[Lustre-devel] OFLAGS change.

Liu, Xuezhao Xuezhao.Liu at emc.com
Fri Dec 21 05:44:08 PST 2012


I agree about that is change is good example - but focusing on vanila kernel development should be don't break something else, a specially add's problems with supporting a RHEL, SuSe.

[Xuezhao] Sure, we must ensure the added code will not break any supporting or to-be-supported vendor kernels, I will make a fixing patch if the break really happens.

>>
        # the implementation of cancel_dirty_page in OFED 1.4.1's SLES10 SP2
        # backport is broken, so ignore it
        if test -f $OFED_BACKPORT_PATH/linux/mm.h &&
           test "$(sed -ne '/^static inline void cancel_dirty_page(struct page \*page, unsigned int account_size)$/,/^}$/p' $OFED_BACKPORT_PATH/linux/mm.h | md5sum)" = "c518cb32d6394760c5bc
>>

that is from way you suggest.
that backport produce a live lock in lustre due impossible to have correct emulation for a cancel_dirty_pages in OFED.
and sometimes be easy create a implementation for old API via new when rewrite a lustre code to use new API and emulation for older kernels.

[Xuezhao] I may don't know the full story of the above case, but I understand your point. 
          OFED emulates new API based on old API on old kernel, and emulates linux API based on other OS' primitives on other OS. 
          But I think that in the above case it does not matter whether or not have the "cfs_" prefix, even if OFED has "ofd_cancel_dirty_page" it also cannot avoid the above.

Thanks,          
Xuezhao




More information about the lustre-devel mailing list