On 19/08/2025 20:51, Luck, Tony wrote:
> On Tue, Aug 19, 2025 at 01:28:46PM -0400, Yazen Ghannam wrote:
>> On Tue, Aug 19, 2025 at 07:24:34PM +0300, Adrian Hunter wrote:
>>> Commit 8a01ec97dc066 ("x86/mce: Mask out non-address bits from machine
>>> check bank") introduced a new #define MCI_ADDR_PHYSADDR for the mask of
>>> valid physical address bits within the machine check bank address register.
>>>
>>> This is particularly needed in the case of errors in TDX/SEAM non-root mode
>>> because the reported address contains the TDX KeyID. Refer to TDX and
>>> TME-MK documentation for more information about KeyIDs.
>>>
>>> Commit 7911f145de5fe ("x86/mce: Implement recovery for errors in TDX/SEAM
>>> non-root mode") uses the address to mark the affected page as poisoned, but
>>> omits to use the aforementioned mask.
>>>
>>> Investigation of user space expectations has concluded it would be more
>>> correct for the address to contain only address bits in the first place.
>>> Refer https://lore.kernel.org/r/807ff02d-7af0-419d-8d14-a4d6c5d5420d@xxxxxxxxx
>>>
>>> Mask the address when it is read from the machine check bank address
>>> register. Do not use MCI_ADDR_PHYSADDR because that will be removed in a
>>> later patch.
>>>
>>> It is assumed __log_error() in arch/x86/kernel/cpu/mce/amd.c does not need
>>> similar treatment.
>>>
>>> Amend struct mce addr member description slightly to reflect that it is
>>> not, and never has been, an exact copy of the bank's MCi_ADDR MSR.
>>>
>>
>> I think it would be more accurate to say that the MCi_ADDR MSR is not,
>> and never has been, guaranteed to be a system physical address.
>>
>> We could introduce a new field that represents the system physical
>> address, if one exists for the error type. This way we can operate on a
>> value without assumption or additional checks. And we can keep the raw
>> MCi_ADDR MSR value in case it is of value to debug folks or hardware
>> designers. In my experience, they seem to appreciate having the full,
>> unfiltered data. We don't give them that today, but we can work towards
>> that goal.
>
> Having and exact copy of MCi_ADDR might be useful. I recall some angst
> about this code masking off low order bits:
>
> m->addr = mce_rdmsrq(mca_msr_reg(i, MCA_ADDR));
>
> /*
> * Mask the reported address by the reported granularity.
> */
> if (mca_cfg.ser && (m->status & MCI_STATUS_MISCV)) {
> u8 shift = MCI_MISC_ADDR_LSB(m->misc);
> m->addr >>= shift;
> m->addr <<= shift;
> }
>
> this proposal masks some high order bits too.
>
> I second Yazen's suggestion of a new field. One for the raw value,
> another for the massaged phsical address derived from the MSR.
For struct mce? Maybe that should be 2 new fields:
__u64 addr; /* Deprecated */
...
__u64 mci_addr; /* Bank's MCi_ADDR MSR */
__u64 phys_addr; /* Physical address */
