| Message ID | 20241010061024.1846220-3-wangkefeng.wang@huawei.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | mm: cleanup thp and shmem allowable order check | expand |
On 10.10.24 08:10, Kefeng Wang wrote: > Add thp_vma_disabled() helper to shared by shmem_allowable_huge_orders() > and __thp_vma_allowable_orders(). > > Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> > --- > include/linux/huge_mm.h | 19 +++++++++++++++++++ > mm/huge_memory.c | 13 +------------ > mm/shmem.c | 7 +------ > 3 files changed, 21 insertions(+), 18 deletions(-) > > diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h > index 795df660efa5..d77891332b35 100644 > --- a/include/linux/huge_mm.h > +++ b/include/linux/huge_mm.h > @@ -309,6 +309,25 @@ struct thpsize { > (transparent_hugepage_flags & \ > (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG)) > > +static inline bool thp_vma_disabled(struct vm_area_struct *vma, > + unsigned long vm_flags) > +{ > + /* > + * Explicitly disabled through madvise or prctl, or some > + * architectures may disable THP for some mappings, for > + * example, s390 kvm. > + */ > + if (vma && ((vm_flags & VM_NOHUGEPAGE) || > + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))) > + return true; > + > + /* If the hardware/firmware marked hugepage support disabled. */ > + if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) > + return true; > + > + return false; > +} > + Should we call this "vma_thp_disabled()" ? Also, I wonder if it would be more natural to check for the opposite ... "vma_thp_enabled()", like we test for "allowed" and "suitable".
On 10.10.24 08:10, Kefeng Wang wrote: > Add thp_vma_disabled() helper to shared by shmem_allowable_huge_orders() > and __thp_vma_allowable_orders(). > > Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> > --- > include/linux/huge_mm.h | 19 +++++++++++++++++++ > mm/huge_memory.c | 13 +------------ > mm/shmem.c | 7 +------ > 3 files changed, 21 insertions(+), 18 deletions(-) > > diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h > index 795df660efa5..d77891332b35 100644 > --- a/include/linux/huge_mm.h > +++ b/include/linux/huge_mm.h > @@ -309,6 +309,25 @@ struct thpsize { > (transparent_hugepage_flags & \ > (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG)) > > +static inline bool thp_vma_disabled(struct vm_area_struct *vma, > + unsigned long vm_flags) > +{ > I might need a patch like this for an independent fix, and the more I look at this the more I hate the separate vm_flags and the optional vma argument. Let me try to improve things. Long term we can get rid of the vm_flags, it just needs some madvise() massaging.
On 10.10.24 16:41, David Hildenbrand wrote: > On 10.10.24 08:10, Kefeng Wang wrote: >> Add thp_vma_disabled() helper to shared by shmem_allowable_huge_orders() >> and __thp_vma_allowable_orders(). >> >> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> >> --- >> include/linux/huge_mm.h | 19 +++++++++++++++++++ >> mm/huge_memory.c | 13 +------------ >> mm/shmem.c | 7 +------ >> 3 files changed, 21 insertions(+), 18 deletions(-) >> >> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h >> index 795df660efa5..d77891332b35 100644 >> --- a/include/linux/huge_mm.h >> +++ b/include/linux/huge_mm.h >> @@ -309,6 +309,25 @@ struct thpsize { >> (transparent_hugepage_flags & \ >> (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG)) >> >> +static inline bool thp_vma_disabled(struct vm_area_struct *vma, >> + unsigned long vm_flags) >> +{ >> > > I might need a patch like this for an independent fix, and the more I > look at this the more I hate the separate vm_flags and the optional vma > argument. > > Let me try to improve things. > > Long term we can get rid of the vm_flags, it just needs some madvise() > massaging. For the time being I suggest this: From 318c25742380cdf15c8c807e5e8a52cabc217ef4 Mon Sep 17 00:00:00 2001 From: Kefeng Wang <wangkefeng.wang@huawei.com> Date: Thu, 10 Oct 2024 14:10:23 +0800 Subject: [PATCH] mm: huge_memory: add vma_thp_disabled() and thp_disabled_by_hw() Add vma_thp_disabled() and thp_disabled_by_hw() helpers to be shared by shmem_allowable_huge_orders() and __thp_vma_allowable_orders(). Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> [ rename to vma_thp_disabled(), split out thp_disabled_by_hw() ] Signed-off-by: David Hildenbrand <david@redhat.com> --- include/linux/huge_mm.h | 18 ++++++++++++++++++ mm/huge_memory.c | 13 +------------ mm/shmem.c | 7 +------ 3 files changed, 20 insertions(+), 18 deletions(-) diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index 67d0ab3c3bba..57b62fd1ccb4 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -322,6 +322,24 @@ struct thpsize { (transparent_hugepage_flags & \ (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG)) +static inline bool vma_thp_disabled(struct vm_area_struct *vma, + unsigned long vm_flags) +{ + /* + * Explicitly disabled through madvise or prctl, or some + * architectures may disable THP for some mappings, for + * example, s390x kvm. + */ + return (vm_flags & VM_NOHUGEPAGE) || + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags); +} + +static inline bool thp_disabled_by_hw(void) +{ + /* If the hardware/firmware marked hugepage support disabled. */ + return transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED); +} + unsigned long thp_get_unmapped_area(struct file *filp, unsigned long addr, unsigned long len, unsigned long pgoff, unsigned long flags); unsigned long thp_get_unmapped_area_vmflags(struct file *filp, unsigned long addr, diff --git a/mm/huge_memory.c b/mm/huge_memory.c index 3ca89e0279a7..ffbf0add2a82 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -109,18 +109,7 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma, if (!vma->vm_mm) /* vdso */ return 0; - /* - * Explicitly disabled through madvise or prctl, or some - * architectures may disable THP for some mappings, for - * example, s390 kvm. - * */ - if ((vm_flags & VM_NOHUGEPAGE) || - test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) - return 0; - /* - * If the hardware/firmware marked hugepage support disabled. - */ - if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + if (thp_disabled_by_hw() || vma_thp_disabled(vma, vm_flags)) return 0; /* khugepaged doesn't collapse DAX vma, but page fault is fine. */ diff --git a/mm/shmem.c b/mm/shmem.c index 4f11b5506363..c5adb987b23c 100644 --- a/mm/shmem.c +++ b/mm/shmem.c @@ -1664,12 +1664,7 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode, loff_t i_size; int order; - if (vma && ((vm_flags & VM_NOHUGEPAGE) || - test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))) - return 0; - - /* If the hardware/firmware marked hugepage support disabled. */ - if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + if (thp_disabled_by_hw() || (vma && vma_thp_disabled(vma, vm_flags))) return 0; global_huge = shmem_huge_global_enabled(inode, index, write_end,
On 2024/10/10 22:53, David Hildenbrand wrote: > On 10.10.24 16:41, David Hildenbrand wrote: >> On 10.10.24 08:10, Kefeng Wang wrote: >>> Add thp_vma_disabled() helper to shared by shmem_allowable_huge_orders() >>> and __thp_vma_allowable_orders(). >>> >>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> >>> --- >>> include/linux/huge_mm.h | 19 +++++++++++++++++++ >>> mm/huge_memory.c | 13 +------------ >>> mm/shmem.c | 7 +------ >>> 3 files changed, 21 insertions(+), 18 deletions(-) >>> >>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h >>> index 795df660efa5..d77891332b35 100644 >>> --- a/include/linux/huge_mm.h >>> +++ b/include/linux/huge_mm.h >>> @@ -309,6 +309,25 @@ struct thpsize { >>> (transparent_hugepage_flags & \ >>> (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG)) >>> +static inline bool thp_vma_disabled(struct vm_area_struct *vma, >>> + unsigned long vm_flags) >>> +{ >>> >> >> I might need a patch like this for an independent fix, and the more I >> look at this the more I hate the separate vm_flags and the optional vma >> argument. Yes, it is a little strange when made this changes, a separate vm_flags and another vma argument, most vm_flags is just vma->vm_flags(madvise is a special case). >> >> Let me try to improve things. >> >> Long term we can get rid of the vm_flags, it just needs some madvise() >> massaging. > Thanks for your improvement, it is more accurate. > For the time being I suggest this: > > From 318c25742380cdf15c8c807e5e8a52cabc217ef4 Mon Sep 17 00:00:00 2001 > From: Kefeng Wang <wangkefeng.wang@huawei.com> > Date: Thu, 10 Oct 2024 14:10:23 +0800 > Subject: [PATCH] mm: huge_memory: add vma_thp_disabled() and > thp_disabled_by_hw() > > Add vma_thp_disabled() and thp_disabled_by_hw() helpers to be shared by > shmem_allowable_huge_orders() and __thp_vma_allowable_orders(). > > Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> > [ rename to vma_thp_disabled(), split out thp_disabled_by_hw() ] > Signed-off-by: David Hildenbrand <david@redhat.com> > --- > include/linux/huge_mm.h | 18 ++++++++++++++++++ > mm/huge_memory.c | 13 +------------ > mm/shmem.c | 7 +------ > 3 files changed, 20 insertions(+), 18 deletions(-) > > diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h > index 67d0ab3c3bba..57b62fd1ccb4 100644 > --- a/include/linux/huge_mm.h > +++ b/include/linux/huge_mm.h > @@ -322,6 +322,24 @@ struct thpsize { > (transparent_hugepage_flags & \ > (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG)) > > +static inline bool vma_thp_disabled(struct vm_area_struct *vma, > + unsigned long vm_flags) > +{ > + /* > + * Explicitly disabled through madvise or prctl, or some > + * architectures may disable THP for some mappings, for > + * example, s390x kvm. > + */ > + return (vm_flags & VM_NOHUGEPAGE) || > + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags); > +} > + > +static inline bool thp_disabled_by_hw(void) > +{ > + /* If the hardware/firmware marked hugepage support disabled. */ > + return transparent_hugepage_flags & (1 << > TRANSPARENT_HUGEPAGE_UNSUPPORTED); > +} > + > unsigned long thp_get_unmapped_area(struct file *filp, unsigned long > addr, > unsigned long len, unsigned long pgoff, unsigned long flags); > unsigned long thp_get_unmapped_area_vmflags(struct file *filp, > unsigned long addr, > diff --git a/mm/huge_memory.c b/mm/huge_memory.c > index 3ca89e0279a7..ffbf0add2a82 100644 > --- a/mm/huge_memory.c > +++ b/mm/huge_memory.c > @@ -109,18 +109,7 @@ unsigned long __thp_vma_allowable_orders(struct > vm_area_struct *vma, > if (!vma->vm_mm) /* vdso */ > return 0; > > - /* > - * Explicitly disabled through madvise or prctl, or some > - * architectures may disable THP for some mappings, for > - * example, s390 kvm. > - * */ > - if ((vm_flags & VM_NOHUGEPAGE) || > - test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) > - return 0; > - /* > - * If the hardware/firmware marked hugepage support disabled. > - */ > - if (transparent_hugepage_flags & (1 << > TRANSPARENT_HUGEPAGE_UNSUPPORTED)) > + if (thp_disabled_by_hw() || vma_thp_disabled(vma, vm_flags)) > return 0; > > /* khugepaged doesn't collapse DAX vma, but page fault is fine. */ > diff --git a/mm/shmem.c b/mm/shmem.c > index 4f11b5506363..c5adb987b23c 100644 > --- a/mm/shmem.c > +++ b/mm/shmem.c > @@ -1664,12 +1664,7 @@ unsigned long shmem_allowable_huge_orders(struct > inode *inode, > loff_t i_size; > int order; > > - if (vma && ((vm_flags & VM_NOHUGEPAGE) || > - test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))) > - return 0; > - > - /* If the hardware/firmware marked hugepage support disabled. */ > - if (transparent_hugepage_flags & (1 << > TRANSPARENT_HUGEPAGE_UNSUPPORTED)) > + if (thp_disabled_by_hw() || (vma && vma_thp_disabled(vma, vm_flags))) > return 0; > > global_huge = shmem_huge_global_enabled(inode, index, write_end,
On 11.10.24 02:40, Kefeng Wang wrote: > > > On 2024/10/10 22:53, David Hildenbrand wrote: >> On 10.10.24 16:41, David Hildenbrand wrote: >>> On 10.10.24 08:10, Kefeng Wang wrote: >>>> Add thp_vma_disabled() helper to shared by shmem_allowable_huge_orders() >>>> and __thp_vma_allowable_orders(). >>>> >>>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> >>>> --- >>>> include/linux/huge_mm.h | 19 +++++++++++++++++++ >>>> mm/huge_memory.c | 13 +------------ >>>> mm/shmem.c | 7 +------ >>>> 3 files changed, 21 insertions(+), 18 deletions(-) >>>> >>>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h >>>> index 795df660efa5..d77891332b35 100644 >>>> --- a/include/linux/huge_mm.h >>>> +++ b/include/linux/huge_mm.h >>>> @@ -309,6 +309,25 @@ struct thpsize { >>>> (transparent_hugepage_flags & \ >>>> (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG)) >>>> +static inline bool thp_vma_disabled(struct vm_area_struct *vma, >>>> + unsigned long vm_flags) >>>> +{ >>>> >>> >>> I might need a patch like this for an independent fix, and the more I >>> look at this the more I hate the separate vm_flags and the optional vma >>> argument. > > Yes, it is a little strange when made this changes, a separate vm_flags > and another vma argument, most vm_flags is just vma->vm_flags(madvise is > a special case). > Yes, we should be able to handle the madvise stuff in a better way (simply check after the vma->vm_flags where modified). >>> >>> Let me try to improve things. >>> >>> Long term we can get rid of the vm_flags, it just needs some madvise() >>> massaging. >> > > Thanks for your improvement, it is more accurate. I will include that in my next small fix-series and CC you.
diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index 795df660efa5..d77891332b35 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -309,6 +309,25 @@ struct thpsize { (transparent_hugepage_flags & \ (1<<TRANSPARENT_HUGEPAGE_USE_ZERO_PAGE_FLAG)) +static inline bool thp_vma_disabled(struct vm_area_struct *vma, + unsigned long vm_flags) +{ + /* + * Explicitly disabled through madvise or prctl, or some + * architectures may disable THP for some mappings, for + * example, s390 kvm. + */ + if (vma && ((vm_flags & VM_NOHUGEPAGE) || + test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))) + return true; + + /* If the hardware/firmware marked hugepage support disabled. */ + if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + return true; + + return false; +} + unsigned long thp_get_unmapped_area(struct file *filp, unsigned long addr, unsigned long len, unsigned long pgoff, unsigned long flags); unsigned long thp_get_unmapped_area_vmflags(struct file *filp, unsigned long addr, diff --git a/mm/huge_memory.c b/mm/huge_memory.c index d23e4aab7511..30912a93f7dc 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -123,18 +123,7 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma, if (!vma->vm_mm) /* vdso */ return 0; - /* - * Explicitly disabled through madvise or prctl, or some - * architectures may disable THP for some mappings, for - * example, s390 kvm. - * */ - if ((vm_flags & VM_NOHUGEPAGE) || - test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)) - return 0; - /* - * If the hardware/firmware marked hugepage support disabled. - */ - if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + if (thp_vma_disabled(vma, vm_flags)) return 0; /* khugepaged doesn't collapse DAX vma, but page fault is fine. */ diff --git a/mm/shmem.c b/mm/shmem.c index 0a2f78c2b919..34a31e7e527c 100644 --- a/mm/shmem.c +++ b/mm/shmem.c @@ -1683,12 +1683,7 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode, loff_t i_size; int order; - if (vma && ((vm_flags & VM_NOHUGEPAGE) || - test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))) - return 0; - - /* If the hardware/firmware marked hugepage support disabled. */ - if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED)) + if (thp_vma_disabled(vma, vm_flags)) return 0; global_huge = shmem_huge_global_enabled(inode, index, write_end,
Add thp_vma_disabled() helper to shared by shmem_allowable_huge_orders() and __thp_vma_allowable_orders(). Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com> --- include/linux/huge_mm.h | 19 +++++++++++++++++++ mm/huge_memory.c | 13 +------------ mm/shmem.c | 7 +------ 3 files changed, 21 insertions(+), 18 deletions(-)