Re: [PATCH v3 30/39] xen/riscv: implement APLIC-IMSIC-hart sync barrier for vCPU migration

[Baptiste Le Duc <baptiste.le-duc@xxxxxxxxxx>]

On 11/5/26 4:51 PM, Oleksii Kurochko wrote:
> On 11/5/26 4:32 PM, Baptiste Le Duc wrote:
> > On 2026-11-05 15:21 +0100, Oleksii Kurochko wrote:
> > > [ ... 348 lines skipped ... ]
> > > > LGTM, but I wonder why, as I said in the original patch, sync_id is
> > > > really needed.
> > > I think that I am not understand your response. Could you please
> > > re-phrase it?
> > I'm questionning the need of the sync_id field as it will always be
> > equal to imsic_cfg.nr_ids, couldn't we drop it?
> It's equal to nr_ids by choice rather than by necessity: which identity
> gets reserved is decided in a single place, imsic_parse_node(), and all
> the users (the genmsi write, the pending bit the barrier spins on, the
> limit on the APLIC sources and the ASSERT() in imsic_irq_enable()) refer
> to it by its role rather than as "the last identity".
>
> Basically if one day someone will decide to come up with EIID allocator 
> instead of using static allocation then it will be harder to find a 
> places where to change.
> So yes, it could be drooped but at the moment I think it is useful to 
> have separate field or ,at least, as an option in asm/imsic.h:
> static inline unsigned int imsic_sync_id(const struct imsic_config *cfg)
> {
>      return cfg->nr_ids;
> }
I prefer this as it would save space in imsic_config struct.

> ~ Oleksii