[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <aYcVu7gs65S1CeST@kernel.org>
Date: Sat, 7 Feb 2026 12:36:43 +0200
From: Mike Rapoport <rppt@...nel.org>
To: Andreas Larsson <andreas@...sler.com>
Cc: Andrew Morton <akpm@...ux-foundation.org>,
Borislav Petkov <bp@...en8.de>, Brian Cain <bcain@...nel.org>,
Catalin Marinas <catalin.marinas@....com>,
"David S. Miller" <davem@...emloft.net>,
Dave Hansen <dave.hansen@...ux.intel.com>,
David Hildenbrand <david@...nel.org>,
Dinh Nguyen <dinguyen@...nel.org>,
Geert Uytterhoeven <geert@...ux-m68k.org>,
Guo Ren <guoren@...nel.org>, Helge Deller <deller@....de>,
Huacai Chen <chenhuacai@...nel.org>, Ingo Molnar <mingo@...hat.com>,
Johannes Berg <johannes@...solutions.net>,
John Paul Adrian Glaubitz <glaubitz@...sik.fu-berlin.de>,
"Liam R. Howlett" <Liam.Howlett@...cle.com>,
Lorenzo Stoakes <lorenzo.stoakes@...cle.com>,
Madhavan Srinivasan <maddy@...ux.ibm.com>,
Magnus Lindholm <linmag7@...il.com>,
Matt Turner <mattst88@...il.com>, Max Filippov <jcmvbkbc@...il.com>,
Michael Ellerman <mpe@...erman.id.au>,
Michal Hocko <mhocko@...e.com>, Michal Simek <monstr@...str.eu>,
Palmer Dabbelt <palmer@...belt.com>,
Richard Weinberger <richard@....at>,
Russell King <linux@...linux.org.uk>,
Stafford Horne <shorne@...il.com>,
Suren Baghdasaryan <surenb@...gle.com>,
Thomas Gleixner <tglx@...nel.org>, Vineet Gupta <vgupta@...nel.org>,
Vlastimil Babka <vbabka@...e.cz>, Will Deacon <will@...nel.org>,
linux-alpha@...r.kernel.org, linux-kernel@...r.kernel.org,
linux-snps-arc@...ts.infradead.org,
linux-arm-kernel@...ts.infradead.org, linux-csky@...r.kernel.org,
linux-hexagon@...r.kernel.org, loongarch@...ts.linux.dev,
linux-m68k@...ts.linux-m68k.org, linux-openrisc@...r.kernel.org,
linux-parisc@...r.kernel.org, linuxppc-dev@...ts.ozlabs.org,
linux-riscv@...ts.infradead.org, linux-sh@...r.kernel.org,
sparclinux@...r.kernel.org, linux-um@...ts.infradead.org,
linux-mm@...ck.org, x86@...nel.org
Subject: Re: [PATCH mm-unstable] arch, mm: consolidate empty_zero_page
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.
Powered by blists - more mailing lists