[v2,2/3] mm/gup.c: Migrate device coherent pages when pinning instead of failing

Currently any attempts to pin a device coherent page will fail. This is
because device coherent pages need to be managed by a device driver, and
pinning them would prevent a driver from migrating them off the device.

However this is no reason to fail pinning of these pages. These are
coherent and accessible from the CPU so can be migrated just like
pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin
them first try migrating them out of ZONE_DEVICE.

Signed-off-by: Alistair Popple <apopple@nvidia.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
---

Changes for v2:

 - Added Felix's Acked-by
 - Fixed missing check for dpage == NULL

 mm/gup.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 95 insertions(+), 10 deletions(-)

On 07.02.22 05:26, Alistair Popple wrote:
> Currently any attempts to pin a device coherent page will fail. This is
> because device coherent pages need to be managed by a device driver, and
> pinning them would prevent a driver from migrating them off the device.
> 
> However this is no reason to fail pinning of these pages. These are
> coherent and accessible from the CPU so can be migrated just like
> pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin
> them first try migrating them out of ZONE_DEVICE.
> 
> Signed-off-by: Alistair Popple <apopple@nvidia.com>
> Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
> ---
> 
> Changes for v2:
> 
>  - Added Felix's Acked-by
>  - Fixed missing check for dpage == NULL
> 
>  mm/gup.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++------
>  1 file changed, 95 insertions(+), 10 deletions(-)
> 
> diff --git a/mm/gup.c b/mm/gup.c
> index 56d9577..5e826db 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -1861,6 +1861,60 @@ struct page *get_dump_page(unsigned long addr)
>  
>  #ifdef CONFIG_MIGRATION
>  /*
> + * Migrates a device coherent page back to normal memory. Caller should have a
> + * reference on page which will be copied to the new page if migration is
> + * successful or dropped on failure.
> + */
> +static struct page *migrate_device_page(struct page *page,
> +					unsigned int gup_flags)
> +{
> +	struct page *dpage;
> +	struct migrate_vma args;
> +	unsigned long src_pfn, dst_pfn = 0;
> +
> +	lock_page(page);
> +	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
> +	args.src = &src_pfn;
> +	args.dst = &dst_pfn;
> +	args.cpages = 1;
> +	args.npages = 1;
> +	args.vma = NULL;
> +	migrate_vma_setup(&args);
> +	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
> +		return NULL;
> +
> +	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
> +
> +	/*
> +	 * get/pin the new page now so we don't have to retry gup after
> +	 * migrating. We already have a reference so this should never fail.
> +	 */
> +	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
> +		__free_pages(dpage, 0);
> +		dpage = NULL;
> +	}
> +
> +	if (dpage) {
> +		lock_page(dpage);
> +		dst_pfn = migrate_pfn(page_to_pfn(dpage));
> +	}
> +
> +	migrate_vma_pages(&args);
> +	if (src_pfn & MIGRATE_PFN_MIGRATE)
> +		copy_highpage(dpage, page);
> +	migrate_vma_finalize(&args);
> +	if (dpage && !(src_pfn & MIGRATE_PFN_MIGRATE)) {
> +		if (gup_flags & FOLL_PIN)
> +			unpin_user_page(dpage);
> +		else
> +			put_page(dpage);
> +		dpage = NULL;
> +	}
> +
> +	return dpage;
> +}
> +
> +/*
>   * Check whether all pages are pinnable, if so return number of pages.  If some
>   * pages are not pinnable, migrate them, and unpin all pages. Return zero if
>   * pages were migrated, or if some pages were not successfully isolated.
> @@ -1888,15 +1942,40 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
>  			continue;
>  		prev_head = head;
>  		/*
> -		 * If we get a movable page, since we are going to be pinning
> -		 * these entries, try to move them out if possible.
> +		 * Device coherent pages are managed by a driver and should not
> +		 * be pinned indefinitely as it prevents the driver moving the
> +		 * page. So when trying to pin with FOLL_LONGTERM instead try
> +		 * migrating page out of device memory.
>  		 */
>  		if (is_dev_private_or_coherent_page(head)) {
> +			/*
> +			 * device private pages will get faulted in during gup
> +			 * so it shouldn't be possible to see one here.
> +			 */
>  			WARN_ON_ONCE(is_device_private_page(head));
> -			ret = -EFAULT;
> -			goto unpin_pages;
> +			WARN_ON_ONCE(PageCompound(head));
> +
> +			/*
> +			 * migration will fail if the page is pinned, so convert
> +			 * the pin on the source page to a normal reference.
> +			 */
> +			if (gup_flags & FOLL_PIN) {
> +				get_page(head);
> +				unpin_user_page(head);
> +			}
> +
> +			pages[i] = migrate_device_page(head, gup_flags);



For ordinary migrate_pages(), we'll unpin all pages and return 0 so the
caller will retry pinning by walking the page tables again.

Why can't we apply the same mechanism here? This "let's avoid another
walk" looks unnecessary complicated to me, but I might be wrong.

On Thursday, 10 February 2022 9:53:38 PM AEDT David Hildenbrand wrote:
> On 07.02.22 05:26, Alistair Popple wrote:
> > Currently any attempts to pin a device coherent page will fail. This is
> > because device coherent pages need to be managed by a device driver, and
> > pinning them would prevent a driver from migrating them off the device.
> > 
> > However this is no reason to fail pinning of these pages. These are
> > coherent and accessible from the CPU so can be migrated just like
> > pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin
> > them first try migrating them out of ZONE_DEVICE.
> > 
> > Signed-off-by: Alistair Popple <apopple@nvidia.com>
> > Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
> > ---
> > 
> > Changes for v2:
> > 
> >  - Added Felix's Acked-by
> >  - Fixed missing check for dpage == NULL
> > 
> >  mm/gup.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++------
> >  1 file changed, 95 insertions(+), 10 deletions(-)
> > 
> > diff --git a/mm/gup.c b/mm/gup.c
> > index 56d9577..5e826db 100644
> > --- a/mm/gup.c
> > +++ b/mm/gup.c
> > @@ -1861,6 +1861,60 @@ struct page *get_dump_page(unsigned long addr)
> >  
> >  #ifdef CONFIG_MIGRATION
> >  /*
> > + * Migrates a device coherent page back to normal memory. Caller should have a
> > + * reference on page which will be copied to the new page if migration is
> > + * successful or dropped on failure.
> > + */
> > +static struct page *migrate_device_page(struct page *page,
> > +					unsigned int gup_flags)
> > +{
> > +	struct page *dpage;
> > +	struct migrate_vma args;
> > +	unsigned long src_pfn, dst_pfn = 0;
> > +
> > +	lock_page(page);
> > +	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
> > +	args.src = &src_pfn;
> > +	args.dst = &dst_pfn;
> > +	args.cpages = 1;
> > +	args.npages = 1;
> > +	args.vma = NULL;
> > +	migrate_vma_setup(&args);
> > +	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
> > +		return NULL;
> > +
> > +	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
> > +
> > +	/*
> > +	 * get/pin the new page now so we don't have to retry gup after
> > +	 * migrating. We already have a reference so this should never fail.
> > +	 */
> > +	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
> > +		__free_pages(dpage, 0);
> > +		dpage = NULL;
> > +	}
> > +
> > +	if (dpage) {
> > +		lock_page(dpage);
> > +		dst_pfn = migrate_pfn(page_to_pfn(dpage));
> > +	}
> > +
> > +	migrate_vma_pages(&args);
> > +	if (src_pfn & MIGRATE_PFN_MIGRATE)
> > +		copy_highpage(dpage, page);
> > +	migrate_vma_finalize(&args);
> > +	if (dpage && !(src_pfn & MIGRATE_PFN_MIGRATE)) {
> > +		if (gup_flags & FOLL_PIN)
> > +			unpin_user_page(dpage);
> > +		else
> > +			put_page(dpage);
> > +		dpage = NULL;
> > +	}
> > +
> > +	return dpage;
> > +}
> > +
> > +/*
> >   * Check whether all pages are pinnable, if so return number of pages.  If some
> >   * pages are not pinnable, migrate them, and unpin all pages. Return zero if
> >   * pages were migrated, or if some pages were not successfully isolated.
> > @@ -1888,15 +1942,40 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
> >  			continue;
> >  		prev_head = head;
> >  		/*
> > -		 * If we get a movable page, since we are going to be pinning
> > -		 * these entries, try to move them out if possible.
> > +		 * Device coherent pages are managed by a driver and should not
> > +		 * be pinned indefinitely as it prevents the driver moving the
> > +		 * page. So when trying to pin with FOLL_LONGTERM instead try
> > +		 * migrating page out of device memory.
> >  		 */
> >  		if (is_dev_private_or_coherent_page(head)) {
> > +			/*
> > +			 * device private pages will get faulted in during gup
> > +			 * so it shouldn't be possible to see one here.
> > +			 */
> >  			WARN_ON_ONCE(is_device_private_page(head));
> > -			ret = -EFAULT;
> > -			goto unpin_pages;
> > +			WARN_ON_ONCE(PageCompound(head));
> > +
> > +			/*
> > +			 * migration will fail if the page is pinned, so convert
> > +			 * the pin on the source page to a normal reference.
> > +			 */
> > +			if (gup_flags & FOLL_PIN) {
> > +				get_page(head);
> > +				unpin_user_page(head);
> > +			}
> > +
> > +			pages[i] = migrate_device_page(head, gup_flags);
> 
> For ordinary migrate_pages(), we'll unpin all pages and return 0 so the
> caller will retry pinning by walking the page tables again.
> 
> Why can't we apply the same mechanism here? This "let's avoid another
> walk" looks unnecessary complicated to me, but I might be wrong.

There's no reason we couldn't. I figured we have the page in the right spot
anyway so it was easy to do, and looking at this rebased on top of Christoph's
ZONE_DEVICE refcount simplification I'm not sure it would be any simpler
anyway.

It would remove the call to try_grab_page(), but we'd still have to return an
error on migration failures whilst also ensuring we putback any non-device
pages that may have been isolated. I might have overlooked something though,
so certainly happy for suggestions.

On 10.02.22 12:39, Alistair Popple wrote:
> On Thursday, 10 February 2022 9:53:38 PM AEDT David Hildenbrand wrote:
>> On 07.02.22 05:26, Alistair Popple wrote:
>>> Currently any attempts to pin a device coherent page will fail. This is
>>> because device coherent pages need to be managed by a device driver, and
>>> pinning them would prevent a driver from migrating them off the device.
>>>
>>> However this is no reason to fail pinning of these pages. These are
>>> coherent and accessible from the CPU so can be migrated just like
>>> pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin
>>> them first try migrating them out of ZONE_DEVICE.
>>>
>>> Signed-off-by: Alistair Popple <apopple@nvidia.com>
>>> Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
>>> ---
>>>
>>> Changes for v2:
>>>
>>>  - Added Felix's Acked-by
>>>  - Fixed missing check for dpage == NULL
>>>
>>>  mm/gup.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++------
>>>  1 file changed, 95 insertions(+), 10 deletions(-)
>>>
>>> diff --git a/mm/gup.c b/mm/gup.c
>>> index 56d9577..5e826db 100644
>>> --- a/mm/gup.c
>>> +++ b/mm/gup.c
>>> @@ -1861,6 +1861,60 @@ struct page *get_dump_page(unsigned long addr)
>>>  
>>>  #ifdef CONFIG_MIGRATION
>>>  /*
>>> + * Migrates a device coherent page back to normal memory. Caller should have a
>>> + * reference on page which will be copied to the new page if migration is
>>> + * successful or dropped on failure.
>>> + */
>>> +static struct page *migrate_device_page(struct page *page,
>>> +					unsigned int gup_flags)
>>> +{
>>> +	struct page *dpage;
>>> +	struct migrate_vma args;
>>> +	unsigned long src_pfn, dst_pfn = 0;
>>> +
>>> +	lock_page(page);
>>> +	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
>>> +	args.src = &src_pfn;
>>> +	args.dst = &dst_pfn;
>>> +	args.cpages = 1;
>>> +	args.npages = 1;
>>> +	args.vma = NULL;
>>> +	migrate_vma_setup(&args);
>>> +	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
>>> +		return NULL;
>>> +
>>> +	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
>>> +
>>> +	/*
>>> +	 * get/pin the new page now so we don't have to retry gup after
>>> +	 * migrating. We already have a reference so this should never fail.
>>> +	 */
>>> +	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
>>> +		__free_pages(dpage, 0);
>>> +		dpage = NULL;
>>> +	}
>>> +
>>> +	if (dpage) {
>>> +		lock_page(dpage);
>>> +		dst_pfn = migrate_pfn(page_to_pfn(dpage));
>>> +	}
>>> +
>>> +	migrate_vma_pages(&args);
>>> +	if (src_pfn & MIGRATE_PFN_MIGRATE)
>>> +		copy_highpage(dpage, page);
>>> +	migrate_vma_finalize(&args);
>>> +	if (dpage && !(src_pfn & MIGRATE_PFN_MIGRATE)) {
>>> +		if (gup_flags & FOLL_PIN)
>>> +			unpin_user_page(dpage);
>>> +		else
>>> +			put_page(dpage);
>>> +		dpage = NULL;
>>> +	}
>>> +
>>> +	return dpage;
>>> +}
>>> +
>>> +/*
>>>   * Check whether all pages are pinnable, if so return number of pages.  If some
>>>   * pages are not pinnable, migrate them, and unpin all pages. Return zero if
>>>   * pages were migrated, or if some pages were not successfully isolated.
>>> @@ -1888,15 +1942,40 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
>>>  			continue;
>>>  		prev_head = head;
>>>  		/*
>>> -		 * If we get a movable page, since we are going to be pinning
>>> -		 * these entries, try to move them out if possible.
>>> +		 * Device coherent pages are managed by a driver and should not
>>> +		 * be pinned indefinitely as it prevents the driver moving the
>>> +		 * page. So when trying to pin with FOLL_LONGTERM instead try
>>> +		 * migrating page out of device memory.
>>>  		 */
>>>  		if (is_dev_private_or_coherent_page(head)) {
>>> +			/*
>>> +			 * device private pages will get faulted in during gup
>>> +			 * so it shouldn't be possible to see one here.
>>> +			 */
>>>  			WARN_ON_ONCE(is_device_private_page(head));
>>> -			ret = -EFAULT;
>>> -			goto unpin_pages;
>>> +			WARN_ON_ONCE(PageCompound(head));
>>> +
>>> +			/*
>>> +			 * migration will fail if the page is pinned, so convert
>>> +			 * the pin on the source page to a normal reference.
>>> +			 */
>>> +			if (gup_flags & FOLL_PIN) {
>>> +				get_page(head);
>>> +				unpin_user_page(head);
>>> +			}
>>> +
>>> +			pages[i] = migrate_device_page(head, gup_flags);
>>
>> For ordinary migrate_pages(), we'll unpin all pages and return 0 so the
>> caller will retry pinning by walking the page tables again.
>>
>> Why can't we apply the same mechanism here? This "let's avoid another
>> walk" looks unnecessary complicated to me, but I might be wrong.
> 
> There's no reason we couldn't. I figured we have the page in the right spot
> anyway so it was easy to do, and looking at this rebased on top of Christoph's
> ZONE_DEVICE refcount simplification I'm not sure it would be any simpler
> anyway.
> 
> It would remove the call to try_grab_page(), but we'd still have to return an
> error on migration failures whilst also ensuring we putback any non-device
> pages that may have been isolated. I might have overlooked something though,
> so certainly happy for suggestions.

Staring at the code, I was wondering if we could either

* build a second list of device coherent pages to migrate and call a
  migrate_device_pages() bulk function
* simply use movable_page_list() and teach migrate_pages() how to handle
  them.

I'd really appreciate as little special casing as possible for the ever
growing list of new DEVICE types all over the place. E.g., just staring
at fork even before the new device coherent made my head spin.

On Thursday, 10 February 2022 10:47:35 PM AEDT David Hildenbrand wrote:
> On 10.02.22 12:39, Alistair Popple wrote:
> > On Thursday, 10 February 2022 9:53:38 PM AEDT David Hildenbrand wrote:
> >> On 07.02.22 05:26, Alistair Popple wrote:
> >>> Currently any attempts to pin a device coherent page will fail. This is
> >>> because device coherent pages need to be managed by a device driver, and
> >>> pinning them would prevent a driver from migrating them off the device.
> >>>
> >>> However this is no reason to fail pinning of these pages. These are
> >>> coherent and accessible from the CPU so can be migrated just like
> >>> pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin
> >>> them first try migrating them out of ZONE_DEVICE.
> >>>
> >>> Signed-off-by: Alistair Popple <apopple@nvidia.com>
> >>> Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
> >>> ---
> >>>
> >>> Changes for v2:
> >>>
> >>>  - Added Felix's Acked-by
> >>>  - Fixed missing check for dpage == NULL
> >>>
> >>>  mm/gup.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++------
> >>>  1 file changed, 95 insertions(+), 10 deletions(-)
> >>>
> >>> diff --git a/mm/gup.c b/mm/gup.c
> >>> index 56d9577..5e826db 100644
> >>> --- a/mm/gup.c
> >>> +++ b/mm/gup.c
> >>> @@ -1861,6 +1861,60 @@ struct page *get_dump_page(unsigned long addr)
> >>>  
> >>>  #ifdef CONFIG_MIGRATION
> >>>  /*
> >>> + * Migrates a device coherent page back to normal memory. Caller should have a
> >>> + * reference on page which will be copied to the new page if migration is
> >>> + * successful or dropped on failure.
> >>> + */
> >>> +static struct page *migrate_device_page(struct page *page,
> >>> +					unsigned int gup_flags)
> >>> +{
> >>> +	struct page *dpage;
> >>> +	struct migrate_vma args;
> >>> +	unsigned long src_pfn, dst_pfn = 0;
> >>> +
> >>> +	lock_page(page);
> >>> +	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
> >>> +	args.src = &src_pfn;
> >>> +	args.dst = &dst_pfn;
> >>> +	args.cpages = 1;
> >>> +	args.npages = 1;
> >>> +	args.vma = NULL;
> >>> +	migrate_vma_setup(&args);
> >>> +	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
> >>> +		return NULL;
> >>> +
> >>> +	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
> >>> +
> >>> +	/*
> >>> +	 * get/pin the new page now so we don't have to retry gup after
> >>> +	 * migrating. We already have a reference so this should never fail.
> >>> +	 */
> >>> +	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
> >>> +		__free_pages(dpage, 0);
> >>> +		dpage = NULL;
> >>> +	}
> >>> +
> >>> +	if (dpage) {
> >>> +		lock_page(dpage);
> >>> +		dst_pfn = migrate_pfn(page_to_pfn(dpage));
> >>> +	}
> >>> +
> >>> +	migrate_vma_pages(&args);
> >>> +	if (src_pfn & MIGRATE_PFN_MIGRATE)
> >>> +		copy_highpage(dpage, page);
> >>> +	migrate_vma_finalize(&args);
> >>> +	if (dpage && !(src_pfn & MIGRATE_PFN_MIGRATE)) {
> >>> +		if (gup_flags & FOLL_PIN)
> >>> +			unpin_user_page(dpage);
> >>> +		else
> >>> +			put_page(dpage);
> >>> +		dpage = NULL;
> >>> +	}
> >>> +
> >>> +	return dpage;
> >>> +}
> >>> +
> >>> +/*
> >>>   * Check whether all pages are pinnable, if so return number of pages.  If some
> >>>   * pages are not pinnable, migrate them, and unpin all pages. Return zero if
> >>>   * pages were migrated, or if some pages were not successfully isolated.
> >>> @@ -1888,15 +1942,40 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
> >>>  			continue;
> >>>  		prev_head = head;
> >>>  		/*
> >>> -		 * If we get a movable page, since we are going to be pinning
> >>> -		 * these entries, try to move them out if possible.
> >>> +		 * Device coherent pages are managed by a driver and should not
> >>> +		 * be pinned indefinitely as it prevents the driver moving the
> >>> +		 * page. So when trying to pin with FOLL_LONGTERM instead try
> >>> +		 * migrating page out of device memory.
> >>>  		 */
> >>>  		if (is_dev_private_or_coherent_page(head)) {
> >>> +			/*
> >>> +			 * device private pages will get faulted in during gup
> >>> +			 * so it shouldn't be possible to see one here.
> >>> +			 */
> >>>  			WARN_ON_ONCE(is_device_private_page(head));
> >>> -			ret = -EFAULT;
> >>> -			goto unpin_pages;
> >>> +			WARN_ON_ONCE(PageCompound(head));
> >>> +
> >>> +			/*
> >>> +			 * migration will fail if the page is pinned, so convert
> >>> +			 * the pin on the source page to a normal reference.
> >>> +			 */
> >>> +			if (gup_flags & FOLL_PIN) {
> >>> +				get_page(head);
> >>> +				unpin_user_page(head);
> >>> +			}
> >>> +
> >>> +			pages[i] = migrate_device_page(head, gup_flags);
> >>
> >> For ordinary migrate_pages(), we'll unpin all pages and return 0 so the
> >> caller will retry pinning by walking the page tables again.
> >>
> >> Why can't we apply the same mechanism here? This "let's avoid another
> >> walk" looks unnecessary complicated to me, but I might be wrong.
> > 
> > There's no reason we couldn't. I figured we have the page in the right spot
> > anyway so it was easy to do, and looking at this rebased on top of Christoph's
> > ZONE_DEVICE refcount simplification I'm not sure it would be any simpler
> > anyway.
> > 
> > It would remove the call to try_grab_page(), but we'd still have to return an
> > error on migration failures whilst also ensuring we putback any non-device
> > pages that may have been isolated. I might have overlooked something though,
> > so certainly happy for suggestions.
> 
> Staring at the code, I was wondering if we could either
> 
> * build a second list of device coherent pages to migrate and call a
>   migrate_device_pages() bulk function
> * simply use movable_page_list() and teach migrate_pages() how to handle
>   them.

I did consider that approach. The problem is zone device pages are not LRU
pages. In particular page->lru is not available to add the page to a list, and
as an external API and internally migrate_pages() relies heavily on moving
pages between lists.

> I'd really appreciate as little special casing as possible for the ever
> growing list of new DEVICE types all over the place. E.g., just staring
> at fork even before the new device coherent made my head spin.

That's fair. We could pull the checks for device pages out into a self
contained function (eg. check_and_migrate_device_pages()) called before
check_and_migrate_movable_pages(). The down side of that is we'd always have an
extra loop over all the pages just to scan for device pages, but perhaps that's
not a concern?

On 2/6/22 20:26, Alistair Popple wrote:
> Currently any attempts to pin a device coherent page will fail. This is
> because device coherent pages need to be managed by a device driver, and
> pinning them would prevent a driver from migrating them off the device.
> 
> However this is no reason to fail pinning of these pages. These are
> coherent and accessible from the CPU so can be migrated just like
> pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin
> them first try migrating them out of ZONE_DEVICE.
> 

Hi Alistair and all,

Here's a possible issue (below) that I really should have spotted the
first time around, sorry for this late-breaking review. And maybe it's
actually just my misunderstanding, anyway.


> Signed-off-by: Alistair Popple <apopple@nvidia.com>
> Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
> ---
> 
> Changes for v2:
> 
>   - Added Felix's Acked-by
>   - Fixed missing check for dpage == NULL
> 
>   mm/gup.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++------
>   1 file changed, 95 insertions(+), 10 deletions(-)
> 
> diff --git a/mm/gup.c b/mm/gup.c
> index 56d9577..5e826db 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -1861,6 +1861,60 @@ struct page *get_dump_page(unsigned long addr)
>   
>   #ifdef CONFIG_MIGRATION
>   /*
> + * Migrates a device coherent page back to normal memory. Caller should have a
> + * reference on page which will be copied to the new page if migration is
> + * successful or dropped on failure.
> + */
> +static struct page *migrate_device_page(struct page *page,
> +					unsigned int gup_flags)
> +{
> +	struct page *dpage;
> +	struct migrate_vma args;
> +	unsigned long src_pfn, dst_pfn = 0;
> +
> +	lock_page(page);
> +	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
> +	args.src = &src_pfn;
> +	args.dst = &dst_pfn;
> +	args.cpages = 1;
> +	args.npages = 1;
> +	args.vma = NULL;
> +	migrate_vma_setup(&args);
> +	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
> +		return NULL;
> +
> +	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
> +
> +	/*
> +	 * get/pin the new page now so we don't have to retry gup after
> +	 * migrating. We already have a reference so this should never fail.
> +	 */
> +	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
> +		__free_pages(dpage, 0);
> +		dpage = NULL;
> +	}
> +
> +	if (dpage) {
> +		lock_page(dpage);
> +		dst_pfn = migrate_pfn(page_to_pfn(dpage));
> +	}
> +
> +	migrate_vma_pages(&args);
> +	if (src_pfn & MIGRATE_PFN_MIGRATE)
> +		copy_highpage(dpage, page);
> +	migrate_vma_finalize(&args);
> +	if (dpage && !(src_pfn & MIGRATE_PFN_MIGRATE)) {
> +		if (gup_flags & FOLL_PIN)
> +			unpin_user_page(dpage);
> +		else
> +			put_page(dpage);
> +		dpage = NULL;
> +	}
> +
> +	return dpage;
> +}
> +
> +/*
>    * Check whether all pages are pinnable, if so return number of pages.  If some
>    * pages are not pinnable, migrate them, and unpin all pages. Return zero if
>    * pages were migrated, or if some pages were not successfully isolated.
> @@ -1888,15 +1942,40 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
>   			continue;
>   		prev_head = head;
>   		/*
> -		 * If we get a movable page, since we are going to be pinning
> -		 * these entries, try to move them out if possible.
> +		 * Device coherent pages are managed by a driver and should not
> +		 * be pinned indefinitely as it prevents the driver moving the
> +		 * page. So when trying to pin with FOLL_LONGTERM instead try
> +		 * migrating page out of device memory.
>   		 */
>   		if (is_dev_private_or_coherent_page(head)) {
> +			/*
> +			 * device private pages will get faulted in during gup
> +			 * so it shouldn't be possible to see one here.
> +			 */
>   			WARN_ON_ONCE(is_device_private_page(head));
> -			ret = -EFAULT;
> -			goto unpin_pages;
> +			WARN_ON_ONCE(PageCompound(head));
> +
> +			/*
> +			 * migration will fail if the page is pinned, so convert
> +			 * the pin on the source page to a normal reference.
> +			 */
> +			if (gup_flags & FOLL_PIN) {
> +				get_page(head);
> +				unpin_user_page(head);

OK...but now gup_flags can no longer be used as a guide for how to
release these pages, right? In other words, up until this point,
FOLL_PIN meant "call unpin_user_page() in order to release". However,
now this page must be released via put_page().

See below...

> +			}
> +
> +			pages[i] = migrate_device_page(head, gup_flags);
> +			if (!pages[i]) {
> +				ret = -EBUSY;
> +				break;
> +			}
> +			continue;
>   		}
>   
> +		/*
> +		 * If we get a movable page, since we are going to be pinning
> +		 * these entries, try to move them out if possible.
> +		 */
>   		if (!is_pinnable_page(head)) {
>   			if (PageHuge(head)) {
>   				if (!isolate_huge_page(head, &movable_page_list))
> @@ -1924,16 +2003,22 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
>   	 * If list is empty, and no isolation errors, means that all pages are
>   	 * in the correct zone.
>   	 */
> -	if (list_empty(&movable_page_list) && !isolation_error_count)
> +	if (!ret && list_empty(&movable_page_list) && !isolation_error_count)
>   		return nr_pages;
>   
> -unpin_pages:
> -	if (gup_flags & FOLL_PIN) {
> -		unpin_user_pages(pages, nr_pages);
> -	} else {
> -		for (i = 0; i < nr_pages; i++)
> +	for (i = 0; i < nr_pages; i++)
> +		if (!pages[i])
> +			continue;
> +		else if (gup_flags & FOLL_PIN)
> +			unpin_user_page(pages[i]);

...and here, for example, we are still trusting gup_flags to decide how
to release the pages.

This reminds me: out of the many things to monitor, the FOLL_PIN counts
in /proc/vmstat are especially helpful, whenever making changes to code
that deals with this:

	nr_foll_pin_acquired
	nr_foll_pin_released

...and those should normally be equal to each other when "at rest".


thanks,

On Saturday, 12 February 2022 1:10:29 PM AEDT John Hubbard wrote:
> On 2/6/22 20:26, Alistair Popple wrote:
> > Currently any attempts to pin a device coherent page will fail. This is
> > because device coherent pages need to be managed by a device driver, and
> > pinning them would prevent a driver from migrating them off the device.
> > 
> > However this is no reason to fail pinning of these pages. These are
> > coherent and accessible from the CPU so can be migrated just like
> > pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin
> > them first try migrating them out of ZONE_DEVICE.
> > 
> 
> Hi Alistair and all,
> 
> Here's a possible issue (below) that I really should have spotted the
> first time around, sorry for this late-breaking review. And maybe it's
> actually just my misunderstanding, anyway.

I think it might be a misunderstanding, see below.

> > Signed-off-by: Alistair Popple <apopple@nvidia.com>
> > Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
> > ---
> > 
> > Changes for v2:
> > 
> >   - Added Felix's Acked-by
> >   - Fixed missing check for dpage == NULL
> > 
> >   mm/gup.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++------
> >   1 file changed, 95 insertions(+), 10 deletions(-)
> > 
> > diff --git a/mm/gup.c b/mm/gup.c
> > index 56d9577..5e826db 100644
> > --- a/mm/gup.c
> > +++ b/mm/gup.c
> > @@ -1861,6 +1861,60 @@ struct page *get_dump_page(unsigned long addr)
> >   
> >   #ifdef CONFIG_MIGRATION
> >   /*
> > + * Migrates a device coherent page back to normal memory. Caller should have a
> > + * reference on page which will be copied to the new page if migration is
> > + * successful or dropped on failure.
> > + */
> > +static struct page *migrate_device_page(struct page *page,
> > +					unsigned int gup_flags)
> > +{
> > +	struct page *dpage;
> > +	struct migrate_vma args;
> > +	unsigned long src_pfn, dst_pfn = 0;
> > +
> > +	lock_page(page);
> > +	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
> > +	args.src = &src_pfn;
> > +	args.dst = &dst_pfn;
> > +	args.cpages = 1;
> > +	args.npages = 1;
> > +	args.vma = NULL;
> > +	migrate_vma_setup(&args);
> > +	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
> > +		return NULL;
> > +
> > +	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
> > +
> > +	/*
> > +	 * get/pin the new page now so we don't have to retry gup after
> > +	 * migrating. We already have a reference so this should never fail.
> > +	 */
> > +	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
> > +		__free_pages(dpage, 0);
> > +		dpage = NULL;
> > +	}
> > +
> > +	if (dpage) {
> > +		lock_page(dpage);
> > +		dst_pfn = migrate_pfn(page_to_pfn(dpage));
> > +	}
> > +
> > +	migrate_vma_pages(&args);
> > +	if (src_pfn & MIGRATE_PFN_MIGRATE)
> > +		copy_highpage(dpage, page);
> > +	migrate_vma_finalize(&args);
> > +	if (dpage && !(src_pfn & MIGRATE_PFN_MIGRATE)) {
> > +		if (gup_flags & FOLL_PIN)
> > +			unpin_user_page(dpage);
> > +		else
> > +			put_page(dpage);
> > +		dpage = NULL;
> > +	}
> > +
> > +	return dpage;
> > +}
> > +
> > +/*
> >    * Check whether all pages are pinnable, if so return number of pages.  If some
> >    * pages are not pinnable, migrate them, and unpin all pages. Return zero if
> >    * pages were migrated, or if some pages were not successfully isolated.
> > @@ -1888,15 +1942,40 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
> >   			continue;
> >   		prev_head = head;
> >   		/*
> > -		 * If we get a movable page, since we are going to be pinning
> > -		 * these entries, try to move them out if possible.
> > +		 * Device coherent pages are managed by a driver and should not
> > +		 * be pinned indefinitely as it prevents the driver moving the
> > +		 * page. So when trying to pin with FOLL_LONGTERM instead try
> > +		 * migrating page out of device memory.
> >   		 */
> >   		if (is_dev_private_or_coherent_page(head)) {
> > +			/*
> > +			 * device private pages will get faulted in during gup
> > +			 * so it shouldn't be possible to see one here.
> > +			 */
> >   			WARN_ON_ONCE(is_device_private_page(head));
> > -			ret = -EFAULT;
> > -			goto unpin_pages;
> > +			WARN_ON_ONCE(PageCompound(head));
> > +
> > +			/*
> > +			 * migration will fail if the page is pinned, so convert
> > +			 * the pin on the source page to a normal reference.
> > +			 */
> > +			if (gup_flags & FOLL_PIN) {
> > +				get_page(head);
> > +				unpin_user_page(head);
> 
> OK...but now gup_flags can no longer be used as a guide for how to
> release these pages, right? In other words, up until this point,
> FOLL_PIN meant "call unpin_user_page() in order to release". However,
> now this page must be released via put_page().

This is the source page (head). We are unpinning it because we can't migrate a
pinned page, however we still need a reference on it for migrate_vma hence the
get_page followed by unpin. In the non-FOLL_PIN case we already have a
reference from gup.

> See below...
> 
> > +			}
> > +
> > +			pages[i] = migrate_device_page(head, gup_flags);

migrate_device_page() will return a new page that has been correctly pinned
with gup_flags by try_grab_page(). Therefore this page can still be released
with unpin_user_page() or put_page() as appropriate for the given gup_flags.

The reference we had on the source page (head) always gets dropped in
migrate_vma_finalize().

> > +			if (!pages[i]) {
> > +				ret = -EBUSY;
> > +				break;
> > +			}
> > +			continue;
> >   		}
> >   
> > +		/*
> > +		 * If we get a movable page, since we are going to be pinning
> > +		 * these entries, try to move them out if possible.
> > +		 */
> >   		if (!is_pinnable_page(head)) {
> >   			if (PageHuge(head)) {
> >   				if (!isolate_huge_page(head, &movable_page_list))
> > @@ -1924,16 +2003,22 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
> >   	 * If list is empty, and no isolation errors, means that all pages are
> >   	 * in the correct zone.
> >   	 */
> > -	if (list_empty(&movable_page_list) && !isolation_error_count)
> > +	if (!ret && list_empty(&movable_page_list) && !isolation_error_count)
> >   		return nr_pages;
> >   
> > -unpin_pages:
> > -	if (gup_flags & FOLL_PIN) {
> > -		unpin_user_pages(pages, nr_pages);
> > -	} else {
> > -		for (i = 0; i < nr_pages; i++)
> > +	for (i = 0; i < nr_pages; i++)
> > +		if (!pages[i])
> > +			continue;
> > +		else if (gup_flags & FOLL_PIN)
> > +			unpin_user_page(pages[i]);
> 
> ...and here, for example, we are still trusting gup_flags to decide how
> to release the pages.

Which unless I've missed something is still the correct thing to do.

> This reminds me: out of the many things to monitor, the FOLL_PIN counts
> in /proc/vmstat are especially helpful, whenever making changes to code
> that deals with this:
> 
> 	nr_foll_pin_acquired
> 	nr_foll_pin_released
> 
> ...and those should normally be equal to each other when "at rest".
> 
> 
> thanks,
> > +		else
> >   			put_page(pages[i]);
> > +
> > +	if (ret && !list_empty(&movable_page_list)) {
> > +		putback_movable_pages(&movable_page_list);
> > +		return ret;
> >   	}
> > +
> >   	if (!list_empty(&movable_page_list)) {
> >   		ret = migrate_pages(&movable_page_list, alloc_migration_target,
> >   				    NULL, (unsigned long)&mtc, MIGRATE_SYNC,
> 
>

On 2/11/22 18:51, Alistair Popple wrote:
...
>>> @@ -1888,15 +1942,40 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
>>>    			continue;
>>>    		prev_head = head;
>>>    		/*
>>> -		 * If we get a movable page, since we are going to be pinning
>>> -		 * these entries, try to move them out if possible.
>>> +		 * Device coherent pages are managed by a driver and should not
>>> +		 * be pinned indefinitely as it prevents the driver moving the
>>> +		 * page. So when trying to pin with FOLL_LONGTERM instead try
>>> +		 * migrating page out of device memory.
>>>    		 */
>>>    		if (is_dev_private_or_coherent_page(head)) {
>>> +			/*
>>> +			 * device private pages will get faulted in during gup
>>> +			 * so it shouldn't be possible to see one here.
>>> +			 */
>>>    			WARN_ON_ONCE(is_device_private_page(head));
>>> -			ret = -EFAULT;
>>> -			goto unpin_pages;
>>> +			WARN_ON_ONCE(PageCompound(head));
>>> +
>>> +			/*
>>> +			 * migration will fail if the page is pinned, so convert
>>> +			 * the pin on the source page to a normal reference.
>>> +			 */
>>> +			if (gup_flags & FOLL_PIN) {
>>> +				get_page(head);
>>> +				unpin_user_page(head);
>>
>> OK...but now gup_flags can no longer be used as a guide for how to
>> release these pages, right? In other words, up until this point,
>> FOLL_PIN meant "call unpin_user_page() in order to release". However,
>> now this page must be released via put_page().
> 
> This is the source page (head). We are unpinning it because we can't migrate a
> pinned page, however we still need a reference on it for migrate_vma hence the
> get_page followed by unpin. In the non-FOLL_PIN case we already have a
> reference from gup.
> 
>> See below...
>>
>>> +			}
>>> +
>>> +			pages[i] = migrate_device_page(head, gup_flags);
> 
> migrate_device_page() will return a new page that has been correctly pinned
> with gup_flags by try_grab_page(). Therefore this page can still be released
> with unpin_user_page() or put_page() as appropriate for the given gup_flags.
> 
> The reference we had on the source page (head) always gets dropped in
> migrate_vma_finalize().

OK. Good.

The above would be good to have in a comment, right around here, imho.
Because we have this marvelous mix of references for migration (get_page())
and other, and it's a bit hard to see that it's all correct without a
hint or two.

...
> 
> Which unless I've missed something is still the correct thing to do.
> 
>> This reminds me: out of the many things to monitor, the FOLL_PIN counts
>> in /proc/vmstat are especially helpful, whenever making changes to code
>> that deals with this:
>>
>> 	nr_foll_pin_acquired
>> 	nr_foll_pin_released
>>
>> ...and those should normally be equal to each other when "at rest".
>>

I hope this is/was run, just to be sure?


thanks,

John Hubbard <jhubbard@nvidia.com> writes:

> On 2/11/22 18:51, Alistair Popple wrote:

[…]

>>> See below…
>>>
>>>> +			}
>>>> +
>>>> +			pages[i] = migrate_device_page(head, gup_flags);
>> migrate_device_page() will return a new page that has been correctly pinned
>> with gup_flags by try_grab_page(). Therefore this page can still be released
>> with unpin_user_page() or put_page() as appropriate for the given gup_flags.
>> The reference we had on the source page (head) always gets dropped in
>> migrate_vma_finalize().
>
> OK. Good.
>
> The above would be good to have in a comment, right around here, imho.
> Because we have this marvelous mix of references for migration (get_page())
> and other, and it’s a bit hard to see that it’s all correct without a
> hint or two.

Ok, will do.

>
> …
>> Which unless I’ve missed something is still the correct thing to do.
>>
>>> This reminds me: out of the many things to monitor, the FOLL_PIN counts
>>> in /proc/vmstat are especially helpful, whenever making changes to code
>>> that deals with this:
>>>
>>> 	nr_foll_pin_acquired
>>> 	nr_foll_pin_released
>>>
>>> …and those should normally be equal to each other when “at rest”.
>>>
>
> I hope this is/was run, just to be sure?

Thanks for the suggestion, these remain equal to each other after running
hmm-tests which confirms everything is working as expected.

>
> thanks,

On 11.02.22 00:41, Alistair Popple wrote:
> On Thursday, 10 February 2022 10:47:35 PM AEDT David Hildenbrand wrote:
>> On 10.02.22 12:39, Alistair Popple wrote:
>>> On Thursday, 10 February 2022 9:53:38 PM AEDT David Hildenbrand wrote:
>>>> On 07.02.22 05:26, Alistair Popple wrote:
>>>>> Currently any attempts to pin a device coherent page will fail. This is
>>>>> because device coherent pages need to be managed by a device driver, and
>>>>> pinning them would prevent a driver from migrating them off the device.
>>>>>
>>>>> However this is no reason to fail pinning of these pages. These are
>>>>> coherent and accessible from the CPU so can be migrated just like
>>>>> pinning ZONE_MOVABLE pages. So instead of failing all attempts to pin
>>>>> them first try migrating them out of ZONE_DEVICE.
>>>>>
>>>>> Signed-off-by: Alistair Popple <apopple@nvidia.com>
>>>>> Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
>>>>> ---
>>>>>
>>>>> Changes for v2:
>>>>>
>>>>>  - Added Felix's Acked-by
>>>>>  - Fixed missing check for dpage == NULL
>>>>>
>>>>>  mm/gup.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++------
>>>>>  1 file changed, 95 insertions(+), 10 deletions(-)
>>>>>
>>>>> diff --git a/mm/gup.c b/mm/gup.c
>>>>> index 56d9577..5e826db 100644
>>>>> --- a/mm/gup.c
>>>>> +++ b/mm/gup.c
>>>>> @@ -1861,6 +1861,60 @@ struct page *get_dump_page(unsigned long addr)
>>>>>  
>>>>>  #ifdef CONFIG_MIGRATION
>>>>>  /*
>>>>> + * Migrates a device coherent page back to normal memory. Caller should have a
>>>>> + * reference on page which will be copied to the new page if migration is
>>>>> + * successful or dropped on failure.
>>>>> + */
>>>>> +static struct page *migrate_device_page(struct page *page,
>>>>> +					unsigned int gup_flags)
>>>>> +{
>>>>> +	struct page *dpage;
>>>>> +	struct migrate_vma args;
>>>>> +	unsigned long src_pfn, dst_pfn = 0;
>>>>> +
>>>>> +	lock_page(page);
>>>>> +	src_pfn = migrate_pfn(page_to_pfn(page)) | MIGRATE_PFN_MIGRATE;
>>>>> +	args.src = &src_pfn;
>>>>> +	args.dst = &dst_pfn;
>>>>> +	args.cpages = 1;
>>>>> +	args.npages = 1;
>>>>> +	args.vma = NULL;
>>>>> +	migrate_vma_setup(&args);
>>>>> +	if (!(src_pfn & MIGRATE_PFN_MIGRATE))
>>>>> +		return NULL;
>>>>> +
>>>>> +	dpage = alloc_pages(GFP_USER | __GFP_NOWARN, 0);
>>>>> +
>>>>> +	/*
>>>>> +	 * get/pin the new page now so we don't have to retry gup after
>>>>> +	 * migrating. We already have a reference so this should never fail.
>>>>> +	 */
>>>>> +	if (dpage && WARN_ON_ONCE(!try_grab_page(dpage, gup_flags))) {
>>>>> +		__free_pages(dpage, 0);
>>>>> +		dpage = NULL;
>>>>> +	}
>>>>> +
>>>>> +	if (dpage) {
>>>>> +		lock_page(dpage);
>>>>> +		dst_pfn = migrate_pfn(page_to_pfn(dpage));
>>>>> +	}
>>>>> +
>>>>> +	migrate_vma_pages(&args);
>>>>> +	if (src_pfn & MIGRATE_PFN_MIGRATE)
>>>>> +		copy_highpage(dpage, page);
>>>>> +	migrate_vma_finalize(&args);
>>>>> +	if (dpage && !(src_pfn & MIGRATE_PFN_MIGRATE)) {
>>>>> +		if (gup_flags & FOLL_PIN)
>>>>> +			unpin_user_page(dpage);
>>>>> +		else
>>>>> +			put_page(dpage);
>>>>> +		dpage = NULL;
>>>>> +	}
>>>>> +
>>>>> +	return dpage;
>>>>> +}
>>>>> +
>>>>> +/*
>>>>>   * Check whether all pages are pinnable, if so return number of pages.  If some
>>>>>   * pages are not pinnable, migrate them, and unpin all pages. Return zero if
>>>>>   * pages were migrated, or if some pages were not successfully isolated.
>>>>> @@ -1888,15 +1942,40 @@ static long check_and_migrate_movable_pages(unsigned long nr_pages,
>>>>>  			continue;
>>>>>  		prev_head = head;
>>>>>  		/*
>>>>> -		 * If we get a movable page, since we are going to be pinning
>>>>> -		 * these entries, try to move them out if possible.
>>>>> +		 * Device coherent pages are managed by a driver and should not
>>>>> +		 * be pinned indefinitely as it prevents the driver moving the
>>>>> +		 * page. So when trying to pin with FOLL_LONGTERM instead try
>>>>> +		 * migrating page out of device memory.
>>>>>  		 */
>>>>>  		if (is_dev_private_or_coherent_page(head)) {
>>>>> +			/*
>>>>> +			 * device private pages will get faulted in during gup
>>>>> +			 * so it shouldn't be possible to see one here.
>>>>> +			 */
>>>>>  			WARN_ON_ONCE(is_device_private_page(head));
>>>>> -			ret = -EFAULT;
>>>>> -			goto unpin_pages;
>>>>> +			WARN_ON_ONCE(PageCompound(head));
>>>>> +
>>>>> +			/*
>>>>> +			 * migration will fail if the page is pinned, so convert
>>>>> +			 * the pin on the source page to a normal reference.
>>>>> +			 */
>>>>> +			if (gup_flags & FOLL_PIN) {
>>>>> +				get_page(head);
>>>>> +				unpin_user_page(head);
>>>>> +			}
>>>>> +
>>>>> +			pages[i] = migrate_device_page(head, gup_flags);
>>>>
>>>> For ordinary migrate_pages(), we'll unpin all pages and return 0 so the
>>>> caller will retry pinning by walking the page tables again.
>>>>
>>>> Why can't we apply the same mechanism here? This "let's avoid another
>>>> walk" looks unnecessary complicated to me, but I might be wrong.
>>>
>>> There's no reason we couldn't. I figured we have the page in the right spot
>>> anyway so it was easy to do, and looking at this rebased on top of Christoph's
>>> ZONE_DEVICE refcount simplification I'm not sure it would be any simpler
>>> anyway.
>>>
>>> It would remove the call to try_grab_page(), but we'd still have to return an
>>> error on migration failures whilst also ensuring we putback any non-device
>>> pages that may have been isolated. I might have overlooked something though,
>>> so certainly happy for suggestions.
>>
>> Staring at the code, I was wondering if we could either
>>
>> * build a second list of device coherent pages to migrate and call a
>>   migrate_device_pages() bulk function
>> * simply use movable_page_list() and teach migrate_pages() how to handle
>>   them.
> 

(sorry for the late reply)

> I did consider that approach. The problem is zone device pages are not LRU
> pages. In particular page->lru is not available to add the page to a list, and
> as an external API and internally migrate_pages() relies heavily on moving
> pages between lists.

I see, so I assume there is no way we could add them to a list? We could
use a temporary array we'd try allocating once we stumble over over such
a device page that needs migration.

The you'd teach is_pinnable_page() to reject
is_dev_private_or_coherent_page() and special case
is_dev_private_or_coherent_page() under the "if
(!is_pinnable_page(head))" check.

You'd grab an additional reference and add them to the temp array. The
you'd just proceed as normal towards the end of the function (reverting
the pin/ref from the input array) and would try to migrate all device
pages in the temp array just before migrate_pages() --
migrate_device_pages(), properly handling/indicating if either migration
path fails.

Instead of putback_movable_pages() on the list you'd had
unref_device_pages() and supply the array.


Just a thought to limit the impact and eventually make it a bit nicer to
read, avoiding modifications of the input array.

> 
>> I'd really appreciate as little special casing as possible for the ever
>> growing list of new DEVICE types all over the place. E.g., just staring
>> at fork even before the new device coherent made my head spin.
> 
> That's fair. We could pull the checks for device pages out into a self
> contained function (eg. check_and_migrate_device_pages()) called before
> check_and_migrate_movable_pages(). The down side of that is we'd always have an
> extra loop over all the pages just to scan for device pages, but perhaps that's
> not a concern?

I mean, they are movable ... just not "ordinarily" movable, so it smells
like this logic belongs into check_and_migrate_movable_pages() :)