FT> My point is that the kernel module always sets DMU_FLAG_SYNC flag FT> regardless of user space. As the patch does, should dmu_cpy_flag FT> be used instead of cmu_set_flag? This enables a user-space FT> programs to choose the behavior. This is what I did previously; I let userspace set the sync flag if it wanted to catch the endio. However, during some debugging, I decided to just send every endio to userspace. I think I did this because I was getting some corruption and thought it might be related to endio. I think you are right, however, that because we can complete the IOs in any order, we can delay some endios and not others. I'll change it back to copy the flag and do some more testing Monday to verify. -- Dan Smith IBM Linux Technology Center Open Hypervisor Team email: danms@xxxxxxxxxx
Attachment:
pgpFtokQgEbFi.pgp
Description: PGP signature
-- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
