Re: [PATCH v5 11/15] coco: host: arm64: Connect RMM pdev streams for IDE devices

[Aneesh Kumar K . V]

Ankit Agrawal <ankita@xxxxxxxxxx> writes:

> Hi Aneesh
>
> As I read it, the address ranges handed to RMM in this patch are captured
> once and at the point cca_tsm_connect() runs. That becomes the address
> range RMM has on record for the pdev/stream going forward.
>
> So a couple of questions on that..
> 1.  Is connect -> {driver probe / vfio-pci bind } the intended
>      sequence or is that ordering left entirely to userspace/VMM policy?
>

We can tsm connect the device even while its driver is bound.

> 2. Are the resources/BARs captured at connect time assumed to be
>     fixed for the lifetime of the PDEV? If a PCI rescan (or resource
>     reassignment such as pci=realloc, a hotplug re-enumeration
>     happens after connect, is that expected to be rejected/blocked?


IIUC, a normal rescan neither rereads nor relocates an already known and
assigned BAR? To discover an externally changed BAR, the device would
normally need to be removed before rescanning. Device removal reaches
pci_destroy_dev(), which calls pci_tsm_destroy() -> pci_tsm_disconnect()


-aneesh