s390/mm: Add PTE_MARKER support for hugetlbfs mappings

JIRA: https://issues.redhat.com/browse/RHEL-74362
Tested: by me
Conflicts: due b20c821 patchset not being backported

commit 487ef5d
Author: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Date:   Thu Nov 21 18:45:23 2024 +0100

    s390/mm: Add PTE_MARKER support for hugetlbfs mappings

    Commit 8a13897 ("mm: userfaultfd: support UFFDIO_POISON for
    hugetlbfs") added support for PTE_MARKER_POISONED for hugetlbfs, but
    PTE_MARKER also needs support for swap entries. For s390, swap entries
    were only supported on PTE level, not on the PMD/PUD levels that are used
    for large hugetlbfs mappings.

    Therefore, when writing a PTE_MARKER_POISONED entry, the resulting entry
    on PMD/PUD level would be an invalid / empty entry. Further access would
    then generate a pagefault loop, instead of the expected SIGBUS. It is a
    loop inside the kernel, but interruptible and uffd fault handling also
    calls schedule() in between, so at least it won't completely block the
    system.

    Previous commits prepared support for swap entries on PMD/PUD levels.
    PTE_MARKER support for hugetlbfs can now be enabled by simply adding an
    extra is_pte_marker() check to huge_pte_none_mostly(). Fault handling
    code also needs to be adjusted to expect the VM_FAULT_HWPOISON_LARGE
    fault flag, which was not possible on s390 before.

    Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com>
    Signed-off-by: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
    Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

Signed-off-by: Aristeu Rozanski <arozansk@redhat.com>

Author: aristeu

arch/s390/include/asm/hugetlb.h

@@ -91,7 +91,7 @@ static inline int huge_pte_none(pte_t pte)
 
 static inline int huge_pte_none_mostly(pte_t pte)
 {
-	return huge_pte_none(pte);
+	return huge_pte_none(pte) || is_pte_marker(pte);
 }
 
 static inline int huge_pte_write(pte_t pte)

arch/s390/mm/fault.c

@@ -307,7 +307,8 @@ static void do_fault_error(struct pt_regs *regs, vm_fault_t fault)
 				do_no_context(regs);
 			else
 				do_sigsegv(regs, SEGV_MAPERR);
-		} else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON)) {
+		} else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON |
+				    VM_FAULT_HWPOISON_LARGE)) {
 			/* Kernel mode? Handle exceptions or die */
 			if (!user_mode(regs))
 				do_no_context(regs);