On Thu, Jun 12, 2025 at 12:42:42PM -0300, Jason Gunthorpe wrote:
On Thu, Jun 12, 2025 at 05:23:01PM +0200, Thomas Weißschuh wrote:
On Thu, Jun 12, 2025 at 11:58:01AM -0300, Jason Gunthorpe wrote:
On Thu, Jun 12, 2025 at 04:27:41PM +0200, Thomas Weißschuh wrote:
If the assumption is that this is most likely a kernel bug, shouldn't it be fixed properly rather than worked around? After all the job of a selftest is to detect bugs to be fixed.
I investigated the history for a bit and it seems likely we cannot change the kernel here. Call it an undocumented "feature".
I looked a bit and it seems to be mentioned in mmap(2):
For mmap(), offset must be a multiple of the underlying huge page size. The system automatically aligns length to be a multiple of the underlying huge page size.
Oh there you go then :) Horrible design. No way for userspace to know what the rounded up length actually was and thus no way for userspace to unmap it.
OK. I think we would have to skip those cases then.
MAP_HUGETLBFS rounds up the length to some value, userspace has to figure that out and not pass incorrect lengths. The selftest is doing that wrong.
The selftest would be more robust if MAP_FIXED is replaced by MAP_FIXED_NOREPLACE. Even with the new explicit skip logic it should make debugging easier if something goes wrong.
The point is to replace something that is already mapped there, though I no longer remember why it is working like this.
By replacing MAP_FIXED with MAP_FIXED_NOREPLACE, at the existing two places, the selftest crashed at early setup_sizes...:
iommufd: iommufd.c:53: setup_sizes: Assertion `vrc == buffer' failed. /nicolinc/iommufd_selftest.sh: line 19: 21487 Aborted (core dumped) tools/testing/selftests/iommu/iommufd strace: mmap(0xffff80000000, 1048576, PROT_READ|PROT_WRITE, MAP_SHARED|MAP_ANONYMOUS|MAP_FIXED_NOREPLACE, -1, 0) = -1 EEXIST (File exists)
This one doesn't MAP_HUGETLBFS btw...
Thanks Nicolin