Hi,
> pmd_huge is defined out to false when HUGETLB_PAGE is not configured,
> but the vmap code still installs huge PMDs. This leads to errors
> encountering bad PMDs when vunmapping because it is not seen as a
> huge PTE, and the bad PMD check catches it. The end result may not
> be much more serious than some bad pmd warning messages, because the
> pmd_none_or_clear_bad() does what we wanted and clears the huge PTE
> anyway.
Huh. So vmap seems to key off arch_vmap_p?d_supported which checks for
radix and HAVE_ARCH_HUGE_VMAP.
> Fix this by checking pmd_is_leaf(), which checks for a PTE regardless
> of config options. The whole huge/large/leaf stuff is a tangled mess
> but that's kernel-wide and not something we can improve much in
> arch/powerpc code.
I guess I'm a bit late to the party here because p?d_is_leaf was added
in 2019 in commit d6eacedd1f0e ("powerpc/book3s: Use config independent
helpers for page table walk") but why wouldn't we just make pmd_huge()
not config dependent?
Also, looking at that commit, there are a few places that might still
throw warnings, e.g. find_linux_pte, find_current_mm_pte, pud_page which
seem like they might still throw warnings if they were to encounter a
huge vmap page:
struct page *pud_page(pud_t pud)
{
if (pud_is_leaf(pud)) {
VM_WARN_ON(!pud_huge(pud));
Do these functions need special treatment for huge vmappings()?
Apart from those questions, the patch itself makes sense to me and I can
follow how it would fix a problem.
Reviewed-by: Daniel Axtens <d...@axtens.net>
Kind regards,
Daniel
