2021-11-02 10:26:26

by Thomas Bogendoerfer

[permalink] [raw]
Subject: Re: [PATCH] mips: fix HUGETLB function without THP enabled

On Tue, Nov 02, 2021 at 04:24:37PM +0800, Zhaolong Zhang wrote:
> ltp test futex_wake04 without THP enabled leads to below bt:
> [<ffffffff80a03728>] BUG+0x0/0x8
> [<ffffffff80a0624c>] internal_get_user_pages_fast+0x81c/0x820
> [<ffffffff8093ac18>] get_futex_key+0xa0/0x480
> [<ffffffff8093b074>] futex_wait_setup+0x7c/0x1a8
> [<ffffffff8093b2c0>] futex_wait+0x120/0x228
> [<ffffffff8093dbe8>] do_futex+0x140/0xbd8
> [<ffffffff8093e78c>] sys_futex+0x10c/0x1c0
> [<ffffffff808703d0>] syscall_common+0x34/0x58
>
> Move pmd_write() and pmd_page() from TRANSPARENT_HUGEPAGE scope to
> MIPS_HUGE_TLB_SUPPORT scope, because both THP and HUGETLB will need
> them.
>
> Signed-off-by: Zhaolong Zhang <[email protected]>
> ---
> arch/mips/include/asm/pgtable.h | 40 +++++++++++++++++----------------
> 1 file changed, 21 insertions(+), 19 deletions(-)

applied to mips-next.

>
> diff --git a/arch/mips/include/asm/pgtable.h b/arch/mips/include/asm/pgtable.h
> index 804889b70965..1fcf4be5cd20 100644
> --- a/arch/mips/include/asm/pgtable.h
> +++ b/arch/mips/include/asm/pgtable.h
> @@ -86,10 +86,12 @@ extern void paging_init(void);
> */
> #define pmd_phys(pmd) virt_to_phys((void *)pmd_val(pmd))
>
> +#ifndef CONFIG_MIPS_HUGE_TLB_SUPPORT
> #define __pmd_page(pmd) (pfn_to_page(pmd_phys(pmd) >> PAGE_SHIFT))
> #ifndef CONFIG_TRANSPARENT_HUGEPAGE
> #define pmd_page(pmd) __pmd_page(pmd)
> #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
> +#endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */

While applying I've simplified that hunk to

-#define __pmd_page(pmd) (pfn_to_page(pmd_phys(pmd) >> PAGE_SHIFT))
-#ifndef CONFIG_TRANSPARENT_HUGEPAGE
-#define pmd_page(pmd) __pmd_page(pmd)
-#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
+#ifndef CONFIG_MIPS_HUGE_TLB_SUPPORT
+#define pmd_page(pmd) (pfn_to_page(pmd_phys(pmd) >> PAGE_SHIFT))
+#endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */

Thomas.

--
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea. [ RFC1925, 2.3 ]


2021-11-02 12:48:41

by Zhaolong Zhang

[permalink] [raw]
Subject: Re: [PATCH] mips: fix HUGETLB function without THP enabled


At 2021-11-02 18:20:51, "Thomas Bogendoerfer" <[email protected]> wrote:
>On Tue, Nov 02, 2021 at 04:24:37PM +0800, Zhaolong Zhang wrote:
>> ltp test futex_wake04 without THP enabled leads to below bt:
>> [<ffffffff80a03728>] BUG+0x0/0x8
>> [<ffffffff80a0624c>] internal_get_user_pages_fast+0x81c/0x820
>> [<ffffffff8093ac18>] get_futex_key+0xa0/0x480
>> [<ffffffff8093b074>] futex_wait_setup+0x7c/0x1a8
>> [<ffffffff8093b2c0>] futex_wait+0x120/0x228
>> [<ffffffff8093dbe8>] do_futex+0x140/0xbd8
>> [<ffffffff8093e78c>] sys_futex+0x10c/0x1c0
>> [<ffffffff808703d0>] syscall_common+0x34/0x58
>>
>> Move pmd_write() and pmd_page() from TRANSPARENT_HUGEPAGE scope to
>> MIPS_HUGE_TLB_SUPPORT scope, because both THP and HUGETLB will need
>> them.
>>
>> Signed-off-by: Zhaolong Zhang <[email protected]>
>> ---
>> arch/mips/include/asm/pgtable.h | 40 +++++++++++++++++----------------
>> 1 file changed, 21 insertions(+), 19 deletions(-)
>
>applied to mips-next.
>
>>
>> diff --git a/arch/mips/include/asm/pgtable.h b/arch/mips/include/asm/pgtable.h
>> index 804889b70965..1fcf4be5cd20 100644
>> --- a/arch/mips/include/asm/pgtable.h
>> +++ b/arch/mips/include/asm/pgtable.h
>> @@ -86,10 +86,12 @@ extern void paging_init(void);
>> */
>> #define pmd_phys(pmd) virt_to_phys((void *)pmd_val(pmd))
>>
>> +#ifndef CONFIG_MIPS_HUGE_TLB_SUPPORT
>> #define __pmd_page(pmd) (pfn_to_page(pmd_phys(pmd) >> PAGE_SHIFT))
>> #ifndef CONFIG_TRANSPARENT_HUGEPAGE
>> #define pmd_page(pmd) __pmd_page(pmd)
>> #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
>> +#endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */
>
>While applying I've simplified that hunk to
>
>-#define __pmd_page(pmd) (pfn_to_page(pmd_phys(pmd) >> PAGE_SHIFT))
>-#ifndef CONFIG_TRANSPARENT_HUGEPAGE
>-#define pmd_page(pmd) __pmd_page(pmd)
>-#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
>+#ifndef CONFIG_MIPS_HUGE_TLB_SUPPORT
>+#define pmd_page(pmd) (pfn_to_page(pmd_phys(pmd) >> PAGE_SHIFT))
>+#endif /* CONFIG_MIPS_HUGE_TLB_SUPPORT */
>

It is much better.
Thanks for your kind reply.

Regards,
Zhaolong

>Thomas.
>
>--
>Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
>good idea. [ RFC1925, 2.3 ]