On Wed, Sep 28, 2022 at 11:38:32AM -0400, Thomas Tai wrote: > Inspired by commit c6acb1e7bf46 (x86/sgx: Add hook to error injection > address validation), add a similar code in hwpoison_inject function to > check if the address is located in SGX Memory. The error will then be > handled by the arch_memory_failure function in the SGX driver. > > Signed-off-by: Thomas Tai <thomas.tai@xxxxxxxxxx> Thank you for sending patch. > --- > Documentation/mm/hwpoison.rst | 44 +++++++++++++++++++++++++++++++++++ > mm/hwpoison-inject.c | 4 ++++ > 2 files changed, 48 insertions(+) > > diff --git a/Documentation/mm/hwpoison.rst b/Documentation/mm/hwpoison.rst > index b9d5253c1305..8a542aca4744 100644 > --- a/Documentation/mm/hwpoison.rst > +++ b/Documentation/mm/hwpoison.rst > @@ -162,6 +162,50 @@ Testing > > Some portable hwpoison test programs in mce-test, see below. > > +* Special notes for injection into SGX enclaves > + > + 1) Determine physical address of enclave page > + > + dmesg | grep "sgx: EPC" > + > + sgx: EPC section 0x8000c00000-0x807f7fffff > + sgx: EPC section 0x10000c00000-0x1007fffffff > + > + 2) Convert the EPC address to page frame number. > + > + For 4K page size, the page frame number for 0x8000c00000 is > + 0x8000c00000 / 0x1000 = 0x8000c00. > + > + 3) Trace memory_failure > + > + echo nop > /sys/kernel/tracing/current_tracer > + echo *memory_failure > /sys/kernel/tracing/set_ftrace_filter > + echo function > /sys/kernel/tracing/current_tracer > + > + 4) Inject a memory error > + > + modprobe hwpoison-inject > + echo "0x8000c00" > /sys/kernel/debug/hwpoison/corrupt-pfn > + > + 5) Check the trace output > + > + cat /sys/kernel/tracing/trace > + > + # tracer: function > + # > + # entries-in-buffer/entries-written: 2/2 #P:128 > + # > + # _-----=> irqs-off > + # / _----=> need-resched > + # | / _---=> hardirq/softirq > + # || / _--=> preempt-depth > + # ||| / _-=> migrate-disable > + # |||| / delay > + # TASK-PID CPU# ||||| TIMESTAMP FUNCTION > + # | | | ||||| | | > + bash-12167 [002] ..... 113.136808: memory_failure<-simple_attr_write > + bash-12167 [002] ..... 113.136810: arch_memory_failure<-memory_failure In other cases of page types, memory_failure() leaves some kernel message like "Memory failure: 0x10cf09: recovery action for free buddy page: Recovered", which is printed out by action_result(). So I think it's better to adjust to this convention also in SGX page's case. Then, you don't have to use ftrace to confirm the result of error injection. Thanks, Naoya Horiguchi > + > References > ========== > > diff --git a/mm/hwpoison-inject.c b/mm/hwpoison-inject.c > index 65e242b5a432..bf83111c1d9b 100644 > --- a/mm/hwpoison-inject.c > +++ b/mm/hwpoison-inject.c > @@ -21,6 +21,10 @@ static int hwpoison_inject(void *data, u64 val) > if (!capable(CAP_SYS_ADMIN)) > return -EPERM; > > + /* Inject the error if the page is part of the processor reserved memory */ > + if (arch_is_platform_page(pfn << PAGE_SHIFT)) > + goto inject; > + > if (!pfn_valid(pfn)) > return -ENXIO; > > -- > 2.31.1
