Attached patch fixes FP initialization bug with x86-64 xenlinux. With this patch, # of testcase failures with LTP is < 5 (out of 800), almost same results as the native x86-64 Linux. The other two patches for LTP are coming soon; with these, all the testcases will pass (except the ones that fail on the native x86-64 Linux). Jun --- Intel Open Source Technology Center _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Nakajima, Jun wrote:> Attached patch fixes FP initialization bug with x86-64 xenlinux. >The patch is attached now.> With this patch, # of testcase failures with LTP is < 5 (out of 800), > almost same results as the native x86-64 Linux. The other two patches > for LTP are coming soon; with these, all the testcases will pass > (except the ones that fail on the native x86-64 Linux). > > JunJun --- Intel Open Source Technology Center _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On 6 Jun 2005, at 08:02, Nakajima, Jun wrote:>> With this patch, # of testcase failures with LTP is < 5 (out of 800), >> almost same results as the native x86-64 Linux. The other two patches >> for LTP are coming soon; with these, all the testcases will pass >> (except the ones that fail on the native x86-64 Linux).Consider defining read/write_cr0 macros same as we do in arch/xen/i386. We set them the same as native (i.e., real accesses to %cr0) and then emulate in Xen. This will ensure that things like raid6 will work, which save/restore %cr0 as part of their fpu save/restore code. -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Keir Fraser wrote:> On 6 Jun 2005, at 08:02, Nakajima, Jun wrote: > >>> With this patch, # of testcase failures with LTP is < 5 (out of >>> 800), almost same results as the native x86-64 Linux. The other two >>> patches for LTP are coming soon; with these, all the testcases will >>> pass (except the ones that fail on the native x86-64 Linux). > > Consider defining read/write_cr0 macros same as we do in > arch/xen/i386. We set them the same as native (i.e., real accesses to > %cr0) and then emulate in Xen. This will ensure that things like > raid6 will work, which save/restore %cr0 as part of their fpu > save/restore code. > > -- KeirThat''s better. I''ll send another patch. BTW, they are defined like the below for x86 xenlinux: #define read_cr0() \ BUG(); #define write_cr0(x) \ BUG(); We need to fix this as well. Jun --- Intel Open Source Technology Center _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On 6 Jun 2005, at 09:23, Nakajima, Jun wrote:> That''s better. I''ll send another patch. BTW, they are defined like the > below for x86 xenlinux: > > #define read_cr0() \ > BUG(); > #define write_cr0(x) \ > BUG(); > > We need to fix this as well.Ah. I guess raid6 accesses %cr0 via its own assembly macros. But yes, I think defining the above, and read/write_cr4, the same as native is the right thing to do. We can then add in extra functionality to the emulation as required. I don''t think this will mask bugs in the guest OS -- accesses to %cr4 currently aren;t emulated anyway so they will crash just as they do today. I may tighten up emulation of %cr0 a bit so toggling of unexpected bits at least causes a DPRINTK. -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Keir Fraser wrote:> On 6 Jun 2005, at 08:02, Nakajima, Jun wrote: > >>> With this patch, # of testcase failures with LTP is < 5 (out of >>> 800), almost same results as the native x86-64 Linux. The other two >>> patches for LTP are coming soon; with these, all the testcases will >>> pass (except the ones that fail on the native x86-64 Linux). > > Consider defining read/write_cr0 macros same as we do in > arch/xen/i386. We set them the same as native (i.e., real accesses to > %cr0) and then emulate in Xen. This will ensure that things like > raid6 will work, which save/restore %cr0 as part of their fpu > save/restore code. > > -- KeirUpdated it to use the same as native. Signed-off-by: Jun Nakajima <jun.nakajima@intel.com> Signed-off-by: Nitin A Kamble <nitin.a.kamble@intel.com> Jun --- Intel Open Source Technology Center _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel