[LU-1461] Compile problems on ppc64

• Status: Resolved
• Project: Lustre
• Fix Version/s: Lustre 2.3.0
• Type: Bug
• Priority: Minor
• Reporter: Oleg Drokin
• Assignee: Oleg Drokin
• Resolution: Fixed
• Severity: 3

Description

Testing ppc64 build node I found this issue:
echo_client.c: In function 'echo_md_build_name':
echo_client.c:1415: error: format '%llu' expects type 'long long unsigned int', but argument 3 has type '__u64'

This is due to direct usage of %llu instead of %LPU64

Similar issue in lfs.c:
lfs.c: In function 'lfs_data_version':
lfs.c:2788: error: format '%llu' expects type 'long long unsigned int', but argument 2 has type '__u64'

Comments

[Peter Jones]

http://review.whamcloud.com/#change,2994

[Oleg Drokin]

Another issue, now in server build in ldiskfs, introduced by: http://review.whamcloud.com/1095

/var/lib/jenkins/workspace/lustre-ppc-builder/arch/ppc64/build_type/server/distro/el6/ib_stack/ofa/BUILD/BUILD/lustre-ldiskfs-3.3.0/ldiskfs/mballoc.c: In function 'ldiskfs_mb_init_backend':
/var/lib/jenkins/workspace/lustre-ppc-builder/arch/ppc64/build_type/server/distro/el6/ib_stack/ofa/BUILD/BUILD/lustre-ldiskfs-3.3.0/ldiskfs/mballoc.c:2419: error: implicit declaration of function 'vmalloc'
/var/lib/jenkins/workspace/lustre-ppc-builder/arch/ppc64/build_type/server/distro/el6/ib_stack/ofa/BUILD/BUILD/lustre-ldiskfs-3.3.0/ldiskfs/mballoc.c:2419: warning: assignment makes pointer from integer without a cast
/var/lib/jenkins/workspace/lustre-ppc-builder/arch/ppc64/build_type/server/distro/el6/ib_stack/ofa/BUILD/BUILD/lustre-ldiskfs-3.3.0/ldiskfs/mballoc.c:2461: error: implicit declaration of function 'vfree'
make[5]: *** [/var/lib/jenkins/workspace/lustre-ppc-builder/arch/ppc64/build_type/server/distro/el6/ib_stack/ofa/BUILD/BUILD/lustre-ldiskfs-3.3.0/ldiskfs/mballoc.o] Error 1
make[5]: *** Waiting for unfinished jobs....

Patch tracking at http://review.whamcloud.com/3036

[James A Simmons]

I noticed also the crypto problem on x86 builds if you have a older gcc.

[James A Simmons]

Here is the exact error I get.

crc32-pclmul_asm.S:321: Error: no such instruction: `pextrd $0x01,%xmm1,%eax'

This is on a x86_64 box with gcc 4.1.2.

[Alexander Boyko]

James, I check my Centos5, and have no any trouble with
gcc version 4.1.2 20080704 (Red Hat 4.1.2-52)

[James A Simmons]

Could be what instruction supported on the cpu on the build box messing it up. On the build box I have the cpu flags of:

flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt rdtscp lm 3dnowext 3dnow constant_tsc nonstop_tsc pni cx16 popcnt lahf_lm cmp_legacy svm extapic cr8_legacy altmovcr8 abm sse4a misalignsse 3dnowprefetch osvw

What flags do you have?

This is a Quad-Core AMD Opteron(tm) Processor 2389

[Alexander Boyko]

No, cpuinfo does not show this instruction. And one more thing, if cpu support new instructions old kernels does not show this instructions through /proc/cpuinfo, but instructions work well.
You have the problem with old gcc, it doesn`t know pextrd instruction. Any way, if gcc compiled code successful, when libcfs is loading, it checks the pclmulqdq instruction support by cpu and provide this functionality or not.

[Alexander Boyko]

pextrd exist on intel CPU from SSE4.1

[Alexander Boyko]

James, try this patch
http://review.whamcloud.com/3801

[James A Simmons]

Thanks. Will move reporting to LU-1786

[Jodi Levi (Inactive)]

Please reopen ticket if more work is needed.