Re: [PATCH v2 11/20] hw/core: Implement Remote Port Memory Master bus transactions

[Itai Handler <itai.handler@gmail.com>]

Hi Ruslan,

On 9/15/26 18:00, Ruslan Ruslichenko wrote:
[...]
> +    rp_write(rp, (void *) &pay, len);
> +
> +    if (!rp_timeout) {
> +        rsp_slot = rp_dev_wait_resp(rp, in.dev, in.id);

Thank you for this series.  I have been testing v2 by booting a guest
from flash behind a QSPI controller modelled as a Remote Port peer.
Every guest access to the memory master waits here for a round trip to
the peer, about 16 us, and U-Boot and Linux drain the controller's
FIFO one 32-bit word at a time, so a 15 MiB kernel image takes 64 s to
load.

In my tests, a small read FIFO in the memory master, which the peer
fills, makes such reads 13x faster (the kernel image loads in 4.9 s).
The peer may send a posted write before its response to a read, and
the guest's next reads of that range are then served from the FIFO
without a round trip.

Would you consider including something like this in v3?  I have a
working patch on top of v2 (about 140 lines) that I would be glad to
send you or post as an RFC, whichever you prefer, and I am happy to
test v3.

Thanks,
Itai