On Mon, Jul 15, 2024 at 03:21:34PM -0400, Peter Xu wrote: > [Based on mm-unstable, commit 31334cf98dbd, July 2nd] > > v3: > - Fix a build issue on i386 PAE config > - Moved one line from patch 8 to patch 3 > > v1: https://lore.kernel.org/r/20240621142504.1940209-1-peterx@xxxxxxxxxx > v2: https://lore.kernel.org/r/20240703212918.2417843-1-peterx@xxxxxxxxxx > > Dax supports pud pages for a while, but mprotect on puds was missing since > the start. This series tries to fix that by providing pud handling in > mprotect(). The goal is to add more types of pud mappings like hugetlb or > pfnmaps. This series paves way for it by fixing known pud entries. > > Considering nobody reported this until when I looked at those other types > of pud mappings, I am thinking maybe it doesn't need to be a fix for stable > and this may not need to be backported. I would guess whoever cares about > mprotect() won't care 1G dax puds yet, vice versa. I hope fixing that in > new kernels would be fine, but I'm open to suggestions. > > There're a few small things changed to teach mprotect work on PUDs. E.g. it > will need to start with dropping NUMA_HUGE_PTE_UPDATES which may stop > making sense when there can be more than one type of huge pte. OTOH, we'll > also need to push the mmu notifiers from pmd to pud layers, which might > need some attention but so far I think it's safe. For such details, please > refer to each patch's commit message. > > The mprotect() pud process should be straightforward, as I kept it as > simple as possible. There's no NUMA handled as dax simply doesn't support > that. There's also no userfault involvements as file memory (even if work > with userfault-wp async mode) will need to split a pud, so pud entry > doesn't need to yet know userfault's existance (but hugetlb entries will; > that's also for later). > > Tests > ===== > > What I did test: > > - cross-build tests that I normally cover [1] > > - smoke tested on x86_64 the simplest program [2] on dev_dax 1G PUD > mprotect() using QEMU's nvdimm emulations [3] and ndctl to create > namespaces with proper alignments, which used to throw "bad pud" but now > it'll run through all fine. I checked sigbus happens if with illegal > access on protected puds. > > - vmtests. > > What I didn't test: > > - fsdax: I wanted to also give it a shot, but only until then I noticed it > doesn't seem to be supported (according to dax_iomap_fault(), which will > always fallback on PUD_ORDER). I did remember it was supported before, I > could miss something important there.. please shoot if so. > > - userfault wp-async: I also wanted to test userfault-wp async be able to > split huge puds (here it's simply a clear_pud.. though), but it won't > work for devdax anyway due to not allowed to do smaller than 1G faults in > this case. So skip too. > > - Power, as no hardware on hand. Ping - any review comments or even tests would be greatly welcomed. I'm not sure whether this matters for anyone yet so far. I hope this still makes sense for DAX even if this is an extremely corner case... Just to mention the follow up users of this path: - huge pfnmap 1G may use this, when VM_PFNMAP can be mapped with 1G too, then we should hit similar "bad pud" here. - hugetlb rework will use this, when we want this path to process 1G hugetlb pages too. The 1st user is not a must in my initial plan, as VFIO + VM use case doesn't use mprotect(), so we can keep (1) broken together with DAX 1G here. But for the long term we should still fix this, IMHO. Thanks, -- Peter Xu
