Re: [PATCH mm-unstable] arch, mm: consolidate empty_zero_page

[Mike Rapoport <rppt@kernel.org>]

On Tue, Feb 03, 2026 at 11:17:08AM +0100, Andreas Larsson wrote:
> On 2026-01-27 19:11, Mike Rapoport wrote:
> > On Tue, Jan 27, 2026 at 05:02:39PM +0100, Andreas Larsson wrote:
> >> On 2026-01-24 10:56, Mike Rapoport wrote:
> >>
> >>> Every architecture defines empty_zero_page that way or another, but for the
> >>> most of them it is always a page aligned page in BSS and most definitions
> >>> of ZERO_PAGE do virt_to_page(empty_zero_page).
> >>
> >> Running this in an LDOM on an UltraSparc T4 sparc64, the entire LDOM
> >> hangs after a while during boot.
> >>
> >>> diff --git a/arch/sparc/mm/init_64.c b/arch/sparc/mm/init_64.c
> >>> index c2d19c9a9244..2bd99944176d 100644
> >>> --- a/arch/sparc/mm/init_64.c
> >>> +++ b/arch/sparc/mm/init_64.c
> >>> @@ -177,9 +177,6 @@ extern unsigned long sparc_ramdisk_image64;
> >>>  extern unsigned int sparc_ramdisk_image;
> >>>  extern unsigned int sparc_ramdisk_size;
> >>>  
> >>> -struct page *mem_map_zero __read_mostly;
> >>> -EXPORT_SYMBOL(mem_map_zero);
> >>> -
> >>>  unsigned int sparc64_highest_unlocked_tlb_ent __read_mostly;
> >>>  
> >>>  unsigned long sparc64_kern_pri_context __read_mostly;
> >>> @@ -2506,18 +2503,6 @@ void __init mem_init(void)
> >>>  	 */
> >>>  	register_page_bootmem_info();
> >>>  
> >>> -	/*
> >>> -	 * Set up the zero page, mark it reserved, so that page count
> >>> -	 * is not manipulated when freeing the page from user ptes.
> >>> -	 */
> >>> -	mem_map_zero = alloc_pages(GFP_KERNEL|__GFP_ZERO, 0);
> >>> -	if (mem_map_zero == NULL) {
> >>> -		prom_printf("paging_init: Cannot alloc zero page.\n");
> >>> -		prom_halt();
> >>> -	}
> >>> -	mark_page_reserved(mem_map_zero);
> >>> -
> >>> -
> >>>  	if (tlb_type == cheetah || tlb_type == cheetah_plus)
> >>>  		cheetah_ecache_flush_init();
> >>>  }
> >>
> >> This just removes the mark_page_reserved(mem_map_zero) without 
> >> replacing it with something corresponding to that. Perhaps part
> >> of the problem?
> > 
> > I don't think so, empty_zero_page is in BSS now an it's reserved as a part
> > of the kernel image.
> > 
> > I suspect that virt_to_page() does not work BSS symbols on sparc64. Can you
> > please try with this patch:
> > 
> > diff --git a/arch/sparc/include/asm/pgtable_64.h b/arch/sparc/include/asm/pgtable_64.h
> > index 74ede706fb32..0578c5172d4e 100644
> > --- a/arch/sparc/include/asm/pgtable_64.h
> > +++ b/arch/sparc/include/asm/pgtable_64.h
> > @@ -22,6 +22,7 @@
> >  #include <asm/adi.h>
> >  #include <asm/page.h>
> >  #include <asm/processor.h>
> > +#include <asm/vaddrs.h>
> >  
> >  /* The kernel image occupies 0x4000000 to 0x6000000 (4MB --> 96MB).
> >   * The page copy blockops can use 0x6000000 to 0x8000000.
> > @@ -210,6 +211,11 @@ extern unsigned long _PAGE_CACHE;
> >  extern unsigned long pg_iobits;
> >  extern unsigned long _PAGE_ALL_SZ_BITS;
> >  
> > +extern unsigned long kern_base;
> > +#define ZERO_PAGE(vaddr)						   \
> > +	(virt_to_page(empty_zero_page + ((unsigned long)__va(kern_base)) - \
> > +		      ((unsigned long)KERNBASE)))
> > +
> >  /* PFNs are real physical page numbers.  However, mem_map only begins to record
> >   * per-page information starting at pfn_base.  This is to handle systems where
> >   * the first physical page in the machine is at some huge physical address,
> > diff --git a/arch/sparc/mm/init_64.c b/arch/sparc/mm/init_64.c
> > index 2bd99944176d..d2d724ba4f83 100644
> > --- a/arch/sparc/mm/init_64.c
> > +++ b/arch/sparc/mm/init_64.c
> > @@ -170,6 +170,8 @@ static void __init read_obp_memory(const char *property,
> >  
> >  /* Kernel physical address base and size in bytes.  */
> >  unsigned long kern_base __read_mostly;
> > +EXPORT_SYMBOL(kern_base);
> > +
> >  unsigned long kern_size __read_mostly;
> >  
> >  /* Initial ramdisk setup */
> Hi,
> 
> Unfortunately, that does not help. The LDOM goes down in the same fashion.

Apparently something is wrong with my pointer arithmetics :/

Can you try this one instead?

diff --git a/arch/sparc/include/asm/pgtable_64.h b/arch/sparc/include/asm/pgtable_64.h
index 74ede706fb32..615f460c50af 100644
--- a/arch/sparc/include/asm/pgtable_64.h
+++ b/arch/sparc/include/asm/pgtable_64.h
@@ -210,6 +210,9 @@ extern unsigned long _PAGE_CACHE;
 extern unsigned long pg_iobits;
 extern unsigned long _PAGE_ALL_SZ_BITS;
 
+extern struct page *mem_map_zero;
+#define ZERO_PAGE(vaddr)	(mem_map_zero)
+
 /* PFNs are real physical page numbers.  However, mem_map only begins to record
  * per-page information starting at pfn_base.  This is to handle systems where
  * the first physical page in the machine is at some huge physical address,
diff --git a/arch/sparc/mm/init_64.c b/arch/sparc/mm/init_64.c
index 2bd99944176d..aa1f9f071fb2 100644
--- a/arch/sparc/mm/init_64.c
+++ b/arch/sparc/mm/init_64.c
@@ -177,6 +177,9 @@ extern unsigned long sparc_ramdisk_image64;
 extern unsigned int sparc_ramdisk_image;
 extern unsigned int sparc_ramdisk_size;
 
+struct page *mem_map_zero __read_mostly;
+EXPORT_SYMBOL(mem_map_zero);
+
 unsigned int sparc64_highest_unlocked_tlb_ent __read_mostly;
 
 unsigned long sparc64_kern_pri_context __read_mostly;
@@ -2495,6 +2498,9 @@ static void __init register_page_bootmem_info(void)
 }
 void __init mem_init(void)
 {
+	phys_addr_t zero_page_pa = kern_base +
+		((unsigned long)&empty_zero_page[0] - KERNBASE);
+
 	/*
 	 * Must be done after boot memory is put on freelist, because here we
 	 * might set fields in deferred struct pages that have not yet been
@@ -2503,6 +2509,12 @@ void __init mem_init(void)
 	 */
 	register_page_bootmem_info();
 
+	/*
+	 * Set up the zero page, mark it reserved, so that page count
+	 * is not manipulated when freeing the page from user ptes.
+	 */
+	mem_map_zero = pfn_to_page(PHYS_PFN(zero_page_pa));
+
 	if (tlb_type == cheetah || tlb_type == cheetah_plus)
 		cheetah_ecache_flush_init();
 }

-- 
Sincerely yours,
Mike.