On Thu, May 9, 2024 at 2:05 PM Peter Xu <peterx@xxxxxxxxxx> wrote:
>
> On Thu, May 09, 2024 at 01:39:07PM -0700, Axel Rasmussen wrote:
> > For real MCEs, various architectures print log messages when poisoned
> > memory is accessed (which results in a SIGBUS). These messages can be
> > important for users to understand the issue.
> >
> > On the other hand, we have the userfaultfd UFFDIO_POISON operation,
> > which can "simulate" memory poisoning. That particular process will get
>
> It also coveres swapin errors as we talked before, so not always SIM.
>
> I was thinking we should also do that report for swapin errors, however
> then I noticed it wasn't reported before the replacement of pte markers,
> in commit 15520a3f04, since 2022:
>
> @@ -3727,8 +3731,6 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
> put_page(vmf->page);
> } else if (is_hwpoison_entry(entry)) {
> ret = VM_FAULT_HWPOISON;
> - } else if (is_swapin_error_entry(entry)) {
> - ret = VM_FAULT_SIGBUS;
> } else if (is_pte_marker_entry(entry)) {
> ret = handle_pte_marker(vmf);
> } else {
>
> So I am guessing it could be fine to just turn this report off to syslog.
> There will be a back-and-forth on this behavior, but hopefully this is even
> rarer than hwpoison so nobody will notice.
>
> With that, the idea looks valid to me, but perhaps a rename is needed.
> Maybe _QUIESCE or _SILENT?
Ah, I had forgotten about the swapin error case.
I think it still makes sense to silence the log in that case; if we
consider a scenario like disk error, it could seem weird to get an MCE
message for that, since the physical memory is fine and it wouldn't
show up in mcelog or similar.
I like _SILENT, I'll do the rename and update my comments to better
explain in v2.
>
> > SIGBUS on access to the memory, but this effect is tied to an MM, rather
> > than being global like a real poison event. So, we don't want to log
> > about this case to the global kernel log; instead, let the process
> > itself log or whatever else it wants to do. This avoids spamming the
> > kernel log, and avoids e.g. drowning out real events with simulated
> > ones.
> >
> > To identify this situation, add a new VM_FAULT_HWPOISON_SIM flag. This
> > is expected to be set *in addition to* one of the existing
> > VM_FAULT_HWPOISON or VM_FAULT_HWPOISON_LARGE flags (which are mutually
> > exclusive).
> >
> > Signed-off-by: Axel Rasmussen <axelrasmussen@xxxxxxxxxx>
> > ---
> > arch/parisc/mm/fault.c | 7 +++++--
> > arch/powerpc/mm/fault.c | 6 ++++--
> > arch/x86/mm/fault.c | 6 ++++--
> > include/linux/mm_types.h | 5 +++++
> > mm/hugetlb.c | 3 ++-
> > mm/memory.c | 2 +-
> > 6 files changed, 21 insertions(+), 8 deletions(-)
> >
> > diff --git a/arch/parisc/mm/fault.c b/arch/parisc/mm/fault.c
> > index c39de84e98b0..e5370bcadf27 100644
> > --- a/arch/parisc/mm/fault.c
> > +++ b/arch/parisc/mm/fault.c
> > @@ -400,9 +400,12 @@ void do_page_fault(struct pt_regs *regs, unsigned long code,
> > #ifdef CONFIG_MEMORY_FAILURE
> > if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
> > unsigned int lsb = 0;
> > - printk(KERN_ERR
> > +
> > + if (!(fault & VM_FAULT_HWPOISON_SIM)) {
> > + pr_err(
> > "MCE: Killing %s:%d due to hardware memory corruption fault at %08lx\n",
> > - tsk->comm, tsk->pid, address);
> > + tsk->comm, tsk->pid, address);
> > + }
> > /*
> > * Either small page or large page may be poisoned.
> > * In other words, VM_FAULT_HWPOISON_LARGE and
> > diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
> > index 53335ae21a40..ac5e8a3c7fba 100644
> > --- a/arch/powerpc/mm/fault.c
> > +++ b/arch/powerpc/mm/fault.c
> > @@ -140,8 +140,10 @@ static int do_sigbus(struct pt_regs *regs, unsigned long address,
> > if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
> > unsigned int lsb = 0; /* shutup gcc */
> >
> > - pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
> > - current->comm, current->pid, address);
> > + if (!(fault & VM_FAULT_HWPOISON_SIM)) {
> > + pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
> > + current->comm, current->pid, address);
> > + }
> >
> > if (fault & VM_FAULT_HWPOISON_LARGE)
> > lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
> > diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
> > index e4f3c7721f45..16d077a3ad14 100644
> > --- a/arch/x86/mm/fault.c
> > +++ b/arch/x86/mm/fault.c
> > @@ -928,9 +928,11 @@ do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
> > struct task_struct *tsk = current;
> > unsigned lsb = 0;
> >
> > - pr_err_ratelimited(
> > + if (!(fault & VM_FAULT_HWPOISON_SIM)) {
> > + pr_err_ratelimited(
> > "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
> > - tsk->comm, tsk->pid, address);
> > + tsk->comm, tsk->pid, address);
> > + }
> > if (fault & VM_FAULT_HWPOISON_LARGE)
> > lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
> > if (fault & VM_FAULT_HWPOISON)
> > diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
> > index 5240bd7bca33..7f8fc3efc5b2 100644
> > --- a/include/linux/mm_types.h
> > +++ b/include/linux/mm_types.h
> > @@ -1226,6 +1226,9 @@ typedef __bitwise unsigned int vm_fault_t;
> > * @VM_FAULT_HWPOISON_LARGE: Hit poisoned large page. Index encoded
> > * in upper bits
> > * @VM_FAULT_SIGSEGV: segmentation fault
> > + * @VM_FAULT_HWPOISON_SIM Hit poisoned, PTE marker; this indicates a
> > + * simulated poison (e.g. via usefaultfd's
> > + * UFFDIO_POISON), not a "real" hwerror.
> > * @VM_FAULT_NOPAGE: ->fault installed the pte, not return page
> > * @VM_FAULT_LOCKED: ->fault locked the returned page
> > * @VM_FAULT_RETRY: ->fault blocked, must retry
> > @@ -1245,6 +1248,7 @@ enum vm_fault_reason {
> > VM_FAULT_HWPOISON = (__force vm_fault_t)0x000010,
> > VM_FAULT_HWPOISON_LARGE = (__force vm_fault_t)0x000020,
> > VM_FAULT_SIGSEGV = (__force vm_fault_t)0x000040,
> > + VM_FAULT_HWPOISON_SIM = (__force vm_fault_t)0x000080,
> > VM_FAULT_NOPAGE = (__force vm_fault_t)0x000100,
> > VM_FAULT_LOCKED = (__force vm_fault_t)0x000200,
> > VM_FAULT_RETRY = (__force vm_fault_t)0x000400,
> > @@ -1270,6 +1274,7 @@ enum vm_fault_reason {
> > { VM_FAULT_HWPOISON, "HWPOISON" }, \
> > { VM_FAULT_HWPOISON_LARGE, "HWPOISON_LARGE" }, \
> > { VM_FAULT_SIGSEGV, "SIGSEGV" }, \
> > + { VM_FAULT_HWPOISON_SIM, "HWPOISON_SIM" }, \
> > { VM_FAULT_NOPAGE, "NOPAGE" }, \
> > { VM_FAULT_LOCKED, "LOCKED" }, \
> > { VM_FAULT_RETRY, "RETRY" }, \
> > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > index 65456230cc71..2b4e0173e806 100644
> > --- a/mm/hugetlb.c
> > +++ b/mm/hugetlb.c
> > @@ -6485,7 +6485,8 @@ vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
> > pte_marker_get(pte_to_swp_entry(entry));
> >
> > if (marker & PTE_MARKER_POISONED) {
> > - ret = VM_FAULT_HWPOISON_LARGE |
> > + ret = VM_FAULT_HWPOISON_SIM |
> > + VM_FAULT_HWPOISON_LARGE |
> > VM_FAULT_SET_HINDEX(hstate_index(h));
> > goto out_mutex;
> > }
> > diff --git a/mm/memory.c b/mm/memory.c
> > index d2155ced45f8..29a833b996ae 100644
> > --- a/mm/memory.c
> > +++ b/mm/memory.c
> > @@ -3910,7 +3910,7 @@ static vm_fault_t handle_pte_marker(struct vm_fault *vmf)
> >
> > /* Higher priority than uffd-wp when data corrupted */
> > if (marker & PTE_MARKER_POISONED)
> > - return VM_FAULT_HWPOISON;
> > + return VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_SIM;
> >
> > if (pte_marker_entry_uffd_wp(entry))
> > return pte_marker_handle_uffd_wp(vmf);
> > --
> > 2.45.0.118.g7fe29c98d7-goog
> >
>
> --
> Peter Xu
>
