Re: [PATCH v8 4/5] xen/riscv: probe the A/D update scheme if the DT advertises neither

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

On 2026-10-26 16:35:44+01:00, Oleksii Kurochko wrote:
> On 10/26/26 4:14 PM, Baptiste Le Duc wrote:
> 
> > +/*
> > + * Probe whether hardware updates the PTE A/D bits by writing to a local
> > + * variable through an alias mapping whose A and D bits are clear.
> > + *
> > + * Check the D bit rather than the A bit, as the A bit may be set
> > + * speculatively while the D bit can't.
> > + *
> > + * Under Svade, the store raises a page fault instead, so use the extable 
> > to
> > + * handle it.
> > + */
> > +static bool __init probe_hw_ad_update(void)
> > +{
> > +    unsigned int probe_target;
> > +    vaddr_t va = FIXMAP_ADDR(FIX_MISC) + PAGE_OFFSET(&probe_target);
> > +    pte_t *entry = &xen_fixmap[FIX_MISC];
> > +    bool d_set = false;
> > +
> > +    ASSERT(!pte_is_valid(*entry));
> > +
> > +    /*
> > +     * Any page in use will do, so alias the stack page this code runs on,
> > +     * with A and D clear. Store to a variable of its own, so that nothing
> > +     * the compiler relies on gets overwritten.
> > +     */
> > +    write_pte(entry, 
> > pte_from_mfn(maddr_to_mfn(virt_to_maddr(&probe_target)),
> > +                                  PTE_VALID | PTE_READABLE | 
> > PTE_WRITABLE));
> > +    flush_tlb_one_local(va);
> > +
> > +    asm_inline goto (
> 
> I think that generally this code should be wrapped by "#ifdef 
> CONFIG_CC_HAS_ASM_GOTO_OUTPUT" but we can keep it as is here as we don't 
> use output operands.
>
Yes I didn't see any sense to use it as, from my understanding, its
aimed to be use when we want to have label as output operand which is
not our case here.
> [ ... 81 lines skipped ... ]
> Generally, looks good to me:
Thanks!

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