riscv: handle VM_FAULT_[HWPOISON|HWPOISON_LARGE] faults instead of panicking

Patch series "Fix set_huge_pte_at()".

A recent report [1] from Ryan for arm64 revealed that we do not handle
swap entries when setting a hugepage backed by a NAPOT region (the contpte
riscv equivalent).

As explained in [1], the issue was discovered by a new test in kselftest
which uses poison entries, but the symptoms are different from arm64 though:

- the riscv kernel bugs because we do not handle VM_FAULT_HWPOISON*,
  this is fixed by patch 1,
- after that, the test passes because the first pte_napot() fails (the
  poison entry does not have the N bit set), and then we only set the
  first page table entry covering the NAPOT hugepage, which is enough
  for hugetlb_fault() to correctly raise a VM_FAULT_HWPOISON wherever we
  write in this mapping since only this first page table entry is
  checked
  (see https://elixir.bootlin.com/linux/v6.6-rc3/source/mm/hugetlb.c#L6071).
  But this seems fragile so patch 2 sets all page table entries of a
  NAPOT mapping.

[1]: https://lore.kernel.org/linux-arm-kernel/20230922115804.2043771-1-ryan.roberts@arm.com/

This patch (of 2):

We used to panic when such faults were encountered but we should handle
those faults gracefully for userspace by sending a SIGBUS to the process,
like most architectures do.

Link: https://lkml.kernel.org/r/20230928151846.8229-1-alexghiti@rivosinc.com
Link: https://lkml.kernel.org/r/20230928151846.8229-2-alexghiti@rivosinc.com
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Acked-by: Palmer Dabbelt <palmer@rivosinc.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Andrew Jones <ajones@ventanamicro.com>
Cc: Conor Dooley <conor@kernel.org>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Qinglin Pan <panqinglin2020@iscas.ac.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/arch/riscv/mm/fault.c b/arch/riscv/mm/fault.c
index 6115d75..90d4ba3 100644 (file)
--- a/arch/riscv/mm/fault.c
+++ b/arch/riscv/mm/fault.c
@@ -72,7 +72,7 @@ static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_f
                }
                pagefault_out_of_memory();
                return;
-       } else if (fault & VM_FAULT_SIGBUS) {
+       } else if (fault & (VM_FAULT_SIGBUS | VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE)) {
                /* Kernel mode? Handle exceptions or die */
                if (!user_mode(regs)) {
                        no_context(regs, addr);