[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] x86/MTRR: hook mtrr_bp_restore() back up


  • To: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
  • From: Jan Beulich <jbeulich@xxxxxxxx>
  • Date: Thu, 27 Mar 2025 17:12:54 +0100
  • Autocrypt: addr=jbeulich@xxxxxxxx; keydata= xsDiBFk3nEQRBADAEaSw6zC/EJkiwGPXbWtPxl2xCdSoeepS07jW8UgcHNurfHvUzogEq5xk hu507c3BarVjyWCJOylMNR98Yd8VqD9UfmX0Hb8/BrA+Hl6/DB/eqGptrf4BSRwcZQM32aZK 7Pj2XbGWIUrZrd70x1eAP9QE3P79Y2oLrsCgbZJfEwCgvz9JjGmQqQkRiTVzlZVCJYcyGGsD /0tbFCzD2h20ahe8rC1gbb3K3qk+LpBtvjBu1RY9drYk0NymiGbJWZgab6t1jM7sk2vuf0Py O9Hf9XBmK0uE9IgMaiCpc32XV9oASz6UJebwkX+zF2jG5I1BfnO9g7KlotcA/v5ClMjgo6Gl MDY4HxoSRu3i1cqqSDtVlt+AOVBJBACrZcnHAUSuCXBPy0jOlBhxPqRWv6ND4c9PH1xjQ3NP nxJuMBS8rnNg22uyfAgmBKNLpLgAGVRMZGaGoJObGf72s6TeIqKJo/LtggAS9qAUiuKVnygo 3wjfkS9A3DRO+SpU7JqWdsveeIQyeyEJ/8PTowmSQLakF+3fote9ybzd880fSmFuIEJldWxp Y2ggPGpiZXVsaWNoQHN1c2UuY29tPsJgBBMRAgAgBQJZN5xEAhsDBgsJCAcDAgQVAggDBBYC AwECHgECF4AACgkQoDSui/t3IH4J+wCfQ5jHdEjCRHj23O/5ttg9r9OIruwAn3103WUITZee e7Sbg12UgcQ5lv7SzsFNBFk3nEQQCACCuTjCjFOUdi5Nm244F+78kLghRcin/awv+IrTcIWF hUpSs1Y91iQQ7KItirz5uwCPlwejSJDQJLIS+QtJHaXDXeV6NI0Uef1hP20+y8qydDiVkv6l IreXjTb7DvksRgJNvCkWtYnlS3mYvQ9NzS9PhyALWbXnH6sIJd2O9lKS1Mrfq+y0IXCP10eS FFGg+Av3IQeFatkJAyju0PPthyTqxSI4lZYuJVPknzgaeuJv/2NccrPvmeDg6Coe7ZIeQ8Yj t0ARxu2xytAkkLCel1Lz1WLmwLstV30g80nkgZf/wr+/BXJW/oIvRlonUkxv+IbBM3dX2OV8 AmRv1ySWPTP7AAMFB/9PQK/VtlNUJvg8GXj9ootzrteGfVZVVT4XBJkfwBcpC/XcPzldjv+3 HYudvpdNK3lLujXeA5fLOH+Z/G9WBc5pFVSMocI71I8bT8lIAzreg0WvkWg5V2WZsUMlnDL9 mpwIGFhlbM3gfDMs7MPMu8YQRFVdUvtSpaAs8OFfGQ0ia3LGZcjA6Ik2+xcqscEJzNH+qh8V m5jjp28yZgaqTaRbg3M/+MTbMpicpZuqF4rnB0AQD12/3BNWDR6bmh+EkYSMcEIpQmBM51qM EKYTQGybRCjpnKHGOxG0rfFY1085mBDZCH5Kx0cl0HVJuQKC+dV2ZY5AqjcKwAxpE75MLFkr wkkEGBECAAkFAlk3nEQCGwwACgkQoDSui/t3IH7nnwCfcJWUDUFKdCsBH/E5d+0ZnMQi+G0A nAuWpQkjM1ASeQwSHEeAWPgskBQL
  • Cc: "xen-devel@xxxxxxxxxxxxxxxxxxxx" <xen-devel@xxxxxxxxxxxxxxxxxxxx>, Marek Marczykowski <marmarek@xxxxxxxxxxxxxxxxxxxxxx>, Roger Pau Monné <roger.pau@xxxxxxxxxx>
  • Delivery-date: Thu, 27 Mar 2025 16:12:58 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 27.03.2025 16:23, Andrew Cooper wrote:
> On 27/03/2025 2:20 pm, Jan Beulich wrote:
>> On 27.03.2025 15:10, Roger Pau Monné wrote:
>>> On Thu, Mar 27, 2025 at 02:28:42PM +0100, Jan Beulich wrote:
>>>> On 27.03.2025 13:48, Roger Pau Monné wrote:
>>>>> On Thu, Mar 27, 2025 at 01:30:44PM +0100, Jan Beulich wrote:
>>>>>> On 27.03.2025 12:38, Roger Pau Monné wrote:
>>>>>>> On Thu, Mar 27, 2025 at 12:20:47PM +0100, Jan Beulich wrote:
>>>>>>>> Unlike stated in the offending commit's description,
>>>>>>>> load_system_tables() wasn't the only thing left to retain from the
>>>>>>>> earlier restore_rest_processor_state().
>>>>>>>>
>>>>>>>> While there also do Misra-related tidying for the function itself: The
>>>>>>>> function being used from assembly only means it doesn't need to have a
>>>>>>>> declaration, but wants to be asmlinkage.
>>>>>>> I wonder, maybe the intention was for the MTRR restoring on the BSP to
>>>>>>> also be done by the mtrr_aps_sync_end() call in enter_state()?
>>>>>>>
>>>>>>> AFAICT that will set the MTRRs uniformly on all CPUs, by calling
>>>>>>> mtrr_set_all() just like mtrr_bp_restore(), but later in the restore
>>>>>>> process.
>>>>>> Hmm, yes, that's possible. The comment in set_mtrr() is somewhat 
>>>>>> misleading
>>>>>> then, though, as for the BP the writing then isn't just "okay" but 
>>>>>> necessary.
>>>>>> Question is whether doing this so much later is actually good enough.
>>>>> Hm, no idea really.  We do the device restore ahead of the MTRR
>>>>> restore, so I wonder whether we could have issues by using unexpected
>>>>> effective cache attributes for device memory accesses as a result of
>>>>> MTRRs not being initialized?
>>>> That's just one of the possible problems. The father the MTRRs we run with
>>>> diverged from what firmware puts in place, the bigger the possible trouble.
>>>> I think the restoring better is done as being switched to here again. The
>>>> absence of any discussion of MTRRs in that earlier change leaves me pretty
>>>> certain that the behavioral change there wasn't intended. Andrew is usually
>>>> pretty good at spelling out all intended effects.
>>> No objection, however for the BSP we now end up restoring the MTRRs
>>> twice, as we will also do it in mtrr_aps_sync_end().
>>>
>>> Might be worth to mention in the commit message that the MTRR state
>>> was restored in mtrr_aps_sync_end() for the BSP also, but that it
>>> might be too late.
>> I've added "Note that MTRR state was still reloaded via mtrr_aps_sync_end(),
>> but that happens quite a bit later in the resume process."
> 
> Ah yes, you got here too.
> 
> Yes, I think I simply missed this part of the discussion from the commit
> message.
> 
> The MTRR logic is a giant tangle, and lost of it (I'm pretty sure) is
> only relevant for early 32bit days.  Also since then, I expect firmware
> has gotten better, considering that S3 is ubiquitous on laptops nowadays.
> 
> I expect that we don't need to change MTRRs in most cases.  However, if
> change to the MTRRs actually need to happen, then they probably want
> doing as part of the AP boot, rather than in a rendezvous later.  That
> said, it would be a difference between the normal boot and S3 resume paths.

So in summary - you think we don't need/want the patch here? It feels risky
to me to run the BSP with not-yet-restored MTRRs for an extended period of
time.

Jan



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.