------- Comment From p...@au1.ibm.com 2019-09-27 01:51 EDT------- (In reply to comment #73) > Leonardo, can you elaborate on the 'other possible issues'? We're hesitant > to pull 18 patches into a stable kernel under the assumption that they > *might* fix some *potential* issues, without clear evidence. If you can test > the single-patch kernel and report back that there are still issues then > that's a much stronger case for the other patches. > > Commit 'KVM: PPC: Book3S HV: Avoid crash from THP collapse during radix page > fault' that you're asking for requires all these additional backports to > apply cleanly. Which makes me wonder if we're not actually introducing a > problem with these backports just to fix it again later. Not saying that's > the case, just wondering... > > Also, the following seem to be totally unrelated and unnecessary: > - KVM: PPC: Remove unused kvm_unmap_hva callback > - powerpc/mm/radix: Remove unused code > > While looking through the patches I also noticed that the following is the > second patch of a series of 11 but it's the only one from the series that > you're backporting. > - powerpc/kvm: Switch kvm pmd allocator to custom allocator > Its commit message mentions subsequent patches of that series so I'm > wondering why we need/want only this single patch?? > > Remember that we have to support this kernel for years and years to come so > we only want to backport the absolute necessary. > > Lastly and FYI, the following is the minimal subset of your patches that all > cherry-pick cleanly: > - KVM: PPC: Book3S HV: Avoid crash from THP collapse during radix page fault > - KVM: PPC: Book3S HV: Don't use compound_order to determine host mapping > size > - KVM: PPC: Book3S HV: Use correct pagesize in kvm_unmap_radix() > - KVM: PPC: Book3S HV: radix: Refine IO region partition scope attributes > - KVM: PPC: Book3S HV: Use __gfn_to_pfn_memslot() in page fault handler > - KVM: PPC: Book3S HV: Handle 1GB pages in radix page fault handler > - KVM: PPC: Book3S HV: Streamline setting of reference and change bits > - KVM: PPC: Book3S HV: Radix page fault handler optimizations > > Please provide some context why we need all the above (and potentially more).
OK, so these are the ones *not* included in the above list (oldest to newest, with upstream commit IDs): 39c983ea0f96 KVM: PPC: Remove unused kvm_unmap_hva callback This one is dead code removal, it can be dropped. e2560b108fb1 KVM: PPC: Book3S HV: Make radix use correct tlbie sequence in kvmppc_radix_tlbie_page This one adds barriers which are required according to the architecture specification. It is not strictly related to fixing this bug, but if not included here, another bug should be raised to include it. It is quite safe since it is just adding barrier instructions. Without it there is a possibility of occasional mis-translation of addresses (though perhaps a very small possibility). If another bug is raised for this patch, include df158189dbcc below as well in the same bug. 7e3d9a1d0f2c KVM: PPC: Book3S HV: Make radix clear pte when unmapping This fixes a real bug, though it is not strictly related to the bug in this bugzilla. If it is not included here then another bug should be raised to include it. It is a small, simple and safe change. Without it there is a possibility of guests getting stuck doing continual hypervisor page faults. df158189dbcc KVM: PPC: Book 3S HV: Do ptesync in radix guest exit path This one, like e2560b108fb1 above, adds barriers which are required according to the architecture specification. It is not strictly related to fixing this bug, but if not included here, another bug should be raised to include it. It is quite safe since it is just adding barrier instructions. 21828c99ee91 powerpc/kvm: Switch kvm pmd allocator to custom allocator This one is not needed and can be dropped. 99491e2d0e50 powerpc/mm/radix: Remove unused code This is dead code removal and can be dropped. 0078778a86b1 powerpc/mm/radix: implement LPID based TLB flushes to be used by KVM This is not strictly needed and can be dropped if d91cb39ffa7b and 9a4506e11b97 are being dropped. a5fad1e95952 KVM: PPC: Book3S HV: Use a helper to unmap ptes in the radix fault path This is not strictly needed (code refactoring) and can be dropped. a5704e83aa3d KVM: PPC: Book3S HV: Recursively unmap all page table entries when unmapping This one fixes a memory leak, so is not strictly related to this bug. The memory leak will probably not be apparent unless users are using 1GB huge pages to back guests. d91cb39ffa7b KVM: PPC: Book3S HV: Make radix use the Linux translation flush functions for partition scope This is code refactoring and can be dropped. 9a4506e11b97 KVM: PPC: Book3S HV: Make radix handle process scoped LPID flush in C, with relocation on This is code refactoring and can be dropped. 878cf2bb2d8d KVM: PPC: Book3S HV: radix: Do not clear partition PTE when RC or write bits do not match This one is a performance optimization and can be dropped. So in summary, three of these patches should be included, whether under this bug or under other bugs. The other 9 can be dropped. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1788098 Title: Avoid migration issues with aligned 2MB THB To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu-power-systems/+bug/1788098/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
