[v2,2/5] userfaultfd: introduce access-likely mode for common operations

Commit Message

Nadav Amit July 18, 2022, 11:47 a.m. UTC

From: Nadav Amit <[e-mail]>

Introduce access-hints in userfaultfd. The expectation is that userspace
would set access-hints when a page-fault occurred on a page and would
not provide the access-hint on prefaulted memory. The exact behavior of
the kernel in regard to the hints would not be part of userfaultfd api.

At this time the use of the access-hint is only in setting access-bit
similarly to the way it is done in do_set_pte(). In x86, currently PTEs
are always marked as young, including prefetched ones. But on arm64,
PTEs would be marked as old (when access bit is supported).

If access hints are not enabled, the kernel would behave as if the
access-hint was provided for backward compatibility.

Cc: Mike Kravetz <[e-mail]>
Cc: Hugh Dickins <[e-mail]>
Cc: Andrew Morton <[e-mail]>
Cc: Axel Rasmussen <[e-mail]>
Cc: Peter Xu <[e-mail]>
Cc: David Hildenbrand <[e-mail]>
Cc: Mike Rapoport <[e-mail]>
Signed-off-by: Nadav Amit <[e-mail]>
---
 fs/userfaultfd.c                 | 39 ++++++++++++++++++++++++++++----
 include/linux/userfaultfd_k.h    |  1 +
 include/uapi/linux/userfaultfd.h | 20 +++++++++++++++-
 mm/internal.h                    | 13 +++++++++++
 mm/memory.c                      | 12 ----------
 mm/userfaultfd.c                 | 11 +++++++--
 6 files changed, 77 insertions(+), 19 deletions(-)

Comments

Peter Xu July 18, 2022, 8:05 p.m. UTC | #1

On Mon, Jul 18, 2022 at 04:47:45AM -0700, Nadav Amit wrote:
> @@ -261,6 +272,7 @@ struct uffdio_copy {
>  struct uffdio_zeropage {
>  	struct uffdio_range range;
>  #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((__u64)1<<0)
> +#define UFFDIO_ZEROPAGE_MODE_ACCESS_LIKELY	((__u64)1<<1)

Would access hint help zeropage use case?  I remembered you used to comment
around and said it won't help since we won't reclaim zero page anyway.

It won't help either even if this flag is only used for the follow up
WRITE_HINT (since then there'll be a CoW) because when WRITE_HINT attached
it doesn't make sense to not have ACCESS_HINT, then it seems the WRITE_HINT
itself would be enough for ZEROPAGE to me.

[...]

> diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
> index 421784d26651..c15679f3eb6a 100644
> --- a/mm/userfaultfd.c
> +++ b/mm/userfaultfd.c
> @@ -65,6 +65,7 @@ int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
>  	bool writable = dst_vma->vm_flags & VM_WRITE;
>  	bool vm_shared = dst_vma->vm_flags & VM_SHARED;
>  	bool page_in_cache = page->mapping;
> +	bool prefault = !(uffd_flags & UFFD_FLAGS_ACCESS_LIKELY);

I think it's okay to name it "prefault" as a temp var, but ideally IMHO we
shouldn't assume what the user app is doing - it is only installing some
uffd pgtables with !ACCESS_LIKELY and it does not necessarily need to be a
prefault process..

>  	spinlock_t *ptl;
>  	struct inode *inode;
>  	pgoff_t offset, max_off;
> @@ -92,6 +93,11 @@ int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
>  		 */
>  		_dst_pte = pte_wrprotect(_dst_pte);
>  
> +	if (prefault && arch_wants_old_prefaulted_pte())
> +		_dst_pte = pte_mkold(_dst_pte);
> +	else
> +		_dst_pte = pte_sw_mkyoung(_dst_pte);

Could you explain why we couldn't unconditionally mkold here even for x86?

It'll be a pity if this feature bit will only be useful on arm64 but not
covering x86 (which is so far still the majority I think).

IMHO it's slightly different here comparing to kernel prefaults - the uesr
app may not be aware of kernel prefaults, but here !ACCESS_HINT it's
user-aware, and it's what user app explicitly provided.  IMO it's a
stronger proof of a cold page already.

The other thing I got confused here is arch_wants_old_prefaulted_pte()
returns true if arm64 supports hardware AF.  However for all the rest archs
(including x86_64 which, afaict, support AF too in most models) it'll
constantly return false.  Do you know what's the rational behind?

> +
>  	dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
>  
>  	if (vma_is_shmem(dst_vma)) {
> @@ -202,7 +208,8 @@ static int mcopy_atomic_pte(struct mm_struct *dst_mm,
>  static int mfill_zeropage_pte(struct mm_struct *dst_mm,
>  			      pmd_t *dst_pmd,
>  			      struct vm_area_struct *dst_vma,
> -			      unsigned long dst_addr)
> +			      unsigned long dst_addr,
> +			      uffd_flags_t uffd_flags)
>  {
>  	pte_t _dst_pte, *dst_pte;
>  	spinlock_t *ptl;
> @@ -495,7 +502,7 @@ static __always_inline ssize_t mfill_atomic_pte(struct mm_struct *dst_mm,
>  					       uffd_flags);
>  		else
>  			err = mfill_zeropage_pte(dst_mm, dst_pmd,
> -						 dst_vma, dst_addr);
> +						 dst_vma, dst_addr, uffd_flags);
>  	} else {
>  		err = shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
>  					     dst_addr, src_addr,
> -- 
> 2.25.1
>

Nadav Amit July 18, 2022, 8:59 p.m. UTC | #2

On Jul 18, 2022, at 1:05 PM, Peter Xu <[e-mail]> wrote:

> ⚠ External Email
> 
> On Mon, Jul 18, 2022 at 04:47:45AM -0700, Nadav Amit wrote:
>> @@ -261,6 +272,7 @@ struct uffdio_copy {
>> struct uffdio_zeropage {
>>      struct uffdio_range range;
>> #define UFFDIO_ZEROPAGE_MODE_DONTWAKE                ((__u64)1<<0)
>> +#define UFFDIO_ZEROPAGE_MODE_ACCESS_LIKELY   ((__u64)1<<1)
> 
> Would access hint help zeropage use case?  I remembered you used to comment
> around and said it won't help since we won't reclaim zero page anyway.

I agree that there is no meaning for access bit on zero page. I just think
that it is best to have the flags for consistency. If you ask me, I would
prefer to have all the flags in a fixed place (highest bits?). Anyhow, if we
expose the hints as a feature, I do not think we would later want to say
“here is another feature that enables another hint that we thought is not
needed before”. Userfaultfd’s feature bits are already nuts, IMHO.

> It won't help either even if this flag is only used for the follow up
> WRITE_HINT (since then there'll be a CoW) because when WRITE_HINT attached
> it doesn't make sense to not have ACCESS_HINT, then it seems the WRITE_HINT
> itself would be enough for ZEROPAGE to me.

Agreed. Again, I think it is worthy for consistency.

> [...]
> 
>> diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
>> index 421784d26651..c15679f3eb6a 100644
>> --- a/mm/userfaultfd.c
>> +++ b/mm/userfaultfd.c
>> @@ -65,6 +65,7 @@ int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
>>      bool writable = dst_vma->vm_flags & VM_WRITE;
>>      bool vm_shared = dst_vma->vm_flags & VM_SHARED;
>>      bool page_in_cache = page->mapping;
>> +     bool prefault = !(uffd_flags & UFFD_FLAGS_ACCESS_LIKELY);
> 
> I think it's okay to name it "prefault" as a temp var, but ideally IMHO we
> shouldn't assume what the user app is doing - it is only installing some
> uffd pgtables with !ACCESS_LIKELY and it does not necessarily need to be a
> prefault process..
> 
>>      spinlock_t *ptl;
>>      struct inode *inode;
>>      pgoff_t offset, max_off;
>> @@ -92,6 +93,11 @@ int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
>>               */
>>              _dst_pte = pte_wrprotect(_dst_pte);
>> 
>> +     if (prefault && arch_wants_old_prefaulted_pte())
>> +             _dst_pte = pte_mkold(_dst_pte);
>> +     else
>> +             _dst_pte = pte_sw_mkyoung(_dst_pte);
> 
> Could you explain why we couldn't unconditionally mkold here even for x86?

To answer this question and the previous one, please note that the logic is
“borrowed” from do_set_pte(). If you want me to refactor and extract a
function, please let me know.

Here is the deal: for x86, we don’t do pte_mkold() because setting the
access bit is expensive (>500 cycles). For arm64 that have access-bit we
don’t since (according to arm64 code or commit log), the cost of setting the
access bit on arm is low.

> It'll be a pity if this feature bit will only be useful on arm64 but not
> covering x86 (which is so far still the majority I think).
> 
> IMHO it's slightly different here comparing to kernel prefaults - the uesr
> app may not be aware of kernel prefaults, but here !ACCESS_HINT it's
> user-aware, and it's what user app explicitly provided.  IMO it's a
> stronger proof of a cold page already.

I’m ok with that if that is your choice. I actually prefer to give userspace
more control, but I tried to be consistent with other parts of the kernel.
Having said that, it’s really hard for me to see why young bit would be clear,
but dirty bit would be set...

> The other thing I got confused here is arch_wants_old_prefaulted_pte()
> returns true if arm64 supports hardware AF.  However for all the rest archs
> (including x86_64 which, afaict, support AF too in most models) it'll
> constantly return false.  Do you know what's the rational behind?

All x86 (32/64) since 386 support access-bit in the page-tables (IIRC, 286
had access bit in the segments).

I thought we discussed it before: if you access an old PTE on x86, you pay
>500 cycles; this actually affected UnixBench when people tried to change
this behavior [1]. In contrast, on arm64, which I have never profiled, you
probably saw the comment saying: "Experimentally, it's cheap to set the
access flag in hardware and we benefit from prefaulting mappings as 'old’ to
start with.”.

I do not know what happens on other architectures.

( sorry if I have some repetitions in this email )

[1] https://marc.info/?l=linux-kernel&m=146582237922378&w=2

Peter Xu July 18, 2022, 9:21 p.m. UTC | #3

On Mon, Jul 18, 2022 at 08:59:37PM +0000, Nadav Amit wrote:
> On Jul 18, 2022, at 1:05 PM, Peter Xu <[e-mail]> wrote:
> 
> > ⚠ External Email
> > 
> > On Mon, Jul 18, 2022 at 04:47:45AM -0700, Nadav Amit wrote:
> >> @@ -261,6 +272,7 @@ struct uffdio_copy {
> >> struct uffdio_zeropage {
> >>      struct uffdio_range range;
> >> #define UFFDIO_ZEROPAGE_MODE_DONTWAKE                ((__u64)1<<0)
> >> +#define UFFDIO_ZEROPAGE_MODE_ACCESS_LIKELY   ((__u64)1<<1)
> > 
> > Would access hint help zeropage use case?  I remembered you used to comment
> > around and said it won't help since we won't reclaim zero page anyway.
> 
> I agree that there is no meaning for access bit on zero page. I just think
> that it is best to have the flags for consistency. If you ask me, I would
> prefer to have all the flags in a fixed place (highest bits?). Anyhow, if we
> expose the hints as a feature, I do not think we would later want to say
> “here is another feature that enables another hint that we thought is not
> needed before”. Userfaultfd’s feature bits are already nuts, IMHO.
> 
> > It won't help either even if this flag is only used for the follow up
> > WRITE_HINT (since then there'll be a CoW) because when WRITE_HINT attached
> > it doesn't make sense to not have ACCESS_HINT, then it seems the WRITE_HINT
> > itself would be enough for ZEROPAGE to me.
> 
> Agreed. Again, I think it is worthy for consistency.

I'd be fine if it's kernel internal flags only.  But this is solid kernel
ABI.  Are you.. sure?

We're literally trying to introduce some flags just for "consistency" even
if we know nobody will be using it.  It really dosn't sound very right on
designing good interfaces..

> 
> > [...]
> > 
> >> diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
> >> index 421784d26651..c15679f3eb6a 100644
> >> --- a/mm/userfaultfd.c
> >> +++ b/mm/userfaultfd.c
> >> @@ -65,6 +65,7 @@ int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
> >>      bool writable = dst_vma->vm_flags & VM_WRITE;
> >>      bool vm_shared = dst_vma->vm_flags & VM_SHARED;
> >>      bool page_in_cache = page->mapping;
> >> +     bool prefault = !(uffd_flags & UFFD_FLAGS_ACCESS_LIKELY);
> > 
> > I think it's okay to name it "prefault" as a temp var, but ideally IMHO we
> > shouldn't assume what the user app is doing - it is only installing some
> > uffd pgtables with !ACCESS_LIKELY and it does not necessarily need to be a
> > prefault process..
> > 
> >>      spinlock_t *ptl;
> >>      struct inode *inode;
> >>      pgoff_t offset, max_off;
> >> @@ -92,6 +93,11 @@ int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
> >>               */
> >>              _dst_pte = pte_wrprotect(_dst_pte);
> >> 
> >> +     if (prefault && arch_wants_old_prefaulted_pte())
> >> +             _dst_pte = pte_mkold(_dst_pte);
> >> +     else
> >> +             _dst_pte = pte_sw_mkyoung(_dst_pte);
> > 
> > Could you explain why we couldn't unconditionally mkold here even for x86?
> 
> To answer this question and the previous one, please note that the logic is
> “borrowed” from do_set_pte(). If you want me to refactor and extract a
> function, please let me know.
> 
> Here is the deal: for x86, we don’t do pte_mkold() because setting the
> access bit is expensive (>500 cycles). For arm64 that have access-bit we
> don’t since (according to arm64 code or commit log), the cost of setting the
> access bit on arm is low.
> 
> > It'll be a pity if this feature bit will only be useful on arm64 but not
> > covering x86 (which is so far still the majority I think).
> > 
> > IMHO it's slightly different here comparing to kernel prefaults - the uesr
> > app may not be aware of kernel prefaults, but here !ACCESS_HINT it's
> > user-aware, and it's what user app explicitly provided.  IMO it's a
> > stronger proof of a cold page already.
> 
> I’m ok with that if that is your choice. I actually prefer to give userspace
> more control, but I tried to be consistent with other parts of the kernel.

Ah good to know, then if there's a vote I'll go for your proposal.

I'd suggest we make it a strong semantics.  We used to have similar
discussions around the MADV_COLLAPSE on whether it should be restricted to
khugepaged limitations.  I think it's similar here.

> Having said that, it’s really hard for me to see why young bit would be clear,
> but dirty bit would be set...

Assume one page has both young/dirty set, the reclaim code decides to age
this page, then.. young=0 && dirty=1?

> 
> > The other thing I got confused here is arch_wants_old_prefaulted_pte()
> > returns true if arm64 supports hardware AF.  However for all the rest archs
> > (including x86_64 which, afaict, support AF too in most models) it'll
> > constantly return false.  Do you know what's the rational behind?
> 
> All x86 (32/64) since 386 support access-bit in the page-tables (IIRC, 286
> had access bit in the segments).
> 
> I thought we discussed it before: if you access an old PTE on x86, you pay
> >500 cycles; this actually affected UnixBench when people tried to change
> this behavior [1]. In contrast, on arm64, which I have never profiled, you
> probably saw the comment saying: "Experimentally, it's cheap to set the
> access flag in hardware and we benefit from prefaulting mappings as 'old’ to
> start with.”.

Thanks.  I'm really curious how fast would aarch64 be on setting
hardware-assist young bit and why now.

> 
> I do not know what happens on other architectures.
> 
> ( sorry if I have some repetitions in this email )
> 
> [1] https://marc.info/?l=linux-kernel&m=146582237922378&w=2
>

Mike Rapoport July 23, 2022, 9:16 a.m. UTC | #4

On Mon, Jul 18, 2022 at 04:47:45AM -0700, Nadav Amit wrote:
> From: Nadav Amit <[e-mail]>
> 
> Introduce access-hints in userfaultfd. The expectation is that userspace
> would set access-hints when a page-fault occurred on a page and would
> not provide the access-hint on prefaulted memory. The exact behavior of
> the kernel in regard to the hints would not be part of userfaultfd api.
> 
> At this time the use of the access-hint is only in setting access-bit
> similarly to the way it is done in do_set_pte(). In x86, currently PTEs
> are always marked as young, including prefetched ones. But on arm64,
> PTEs would be marked as old (when access bit is supported).
> 
> If access hints are not enabled, the kernel would behave as if the
> access-hint was provided for backward compatibility.
> 
> Cc: Mike Kravetz <[e-mail]>
> Cc: Hugh Dickins <[e-mail]>
> Cc: Andrew Morton <[e-mail]>
> Cc: Axel Rasmussen <[e-mail]>
> Cc: Peter Xu <[e-mail]>
> Cc: David Hildenbrand <[e-mail]>
> Cc: Mike Rapoport <[e-mail]>
> Signed-off-by: Nadav Amit <[e-mail]>
> ---
>  fs/userfaultfd.c                 | 39 ++++++++++++++++++++++++++++----
>  include/linux/userfaultfd_k.h    |  1 +
>  include/uapi/linux/userfaultfd.h | 20 +++++++++++++++-
>  mm/internal.h                    | 13 +++++++++++
>  mm/memory.c                      | 12 ----------
>  mm/userfaultfd.c                 | 11 +++++++--
>  6 files changed, 77 insertions(+), 19 deletions(-)
> 
> diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> index 2ae24327beec..8d8792b27c53 100644
> --- a/fs/userfaultfd.c
> +++ b/fs/userfaultfd.c
> @@ -1708,13 +1708,21 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
>  	ret = -EINVAL;
>  	if (uffdio_copy.src + uffdio_copy.len <= uffdio_copy.src)
>  		goto out;
> -	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP))
> +	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP|
> +				 UFFDIO_COPY_MODE_ACCESS_LIKELY))
>  		goto out;
>  
>  	mode_wp = uffdio_copy.mode & UFFDIO_COPY_MODE_WP;
>  
>  	uffd_flags = mode_wp ? UFFD_FLAGS_WP : UFFD_FLAGS_NONE;
>  
> +	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS) {
> +		if (uffdio_copy.mode & UFFDIO_COPY_MODE_ACCESS_LIKELY)
> +			uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
> +	} else {
> +		uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
> +	}
> +

This is quite a construct and it gets more complex in the following
patches. How about making it to a static inline function?

>  	if (mmget_not_zero(ctx->mm)) {
>  		ret = mcopy_atomic(ctx->mm, uffdio_copy.dst, uffdio_copy.src,
>  				   uffdio_copy.len, &ctx->mmap_changing,
> @@ -1765,9 +1773,17 @@ static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
>  	if (ret)
>  		goto out;
>  	ret = -EINVAL;
> -	if (uffdio_zeropage.mode & ~UFFDIO_ZEROPAGE_MODE_DONTWAKE)
> +	if (uffdio_zeropage.mode & ~(UFFDIO_ZEROPAGE_MODE_DONTWAKE|
> +				     UFFDIO_ZEROPAGE_MODE_ACCESS_LIKELY))
>  		goto out;
>  
> +	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS) {
> +		if (uffdio_zeropage.mode & UFFDIO_ZEROPAGE_MODE_ACCESS_LIKELY)
> +			uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
> +	} else {
> +		uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
> +	}
> +
>  	if (mmget_not_zero(ctx->mm)) {
>  		ret = mfill_zeropage(ctx->mm, uffdio_zeropage.range.start,
>  				     uffdio_zeropage.range.len,

Nadav Amit July 25, 2022, 5:18 p.m. UTC | #5

> On Jul 23, 2022, at 2:16 AM, Mike Rapoport <[e-mail]> wrote:
> 
> On Mon, Jul 18, 2022 at 04:47:45AM -0700, Nadav Amit wrote:
>> From: Nadav Amit <[e-mail]>
>> 
>> Introduce access-hints in userfaultfd. The expectation is that userspace
>> would set access-hints when a page-fault occurred on a page and would
>> not provide the access-hint on prefaulted memory. The exact behavior of
>> the kernel in regard to the hints would not be part of userfaultfd api.
>> 
>> At this time the use of the access-hint is only in setting access-bit
>> similarly to the way it is done in do_set_pte(). In x86, currently PTEs
>> are always marked as young, including prefetched ones. But on arm64,
>> PTEs would be marked as old (when access bit is supported).
>> 
>> If access hints are not enabled, the kernel would behave as if the
>> access-hint was provided for backward compatibility.
>> 
>> Cc: Mike Kravetz <[e-mail]>
>> Cc: Hugh Dickins <[e-mail]>
>> Cc: Andrew Morton <[e-mail]>
>> Cc: Axel Rasmussen <[e-mail]>
>> Cc: Peter Xu <[e-mail]>
>> Cc: David Hildenbrand <[e-mail]>
>> Cc: Mike Rapoport <[e-mail]>
>> Signed-off-by: Nadav Amit <[e-mail]>
>> ---
>> fs/userfaultfd.c                 | 39 ++++++++++++++++++++++++++++----
>> include/linux/userfaultfd_k.h    |  1 +
>> include/uapi/linux/userfaultfd.h | 20 +++++++++++++++-
>> mm/internal.h                    | 13 +++++++++++
>> mm/memory.c                      | 12 ----------
>> mm/userfaultfd.c                 | 11 +++++++--
>> 6 files changed, 77 insertions(+), 19 deletions(-)
>> 
>> diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
>> index 2ae24327beec..8d8792b27c53 100644
>> --- a/fs/userfaultfd.c
>> +++ b/fs/userfaultfd.c
>> @@ -1708,13 +1708,21 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
>> 	ret = -EINVAL;
>> 	if (uffdio_copy.src + uffdio_copy.len <= uffdio_copy.src)
>> 		goto out;
>> -	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP))
>> +	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP|
>> +				 UFFDIO_COPY_MODE_ACCESS_LIKELY))
>> 		goto out;
>> 
>> 	mode_wp = uffdio_copy.mode & UFFDIO_COPY_MODE_WP;
>> 
>> 	uffd_flags = mode_wp ? UFFD_FLAGS_WP : UFFD_FLAGS_NONE;
>> 
>> +	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS) {
>> +		if (uffdio_copy.mode & UFFDIO_COPY_MODE_ACCESS_LIKELY)
>> +			uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
>> +	} else {
>> +		uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
>> +	}
>> +
> 
> This is quite a construct and it gets more complex in the following
> patches. How about making it to a static inline function?

Possible. There is another option though. I think it would have been
much cleaner if some flags were in common offsets in the different
“mode” fields. It might be too late for some fields (WP), but I can
put these the ACCESS/WRITE fields in the the high bits in fixed
place for all modes, which would allow to at least reuse the logic.

Is that ok?

Mike Rapoport July 26, 2022, 4:02 p.m. UTC | #6

On Mon, Jul 25, 2022 at 10:18:38AM -0700, Nadav Amit wrote:
> 
> > On Jul 23, 2022, at 2:16 AM, Mike Rapoport <[e-mail]> wrote:
> > 
> > On Mon, Jul 18, 2022 at 04:47:45AM -0700, Nadav Amit wrote:
> >> From: Nadav Amit <[e-mail]>
> >> 
> >> Introduce access-hints in userfaultfd. The expectation is that userspace
> >> would set access-hints when a page-fault occurred on a page and would
> >> not provide the access-hint on prefaulted memory. The exact behavior of
> >> the kernel in regard to the hints would not be part of userfaultfd api.
> >> 
> >> At this time the use of the access-hint is only in setting access-bit
> >> similarly to the way it is done in do_set_pte(). In x86, currently PTEs
> >> are always marked as young, including prefetched ones. But on arm64,
> >> PTEs would be marked as old (when access bit is supported).
> >> 
> >> If access hints are not enabled, the kernel would behave as if the
> >> access-hint was provided for backward compatibility.
> >> 
> >> Cc: Mike Kravetz <[e-mail]>
> >> Cc: Hugh Dickins <[e-mail]>
> >> Cc: Andrew Morton <[e-mail]>
> >> Cc: Axel Rasmussen <[e-mail]>
> >> Cc: Peter Xu <[e-mail]>
> >> Cc: David Hildenbrand <[e-mail]>
> >> Cc: Mike Rapoport <[e-mail]>
> >> Signed-off-by: Nadav Amit <[e-mail]>
> >> ---
> >> fs/userfaultfd.c                 | 39 ++++++++++++++++++++++++++++----
> >> include/linux/userfaultfd_k.h    |  1 +
> >> include/uapi/linux/userfaultfd.h | 20 +++++++++++++++-
> >> mm/internal.h                    | 13 +++++++++++
> >> mm/memory.c                      | 12 ----------
> >> mm/userfaultfd.c                 | 11 +++++++--
> >> 6 files changed, 77 insertions(+), 19 deletions(-)
> >> 
> >> diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> >> index 2ae24327beec..8d8792b27c53 100644
> >> --- a/fs/userfaultfd.c
> >> +++ b/fs/userfaultfd.c
> >> @@ -1708,13 +1708,21 @@ static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
> >> 	ret = -EINVAL;
> >> 	if (uffdio_copy.src + uffdio_copy.len <= uffdio_copy.src)
> >> 		goto out;
> >> -	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP))
> >> +	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP|
> >> +				 UFFDIO_COPY_MODE_ACCESS_LIKELY))
> >> 		goto out;
> >> 
> >> 	mode_wp = uffdio_copy.mode & UFFDIO_COPY_MODE_WP;
> >> 
> >> 	uffd_flags = mode_wp ? UFFD_FLAGS_WP : UFFD_FLAGS_NONE;
> >> 
> >> +	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS) {
> >> +		if (uffdio_copy.mode & UFFDIO_COPY_MODE_ACCESS_LIKELY)
> >> +			uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
> >> +	} else {
> >> +		uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
> >> +	}
> >> +
> > 
> > This is quite a construct and it gets more complex in the following
> > patches. How about making it to a static inline function?
> 
> Possible. There is another option though. I think it would have been
> much cleaner if some flags were in common offsets in the different
> “mode” fields. It might be too late for some fields (WP), but I can
> put these the ACCESS/WRITE fields in the the high bits in fixed
> place for all modes, which would allow to at least reuse the logic.

So unless I'm missing something it'll be

	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS)
		uffd_flags |= (uffdio_copy.mode & UFFDIO_COPY_MODE_ACCESS_MASK);
	else
		uffd_flags |= UFFD_FLAGS_ACCESS_MASK;

I still think it's worth wrapping it in static inline with a comments about
common offsets for 'if' clause and backward compatibility for 'else'
clause.

> Is that ok?
>

Patch

diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index 2ae24327beec..8d8792b27c53 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -1708,13 +1708,21 @@  static int userfaultfd_copy(struct userfaultfd_ctx *ctx,
 	ret = -EINVAL;
 	if (uffdio_copy.src + uffdio_copy.len <= uffdio_copy.src)
 		goto out;
-	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP))
+	if (uffdio_copy.mode & ~(UFFDIO_COPY_MODE_DONTWAKE|UFFDIO_COPY_MODE_WP|
+				 UFFDIO_COPY_MODE_ACCESS_LIKELY))
 		goto out;
 
 	mode_wp = uffdio_copy.mode & UFFDIO_COPY_MODE_WP;
 
 	uffd_flags = mode_wp ? UFFD_FLAGS_WP : UFFD_FLAGS_NONE;
 
+	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS) {
+		if (uffdio_copy.mode & UFFDIO_COPY_MODE_ACCESS_LIKELY)
+			uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
+	} else {
+		uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
+	}
+
 	if (mmget_not_zero(ctx->mm)) {
 		ret = mcopy_atomic(ctx->mm, uffdio_copy.dst, uffdio_copy.src,
 				   uffdio_copy.len, &ctx->mmap_changing,
@@ -1765,9 +1773,17 @@  static int userfaultfd_zeropage(struct userfaultfd_ctx *ctx,
 	if (ret)
 		goto out;
 	ret = -EINVAL;
-	if (uffdio_zeropage.mode & ~UFFDIO_ZEROPAGE_MODE_DONTWAKE)
+	if (uffdio_zeropage.mode & ~(UFFDIO_ZEROPAGE_MODE_DONTWAKE|
+				     UFFDIO_ZEROPAGE_MODE_ACCESS_LIKELY))
 		goto out;
 
+	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS) {
+		if (uffdio_zeropage.mode & UFFDIO_ZEROPAGE_MODE_ACCESS_LIKELY)
+			uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
+	} else {
+		uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
+	}
+
 	if (mmget_not_zero(ctx->mm)) {
 		ret = mfill_zeropage(ctx->mm, uffdio_zeropage.range.start,
 				     uffdio_zeropage.range.len,
@@ -1817,7 +1833,8 @@  static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx,
 		return ret;
 
 	if (uffdio_wp.mode & ~(UFFDIO_WRITEPROTECT_MODE_DONTWAKE |
-			       UFFDIO_WRITEPROTECT_MODE_WP))
+			       UFFDIO_WRITEPROTECT_MODE_WP |
+			       UFFDIO_WRITEPROTECT_MODE_ACCESS_LIKELY))
 		return -EINVAL;
 
 	mode_wp = uffdio_wp.mode & UFFDIO_WRITEPROTECT_MODE_WP;
@@ -1827,6 +1844,12 @@  static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx,
 		return -EINVAL;
 
 	uffd_flags = mode_wp ? UFFD_FLAGS_WP : UFFD_FLAGS_NONE;
+	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS) {
+		if (uffdio_wp.mode & UFFDIO_WRITEPROTECT_MODE_ACCESS_LIKELY)
+			uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
+	} else {
+		uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
+	}
 
 	if (mmget_not_zero(ctx->mm)) {
 		ret = mwriteprotect_range(ctx->mm, uffdio_wp.range.start,
@@ -1879,9 +1902,17 @@  static int userfaultfd_continue(struct userfaultfd_ctx *ctx, unsigned long arg)
 	    uffdio_continue.range.start) {
 		goto out;
 	}
-	if (uffdio_continue.mode & ~UFFDIO_CONTINUE_MODE_DONTWAKE)
+	if (uffdio_continue.mode & ~(UFFDIO_CONTINUE_MODE_DONTWAKE|
+				     UFFDIO_CONTINUE_MODE_ACCESS_LIKELY))
 		goto out;
 
+	if (ctx->features & UFFD_FEATURE_ACCESS_HINTS) {
+		if (uffdio_continue.mode & UFFDIO_CONTINUE_MODE_ACCESS_LIKELY)
+			uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
+	} else {
+		uffd_flags |= UFFD_FLAGS_ACCESS_LIKELY;
+	}
+
 	if (mmget_not_zero(ctx->mm)) {
 		ret = mcopy_continue(ctx->mm, uffdio_continue.range.start,
 				     uffdio_continue.range.len,
diff --git a/include/linux/userfaultfd_k.h b/include/linux/userfaultfd_k.h
index a63b61823984..b326798b5677 100644
--- a/include/linux/userfaultfd_k.h
+++ b/include/linux/userfaultfd_k.h
@@ -59,6 +59,7 @@  typedef unsigned int __bitwise uffd_flags_t;
 
 #define UFFD_FLAGS_NONE			((__force uffd_flags_t)0)
 #define UFFD_FLAGS_WP			((__force uffd_flags_t)BIT(0))
+#define UFFD_FLAGS_ACCESS_LIKELY	((__force uffd_flags_t)BIT(1))
 
 extern int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
 				    struct vm_area_struct *dst_vma,
diff --git a/include/uapi/linux/userfaultfd.h b/include/uapi/linux/userfaultfd.h
index 7d32b1e797fb..02e0c1f56939 100644
--- a/include/uapi/linux/userfaultfd.h
+++ b/include/uapi/linux/userfaultfd.h
@@ -34,7 +34,8 @@ 
 			   UFFD_FEATURE_MINOR_HUGETLBFS |	\
 			   UFFD_FEATURE_MINOR_SHMEM |		\
 			   UFFD_FEATURE_EXACT_ADDRESS |		\
-			   UFFD_FEATURE_WP_HUGETLBFS_SHMEM)
+			   UFFD_FEATURE_WP_HUGETLBFS_SHMEM |	\
+			   UFFD_FEATURE_ACCESS_HINTS)
 #define UFFD_API_IOCTLS				\
 	((__u64)1 << _UFFDIO_REGISTER |		\
 	 (__u64)1 << _UFFDIO_UNREGISTER |	\
@@ -199,6 +200,9 @@  struct uffdio_api {
 	 *
 	 * UFFD_FEATURE_WP_HUGETLBFS_SHMEM indicates that userfaultfd
 	 * write-protection mode is supported on both shmem and hugetlbfs.
+	 *
+	 * UFFD_FEATURE_ACCESS_HINTS indicates that the ioctl operations
+	 * support the UFFDIO_*_MODE_ACCESS_LIKELY hints.
 	 */
 #define UFFD_FEATURE_PAGEFAULT_FLAG_WP		(1<<0)
 #define UFFD_FEATURE_EVENT_FORK			(1<<1)
@@ -213,6 +217,7 @@  struct uffdio_api {
 #define UFFD_FEATURE_MINOR_SHMEM		(1<<10)
 #define UFFD_FEATURE_EXACT_ADDRESS		(1<<11)
 #define UFFD_FEATURE_WP_HUGETLBFS_SHMEM		(1<<12)
+#define UFFD_FEATURE_ACCESS_HINTS		(1<<13)
 	__u64 features;
 
 	__u64 ioctls;
@@ -247,8 +252,14 @@  struct uffdio_copy {
 	 * the fly.  UFFDIO_COPY_MODE_WP is available only if the
 	 * write protected ioctl is implemented for the range
 	 * according to the uffdio_register.ioctls.
+	 *
+	 * UFFDIO_COPY_MODE_ACCESS_LIKELY provides a hint to the kernel that the
+	 * page is likely to be access in the near future. Providing the hint
+	 * properly can improve performance.
+	 *
 	 */
 #define UFFDIO_COPY_MODE_WP			((__u64)1<<1)
+#define UFFDIO_COPY_MODE_ACCESS_LIKELY		((__u64)1<<2)
 	__u64 mode;
 
 	/*
@@ -261,6 +272,7 @@  struct uffdio_copy {
 struct uffdio_zeropage {
 	struct uffdio_range range;
 #define UFFDIO_ZEROPAGE_MODE_DONTWAKE		((__u64)1<<0)
+#define UFFDIO_ZEROPAGE_MODE_ACCESS_LIKELY	((__u64)1<<1)
 	__u64 mode;
 
 	/*
@@ -280,6 +292,10 @@  struct uffdio_writeprotect {
  * UFFDIO_WRITEPROTECT_MODE_DONTWAKE: set the flag to avoid waking up
  * any wait thread after the operation succeeds.
  *
+ * UFFDIO_WRITEPROTECT_MODE_ACCESS_LIKELY provides a hint to the kernel
+ * that the page is likely to be access in the near future. Providing
+ * the hint properly can improve performance.
+ *
  * NOTE: Write protecting a region (WP=1) is unrelated to page faults,
  * therefore DONTWAKE flag is meaningless with WP=1.  Removing write
  * protection (WP=0) in response to a page fault wakes the faulting
@@ -287,12 +303,14 @@  struct uffdio_writeprotect {
  */
 #define UFFDIO_WRITEPROTECT_MODE_WP		((__u64)1<<0)
 #define UFFDIO_WRITEPROTECT_MODE_DONTWAKE	((__u64)1<<1)
+#define UFFDIO_WRITEPROTECT_MODE_ACCESS_LIKELY	((__u64)1<<2)
 	__u64 mode;
 };
 
 struct uffdio_continue {
 	struct uffdio_range range;
 #define UFFDIO_CONTINUE_MODE_DONTWAKE		((__u64)1<<0)
+#define UFFDIO_CONTINUE_MODE_ACCESS_LIKELY	((__u64)1<<1)
 	__u64 mode;
 
 	/*
diff --git a/mm/internal.h b/mm/internal.h
index c0f8fbe0445b..d035b77b4f2f 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -12,6 +12,7 @@ 
 #include <linux/pagemap.h>
 #include <linux/rmap.h>
 #include <linux/tracepoint-defs.h>
+#include <linux/pgtable.h>
 
 struct folio_batch;
 
@@ -861,4 +862,16 @@  struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags);
 
 DECLARE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
 
+#ifndef arch_wants_old_prefaulted_pte
+static inline bool arch_wants_old_prefaulted_pte(void)
+{
+	/*
+	 * Transitioning a PTE from 'old' to 'young' can be expensive on
+	 * some architectures, even if it's performed in hardware. By
+	 * default, "false" means prefaulted entries will be 'young'.
+	 */
+	return false;
+}
+#endif
+
 #endif	/* __MM_INTERNAL_H */
diff --git a/mm/memory.c b/mm/memory.c
index 580c62febe42..31ec3f0071a2 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -137,18 +137,6 @@  static inline bool arch_faults_on_old_pte(void)
 }
 #endif
 
-#ifndef arch_wants_old_prefaulted_pte
-static inline bool arch_wants_old_prefaulted_pte(void)
-{
-	/*
-	 * Transitioning a PTE from 'old' to 'young' can be expensive on
-	 * some architectures, even if it's performed in hardware. By
-	 * default, "false" means prefaulted entries will be 'young'.
-	 */
-	return false;
-}
-#endif
-
 static int __init disable_randmaps(char *s)
 {
 	randomize_va_space = 0;
diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index 421784d26651..c15679f3eb6a 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -65,6 +65,7 @@  int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
 	bool writable = dst_vma->vm_flags & VM_WRITE;
 	bool vm_shared = dst_vma->vm_flags & VM_SHARED;
 	bool page_in_cache = page->mapping;
+	bool prefault = !(uffd_flags & UFFD_FLAGS_ACCESS_LIKELY);
 	spinlock_t *ptl;
 	struct inode *inode;
 	pgoff_t offset, max_off;
@@ -92,6 +93,11 @@  int mfill_atomic_install_pte(struct mm_struct *dst_mm, pmd_t *dst_pmd,
 		 */
 		_dst_pte = pte_wrprotect(_dst_pte);
 
+	if (prefault && arch_wants_old_prefaulted_pte())
+		_dst_pte = pte_mkold(_dst_pte);
+	else
+		_dst_pte = pte_sw_mkyoung(_dst_pte);
+
 	dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
 
 	if (vma_is_shmem(dst_vma)) {
@@ -202,7 +208,8 @@  static int mcopy_atomic_pte(struct mm_struct *dst_mm,
 static int mfill_zeropage_pte(struct mm_struct *dst_mm,
 			      pmd_t *dst_pmd,
 			      struct vm_area_struct *dst_vma,
-			      unsigned long dst_addr)
+			      unsigned long dst_addr,
+			      uffd_flags_t uffd_flags)
 {
 	pte_t _dst_pte, *dst_pte;
 	spinlock_t *ptl;
@@ -495,7 +502,7 @@  static __always_inline ssize_t mfill_atomic_pte(struct mm_struct *dst_mm,
 					       uffd_flags);
 		else
 			err = mfill_zeropage_pte(dst_mm, dst_pmd,
-						 dst_vma, dst_addr);
+						 dst_vma, dst_addr, uffd_flags);
 	} else {
 		err = shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
 					     dst_addr, src_addr,