com http://lists.xensource.com/xen-ia64-devel
Re: [Xen-ia64-devel] Re: [patch] ia64: kexec: add __va_efi
On Fri, Mar 14, 2008 at 02:42:55PM -0600, Alex Williamson wrote: On Thu, 2008-03-13 at 21:03 -0600, Alex Williamson wrote: On Fri, 2008-03-14 at 10:51 +0900, Simon Horman wrote: On an HP rx2600 this eliminates the problem where the SAL rendezvous address can't be registered and subsequently the boot fails. I suspect it solves similar problems that have been seen on other HP machines too. Actually, its rather amazing that the HP rx2620 and Tiger2 that I have been using work without this fix. Wow, that is rather surprising. Congratulations on tracking this down. I'll try to verify if this also solves the problem on superdome tomorrow. Thanks, Hi Alex, Thanks, it was certainly a moment of insight when I saw that. Hi Simon, Good news/bad news on the superdome. It was definitely affected by the issue you solved on the rx2600 and no longer hangs at boot. However, PAL calls are still broken. It does make it all the way up to dom0 userspace before it falls over completely though. If you have any test code I'd be happy to try it, otherwise I'll see if I can find a system I can share with you the exhibits the problem. FWIW, I didn't prune down the patch series for this test, I just added this on top of the previous 15 patches. Are you as concerned as I am about how many other machine specific issues might still be hidden in this code? Thanks, Patch 13/15 seems nasty, so I think that ought to be left out. But the other patches that I suggested omitting seem harmless enough. I am concerned that there might be other areas where __va_efi() needs to be used and I intend to instrument and inspect that shortly - Friday end up being meetings all afternoon and believe it or not the keyboard of my laptop broke, so I wasn't able to do anything more after sending my previous email. As for machine specific issues. Yes, I am very concerned about that. I had it in mind to try and get some good test coverage - somehow. Some auditing - if someone has the patience and time might help - it seems that these bugs often turn out to be well how on earth was that working at all? cases, like one. With regards to the superdome in particular, its good to hear that there is progress :-) Though not particularly surprising that there is another hurdle. Is it worth me going through the logs that you sent last time? If not, as you probably know I'm not real good with working things out from logs, but it might be worth sending me some and giving it a shot. I think its best to attack these PAL problems, certainly things can't be healthy while they remain. And they may lead to all manner of problems that manifest elsewhere in the boot. -- Horms ___ Xen-ia64-devel mailing list Xen-ia64-devel@lists.xensource.com http://lists.xensource.com/xen-ia64-devel
[Xen-ia64-devel] Re: [patch] ia64: kexec: add __va_efi
On Fri, 2008-03-14 at 10:51 +0900, Simon Horman wrote: On an HP rx2600 this eliminates the problem where the SAL rendezvous address can't be registered and subsequently the boot fails. I suspect it solves similar problems that have been seen on other HP machines too. Actually, its rather amazing that the HP rx2620 and Tiger2 that I have been using work without this fix. Wow, that is rather surprising. Congratulations on tracking this down. I'll try to verify if this also solves the problem on superdome tomorrow. Thanks, Alex -- Alex Williamson HP Open Source Linux Org. ___ Xen-ia64-devel mailing list Xen-ia64-devel@lists.xensource.com http://lists.xensource.com/xen-ia64-devel