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

Re: [PATCH v2 2/3] x86/msi: don't use cached address and data fields in msi_desc for dump_msi()


  • To: Roger Pau Monné <roger.pau@xxxxxxxxxx>
  • From: Jan Beulich <jbeulich@xxxxxxxx>
  • Date: Fri, 7 Mar 2025 11:18:01 +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: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>, xen-devel@xxxxxxxxxxxxxxxxxxxx
  • Delivery-date: Fri, 07 Mar 2025 10:18:10 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 06.03.2025 18:56, Roger Pau Monné wrote:
> On Thu, Mar 06, 2025 at 05:45:27PM +0100, Jan Beulich wrote:
>> On 06.03.2025 15:57, Roger Pau Monne wrote:
>>> Instead compose a dummy MSI message just for the purpose of getting the
>>> delivery attributes, which are the same for all messages.  Note that the
>>> previous usage of the cached MSI message wasn't fetching the hardware MSI
>>> fields either.
>>
>> This feels not future proof. There's no guarantee special IRQs (HPET, IOMMU)
>> would necessarily use msi_compose_msg() (or any open-coded subset thereof).
> 
> Hm, even if not using msi_compose_msg() I don't see how any device
> would use a different MSI settings from physical delivery and fixed
> destination.  I think it's unlikely for a device to use anything
> different from the current values set by msi_compose_msg().

I'm not entirely sure about this. If e.g. broadcast was possible to send via
MSI, I could see use cases. Potentially also for e.g. NMI.

> Otherwise I can see about returning whether the entry needs to be
> updated from iommu_update_ire_from_msi() (if the offset into the IRT
> for the entry has changed).  However that requires adding code to both
> AMD and Intel IOMMU implementations, and will need at least a way to
> signal that the MSI fields must forcefully be written on resume.

While that may indeed be more intrusive, it feels like it may be the less
risky approach (as to overlooking yet another case where we rely on the
untranslated message to be stored). Overall I can only say that I'm
uncertain which of the two approaches would end up being better.

>>> @@ -1512,8 +1520,7 @@ static void cf_check dump_msi(unsigned char key)
>>>              mask = '?';
>>>          printk(" %-6s%4u vec=%02x%7s%6s%3sassert%5s%7s"
>>>                 " dest=%08x mask=%d/%c%c/%c\n",
>>> -               type, irq,
>>> -               (data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT,
>>> +               type, irq, desc->arch.vector,
>>
>> We've already dropped desc's lock, so shouldn't be de-referencing desc 
>> anymore.
> 
> Right, I need to cache it before dropping the lock.
> 
>>>                 data & MSI_DATA_DELIVERY_LOWPRI ? "lowest" : "fixed",
>>>                 data & MSI_DATA_TRIGGER_LEVEL ? "level" : "edge",
>>>                 data & MSI_DATA_LEVEL_ASSERT ? "" : "de",
>>
>> To add to the comment at the top, plus taking patch 1 into account: If we
>> uniformly used the output of the dummy msi_compose_msg() invocation, why 
>> would
>> we even bother to log information conditionally upon what is in data/addr?
> 
> We could change what's set by msi_compose_msg(), and then the
> information here would be incorrect (if hardcoded).

Hmm, yes, that could happen, but would hopefully be caught in review (much
like it happened this time round).

Jan



 


Rackspace

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