| Message ID | 1532532119-59955-1-git-send-email-zhongjiang@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [RFC] mm: Do not look up the next level table when large page exist. | expand |
On Wed, Jul 25, 2018 at 11:21:59PM +0800, zhong jiang wrote: > show_pte can show the kernel page table and user page table. user > page table is totally four level table. Well, you can have huge pages in user space. > but kernel page table can > be a large page. Therefore, it need to check the page table. > > Signed-off-by: zhong jiang <zhongjiang@huawei.com> > --- > arch/arm64/mm/fault.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c > index 1b40676..a397a54 100644 > --- a/arch/arm64/mm/fault.c > +++ b/arch/arm64/mm/fault.c > @@ -169,13 +169,13 @@ void show_pte(unsigned long addr) > pudp = pud_offset(pgdp, addr); > pud = READ_ONCE(*pudp); > pr_cont(", pud=%016llx", pud_val(pud)); > - if (pud_none(pud) || pud_bad(pud)) > + if (pud_none(pud) || pud_bad(pud) || pud_sect(pud)) Have you actually hit a problem with this code? pud_sect() checks that the bottom two bits of the pud are 0b01. pud_bad() is true when bit 1 is 0, which is the case with a section mapping.
On 2018/7/26 0:53, Catalin Marinas wrote: > On Wed, Jul 25, 2018 at 11:21:59PM +0800, zhong jiang wrote: >> show_pte can show the kernel page table and user page table. user >> page table is totally four level table. > Well, you can have huge pages in user space. No, I have not. >> but kernel page table can >> be a large page. Therefore, it need to check the page table. >> >> Signed-off-by: zhong jiang <zhongjiang@huawei.com> >> --- >> arch/arm64/mm/fault.c | 4 ++-- >> 1 file changed, 2 insertions(+), 2 deletions(-) >> >> diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c >> index 1b40676..a397a54 100644 >> --- a/arch/arm64/mm/fault.c >> +++ b/arch/arm64/mm/fault.c >> @@ -169,13 +169,13 @@ void show_pte(unsigned long addr) >> pudp = pud_offset(pgdp, addr); >> pud = READ_ONCE(*pudp); >> pr_cont(", pud=%016llx", pud_val(pud)); >> - if (pud_none(pud) || pud_bad(pud)) >> + if (pud_none(pud) || pud_bad(pud) || pud_sect(pud)) > Have you actually hit a problem with this code? Not yet. Maybe I use the function mistake. It is only used for user and vmalloc pagefault. It should be not handled with large level table. thank you for time. Thanks, zhong jiang > pud_sect() checks that the bottom two bits of the pud are 0b01. > pud_bad() is true when bit 1 is 0, which is the case with a section > mapping. >
diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c index 1b40676..a397a54 100644 --- a/arch/arm64/mm/fault.c +++ b/arch/arm64/mm/fault.c @@ -169,13 +169,13 @@ void show_pte(unsigned long addr) pudp = pud_offset(pgdp, addr); pud = READ_ONCE(*pudp); pr_cont(", pud=%016llx", pud_val(pud)); - if (pud_none(pud) || pud_bad(pud)) + if (pud_none(pud) || pud_bad(pud) || pud_sect(pud)) break; pmdp = pmd_offset(pudp, addr); pmd = READ_ONCE(*pmdp); pr_cont(", pmd=%016llx", pmd_val(pmd)); - if (pmd_none(pmd) || pmd_bad(pmd)) + if (pmd_none(pmd) || pmd_bad(pmd) || pmd_sect(pmd)) break; ptep = pte_offset_map(pmdp, addr);
show_pte can show the kernel page table and user page table. user page table is totally four level table. but kernel page table can be a large page. Therefore, it need to check the page table. Signed-off-by: zhong jiang <zhongjiang@huawei.com> --- arch/arm64/mm/fault.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)