On 21/06/2024 9:51 pm, Saravana Kannan wrote:
On Fri, Jun 21, 2024 at 8:46 PM Robin Murphy <robin.murphy@xxxxxxx> wrote:
v1: https://lore.kernel.org/linux-iommu/cover.1713523251.git.robin.murphy@xxxxxxx
Hi all,
Here's v2 of this little cleanup, with acks and the additional cosmetic
tweak suggested by Andy. There were some slightly non-trivial changes in
the rebase so I've left off Jean-Philippe's tested-by from v1, but I've
given it a quick spin on arm64 ACPI and DT and all seems well still.
Hi Robin,
I see in this series you talk about figuring out if a device has a
driver that could match. There has been a "can_match" flag in every
device that's set if a driver that match it is present, but hasn't
probed the device yet (for whatever reason). Just pointing that out in
case that makes things a lot easier for you. As of now, we don't
handle clearing it when the driver is unregistered, but if that really
needs to be handled, that shouldn't be too difficult.
Thanks, that's interesting to know. I'm not sure it's directly
applicable here since we have the more general case where the IOMMU
driver may also be a module that's not even loaded yet. What ultimately
matters is whether someone has called iommu_device_register() for a
matching IOMMU instance, or may do within a reasonable timeframe, so as
long as we can keep relying on fw_devlink and deferred_probe_timeout to
do most of the heavy lifting then I'd actually hope we can avoid getting
into the low-level details here.
Cheers,
Robin.
