On Wed, May 14, 2025 at 01:33:38PM +0800, Chao Gao wrote:
> >+static void tdx_pamt_put(struct page *page)
> >+{
> >+ unsigned long hpa = page_to_phys(page);
> >+ atomic_t *pamt_refcount;
> >+ LIST_HEAD(pamt_pages);
> >+ u64 err;
> >+
> >+ if (!tdx_supports_dynamic_pamt(tdx_sysinfo))
> >+ return;
> >+
> >+ hpa = ALIGN_DOWN(hpa, SZ_2M);
> >+
> >+ pamt_refcount = tdx_get_pamt_refcount(hpa);
> >+ if (!atomic_dec_and_test(pamt_refcount))
> >+ return;
> >+
> >+ spin_lock(&pamt_lock);
> >+
> >+ /* Lost race against tdx_pamt_add()? */
> >+ if (atomic_read(pamt_refcount) != 0) {
> >+ spin_unlock(&pamt_lock);
> >+ return;
> >+ }
> >+
> >+ err = tdh_phymem_pamt_remove(hpa | TDX_PS_2M, &pamt_pages);
> >+ spin_unlock(&pamt_lock);
> >+
> >+ if (err) {
> >+ pr_tdx_error(TDH_PHYMEM_PAMT_REMOVE, err);
>
> Should the refcount be increased here, since the PAMT pages are not removed?
Right. Thanks.
--
Kiryl Shutsemau / Kirill A. Shutemov
