[permalink] [raw]

Subject: Re: [PATCH v10 05/22] mips: mm: Add p?d_leaf() definitions

Hi Steven,

On Wed, Jul 31, 2019 at 04:45:46PM +0100, Steven Price wrote:
> walk_page_range() is going to be allowed to walk page tables other than
> those of user space. For this it needs to know when it has reached a
> 'leaf' entry in the page tables. This information is provided by the
> p?d_leaf() functions/macros.
>
> If _PAGE_HUGE is defined we can simply look for it. When not defined we
> can be confident that there are no leaf pages in existence and fall back
> on the generic implementation (added in a later patch) which returns 0.
>
> CC: Ralf Baechle <[email protected]>
> CC: Paul Burton <[email protected]>
> CC: James Hogan <[email protected]>
> CC: [email protected]
> Signed-off-by: Steven Price <[email protected]>

Acked-by: Paul Burton <[email protected]>

Thanks,
Paul

> ---
> arch/mips/include/asm/pgtable.h | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/arch/mips/include/asm/pgtable.h b/arch/mips/include/asm/pgtable.h
> index 7d27194e3b45..238ca243ad31 100644
> --- a/arch/mips/include/asm/pgtable.h
> +++ b/arch/mips/include/asm/pgtable.h
> @@ -627,6 +627,11 @@ static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
>
> #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
>
> +#ifdef _PAGE_HUGE
> +#define pmd_leaf(pmd) ((pmd_val(pmd) & _PAGE_HUGE) != 0)
> +#define pud_leaf(pud) ((pud_val(pud) & _PAGE_HUGE) != 0)
> +#endif
> +
> #define gup_fast_permitted(start, end) (!cpu_has_dc_aliases)
>
> #include <asm-generic/pgtable.h>
> --
> 2.20.1
>

2019-08-07 00:00:34

by Andrew Morton

[permalink] [raw]

Subject: Re: [PATCH v10 20/22] x86: mm: Convert dump_pagetables to use walk_page_range

On Wed, 31 Jul 2019 16:46:01 +0100 Steven Price <[email protected]> wrote:

> Make use of the new functionality in walk_page_range to remove the
> arch page walking code and use the generic code to walk the page tables.
>
> The effective permissions are passed down the chain using new fields
> in struct pg_state.
>
> The KASAN optimisation is implemented by including test_p?d callbacks
> which can decide to skip an entire tree of entries
>
> ...
>
> +static const struct ptdump_range ptdump_ranges[] = {
> +#ifdef CONFIG_X86_64
>
> -#define pgd_large(a) (pgtable_l5_enabled() ? pgd_large(a) : p4d_large(__p4d(pgd_val(a))))
> -#define pgd_none(a) (pgtable_l5_enabled() ? pgd_none(a) : p4d_none(__p4d(pgd_val(a))))
> +#define normalize_addr_shift (64 - (__VIRTUAL_MASK_SHIFT + 1))
> +#define normalize_addr(u) ((signed long)(u << normalize_addr_shift) \
> + >> normalize_addr_shift)
>
> -static inline bool is_hypervisor_range(int idx)
> -{
> -#ifdef CONFIG_X86_64
> - /*
> - * A hole in the beginning of kernel address space reserved
> - * for a hypervisor.
> - */
> - return (idx >= pgd_index(GUARD_HOLE_BASE_ADDR)) &&
> - (idx < pgd_index(GUARD_HOLE_END_ADDR));
> + {0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
> + {normalize_addr(PTRS_PER_PGD * PGD_LEVEL_MULT / 2), ~0UL},

This blows up because PGD_LEVEL_MULT is sometimes not a constant.

x86_64 allmodconfig:

In file included from ./arch/x86/include/asm/pgtable_types.h:249:0,
from ./arch/x86/include/asm/paravirt_types.h:45,
from ./arch/x86/include/asm/ptrace.h:94,
from ./arch/x86/include/asm/math_emu.h:5,
from ./arch/x86/include/asm/processor.h:12,
from ./arch/x86/include/asm/cpufeature.h:5,
from ./arch/x86/include/asm/thread_info.h:53,
from ./include/linux/thread_info.h:38,
from ./arch/x86/include/asm/preempt.h:7,
from ./include/linux/preempt.h:78,
from ./include/linux/spinlock.h:51,
from ./include/linux/wait.h:9,
from ./include/linux/wait_bit.h:8,
from ./include/linux/fs.h:6,
from ./include/linux/debugfs.h:15,
from arch/x86/mm/dump_pagetables.c:11:
./arch/x86/include/asm/pgtable_64_types.h:56:22: error: initializer element is not constant
#define PTRS_PER_PGD 512
^
arch/x86/mm/dump_pagetables.c:363:6: note: in expansion of macro ‘PTRS_PER_PGD’
{0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
^~~~~~~~~~~~
./arch/x86/include/asm/pgtable_64_types.h:56:22: note: (near initialization for ‘ptdump_ranges[0].end’)
#define PTRS_PER_PGD 512
^
arch/x86/mm/dump_pagetables.c:363:6: note: in expansion of macro ‘PTRS_PER_PGD’
{0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
^~~~~~~~~~~~
arch/x86/mm/dump_pagetables.c:360:27: error: initializer element is not constant
#define normalize_addr(u) ((signed long)(u << normalize_addr_shift) \
^
arch/x86/mm/dump_pagetables.c:364:3: note: in expansion of macro ‘normalize_addr’
{normalize_addr(PTRS_PER_PGD * PGD_LEVEL_MULT / 2), ~0UL},
^~~~~~~~~~~~~~
arch/x86/mm/dump_pagetables.c:360:27: note: (near initialization for ‘ptdump_ranges[1].start’)
#define normalize_addr(u) ((signed long)(u << normalize_addr_shift) \
^
arch/x86/mm/dump_pagetables.c:364:3: note: in expansion of macro ‘normalize_addr’
{normalize_addr(PTRS_PER_PGD * PGD_LEVEL_MULT / 2), ~0UL},

I don't know what to do about this so I'll drop the series.

2019-08-07 13:00:03

by Steven Price

[permalink] [raw]

Subject: Re: [PATCH v10 20/22] x86: mm: Convert dump_pagetables to use walk_page_range

On 07/08/2019 00:58, Andrew Morton wrote:
> On Wed, 31 Jul 2019 16:46:01 +0100 Steven Price <[email protected]> wrote:
>
>> Make use of the new functionality in walk_page_range to remove the
>> arch page walking code and use the generic code to walk the page tables.
>>
>> The effective permissions are passed down the chain using new fields
>> in struct pg_state.
>>
>> The KASAN optimisation is implemented by including test_p?d callbacks
>> which can decide to skip an entire tree of entries
>>
>> ...
>>
>> +static const struct ptdump_range ptdump_ranges[] = {
>> +#ifdef CONFIG_X86_64
>>
>> -#define pgd_large(a) (pgtable_l5_enabled() ? pgd_large(a) : p4d_large(__p4d(pgd_val(a))))
>> -#define pgd_none(a) (pgtable_l5_enabled() ? pgd_none(a) : p4d_none(__p4d(pgd_val(a))))
>> +#define normalize_addr_shift (64 - (__VIRTUAL_MASK_SHIFT + 1))
>> +#define normalize_addr(u) ((signed long)(u << normalize_addr_shift) \
>> + >> normalize_addr_shift)
>>
>> -static inline bool is_hypervisor_range(int idx)
>> -{
>> -#ifdef CONFIG_X86_64
>> - /*
>> - * A hole in the beginning of kernel address space reserved
>> - * for a hypervisor.
>> - */
>> - return (idx >= pgd_index(GUARD_HOLE_BASE_ADDR)) &&
>> - (idx < pgd_index(GUARD_HOLE_END_ADDR));
>> + {0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
>> + {normalize_addr(PTRS_PER_PGD * PGD_LEVEL_MULT / 2), ~0UL},
>
> This blows up because PGD_LEVEL_MULT is sometimes not a constant.
>
> x86_64 allmodconfig:
>
> In file included from ./arch/x86/include/asm/pgtable_types.h:249:0,
> from ./arch/x86/include/asm/paravirt_types.h:45,
> from ./arch/x86/include/asm/ptrace.h:94,
> from ./arch/x86/include/asm/math_emu.h:5,
> from ./arch/x86/include/asm/processor.h:12,
> from ./arch/x86/include/asm/cpufeature.h:5,
> from ./arch/x86/include/asm/thread_info.h:53,
> from ./include/linux/thread_info.h:38,
> from ./arch/x86/include/asm/preempt.h:7,
> from ./include/linux/preempt.h:78,
> from ./include/linux/spinlock.h:51,
> from ./include/linux/wait.h:9,
> from ./include/linux/wait_bit.h:8,
> from ./include/linux/fs.h:6,
> from ./include/linux/debugfs.h:15,
> from arch/x86/mm/dump_pagetables.c:11:
> ./arch/x86/include/asm/pgtable_64_types.h:56:22: error: initializer element is not constant
> #define PTRS_PER_PGD 512
> ^

This is very unhelpful of GCC - it's actually PTRS_PER_P4D which isn't
constant!

> arch/x86/mm/dump_pagetables.c:363:6: note: in expansion of macro ‘PTRS_PER_PGD’
> {0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
> ^~~~~~~~~~~~
> ./arch/x86/include/asm/pgtable_64_types.h:56:22: note: (near initialization for ‘ptdump_ranges[0].end’)
> #define PTRS_PER_PGD 512
> ^
> arch/x86/mm/dump_pagetables.c:363:6: note: in expansion of macro ‘PTRS_PER_PGD’
> {0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
> ^~~~~~~~~~~~
> arch/x86/mm/dump_pagetables.c:360:27: error: initializer element is not constant
> #define normalize_addr(u) ((signed long)(u << normalize_addr_shift) \
> ^
> arch/x86/mm/dump_pagetables.c:364:3: note: in expansion of macro ‘normalize_addr’
> {normalize_addr(PTRS_PER_PGD * PGD_LEVEL_MULT / 2), ~0UL},
> ^~~~~~~~~~~~~~
> arch/x86/mm/dump_pagetables.c:360:27: note: (near initialization for ‘ptdump_ranges[1].start’)
> #define normalize_addr(u) ((signed long)(u << normalize_addr_shift) \
> ^
> arch/x86/mm/dump_pagetables.c:364:3: note: in expansion of macro ‘normalize_addr’
> {normalize_addr(PTRS_PER_PGD * PGD_LEVEL_MULT / 2), ~0UL},
>
> I don't know what to do about this so I'll drop the series.

My best solution to this is to simply make ptdump_ranges dynamic (see
below). But there are other problems with this series (thanks for
spotting them), so I'll send out another version later.

Thanks,

Steve

----8<-----
diff --git a/arch/x86/mm/dump_pagetables.c b/arch/x86/mm/dump_pagetables.c
index 998c7f46763c..8fc129ff985e 100644
--- a/arch/x86/mm/dump_pagetables.c
+++ b/arch/x86/mm/dump_pagetables.c
@@ -353,7 +353,10 @@ static void note_page(struct ptdump_state *pt_st,
unsigned long addr, int level,
}
}

-static const struct ptdump_range ptdump_ranges[] = {
+static void ptdump_walk_pgd_level_core(struct seq_file *m, struct
mm_struct *mm,
+ bool checkwx, bool dmesg)
+{
+ const struct ptdump_range ptdump_ranges[] = {
#ifdef CONFIG_X86_64

#define normalize_addr_shift (64 - (__VIRTUAL_MASK_SHIFT + 1))
@@ -368,9 +371,6 @@ static const struct ptdump_range ptdump_ranges[] = {
{0, 0}
};

-static void ptdump_walk_pgd_level_core(struct seq_file *m, struct
mm_struct *mm,
- bool checkwx, bool dmesg)
-{
struct pg_state st = {
.ptdump = {
.note_page = note_page,

2019-08-07 20:19:18

by Andrew Morton

[permalink] [raw]

Subject: Re: [PATCH v10 20/22] x86: mm: Convert dump_pagetables to use walk_page_range

On Wed, 7 Aug 2019 13:58:21 +0100 Steven Price <[email protected]> wrote:

> > ./arch/x86/include/asm/pgtable_64_types.h:56:22: error: initializer element is not constant
> > #define PTRS_PER_PGD 512
> > ^
>
> This is very unhelpful of GCC - it's actually PTRS_PER_P4D which isn't
> constant!

Well. You had every right to assume that an all-caps macro is a
compile-time constant.

We are innocent victims of Kirill's c65e774fb3f6af2 ("x86/mm: Make
PGDIR_SHIFT and PTRS_PER_P4D variable") which lazily converted these
macros into runtime-only, under some Kconfig settings. It should have
changed those macros into static inlined lower-case functions.