[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] RE: [Xen-ia64-devel] 2 LTP math cases failed in xen0
Hi Anthony, I rerun the program. The exception happens at 1023. The following is the screen output and the dmesg from /var/log/message: ====== screen output====== iteration 1020 : a=(0, 0, -1020) : iteration 1021 : a=(0, 0, -1021) : iteration 1022 : a=(0, 0, -1022) : Floating point exception [root@localhost FPSWA]# ======/var/log/messages===== Feb 6 11:21:13 localhost double_test: iteration #1020 Feb 6 11:21:13 localhost double_test: iteration #1021 Feb 6 11:21:13 localhost double_test: iteration #1022 Feb 6 11:21:13 localhost double_test: iteration #1023 Feb 6 11:21:13 localhost kernel: trigger_fpswa(463): floating-point assist fault at ip 4000000000000d32, isr 0000020000001001 Feb 6 11:21:13 localhost kernel: handle_fpu_swa: fp_emulate() returned -1 The attachment is the program. Best Regards, Yongkang (Kangkang) 永康 >-----Original Message----- >From: Xu, Anthony >Sent: 2006年2月6日 10:14 >To: 'Magenheimer, Dan (HP Labs Fort Collins)'; You, Yongkang; >xen-ia64-devel@xxxxxxxxxxxxxxxxxxx >Subject: RE: [Xen-ia64-devel] 2 LTP math cases failed in xen0 > >Kangkang, >Can you post the values of fraction and exponent when iteration equal to 1020, >1021, 1023, 1024 in case there are other floating point issues? > >Thanks, >-Anthony > >>-----Original Message----- >>From: xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx >>[mailto:xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of >Magenheimer, >>Dan (HP Labs Fort Collins) >>Sent: 2006年1月27日 1:23 >>To: You, Yongkang; xen-ia64-devel@xxxxxxxxxxxxxxxxxxx >>Subject: RE: [Xen-ia64-devel] 2 LTP math cases failed in xen0 >> >>Thanks Yongkang. >> >>So those tests definitely require FPSWA and FPSWA is >>not (yet) implemented for Xen/ia64 (non-VTI) for either >>dom0 or domU... search for "fpswa" in xen/arch/ia64/xen/dom_fw.c >> >>> Native will get FPSWA fault as the doc said when iteration >>> equals to 1024. But Xen0 will get FPSWA fault and Floating >>> point exception, when iteration only equals to 1022. >> >>Interesting. This may be another bug, but let's try >>again after fpswa is implemented. >> >>Dan >> >>> -----Original Message----- >>> From: You, Yongkang [mailto:yongkang.you@xxxxxxxxx] >>> Sent: Thursday, January 26, 2006 1:17 AM >>> To: Magenheimer, Dan (HP Labs Fort Collins); >>> xen-ia64-devel@xxxxxxxxxxxxxxxxxxx >>> Subject: RE: [Xen-ia64-devel] 2 LTP math cases failed in xen0 >>> >>> Hi Dan, >>> >>> I have read the doc your provided. Before I try it, I can >>> catch "floating-point assist fault" in /var/log/messages both >>> IA64 Native and Xen0. >>> >>> Then I do the test: use compile flag -ffast-math to rebuild >>> float cases, there are no FPSWA fault in the dmesg, while the >>> case can still pass in the Native Linux. So ... I am not sure >>> if am I right. >>> >>> I also try the program provided in the doc. But I get >>> interesting different results between Native and Xen0. >>> >>> Native will get FPSWA fault as the doc said when iteration >>> equals to 1024. But Xen0 will get FPSWA fault and Floating >>> point exception, when iteration only equals to 1022. >>> >>> Best Regards, >>> Yongkang (Kangkang) 永康 >>> >>> >-----Original Message----- >>> >From: Magenheimer, Dan (HP Labs Fort Collins) >>> >[mailto:dan.magenheimer@xxxxxx] >>> >Sent: 2006年1月26日 8:11 >>> >To: You, Yongkang; xen-ia64-devel@xxxxxxxxxxxxxxxxxxx >>> >Subject: RE: [Xen-ia64-devel] 2 LTP math cases failed in xen0 >>> > >>> >An easier way to check is described here: >>> > >>> >http://i-cluster2.inrialpes.fr/doc/misc/fpswa.txt >>> > >>> >> -----Original Message----- >>> >> From: Magenheimer, Dan (HP Labs Fort Collins) >>> >> Sent: Wednesday, January 25, 2006 3:23 PM >>> >> To: 'You, Yongkang'; xen-ia64-devel@xxxxxxxxxxxxxxxxxxx >>> >> Subject: RE: [Xen-ia64-devel] 2 LTP math cases failed in xen0 >>> >> >>> >> There may be better ways of turning off FPSWA, but >>> >> here's one that should work: >>> >> >>> >> Change linux/arch/ia64/kernel/traps.c:trap_init() so >>> >> that fpswa_interface is always set to 0. >>> >> >>> >> If a native kernel without this change passes the >>> >> two LTP float_* tests but the same native kernel with >>> >> this change fails, it would explain why Xen/ia64 doesn't >>> >> work on the tests... Xen/ia64 doesn't support FPSWA. >>> >> >>> >> > -----Original Message----- >>> >> > From: You, Yongkang [mailto:yongkang.you@xxxxxxxxx] >>> >> > Sent: Sunday, January 22, 2006 6:54 PM >>> >> > To: Magenheimer, Dan (HP Labs Fort Collins); >>> >> > xen-ia64-devel@xxxxxxxxxxxxxxxxxxx >>> >> > Subject: RE: [Xen-ia64-devel] 2 LTP math cases failed in xen0 >>> >> > >>> >> > Hi Dan, >>> >> > >>> >> > I am not sure about how to check if those math cases are >>> >> > using FPSWA. I greped the key word in float math cases, but >>> >> > nothing found. >>> >> > >>> >> > 1 information is that these 2 cases (float_exp_log and >>> >> > float_power) can pass in the VTi and IA32 testing. I paste >>> >> > the float codes in the attachment tar ball. Hope it can give >>> >> > some help. If need any information, please let me know. >>> >> > >>> >> > Best Regards, >>> >> > Yongkang (Kangkang) 永康 >>> >> > >>> >> > >-----Original Message----- >>> >> > >From: Magenheimer, Dan (HP Labs Fort Collins) >>> >> > >[mailto:dan.magenheimer@xxxxxx] >>> >> > >Sent: 2006年1月21日 1:23 >>> >> > >To: You, Yongkang; xen-ia64-devel@xxxxxxxxxxxxxxxxxxx >>> >> > >Subject: RE: [Xen-ia64-devel] 2 LTP math cases failed in xen0 >>> >> > > >>> >> > >Hi Yongkang -- >>> >> > > >>> >> > >Could you see if the math tests that fail are using >>> >> > >FPSWA (the EFI floating point software assist module)? >>> >> > >I think it is possible to boot Linux with FPSWA disabled. >>> >> > >If this also fails these math tests, it would explain >>> >> > >why Xen/ia64 fails these tests as Xen/ia64 does not >>> >> > >currently support FPSWA (or any EFI modules). >>> >> > > >>> >> > >Thanks, >>> >> > >Dan >>> >> > > >>> >> > >> -----Original Message----- >>> >> > >> From: xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx >>> >> > >> [mailto:xen-ia64-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf >>> >> > >> Of You, Yongkang >>> >> > >> Sent: Friday, January 20, 2006 12:35 AM >>> >> > >> To: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx >>> >> > >> Subject: [Xen-ia64-devel] 2 LTP math cases failed in xen0 >>> >> > >> >>> >> > >> Hi all, >>> >> > >> >>> >> > >> I have run a basic LTP testing on Xen0. Except for 1 system >>> >> > >> call case failed has reported, there are 2 math cases failed >>> >> > >> too. But they can pass in the Native Linux of the same >>> >> > >> hardware environment. These two cases are related to float >>> >> > >> calculation. >>> >> > >> >>> >> > >> The testing environment is: >>> >> > >> Xen source is xen-ia64-unstable tree Changeset 8476. >>> >> > >> CPU is Intel Montecito. >>> >> > >> Native and service OS Linux is RHEL4u2. >>> >> > >> The LTP is the one of 2005-12-5. >>> >> > >> >>> >> > >> The attachment is the summary of testing results. This time I >>> >> > >> run these test suites: >>> >> > >> systemcall, memory test, disk io, nptl, math, ipc, pty, >>> >> sched, fs. >>> >> > >> >>> >> > >> I will try to run more testing of LTP and report issues. Hope >>> >> > >> it can help to stable xen0. :) >>> >> > >> >>> >> > >> Best Regards, >>> >> > >> Yongkang (Kangkang) 永康 >>> >> > >> >>> >> > >> >>> >> > >>> >> >>> Attachment:
trigger_fpswa.c _______________________________________________ Xen-ia64-devel mailing list Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-ia64-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |