Re: [PATCH v5 8/9] vpci: allow queueing of mapping operations

On 08.07.2026 23:02, Stewart Hildebrand wrote:
> +static struct map_task *alloc_map_task(const struct pci_dev *pdev, bool map,
> +                                       bool rom_only)
> +{
> +    struct map_task *task;
> +    unsigned int i;
> +
> +    task = xvzalloc(struct map_task);
> +    if ( !task )
> +        return NULL;

Just one remark here, as generally I'm trying to stay away from this: From what
Stefano said on earlier occasions (more generally, and not on any vPCI-specific
thread) new runtime allocations want avoiding for the purposes that you have.
How come such a crucial one, required for correct operation at potentially any
point in time, is then deemed acceptable? Even more so as, from the description,
only single items will be queued right now anyway, i.e. this apparently being a
step towards worse overall behavior? This aspect imo needs explicitly discussing
and justifying in the description.

Jan