On Mon, 3 May 2021 at 10:09, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote: > > On 03/05/21 17:25, Mathieu Poirier wrote: > >> Mathieu, can you confirm that your coresight branch will*not* be sent by > >> the ARM maintainers as well? > > Confirmed. Marc's tree is the only place where the ETE-TRBE functionality has > > been added. It was specifically done that way to avoid having the same code in > > multiple branches and prevent merge conflicts. > > Thanks for confirming! > > Generally, what we do for x86 is exactly the opposite: the basic > functionality is committed to the x86 tree, and then merged in _also_ by > myself. For example, this pull request includes a topic branch provided > by the cgroup maintainer and one provided by the x86 maintainers, but in > both cases they _also_ sent exactly the same commits to Linus. The above works if all subsystems are pulled-in directly by Linus. But for CoreSight patches flow through Greg's char-misc tree, which would not have worked for the ETR-TRBE patchset due to dependencies with the KVM/ARM tree. As such I don't think we could have done things differently. > > It works well because git is pretty good at avoiding conflicts when the > same branch is present in multiple branches. Instead, cherry-picking > introduces lots of merge conflicts. > > There are other advantages in doing that. For example, in this case I > didn't (and don't) quite know what ETE and TRBE are, beyond what a quick > Internet search tells me. Sending this functionality to an ARM > maintainer that is more acquainted with the feature would ensure that > the new functionality is documented properly in the tags and therefore > in the commit messages. > > This is what Linux was mentioning when he said "Pull requests need to > have explanations of what they pull - not just because it needs to go > into the merge message, but because the maintainer needs to keep track > of what's happening". In this case, the maintainer was me; based on my > own workflow and due to the lack of commit message I assumed that the > branch was also going to go through the ARM tree (and doubted my > assumption only when sending the pull request to Linus, i.e. way too late). > > I am also guilty as charged of the "Merge branch 'kvm-sev-cgroup' into > HEAD" commit message, where I should have pointed out that Tejun had a > later branchpoint from 5.12-rc than I did, resulting in conflicts. > > So Marc, let's heed Linus's advice and document all topic branches that > we merge into the KVM/ARM and KVM/x86 trees, including whether they also > go into other trees---which they should do almost all the time. > > Thanks, > > Paolo > > > Let me know if you need more information. >
