Hi Shameer,
On 9/4/26 8:05 PM, Shameer Kolothum wrote:
> Extend the accel subsection post_load, after the IDR check, to bring the
> host side back in line. It allocates the vEVENTQ, which the A_CR0 write
> handler normally does, using the restored s->eventq.log2size. It attaches
> the HWPT for the restored GBPA to every device. Then it replays the STEs,
> which overrides that HWPT wherever the guest has a usable entry.
>
> Signed-off-by: Shameer Kolothum <[email protected]>
> ---
> hw/arm/smmuv3-accel.c | 22 ++++++++++++++++++++++
> 1 file changed, 22 insertions(+)
>
> diff --git a/hw/arm/smmuv3-accel.c b/hw/arm/smmuv3-accel.c
> index 6678e49f6e..b653f3f4ce 100644
> --- a/hw/arm/smmuv3-accel.c
> +++ b/hw/arm/smmuv3-accel.c
> @@ -1165,6 +1165,28 @@ static bool smmuv3_accel_vmstate_post_load(void
> *opaque, int version_id,
> return false;
> }
>
> + if (!s->s_accel->viommu) {
> + return true;
> + }
> +
> + /* Normally done by the A_CR0 write handler, which is not replayed. */
> + if (!smmuv3_accel_alloc_veventq(s, errp)) {
> + error_prepend(errp, "Failed to restore the vEVENTQ: ");
> + return false;
> + }
> +
> + /* smmuv3/gbpa is listed before this subsection, so it is in place */
> + if (!smmuv3_accel_attach_gbpa_hwpt(s, errp)) {
> + error_prepend(errp, "Failed to restore the GBPA HWPT: ");
> + return false;
> + }
I would recommend to document the order of operations somewhere, ie.
registers first, data structures in guest RAM and also list all the host
resources that need to be reconstructed
Looks almost too simple :-)
Eric
> +
> + /* Everything else comes from guest memory, which is already restored. */
> + if (!smmuv3_accel_replay_stes(s, errp)) {
> + error_prepend(errp, "Failed to replay the stream table: ");
> + return false;
> + }
> +
> return true;
> }
>
Eric
