There are vma_merge() and do_brk_flags() pass mm to vma_complete(), others
would pass the vma->vm_mm. The following explains that the mm is the
vma->vm_mm in vma_merge() and do_brk_flags().
All vma will point to the same mm struct if the vma_merge() is successful.
So the mm and the vma->mm are the same.
vm_brk_flags() and brk syscall will initialize vmi with current->mm,
so the vma->vm_mm and the current->mm are the same if vma exists in
do_brk_flags().
Remove the mm parameter in vma_complete() and get mm from the vma in vp.
Signed-off-by: Yajun Deng <[email protected]>
---
mm/mmap.c | 16 ++++++++--------
1 file changed, 8 insertions(+), 8 deletions(-)
diff --git a/mm/mmap.c b/mm/mmap.c
index e97b9144c61a..9b968d1edf55 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -509,11 +509,11 @@ static inline void vma_prepare(struct vma_prepare *vp)
*
* @vp: The vma_prepare struct
* @vmi: The vma iterator
- * @mm: The mm_struct
*/
-static inline void vma_complete(struct vma_prepare *vp,
- struct vma_iterator *vmi, struct mm_struct *mm)
+static inline void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi)
{
+ struct mm_struct *mm = vp->vma->vm_mm;
+
if (vp->file) {
if (vp->adj_next)
vma_interval_tree_insert(vp->adj_next,
@@ -666,7 +666,7 @@ int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
vma_set_range(vma, start, end, pgoff);
vma_iter_store(vmi, vma);
- vma_complete(&vp, vmi, vma->vm_mm);
+ vma_complete(&vp, vmi);
return 0;
nomem:
@@ -707,7 +707,7 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
vma_iter_clear(vmi);
vma_set_range(vma, start, end, pgoff);
- vma_complete(&vp, vmi, vma->vm_mm);
+ vma_complete(&vp, vmi);
return 0;
}
@@ -1030,7 +1030,7 @@ static struct vm_area_struct
}
}
- vma_complete(&vp, vmi, mm);
+ vma_complete(&vp, vmi);
khugepaged_enter_vma(res, vm_flags);
return res;
@@ -2377,7 +2377,7 @@ static int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
}
/* vma_complete stores the new vma */
- vma_complete(&vp, vmi, vma->vm_mm);
+ vma_complete(&vp, vmi);
/* Success. */
if (new_below)
@@ -3145,7 +3145,7 @@ static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
vm_flags_set(vma, VM_SOFTDIRTY);
vma_iter_store(vmi, vma);
- vma_complete(&vp, vmi, mm);
+ vma_complete(&vp, vmi);
khugepaged_enter_vma(vma, flags);
goto out;
}
--
2.25.1
* Yajun Deng <[email protected]> [240129 02:53]:
> There are vma_merge() and do_brk_flags() pass mm to vma_complete(), others
> would pass the vma->vm_mm. The following explains that the mm is the
> vma->vm_mm in vma_merge() and do_brk_flags().
>
> All vma will point to the same mm struct if the vma_merge() is successful.
> So the mm and the vma->mm are the same.
Absolutely, they must be the same. I don't think vma_merge() checks
this, but it is true.
>
> vm_brk_flags() and brk syscall will initialize vmi with current->mm,
> so the vma->vm_mm and the current->mm are the same if vma exists in
> do_brk_flags().
>
> Remove the mm parameter in vma_complete() and get mm from the vma in vp.
You have added a dereference to the two paths that don't need it to
reduce the argument list from 3 to 2. It's the same number of lines as
well. vma_shrink() is only used on process creation, but brk is more
common. Note that this function is marked as inline.
I'm not sure this change is worth making.
>
> Signed-off-by: Yajun Deng <[email protected]>
> ---
> mm/mmap.c | 16 ++++++++--------
> 1 file changed, 8 insertions(+), 8 deletions(-)
>
> diff --git a/mm/mmap.c b/mm/mmap.c
> index e97b9144c61a..9b968d1edf55 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -509,11 +509,11 @@ static inline void vma_prepare(struct vma_prepare *vp)
> *
> * @vp: The vma_prepare struct
> * @vmi: The vma iterator
> - * @mm: The mm_struct
> */
> -static inline void vma_complete(struct vma_prepare *vp,
> - struct vma_iterator *vmi, struct mm_struct *mm)
> +static inline void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi)
> {
> + struct mm_struct *mm = vp->vma->vm_mm;
> +
> if (vp->file) {
> if (vp->adj_next)
> vma_interval_tree_insert(vp->adj_next,
> @@ -666,7 +666,7 @@ int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
> vma_set_range(vma, start, end, pgoff);
> vma_iter_store(vmi, vma);
>
> - vma_complete(&vp, vmi, vma->vm_mm);
> + vma_complete(&vp, vmi);
> return 0;
>
> nomem:
> @@ -707,7 +707,7 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
>
> vma_iter_clear(vmi);
> vma_set_range(vma, start, end, pgoff);
> - vma_complete(&vp, vmi, vma->vm_mm);
> + vma_complete(&vp, vmi);
> return 0;
> }
>
> @@ -1030,7 +1030,7 @@ static struct vm_area_struct
> }
> }
>
> - vma_complete(&vp, vmi, mm);
> + vma_complete(&vp, vmi);
> khugepaged_enter_vma(res, vm_flags);
> return res;
>
> @@ -2377,7 +2377,7 @@ static int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
> }
>
> /* vma_complete stores the new vma */
> - vma_complete(&vp, vmi, vma->vm_mm);
> + vma_complete(&vp, vmi);
>
> /* Success. */
> if (new_below)
> @@ -3145,7 +3145,7 @@ static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
> vm_flags_set(vma, VM_SOFTDIRTY);
> vma_iter_store(vmi, vma);
>
> - vma_complete(&vp, vmi, mm);
> + vma_complete(&vp, vmi);
> khugepaged_enter_vma(vma, flags);
> goto out;
> }
> --
> 2.25.1
>
>
Adding Vlastimil and Lorenzo to discuss this patch.
On 2024/1/29 23:04, Liam R. Howlett wrote:
> * Yajun Deng <[email protected]> [240129 02:53]:
>> There are vma_merge() and do_brk_flags() pass mm to vma_complete(), others
>> would pass the vma->vm_mm. The following explains that the mm is the
>> vma->vm_mm in vma_merge() and do_brk_flags().
>>
>> All vma will point to the same mm struct if the vma_merge() is successful.
>> So the mm and the vma->mm are the same.
> Absolutely, they must be the same. I don't think vma_merge() checks
> this, but it is true.
>
>> vm_brk_flags() and brk syscall will initialize vmi with current->mm,
>> so the vma->vm_mm and the current->mm are the same if vma exists in
>> do_brk_flags().
>>
>> Remove the mm parameter in vma_complete() and get mm from the vma in vp.
> You have added a dereference to the two paths that don't need it to
> reduce the argument list from 3 to 2. It's the same number of lines as
> well. vma_shrink() is only used on process creation, but brk is more
> common. Note that this function is marked as inline.
>
> I'm not sure this change is worth making.
If we can make sure the mm is vma->vm_mm, I don't think we need to pass
the mm.
If we can't make sure that, this change is not worth it.
>> Signed-off-by: Yajun Deng <[email protected]>
>> ---
>> mm/mmap.c | 16 ++++++++--------
>> 1 file changed, 8 insertions(+), 8 deletions(-)
>>
>> diff --git a/mm/mmap.c b/mm/mmap.c
>> index e97b9144c61a..9b968d1edf55 100644
>> --- a/mm/mmap.c
>> +++ b/mm/mmap.c
>> @@ -509,11 +509,11 @@ static inline void vma_prepare(struct vma_prepare *vp)
>> *
>> * @vp: The vma_prepare struct
>> * @vmi: The vma iterator
>> - * @mm: The mm_struct
>> */
>> -static inline void vma_complete(struct vma_prepare *vp,
>> - struct vma_iterator *vmi, struct mm_struct *mm)
>> +static inline void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi)
>> {
>> + struct mm_struct *mm = vp->vma->vm_mm;
>> +
>> if (vp->file) {
>> if (vp->adj_next)
>> vma_interval_tree_insert(vp->adj_next,
>> @@ -666,7 +666,7 @@ int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
>> vma_set_range(vma, start, end, pgoff);
>> vma_iter_store(vmi, vma);
>>
>> - vma_complete(&vp, vmi, vma->vm_mm);
>> + vma_complete(&vp, vmi);
>> return 0;
>>
>> nomem:
>> @@ -707,7 +707,7 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
>>
>> vma_iter_clear(vmi);
>> vma_set_range(vma, start, end, pgoff);
>> - vma_complete(&vp, vmi, vma->vm_mm);
>> + vma_complete(&vp, vmi);
>> return 0;
>> }
>>
>> @@ -1030,7 +1030,7 @@ static struct vm_area_struct
>> }
>> }
>>
>> - vma_complete(&vp, vmi, mm);
>> + vma_complete(&vp, vmi);
>> khugepaged_enter_vma(res, vm_flags);
>> return res;
>>
>> @@ -2377,7 +2377,7 @@ static int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
>> }
>>
>> /* vma_complete stores the new vma */
>> - vma_complete(&vp, vmi, vma->vm_mm);
>> + vma_complete(&vp, vmi);
>>
>> /* Success. */
>> if (new_below)
>> @@ -3145,7 +3145,7 @@ static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
>> vm_flags_set(vma, VM_SOFTDIRTY);
>> vma_iter_store(vmi, vma);
>>
>> - vma_complete(&vp, vmi, mm);
>> + vma_complete(&vp, vmi);
>> khugepaged_enter_vma(vma, flags);
>> goto out;
>> }
>> --
>> 2.25.1
>>
>>
* Yajun Deng <[email protected]> [240222 05:26]:
> Adding Vlastimil and Lorenzo to discuss this patch.
>
>
> On 2024/1/29 23:04, Liam R. Howlett wrote:
> > * Yajun Deng <[email protected]> [240129 02:53]:
> > > There are vma_merge() and do_brk_flags() pass mm to vma_complete(), others
> > > would pass the vma->vm_mm. The following explains that the mm is the
> > > vma->vm_mm in vma_merge() and do_brk_flags().
> > >
> > > All vma will point to the same mm struct if the vma_merge() is successful.
> > > So the mm and the vma->mm are the same.
> > Absolutely, they must be the same. I don't think vma_merge() checks
> > this, but it is true.
> >
> > > vm_brk_flags() and brk syscall will initialize vmi with current->mm,
> > > so the vma->vm_mm and the current->mm are the same if vma exists in
> > > do_brk_flags().
> > >
> > > Remove the mm parameter in vma_complete() and get mm from the vma in vp.
> > You have added a dereference to the two paths that don't need it to
> > reduce the argument list from 3 to 2. It's the same number of lines as
> > well. vma_shrink() is only used on process creation, but brk is more
> > common. Note that this function is marked as inline.
> >
> > I'm not sure this change is worth making.
>
> If we can make sure the mm is? vma->vm_mm, I don't think we need to pass the
> mm.
>
> If we can't make sure that, this change is not worth it.
We can be quite confident the mm struct is the same. The point is that
you are causing more instructions for zero gain. There isn't a lot of
arguments and this is marked inline. For most of the cases, we are
already causing 1/2 the dereferences you are moving - except
brk_flags(), which already has the pointer available. But instead of
using the pointer already in a register, you are adding two new
dereferences inside an inline function.
This is like writing:
struct mm_struct *mm = current->mm;
struct vm_area_stuuct *vma = find_vma(mm, 0);
..
use_the_mm(vma->vm_mm);
. only it's worse than that because the compiler will replace
use_the_mm() with the actual code in use_the_mm(), so we have
effectively told the compiler to set another register up by
dereferencing twice instead of using the value already available.
It's a change for the sake of changing.
You are not reducing the code size, you are not increasing the
readability. You are adding two dereferences to brk() and one to all
other callers. Why do this change?
>
> > > Signed-off-by: Yajun Deng <[email protected]>
> > > ---
> > > mm/mmap.c | 16 ++++++++--------
> > > 1 file changed, 8 insertions(+), 8 deletions(-)
> > >
> > > diff --git a/mm/mmap.c b/mm/mmap.c
> > > index e97b9144c61a..9b968d1edf55 100644
> > > --- a/mm/mmap.c
> > > +++ b/mm/mmap.c
> > > @@ -509,11 +509,11 @@ static inline void vma_prepare(struct vma_prepare *vp)
> > > *
> > > * @vp: The vma_prepare struct
> > > * @vmi: The vma iterator
> > > - * @mm: The mm_struct
> > > */
> > > -static inline void vma_complete(struct vma_prepare *vp,
> > > - struct vma_iterator *vmi, struct mm_struct *mm)
> > > +static inline void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi)
> > > {
> > > + struct mm_struct *mm = vp->vma->vm_mm;
> > > +
> > > if (vp->file) {
> > > if (vp->adj_next)
> > > vma_interval_tree_insert(vp->adj_next,
> > > @@ -666,7 +666,7 @@ int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
> > > vma_set_range(vma, start, end, pgoff);
> > > vma_iter_store(vmi, vma);
> > > - vma_complete(&vp, vmi, vma->vm_mm);
> > > + vma_complete(&vp, vmi);
> > > return 0;
> > > nomem:
> > > @@ -707,7 +707,7 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
> > > vma_iter_clear(vmi);
> > > vma_set_range(vma, start, end, pgoff);
> > > - vma_complete(&vp, vmi, vma->vm_mm);
> > > + vma_complete(&vp, vmi);
> > > return 0;
> > > }
> > > @@ -1030,7 +1030,7 @@ static struct vm_area_struct
> > > }
> > > }
> > > - vma_complete(&vp, vmi, mm);
> > > + vma_complete(&vp, vmi);
> > > khugepaged_enter_vma(res, vm_flags);
> > > return res;
> > > @@ -2377,7 +2377,7 @@ static int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
> > > }
> > > /* vma_complete stores the new vma */
> > > - vma_complete(&vp, vmi, vma->vm_mm);
> > > + vma_complete(&vp, vmi);
> > > /* Success. */
> > > if (new_below)
> > > @@ -3145,7 +3145,7 @@ static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
> > > vm_flags_set(vma, VM_SOFTDIRTY);
> > > vma_iter_store(vmi, vma);
> > > - vma_complete(&vp, vmi, mm);
> > > + vma_complete(&vp, vmi);
> > > khugepaged_enter_vma(vma, flags);
> > > goto out;
> > > }
> > > --
> > > 2.25.1
> > >
> > >
On 2024/2/23 00:22, Liam R. Howlett wrote:
> * Yajun Deng <[email protected]> [240222 05:26]:
>> Adding Vlastimil and Lorenzo to discuss this patch.
>>
>>
>> On 2024/1/29 23:04, Liam R. Howlett wrote:
>>> * Yajun Deng <[email protected]> [240129 02:53]:
>>>> There are vma_merge() and do_brk_flags() pass mm to vma_complete(), others
>>>> would pass the vma->vm_mm. The following explains that the mm is the
>>>> vma->vm_mm in vma_merge() and do_brk_flags().
>>>>
>>>> All vma will point to the same mm struct if the vma_merge() is successful.
>>>> So the mm and the vma->mm are the same.
>>> Absolutely, they must be the same. I don't think vma_merge() checks
>>> this, but it is true.
>>>
>>>> vm_brk_flags() and brk syscall will initialize vmi with current->mm,
>>>> so the vma->vm_mm and the current->mm are the same if vma exists in
>>>> do_brk_flags().
>>>>
>>>> Remove the mm parameter in vma_complete() and get mm from the vma in vp.
>>> You have added a dereference to the two paths that don't need it to
>>> reduce the argument list from 3 to 2. It's the same number of lines as
>>> well. vma_shrink() is only used on process creation, but brk is more
>>> common. Note that this function is marked as inline.
>>>
>>> I'm not sure this change is worth making.
>> If we can make sure the mm is vma->vm_mm, I don't think we need to pass the
>> mm.
>>
>> If we can't make sure that, this change is not worth it.
> We can be quite confident the mm struct is the same. The point is that
> you are causing more instructions for zero gain. There isn't a lot of
> arguments and this is marked inline. For most of the cases, we are
> already causing 1/2 the dereferences you are moving - except
> brk_flags(), which already has the pointer available. But instead of
> using the pointer already in a register, you are adding two new
> dereferences inside an inline function.
>
> This is like writing:
>
> struct mm_struct *mm = current->mm;
> struct vm_area_stuuct *vma = find_vma(mm, 0);
>
> ...
>
> use_the_mm(vma->vm_mm);
>
> .. only it's worse than that because the compiler will replace
> use_the_mm() with the actual code in use_the_mm(), so we have
> effectively told the compiler to set another register up by
> dereferencing twice instead of using the value already available.
>
> It's a change for the sake of changing.
>
> You are not reducing the code size, you are not increasing the
> readability. You are adding two dereferences to brk() and one to all
> other callers. Why do this change?
Thank you for your explanation.
>>>> Signed-off-by: Yajun Deng <[email protected]>
>>>> ---
>>>> mm/mmap.c | 16 ++++++++--------
>>>> 1 file changed, 8 insertions(+), 8 deletions(-)
>>>>
>>>> diff --git a/mm/mmap.c b/mm/mmap.c
>>>> index e97b9144c61a..9b968d1edf55 100644
>>>> --- a/mm/mmap.c
>>>> +++ b/mm/mmap.c
>>>> @@ -509,11 +509,11 @@ static inline void vma_prepare(struct vma_prepare *vp)
>>>> *
>>>> * @vp: The vma_prepare struct
>>>> * @vmi: The vma iterator
>>>> - * @mm: The mm_struct
>>>> */
>>>> -static inline void vma_complete(struct vma_prepare *vp,
>>>> - struct vma_iterator *vmi, struct mm_struct *mm)
>>>> +static inline void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi)
>>>> {
>>>> + struct mm_struct *mm = vp->vma->vm_mm;
>>>> +
>>>> if (vp->file) {
>>>> if (vp->adj_next)
>>>> vma_interval_tree_insert(vp->adj_next,
>>>> @@ -666,7 +666,7 @@ int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
>>>> vma_set_range(vma, start, end, pgoff);
>>>> vma_iter_store(vmi, vma);
>>>> - vma_complete(&vp, vmi, vma->vm_mm);
>>>> + vma_complete(&vp, vmi);
>>>> return 0;
>>>> nomem:
>>>> @@ -707,7 +707,7 @@ int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
>>>> vma_iter_clear(vmi);
>>>> vma_set_range(vma, start, end, pgoff);
>>>> - vma_complete(&vp, vmi, vma->vm_mm);
>>>> + vma_complete(&vp, vmi);
>>>> return 0;
>>>> }
>>>> @@ -1030,7 +1030,7 @@ static struct vm_area_struct
>>>> }
>>>> }
>>>> - vma_complete(&vp, vmi, mm);
>>>> + vma_complete(&vp, vmi);
>>>> khugepaged_enter_vma(res, vm_flags);
>>>> return res;
>>>> @@ -2377,7 +2377,7 @@ static int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
>>>> }
>>>> /* vma_complete stores the new vma */
>>>> - vma_complete(&vp, vmi, vma->vm_mm);
>>>> + vma_complete(&vp, vmi);
>>>> /* Success. */
>>>> if (new_below)
>>>> @@ -3145,7 +3145,7 @@ static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
>>>> vm_flags_set(vma, VM_SOFTDIRTY);
>>>> vma_iter_store(vmi, vma);
>>>> - vma_complete(&vp, vmi, mm);
>>>> + vma_complete(&vp, vmi);
>>>> khugepaged_enter_vma(vma, flags);
>>>> goto out;
>>>> }
>>>> --
>>>> 2.25.1
>>>>
>>>>