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

Re: [PATCH v4] x86/msi: fix locking for SR-IOV devices


  • To: Stewart Hildebrand <stewart.hildebrand@xxxxxxx>
  • From: Jan Beulich <jbeulich@xxxxxxxx>
  • Date: Thu, 10 Oct 2024 08:57:59 +0200
  • 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>, Roger Pau Monné <roger.pau@xxxxxxxxxx>, Julien Grall <julien@xxxxxxx>, Stefano Stabellini <sstabellini@xxxxxxxxxx>, Teddy Astie <teddy.astie@xxxxxxxxxx>, xen-devel@xxxxxxxxxxxxxxxxxxxx
  • Delivery-date: Thu, 10 Oct 2024 06:58:10 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 09.10.2024 21:44, Stewart Hildebrand wrote:
> On 8/28/24 06:36, Jan Beulich wrote:
>> On 27.08.2024 05:59, Stewart Hildebrand wrote:
>>> --- a/xen/drivers/passthrough/pci.c
>>> +++ b/xen/drivers/passthrough/pci.c
>>> @@ -341,6 +341,8 @@ static struct pci_dev *alloc_pdev(struct pci_seg *pseg, 
>>> u8 bus, u8 devfn)
>>>  
>>>      list_add(&pdev->alldevs_list, &pseg->alldevs_list);
>>>  
>>> +    INIT_LIST_HEAD(&pdev->physfn.vf_list);
>>
>> There is a certain risk with doing such uniformly when the field is part
>> of a union. Yes, little initialization has happened up to here, but I'm
>> still concerned. (One of the reasons I don't like the struct list_head
>> instances to be split, despite your legitimate point regarding naming.)
>> At the very least this wants moving yet earlier in the function, before
>> the new struct is passed anywhere else.
> 
> Understood. I personally have a slight preference for keeping the entry
> and head names distinct, so I'll plan to move the initialization
> earlier. However, I could easily be convinced to un-split the struct
> list_head instances if that's your preference. Let me know.

I indicated before that this would be my preference, without meaning to
insist on this folding.

>>> +                list_for_each_entry_safe(vf_pdev, tmp, 
>>> &pdev->physfn.vf_list,
>>> +                                         virtfn.entry)
>>> +                    ret = pci_remove_device(vf_pdev->sbdf.seg,
>>> +                                            vf_pdev->sbdf.bus,
>>> +                                            vf_pdev->sbdf.devfn) ?: ret;
>>
>> And if this fails, the VF will still remain orphaned. I think in the
>> model I had suggested no such risk would exist.
>>
>> Misra also isn't going to like the recursion here.
> 
> With the ASSERTs being addressed directly, there's no need to remove
> the VFs right away with the PF.
> 
> BTW, I don't think refusing a removal "request" would be a good idea.
> Dom0 isn't really requesting the device to be removed. Dom0 has already
> removed the device (e.g. in response to hot-unplug or SR-IOV disable),
> and is merely informing Xen of the removal.

Just to clarify: I don't mean returning an error here to indicate
"refusal". As you say, this is merely a notification. Yet I think it's
still legitimate to pass back an error. Whether the Dom0 kernel can
do anything useful with that error is a separate question.

> So during PF removal, I'll plan (for v5) to unlink the the VFs from the
> PF, and continue to rely on dom0 to inform Xen of PF and VF removal
> individually. By unlink, I mean set vf_pdev->virtfn.pf_pdev = NULL and
> remove the VFs from the list. Probably also set vf_pdev->broken = true.

As to the latter - yes. For the rest I guess I need to see the new code.

Jan



 


Rackspace

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