[LU-5611] WARNING: kernel missing dev_set_rdonly patch for testing

• Type: Bug
• Resolution: Duplicate
• Priority: Minor
• Severity: 3
• Assignee: WC Triage
• Reporter: Stephen Champion

Description

This is a build failure found preparing SGI Lustre 2.4.1 :

checking for /usr/src/kernel-source/2.6.32-358.18.1.el6sgi241a15.rhel6.x86_64.lustre/include/linux/migrate.h... yes
checking if address_space_operations.migratepage has 4 args... no
checking if super_operations use dentry as parameter... no
checking if inode_operations use umode_t as parameter... no
checking if touch_atime use one argument... no
checking if have d_make_root... no
checking if kmap_atomic has only 1 argument... no
checking if have clear_inode... yes
checking if encode_fh have parent inode as parameter... no
checking if kernel has generic_file_llseek_size with 5 args... no
checking if i_dentry/d_alias uses hlist... no
checking if dentry_open uses struct path as first argument... no
checking if iop has atomic_open... no
checking if Linux was built with symbol dev_set_rdonly exported... no
configure: WARNING: kernel missing dev_set_rdonly patch for testing

This is the test:
grep -q -E '[[:space:]]dev_set_rdonly[[:space:]]' $LINUX/$SYMVERFILE 2>/dev/null

And we do have the symbol:

schamp@ltest-admin:/home/lwork/schamp/stout7-sgilustre-2.4.1/bldroot.rhel64/boot $ gunzip -c symvers-2.6.32-358.18.1.el6sgi241a15.rhel6.x86_64.lustre.gz | grep '[[:space:]]dev_set_rdonly[[:space:]]'
0x6e7378b8 dev_set_rdonly vmlinux EXPORT_SYMBOL

I bet it screws up the symvers:
checking name of module symbol version file... Module.symvers

--with-linux=/usr/src/kernel-source/2.6.32-358.18.1.el6sgi241a15.rhel6.x86_64.lustre/
so it looks at
/usr/src/kernel-source/2.6.32-358.18.1.el6sgi241a15.rhel6.x86_64.lustre/Module.symvers

Which doesn't exist, because the symver is in the object directory, not
the source directory - a problem which doesn't show up in the Intel build
process because it uses the same directory for both, even though vendors package them separately.

This is likely a problem not only for SGI's internal build process, but for anyone who builds server packages against installed kernel packages instead of the intermediate kernel build objects.

Patch coming right up.

Issue Links

• duplicates: LU-4470 replay-dual test_21b: FAIL: lustre-MDT0000: ffff88007f581100 CANNOT BE SET READONLY: rc = -95