Re: [PATCH] x86/mm/init: respect memblock reserved regions whendestroying mappings

From: Stefano Stabellini
Date: Tue Feb 08 2011 - 09:01:37 EST

Next message: Bartlomiej Zolnierkiewicz: "Re: [PATCH 16/20] ata_piix: add EFAR SLC90E66 support"
Previous message: martin f krafft: "Re: Opteron ECC/ChipKill error"
In reply to: Yinghai Lu: "Re: [PATCH] x86/mm/init: respect memblock reserved regions whendestroying mappings"
Next in thread: Yinghai Lu: "Re: [PATCH] x86/mm/init: respect memblock reserved regions when destroyingmappings"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Tue, 8 Feb 2011, Yinghai Lu wrote:
> On Mon, Feb 7, 2011 at 11:00 AM, Stefano Stabellini
> <stefano.stabellini@xxxxxxxxxxxxx> wrote:
> > On Mon, 7 Feb 2011, Stefano Stabellini wrote:
> >> On Sun, 6 Feb 2011, Yinghai Lu wrote:
> >> > On 02/05/2011 11:30 PM, H. Peter Anvin wrote:
> >> > > On 02/05/2011 11:02 PM, Yinghai Lu wrote:
> >> > >> why not just move calling cleanup_highmap down?
> >> > >>
> >> > >> something like attached patch.
> >> > >
> >> > > This patch looks very clean and looks on the surface of it like it is
> >> > > removing some ugly ad hoc code, but (as always) it needs a description
> >> > > about the problem it solves and why it is correct.
> >> >
> >> > Sure.
> >> >
> >> >
> >> > Jeremy and xen guys, can you please check if it works well with xen ?
> >> >
> >>
> >> Actually this patch makes things worse on xen, because before
> >> cleanup_highmap() wasn't called at all on xen (on purpose) and now it
> >> is, fully destroying all the mappings we have at _end.
> >>
> >> Can we add a check on memblock reserved regions in cleanup_highmap()?
> >
> > In case you are wondering how Yinghai Lu's patch would look like with
> > the added check, here it is:
> >
> >
> > diff --git a/arch/x86/include/asm/memblock.h b/arch/x86/include/asm/memblock.h
> > index 19ae14b..184f778 100644
> > --- a/arch/x86/include/asm/memblock.h
> > +++ b/arch/x86/include/asm/memblock.h
> > @@ -3,6 +3,7 @@
> >
> > Â#define ARCH_DISCARD_MEMBLOCK
> >
> > +bool memblock_check_reserved_size(u64 *addrp, u64 *sizep, u64 align);
> > Âu64 memblock_x86_find_in_range_size(u64 start, u64 *sizep, u64 align);
> > Âvoid memblock_x86_to_bootmem(u64 start, u64 end);
> >
> > diff --git a/arch/x86/include/asm/pgtable_64.h b/arch/x86/include/asm/pgtable_64.h
> > index 975f709..28686b6 100644
> > --- a/arch/x86/include/asm/pgtable_64.h
> > +++ b/arch/x86/include/asm/pgtable_64.h
> > @@ -165,7 +165,7 @@ static inline int pgd_large(pgd_t pgd) { return 0; }
> > Â#define __swp_entry_to_pte(x) Â Â Â Â Â((pte_t) { .pte = (x).val })
> >
> > Âextern int kern_addr_valid(unsigned long addr);
> > -extern void cleanup_highmap(void);
> > +extern void cleanup_highmap(unsigned long end);
> >
> > Â#define HAVE_ARCH_UNMAPPED_AREA
> > Â#define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
> > diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
> > index 2d2673c..5655c22 100644
> > --- a/arch/x86/kernel/head64.c
> > +++ b/arch/x86/kernel/head64.c
> > @@ -77,9 +77,6 @@ void __init x86_64_start_kernel(char * real_mode_data)
> > Â Â Â Â/* Make NULL pointers segfault */
> > Â Â Â Âzap_identity_mappings();
> >
> > - Â Â Â /* Cleanup the over mapped high alias */
> > - Â Â Â cleanup_highmap();
> > -
> > Â Â Â Âmax_pfn_mapped = KERNEL_IMAGE_SIZE >> PAGE_SHIFT;
> >
> > Â Â Â Âfor (i = 0; i < NUM_EXCEPTION_VECTORS; i++) {
> > diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
> > index d3cfe26..91afde6 100644
> > --- a/arch/x86/kernel/setup.c
> > +++ b/arch/x86/kernel/setup.c
> > @@ -297,6 +297,9 @@ static void __init init_gbpages(void)
> > Âstatic inline void init_gbpages(void)
> > Â{
> > Â}
> > +static void __init cleanup_highmap(unsigned long end)
> > +{
> > +}
> > Â#endif
> >
> > Âstatic void __init reserve_brk(void)
> > @@ -922,6 +925,9 @@ void __init setup_arch(char **cmdline_p)
> > Â Â Â Â */
> > Â Â Â Âreserve_brk();
> >
> > + Â Â Â /* Cleanup the over mapped high alias after _brk_end*/
> > + Â Â Â cleanup_highmap(_brk_end);
> > +
> > Â Â Â Âmemblock.current_limit = get_max_mapped();
> > Â Â Â Âmemblock_x86_fill();
> >
> > diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
> > index 947f42a..f13ff3a 100644
> > --- a/arch/x86/mm/init.c
> > +++ b/arch/x86/mm/init.c
> > @@ -279,25 +279,6 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
> > Â Â Â Âload_cr3(swapper_pg_dir);
> > Â#endif
> >
> > -#ifdef CONFIG_X86_64
> > - Â Â Â if (!after_bootmem && !start) {
> > - Â Â Â Â Â Â Â pud_t *pud;
> > - Â Â Â Â Â Â Â pmd_t *pmd;
> > -
> > - Â Â Â Â Â Â Â mmu_cr4_features = read_cr4();
> > -
> > - Â Â Â Â Â Â Â /*
> > - Â Â Â Â Â Â Â Â* _brk_end cannot change anymore, but it and _end may be
> > - Â Â Â Â Â Â Â Â* located on different 2M pages. cleanup_highmap(), however,
> > - Â Â Â Â Â Â Â Â* can only consider _end when it runs, so destroy any
> > - Â Â Â Â Â Â Â Â* mappings beyond _brk_end here.
> > - Â Â Â Â Â Â Â Â*/
> > - Â Â Â Â Â Â Â pud = pud_offset(pgd_offset_k(_brk_end), _brk_end);
> > - Â Â Â Â Â Â Â pmd = pmd_offset(pud, _brk_end - 1);
> > - Â Â Â Â Â Â Â while (++pmd <= pmd_offset(pud, (unsigned long)_end - 1))
> > - Â Â Â Â Â Â Â Â Â Â Â pmd_clear(pmd);
> > - Â Â Â }
> > -#endif
> > Â Â Â Â__flush_tlb_all();
> >
> > Â Â Â Âif (!after_bootmem && e820_table_end > e820_table_start)
> > diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
> > index 71a5929..028c49e 100644
> > --- a/arch/x86/mm/init_64.c
> > +++ b/arch/x86/mm/init_64.c
> > @@ -297,18 +297,26 @@ void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
> > Â* rounded up to the 2MB boundary. This catches the invalid pmds as
> > Â* well, as they are located before _text:
> > Â*/
> > -void __init cleanup_highmap(void)
> > +void __init cleanup_highmap(unsigned long end)
> > Â{
> > Â Â Â Âunsigned long vaddr = __START_KERNEL_map;
> > - Â Â Â unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
> > Â Â Â Âpmd_t *pmd = level2_kernel_pgt;
> > Â Â Â Âpmd_t *last_pmd = pmd + PTRS_PER_PMD;
> > + Â Â Â u64 size, addrp;
> > + Â Â Â bool changed;
> > +
> > + Â Â Â end = roundup(end, PMD_SIZE) - 1;
> >
> > Â Â Â Âfor (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
> > Â Â Â Â Â Â Â Âif (pmd_none(*pmd))
> > Â Â Â Â Â Â Â Â Â Â Â Âcontinue;
> > - Â Â Â Â Â Â Â if (vaddr < (unsigned long) _text || vaddr > end)
> > - Â Â Â Â Â Â Â Â Â Â Â set_pmd(pmd, __pmd(0));
> > + Â Â Â Â Â Â Â if (vaddr < (unsigned long) _text || vaddr > end) {
> > + Â Â Â Â Â Â Â Â Â Â Â addrp = __pa(vaddr);
> > + Â Â Â Â Â Â Â Â Â Â Â size = PMD_SIZE;
> > + Â Â Â Â Â Â Â Â Â Â Â changed = memblock_check_reserved_size(&addrp, &size, PMD_SIZE);
> > + Â Â Â Â Â Â Â Â Â Â Â if (!changed && size)
> > + Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â set_pmd(pmd, __pmd(0));
> > + Â Â Â Â Â Â Â }
>
> for native path, memblock_check_reserved_size() are called 256 times
> without obvious reasons.

what about this patch, does it look like a reasonable solution?

diff --git a/arch/x86/include/asm/memblock.h b/arch/x86/include/asm/memblock.h
index 19ae14b..184f778 100644
--- a/arch/x86/include/asm/memblock.h
+++ b/arch/x86/include/asm/memblock.h
@@ -3,6 +3,7 @@

#define ARCH_DISCARD_MEMBLOCK

+bool memblock_check_reserved_size(u64 *addrp, u64 *sizep, u64 align);
u64 memblock_x86_find_in_range_size(u64 start, u64 *sizep, u64 align);
void memblock_x86_to_bootmem(u64 start, u64 end);

diff --git a/arch/x86/include/asm/pgtable_64.h b/arch/x86/include/asm/pgtable_64.h
index 975f709..28686b6 100644
--- a/arch/x86/include/asm/pgtable_64.h
+++ b/arch/x86/include/asm/pgtable_64.h
@@ -165,7 +165,7 @@ static inline int pgd_large(pgd_t pgd) { return 0; }
#define __swp_entry_to_pte(x) ((pte_t) { .pte = (x).val })

extern int kern_addr_valid(unsigned long addr);
-extern void cleanup_highmap(void);
+extern void cleanup_highmap(unsigned long end);

#define HAVE_ARCH_UNMAPPED_AREA
#define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index 2d2673c..5655c22 100644
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -77,9 +77,6 @@ void __init x86_64_start_kernel(char * real_mode_data)
/* Make NULL pointers segfault */
zap_identity_mappings();

- /* Cleanup the over mapped high alias */
- cleanup_highmap();
-
max_pfn_mapped = KERNEL_IMAGE_SIZE >> PAGE_SHIFT;

for (i = 0; i < NUM_EXCEPTION_VECTORS; i++) {
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index d3cfe26..91afde6 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -297,6 +297,9 @@ static void __init init_gbpages(void)
static inline void init_gbpages(void)
{
}
+static void __init cleanup_highmap(unsigned long end)
+{
+}
#endif

static void __init reserve_brk(void)
@@ -922,6 +925,9 @@ void __init setup_arch(char **cmdline_p)
*/
reserve_brk();

+ /* Cleanup the over mapped high alias after _brk_end*/
+ cleanup_highmap(_brk_end);
+
memblock.current_limit = get_max_mapped();
memblock_x86_fill();

diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
index 947f42a..f13ff3a 100644
--- a/arch/x86/mm/init.c
+++ b/arch/x86/mm/init.c
@@ -279,25 +279,6 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
load_cr3(swapper_pg_dir);
#endif

-#ifdef CONFIG_X86_64
- if (!after_bootmem && !start) {
- pud_t *pud;
- pmd_t *pmd;
-
- mmu_cr4_features = read_cr4();
-
- /*
- * _brk_end cannot change anymore, but it and _end may be
- * located on different 2M pages. cleanup_highmap(), however,
- * can only consider _end when it runs, so destroy any
- * mappings beyond _brk_end here.
- */
- pud = pud_offset(pgd_offset_k(_brk_end), _brk_end);
- pmd = pmd_offset(pud, _brk_end - 1);
- while (++pmd <= pmd_offset(pud, (unsigned long)_end - 1))
- pmd_clear(pmd);
- }
-#endif
__flush_tlb_all();

if (!after_bootmem && e820_table_end > e820_table_start)
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 71a5929..90a64de 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -297,12 +297,25 @@ void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
* rounded up to the 2MB boundary. This catches the invalid pmds as
* well, as they are located before _text:
*/
-void __init cleanup_highmap(void)
+void __init cleanup_highmap(unsigned long end)
{
unsigned long vaddr = __START_KERNEL_map;
- unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
pmd_t *pmd = level2_kernel_pgt;
pmd_t *last_pmd = pmd + PTRS_PER_PMD;
+ u64 size, addrp;
+ bool changed;
+
+ end = roundup(end, PMD_SIZE) - 1;
+
+ /* check for reserved regions after end */
+ addrp = __pa(end);
+ size = (PTRS_PER_PMD * PMD_SIZE + vaddr) - end;
+ changed = memblock_check_reserved_size(&addrp, &size, PMD_SIZE);
+ if (changed || !size) {
+ /* reserved regions found, avoid removing mappings after end */
+ pud_t *pud = pud_offset(pgd_offset_k(end), end);
+ last_pmd = pmd_offset(pud, end);
+ }

for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
if (pmd_none(*pmd))
diff --git a/arch/x86/mm/memblock.c b/arch/x86/mm/memblock.c
index aa11693..fac21d4 100644
--- a/arch/x86/mm/memblock.c
+++ b/arch/x86/mm/memblock.c
@@ -8,7 +8,7 @@
#include <linux/range.h>

/* Check for already reserved areas */
-static bool __init check_with_memblock_reserved_size(u64 *addrp, u64 *sizep, u64 align)
+bool __init memblock_check_reserved_size(u64 *addrp, u64 *sizep, u64 align)
{
struct memblock_region *r;
u64 addr = *addrp, last;
@@ -59,7 +59,7 @@ u64 __init memblock_x86_find_in_range_size(u64 start, u64 *sizep, u64 align)
if (addr >= ei_last)
continue;
*sizep = ei_last - addr;
- while (check_with_memblock_reserved_size(&addr, sizep, align))
+ while (memblock_check_reserved_size(&addr, sizep, align))
;

if (*sizep)

Next message: Bartlomiej Zolnierkiewicz: "Re: [PATCH 16/20] ata_piix: add EFAR SLC90E66 support"
Previous message: martin f krafft: "Re: Opteron ECC/ChipKill error"
In reply to: Yinghai Lu: "Re: [PATCH] x86/mm/init: respect memblock reserved regions whendestroying mappings"
Next in thread: Yinghai Lu: "Re: [PATCH] x86/mm/init: respect memblock reserved regions when destroyingmappings"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]