The only reason vmap_range_noflush() can sleep is because of pagetable allocations. The actual allocation mechanism is arch-specific so might_alloc() doesn't work here (what GFP flags would be used?). Hence, just add a comment. Also note that this might do a TLB shootdown. This is not actually sleeping but it requires IRQs on for x86, and might_sleep() incidentally serves to detect violations of that too. Signed-off-by: Brendan Jackman --- Applies to mm-new. Changes in v3: - Added note about TLB shootdowns and x86. - Since the comment no longer fit on one line anyway, also be a bit more verbose in general. - Link to v2: https://lore.kernel.org/r/20251209-b4-vmalloc-might_alloc-v2-1-2462ba26ead1@google.com Changes in v2: - Switched from might_alloc() to keeping might_sleep() + comment - Link to v1: https://lore.kernel.org/r/20251208-b4-vmalloc-might_alloc-v1-1-94a9bb8ecb08@google.com --- mm/vmalloc.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/mm/vmalloc.c b/mm/vmalloc.c index 41dd01e8430c548167d6123f6ec66bf798f20eec..be4e5f2566a4490d45de29568df538b7c8515b01 100644 --- a/mm/vmalloc.c +++ b/mm/vmalloc.c @@ -305,6 +305,11 @@ static int vmap_range_noflush(unsigned long addr, unsigned long end, int err; pgtbl_mod_mask mask = 0; + /* + * Might allocate pagetables (for most archs a more precise annotation + * would be might_alloc(GFP_PGTABLE_KERNEL)). Also might shootdown TLB + * (requires IRQs enabled on x86). + */ might_sleep(); BUG_ON(addr >= end); --- base-commit: d0a24447990a9d8212bfb3a692d59efa74ce9f86 change-id: 20251208-b4-vmalloc-might_alloc-754a791e4e10 Best regards, -- Brendan Jackman