As preparation for fixing mfn_to_gfn() on ARM, we extend the existing GFN
field in page_info's type_info to be usable for not only xenheap ones.
Another usage will be introduced later for stolen pages in memory exchaging.

Introduce general-purpose page_get_gfn() and page_set_gfn() helpers
that read and write the GFN stored in type_info. The old
page_get_xenheap_gfn() and page_set_xenheap_gfn() are retained as thin
wrappers with their xenheap ASSERTs, so all current callers remain unchanged.

Also introduce PGT_INVALID_GFN as the general sentinel, with
PGT_INVALID_XENHEAP_GFN aliased to it for backward compatibility.

Signed-off-by: Penny Zheng <penny.zheng@amd.com>
---
 xen/arch/arm/include/asm/mm.h | 37 +++++++++++++++++++++++------------
 1 file changed, 25 insertions(+), 12 deletions(-)

diff --git a/xen/arch/arm/include/asm/mm.h b/xen/arch/arm/include/asm/mm.h
index 72a6928624..d1873ec212 100644
--- a/xen/arch/arm/include/asm/mm.h
+++ b/xen/arch/arm/include/asm/mm.h
@@ -113,18 +113,21 @@ struct page_info
 #define PGT_count_mask    PG_mask(3, 3)
 
 /*
- * Stored in bits [28:0] (arm32) or [60:0] (arm64) GFN if page is xenheap page.
+ * Stored in bits [28:0] (arm32) or [60:0] (arm64) GFN if page is xenheap page,
+ * or stolen ones in memory exchanging.
  */
 #define PGT_gfn_width     PG_shift(3)
 #define PGT_gfn_mask      ((1UL<<PGT_gfn_width)-1)
 
-#define PGT_INVALID_XENHEAP_GFN   _gfn(PGT_gfn_mask)
+#define PGT_INVALID_GFN           _gfn(PGT_gfn_mask)
+#define PGT_INVALID_XENHEAP_GFN   PGT_INVALID_GFN
 
 /*
  * An arch-specific initialization pattern is needed for the type_info field
- * as it's GFN portion can contain the valid GFN if page is xenheap page.
+ * as it's GFN portion can contain the valid GFN if page is xenheap page, or
+ * stolen in memory exchanging.
  */
-#define PGT_TYPE_INFO_INITIALIZER   gfn_x(PGT_INVALID_XENHEAP_GFN)
+#define PGT_TYPE_INFO_INITIALIZER   gfn_x(PGT_INVALID_GFN)
 
  /* Cleared when the owning guest 'frees' this page. */
 #define _PGC_allocated    PG_shift(1)
@@ -338,28 +341,38 @@ unsigned int arch_get_dma_bitsize(void);
  * that requirement (risk of deadlock, lock inversion, etc) it is important
  * to make sure that all non-protected updates to this field are atomic.
  */
-static inline gfn_t page_get_xenheap_gfn(const struct page_info *p)
+static inline gfn_t page_get_gfn(const struct page_info *p)
 {
     gfn_t gfn_ = _gfn(ACCESS_ONCE(p->u.inuse.type_info) & PGT_gfn_mask);
 
-    ASSERT(is_xen_heap_page(p));
-
-    return gfn_eq(gfn_, PGT_INVALID_XENHEAP_GFN) ? INVALID_GFN : gfn_;
+    return gfn_eq(gfn_, PGT_INVALID_GFN) ? INVALID_GFN : gfn_;
 }
 
-static inline void page_set_xenheap_gfn(struct page_info *p, gfn_t gfn)
+static inline void page_set_gfn(struct page_info *p, gfn_t gfn)
 {
-    gfn_t gfn_ = gfn_eq(gfn, INVALID_GFN) ? PGT_INVALID_XENHEAP_GFN : gfn;
+    gfn_t gfn_ = gfn_eq(gfn, INVALID_GFN) ? PGT_INVALID_GFN : gfn;
     unsigned long x, nx, y = p->u.inuse.type_info;
 
-    ASSERT(is_xen_heap_page(p));
-
     do {
         x = y;
         nx = (x & ~PGT_gfn_mask) | gfn_x(gfn_);
     } while ( (y = cmpxchg(&p->u.inuse.type_info, x, nx)) != x );
 }
 
+static inline gfn_t page_get_xenheap_gfn(const struct page_info *p)
+{
+    ASSERT(is_xen_heap_page(p));
+
+    return page_get_gfn(p);
+}
+
+static inline void page_set_xenheap_gfn(struct page_info *p, gfn_t gfn)
+{
+    ASSERT(is_xen_heap_page(p));
+
+    page_set_gfn(p, gfn);
+}
+
 #endif /*  __ARCH_ARM_MM__ */
 /*
  * Local variables:
-- 
2.43.0

On 27.03.2026 08:50, Penny Zheng wrote:
> As preparation for fixing mfn_to_gfn() on ARM, we extend the existing GFN
> field in page_info's type_info to be usable for not only xenheap ones.
> Another usage will be introduced later for stolen pages in memory exchaging.
> 
> Introduce general-purpose page_get_gfn() and page_set_gfn() helpers
> that read and write the GFN stored in type_info. The old
> page_get_xenheap_gfn() and page_set_xenheap_gfn() are retained as thin
> wrappers with their xenheap ASSERTs, so all current callers remain unchanged.

Why was this GFN setting limited to Xenheap pages back at the time? Depending
on the reasons, retaining the old accessors may or may not be a good idea.

> Also introduce PGT_INVALID_GFN as the general sentinel, with
> PGT_INVALID_XENHEAP_GFN aliased to it for backward compatibility.

This I view as unnecessary, if not confusing.

> --- a/xen/arch/arm/include/asm/mm.h
> +++ b/xen/arch/arm/include/asm/mm.h
> @@ -113,18 +113,21 @@ struct page_info
>  #define PGT_count_mask    PG_mask(3, 3)
>  
>  /*
> - * Stored in bits [28:0] (arm32) or [60:0] (arm64) GFN if page is xenheap page.
> + * Stored in bits [28:0] (arm32) or [60:0] (arm64) GFN if page is xenheap page,
> + * or stolen ones in memory exchanging.
>   */

Does the purpose really need limiting like this? If the field covered by PGT_gfn_*
is uniformly available (see the question above), I don't see why a new constraint
would need spelling out. If it's not uniformly available, then likely the
description needs expanding as to when the new accessors are okay to use. If
uniformly available, what may want spelling out is under what conditions one can
expect the field to be properly set (until such time where it's set correctly on
all guest-owned pages).

Jan

[Public]

Hi,

> -----Original Message-----
> From: Jan Beulich <jbeulich@suse.com>
> Sent: Monday, March 30, 2026 9:49 PM
> To: Penny, Zheng <penny.zheng@amd.com>
> Cc: Huang, Ray <Ray.Huang@amd.com>; Stefano Stabellini
> <sstabellini@kernel.org>; Julien Grall <julien@xen.org>; Bertrand Marquis
> <bertrand.marquis@arm.com>; Orzel, Michal <Michal.Orzel@amd.com>; Volodymyr
> Babchuk <Volodymyr_Babchuk@epam.com>; xen-devel@lists.xenproject.org;
> Garcia Vallejo, Alejandro <Alejandro.GarciaVallejo@amd.com>
> Subject: Re: [PATCH v1 1/3] xen/arm: generalize per-page GFN storage beyond
> xenheap pages
>
> On 27.03.2026 08:50, Penny Zheng wrote:
> > As preparation for fixing mfn_to_gfn() on ARM, we extend the existing
> > GFN field in page_info's type_info to be usable for not only xenheap ones.
> > Another usage will be introduced later for stolen pages in memory exchaging.
> >
> > Introduce general-purpose page_get_gfn() and page_set_gfn() helpers
> > that read and write the GFN stored in type_info. The old
> > page_get_xenheap_gfn() and page_set_xenheap_gfn() are retained as thin
> > wrappers with their xenheap ASSERTs, so all current callers remain unchanged.
>
> Why was this GFN setting limited to Xenheap pages back at the time? Depending on
> the reasons, retaining the old accessors may or may not be a good idea.
>

The only call site is to set shared_info gfn. I assumed GFN setting could be uniformly available.
FWIT, the only limitation is that it shall not be applied to shared pages. In arm, I think it's static shared memory.

> > Also introduce PGT_INVALID_GFN as the general sentinel, with
> > PGT_INVALID_XENHEAP_GFN aliased to it for backward compatibility.
>
> This I view as unnecessary, if not confusing.
>
> > --- a/xen/arch/arm/include/asm/mm.h
> > +++ b/xen/arch/arm/include/asm/mm.h
> > @@ -113,18 +113,21 @@ struct page_info
> >  #define PGT_count_mask    PG_mask(3, 3)
> >
> >  /*
> > - * Stored in bits [28:0] (arm32) or [60:0] (arm64) GFN if page is xenheap page.
> > + * Stored in bits [28:0] (arm32) or [60:0] (arm64) GFN if page is
> > + xenheap page,
> > + * or stolen ones in memory exchanging.
> >   */
>
> Does the purpose really need limiting like this? If the field covered by PGT_gfn_* is
> uniformly available (see the question above), I don't see why a new constraint would
> need spelling out. If it's not uniformly available, then likely the description needs
> expanding as to when the new accessors are okay to use. If uniformly available,
> what may want spelling out is under what conditions one can expect the field to be
> properly set (until such time where it's set correctly on all guest-owned pages).
>
> Jan