Message ID | 20201012160911.6282-1-sh1r4s3@mail.si-head.nl (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | mm/msync: exit early when the flags is an MS_ASYNC and start < vm_start | expand |
On 10/12/20 6:09 PM, Nikita Ermakov wrote: > Exit from the loop over the VMA in the case when the flags > contain only an MS_ASYNC and start < vm_start. In this case msync() > would return with -ENOMEM anyway so make it return early. > > Signed-off-by: Nikita Ermakov <sh1r4s3@mail.si-head.nl> AFAICS it can still return -EBUSY if there's MS_INVALIDATE and a mlocked vma. This is all subtle and I don't think we should risk breaking something for this optimization. > --- > mm/msync.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/mm/msync.c b/mm/msync.c > index 69c6d2029531..ed20c3621d4c 100644 > --- a/mm/msync.c > +++ b/mm/msync.c > @@ -69,6 +69,8 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags) > goto out_unlock; > /* Here start < vma->vm_end. */ > if (start < vma->vm_start) { > + if (flags == MS_ASYNC) > + goto out_unlock; > start = vma->vm_start; > if (start >= end) > goto out_unlock; > > base-commit: 6824a8a9b4861d7df7ee132a952bdf6f84a99cb8 >
Hi! > On 20 Oct 2020, at 13:19, Vlastimil Babka <vbabka@suse.cz> wrote: > > On 10/12/20 6:09 PM, Nikita Ermakov wrote: >> Exit from the loop over the VMA in the case when the flags >> contain only an MS_ASYNC and start < vm_start. In this case msync() >> would return with -ENOMEM anyway so make it return early. >> Signed-off-by: Nikita Ermakov <sh1r4s3@mail.si-head.nl> > > AFAICS it can still return -EBUSY if there's MS_INVALIDATE and a mlocked vma. This is all subtle and I don't think we should risk breaking something for this optimization. > Yes, it could. But in this patch the optimization works only in the case if the flag is MS_ASYNC. If the flag is (MS_ASYNC | MS_INVALIDATE). >> --- >> mm/msync.c | 2 ++ >> 1 file changed, 2 insertions(+) >> diff --git a/mm/msync.c b/mm/msync.c >> index 69c6d2029531..ed20c3621d4c 100644 >> --- a/mm/msync.c >> +++ b/mm/msync.c >> @@ -69,6 +69,8 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags) >> goto out_unlock; >> /* Here start < vma->vm_end. */ >> if (start < vma->vm_start) { >> + if (flags == MS_ASYNC) >> + goto out_unlock; >> start = vma->vm_start; >> if (start >= end) >> goto out_unlock; >> base-commit: 6824a8a9b4861d7df7ee132a952bdf6f84a99cb8 >
On 10/20/20 4:38 PM, Nikita wrote: > Hi! > >> On 20 Oct 2020, at 13:19, Vlastimil Babka <vbabka@suse.cz> wrote: >> >> On 10/12/20 6:09 PM, Nikita Ermakov wrote: >>> Exit from the loop over the VMA in the case when the flags >>> contain only an MS_ASYNC and start < vm_start. In this case msync() >>> would return with -ENOMEM anyway so make it return early. >>> Signed-off-by: Nikita Ermakov <sh1r4s3@mail.si-head.nl> >> >> AFAICS it can still return -EBUSY if there's MS_INVALIDATE and a mlocked vma. This is all subtle and I don't think we should risk breaking something for this optimization. >> > Yes, it could. But in this patch the optimization works only in the case if the flag is MS_ASYNC. If the flag is (MS_ASYNC | MS_INVALIDATE). Ah, right, it compares "== MS_ASYNC" not "& MS_ASYNC" so it's correct. A comment would be nice, such as: We found an unmapped range and with MS_ASYNC and no MS_INVALIDATE there's nothing to do and the result will always be -ENOMEM, so we can return immediately. >>> --- >>> mm/msync.c | 2 ++ >>> 1 file changed, 2 insertions(+) >>> diff --git a/mm/msync.c b/mm/msync.c >>> index 69c6d2029531..ed20c3621d4c 100644 >>> --- a/mm/msync.c >>> +++ b/mm/msync.c >>> @@ -69,6 +69,8 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags) >>> goto out_unlock; >>> /* Here start < vma->vm_end. */ >>> if (start < vma->vm_start) { >>> + if (flags == MS_ASYNC) >>> + goto out_unlock; >>> start = vma->vm_start; >>> if (start >= end) >>> goto out_unlock; >>> base-commit: 6824a8a9b4861d7df7ee132a952bdf6f84a99cb8 >> >
> On 20 Oct 2020, at 18:03, Vlastimil Babka <vbabka@suse.cz> wrote: > > On 10/20/20 4:38 PM, Nikita wrote: >> Hi! >>> On 20 Oct 2020, at 13:19, Vlastimil Babka <vbabka@suse.cz> wrote: >>> On 10/12/20 6:09 PM, Nikita Ermakov wrote: >>>> Exit from the loop over the VMA in the case when the flags >>>> contain only an MS_ASYNC and start < vm_start. In this case msync() >>>> would return with -ENOMEM anyway so make it return early. >>>> Signed-off-by: Nikita Ermakov <sh1r4s3@mail.si-head.nl> >>> AFAICS it can still return -EBUSY if there's MS_INVALIDATE and a mlocked vma. This is all subtle and I don't think we should risk breaking something for this optimization. >> Yes, it could. But in this patch the optimization works only in the case if the flag is MS_ASYNC. If the flag is (MS_ASYNC | MS_INVALIDATE). > > Ah, right, it compares "== MS_ASYNC" not "& MS_ASYNC" so it's correct. A comment would be nice, such as: > > We found an unmapped range and with MS_ASYNC and no MS_INVALIDATE there's nothing to do and the result will always be -ENOMEM, so we can return immediately. > Thanks for the comment! I should probably implement this comment to the patch description and resubmit it as v2. >>>> --- >>>> mm/msync.c | 2 ++ >>>> 1 file changed, 2 insertions(+) >>>> diff --git a/mm/msync.c b/mm/msync.c >>>> index 69c6d2029531..ed20c3621d4c 100644 >>>> --- a/mm/msync.c >>>> +++ b/mm/msync.c >>>> @@ -69,6 +69,8 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags) >>>> goto out_unlock; >>>> /* Here start < vma->vm_end. */ >>>> if (start < vma->vm_start) { >>>> + if (flags == MS_ASYNC) >>>> + goto out_unlock; >>>> start = vma->vm_start; >>>> if (start >= end) >>>> goto out_unlock; >>>> base-commit: 6824a8a9b4861d7df7ee132a952bdf6f84a99cb8 >
diff --git a/mm/msync.c b/mm/msync.c index 69c6d2029531..ed20c3621d4c 100644 --- a/mm/msync.c +++ b/mm/msync.c @@ -69,6 +69,8 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags) goto out_unlock; /* Here start < vma->vm_end. */ if (start < vma->vm_start) { + if (flags == MS_ASYNC) + goto out_unlock; start = vma->vm_start; if (start >= end) goto out_unlock;
Exit from the loop over the VMA in the case when the flags contain only an MS_ASYNC and start < vm_start. In this case msync() would return with -ENOMEM anyway so make it return early. Signed-off-by: Nikita Ermakov <sh1r4s3@mail.si-head.nl> --- mm/msync.c | 2 ++ 1 file changed, 2 insertions(+) base-commit: 6824a8a9b4861d7df7ee132a952bdf6f84a99cb8