Re: [PATCH] iommu/vt-d: avoid sending explicit ATS invalidation request to released device

[Ethan Zhao <haifeng.zhao@linux.intel.com>]

On 3/2/2024 5:56 AM, Bjorn Helgaas wrote:
> On Fri, Mar 01, 2024 at 09:50:36AM +0800, Ethan Zhao wrote:
>> On 3/1/2024 5:06 AM, Bjorn Helgaas wrote:
>>> On Wed, Feb 28, 2024 at 10:31:38PM -0500, Ethan Zhao wrote:
>>>> The introduction of per iommu device rbtree also defines the lifetime of
>>>> interoperation between iommu and devices, if the device has been released
>>>> from device rbtree, no need to send ATS invalidation request to it anymore,
>>>> thus avoid the possibility of later ITE fault to be triggered.
>>>>
>>>> This is part of the followup of prior proposed patchset
>>>>
>>>> https://do-db2.lkml.org/lkml/2024/2/22/350
>>> Please use https://lore.kernel.org/ URLs instead.  This one looks like
>>> https://lore.kernel.org/r/20240222090251.2849702-1-haifeng.zhao@linux.intel.com/
>>>
>>>> To make sure all the devTLB entries to be invalidated in the device release
>>>> path, do implict invalidation by fapping the E bit of ATS control register.
>>>> see PCIe spec v6.2, sec 10.3.7 implicit invalidation events.
>>> s/implict/implicit/
>>>
>>> s/fapping/?/  (no idea :)  "flipping"?  Oh, probably "flapping" per the
>>> comment below.  But I think "flapping" is ambiguous; "setting" would be
>>> better)
>> Yup, like the memory bit flipping, no idea what is the right word,
>> setting one bit to 0, then 1, then back to 0. perhaps details the
>> setting action 0-->1-->0 ?
> In PCIe spec-speak, "Set" means "assign 1 to this", and "Clear" means
> "assign 0 to this".
>
> Maybe you could copy the spec language like this:
>
>    Invalidate all ATC entries by changing the E field in the ATS
>    Capability from Clear to Set, which causes an implicit invalidation
>    event.

Fair enough.

Thanks,
Ethan

>
> Bjorn
>