Re: [PATCH v3 08/13] selftests/mm: ensure destination is hugetlb-backed in hugepage-mremap

[Sayali Patil <sayalip@linux.ibm.com>]

On 01/04/26 20:10, Lorenzo Stoakes (Oracle) wrote:
> On Wed, Apr 01, 2026 at 04:21:55PM +0200, David Hildenbrand (Arm) wrote:
>> On 3/27/26 08:16, Sayali Patil wrote:
>>> The hugepage-mremap selftest reserves the destination address using a
>>> anonymous base-page mapping before calling mremap() with MREMAP_FIXED,
>>> while the source region is hugetlb-backed.
>>>
>>> When remapping a hugetlb mapping into a base-page VMA may fail with:
>>>
>>>      mremap: Device or resource busy
>>>
>>> This is observed on powerpc hash MMU systems where slice constraints
>>> and page size incompatibilities prevent the remap.
> 
> OK so digging in:
> 
> mremap -> ... -> vrm_set_new_addr() -> get_unmapped_area() -> ... (in ppc arch
> code) -> slice_get_unmapped_area():
> 
> unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
> 				      unsigned long flags, unsigned int psize,
> 				      int topdown)
> {
> 	...
> 	/* bunch of checks */
> 
> 	/* If we have MAP_FIXED and failed the above steps, then error out */
> 	if (fixed)
> 		return -EBUSY;
> 
> 	...
> }
> 
> Is presumably where we hit the issue.
> 
>>>
>>
>> That is weird. An mremap(MREMAP_FIXED) is really just an munmap() + move.
> 
> Yeah the weird bit I guess is that we _still_ invoke get_unmapped_area() but
> with MAP_FIXED set to indicate that we want the specific address, so it's
> subject to the above checks.
> 
>>
>> Are we sure this is not some actual problem in the hugetlb implementation?
> 
> It seems the 'slices' check sees if the _target address_ has an equivalent page
> size, presumably hugetlb-mandated, and fails if they're not equivalent, so this
> change is just accounting for that.
> 
Yes, this change accounts for that by ensuring the destination is 
created with MAP_HUGETLB so it has the same page size as the source.
> 
>>
>>> Ensure the destination region is created using MAP_HUGETLB so that both
>>> source and destination VMAs are hugetlb-backed and compatible. Also add
>>> MAP_POPULATE to the destination mapping to prefault hugepages,
>>> matching the behaviour used for other hugetlb mapping in the test and
>>> ensuring deterministic behaviour.
>>
>> But then the test suddenly requires more hugetlb pages, no? I don't see
>> a good reason for the MAP_POPULATE, really. It will be discarded either way.
> 
> Yeah I'm not sure about the MAP_POPULATE being all that important here.
> 
As far as I understand, without MAP_POPULATE, memory accesses would 
trigger userfaults, and since the test is single-threaded and has no 
background handler for the uffd, it would deadlock. MAP_POPULATE ensures 
the test runs correctly by prefaulting all pages, but please let me know 
if I’m mistaken.
>>
>>>
>>> Update the FLAGS macro to include MAP_HUGETLB | MAP_SHARED |
>>> MAP_POPULATE so that both mappings are hugetlb-backed and compatible.
>>> Also use the macro for the mmap() calls to avoid repeating
>>> the flag combination.
>>>
>>> This ensures the test reliably exercises hugetlb mremap instead of
>>> failing due to VMA type mismatch.
>>>
>>> Fixes: 12b613206474 ("mm, hugepages: add hugetlb vma mremap() test")
>>> Acked-by: Zi Yan <ziy@nvidia.com>
>>> Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
>>> Signed-off-by: Sayali Patil <sayalip@linux.ibm.com>
>>> ---
>>>   tools/testing/selftests/mm/hugepage-mremap.c | 11 ++++-------
>>>   1 file changed, 4 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/tools/testing/selftests/mm/hugepage-mremap.c b/tools/testing/selftests/mm/hugepage-mremap.c
>>> index e611249080d6..48c24a4ba9a7 100644
>>> --- a/tools/testing/selftests/mm/hugepage-mremap.c
>>> +++ b/tools/testing/selftests/mm/hugepage-mremap.c
>>> @@ -31,7 +31,7 @@
>>>   #define MB_TO_BYTES(x) (x * 1024 * 1024)
>>>
>>>   #define PROTECTION (PROT_READ | PROT_WRITE | PROT_EXEC)
>>> -#define FLAGS (MAP_SHARED | MAP_ANONYMOUS)
>>> +#define FLAGS (MAP_HUGETLB | MAP_SHARED | MAP_POPULATE)
>>>
>>>   static void check_bytes(char *addr)
>>>   {
>>> @@ -121,23 +121,20 @@ int main(int argc, char *argv[])
>>>
>>>   	/* mmap to a PUD aligned address to hopefully trigger pmd sharing. */
>>>   	unsigned long suggested_addr = 0x7eaa40000000;
>>> -	void *haddr = mmap((void *)suggested_addr, length, PROTECTION,
>>> -			   MAP_HUGETLB | MAP_SHARED | MAP_POPULATE, fd, 0);
>>> +	void *haddr = mmap((void *)suggested_addr, length, PROTECTION, FLAGS, fd, 0);
>>>   	ksft_print_msg("Map haddr: Returned address is %p\n", haddr);
>>>   	if (haddr == MAP_FAILED)
>>>   		ksft_exit_fail_msg("mmap1: %s\n", strerror(errno));
>>>
>>>   	/* mmap again to a dummy address to hopefully trigger pmd sharing. */
>>>   	suggested_addr = 0x7daa40000000;
>>> -	void *daddr = mmap((void *)suggested_addr, length, PROTECTION,
>>> -			   MAP_HUGETLB | MAP_SHARED | MAP_POPULATE, fd, 0);
>>> +	void *daddr = mmap((void *)suggested_addr, length, PROTECTION, FLAGS, fd, 0);
>>>   	ksft_print_msg("Map daddr: Returned address is %p\n", daddr);
>>>   	if (daddr == MAP_FAILED)
>>>   		ksft_exit_fail_msg("mmap3: %s\n", strerror(errno));
>>>
>>>   	suggested_addr = 0x7faa40000000;
>>> -	void *vaddr =
>>> -		mmap((void *)suggested_addr, length, PROTECTION, FLAGS, -1, 0);
>>> +	void *vaddr = mmap((void *)suggested_addr, length, PROTECTION, FLAGS, fd, 0);
>>>   	ksft_print_msg("Map vaddr: Returned address is %p\n", vaddr);
>>>   	if (vaddr == MAP_FAILED)
>>>   		ksft_exit_fail_msg("mmap2: %s\n", strerror(errno));
>>
>>
>> --
>> Cheers,
>>
>> David
> 
> Cheers, Lorenzo