Patchwork PCI/P2PDMA: Match interface changes to devm_memremap_pages()

login
register
mail settings
Submitter Logan Gunthorpe
Date Nov. 30, 2018, 10:59 p.m.
Message ID <20181130225911.2900-1-logang@deltatee.com>
Download mbox | patch
Permalink /patch/669607/
State New
Headers show

Comments

Logan Gunthorpe - Nov. 30, 2018, 10:59 p.m.
"mm-hmm-mark-hmm_devmem_add-add_resource-export_symbol_gpl.patch" in the
mm tree breaks p2pdma. The patch was written and reviewed before p2pdma
was merged so the necessary changes were not done to the call site in
that code.

Without this patch, all drivers will fail to register P2P resources
because devm_memremap_pages() will return -EINVAL due to the 'kill'
member of the pagemap structure not yet being set.

Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Bjorn Helgaas <bhelgaas@google.com>
---

Ideally this patch should be squashed with the one mentioned above to
avoid a bisect regression point.

drivers/pci/p2pdma.c | 10 ++--------
 1 file changed, 2 insertions(+), 8 deletions(-)

--
2.19.0
Dan Williams - Dec. 1, 2018, 12:52 a.m.
On Fri, Nov 30, 2018 at 2:59 PM Logan Gunthorpe <logang@deltatee.com> wrote:
>
> "mm-hmm-mark-hmm_devmem_add-add_resource-export_symbol_gpl.patch" in the
> mm tree breaks p2pdma. The patch was written and reviewed before p2pdma
> was merged so the necessary changes were not done to the call site in
> that code.
>
> Without this patch, all drivers will fail to register P2P resources
> because devm_memremap_pages() will return -EINVAL due to the 'kill'
> member of the pagemap structure not yet being set.
>
> Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Dan Williams <dan.j.williams@intel.com>

Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Bjorn Helgaas - Dec. 6, 2018, 8:46 p.m.
On Fri, Nov 30, 2018 at 03:59:11PM -0700, Logan Gunthorpe wrote:
> "mm-hmm-mark-hmm_devmem_add-add_resource-export_symbol_gpl.patch" in the
> mm tree breaks p2pdma. The patch was written and reviewed before p2pdma
> was merged so the necessary changes were not done to the call site in
> that code.
> 
> Without this patch, all drivers will fail to register P2P resources
> because devm_memremap_pages() will return -EINVAL due to the 'kill'
> member of the pagemap structure not yet being set.
> 
> Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Dan Williams <dan.j.williams@intel.com>
> Cc: Bjorn Helgaas <bhelgaas@google.com>

Applied with Dan's reviewed-by to pci/peer-to-peer for v4.21, thanks!

If the mm patch you mention gets merged for v4.20, let me know and I can
promote this to for-linus so v4.20 doesn't end up broken.

> ---
> 
> Ideally this patch should be squashed with the one mentioned above to
> avoid a bisect regression point.
> 
> drivers/pci/p2pdma.c | 10 ++--------
>  1 file changed, 2 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c
> index ae3c5b25dcc7..a2eb25271c96 100644
> --- a/drivers/pci/p2pdma.c
> +++ b/drivers/pci/p2pdma.c
> @@ -82,10 +82,8 @@ static void pci_p2pdma_percpu_release(struct percpu_ref *ref)
>  	complete_all(&p2p->devmap_ref_done);
>  }
> 
> -static void pci_p2pdma_percpu_kill(void *data)
> +static void pci_p2pdma_percpu_kill(struct percpu_ref *ref)
>  {
> -	struct percpu_ref *ref = data;
> -
>  	/*
>  	 * pci_p2pdma_add_resource() may be called multiple times
>  	 * by a driver and may register the percpu_kill devm action multiple
> @@ -198,6 +196,7 @@ int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size,
>  	pgmap->type = MEMORY_DEVICE_PCI_P2PDMA;
>  	pgmap->pci_p2pdma_bus_offset = pci_bus_address(pdev, bar) -
>  		pci_resource_start(pdev, bar);
> +	pgmap->kill = pci_p2pdma_percpu_kill;
> 
>  	addr = devm_memremap_pages(&pdev->dev, pgmap);
>  	if (IS_ERR(addr)) {
> @@ -211,11 +210,6 @@ int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size,
>  	if (error)
>  		goto pgmap_free;
> 
> -	error = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_percpu_kill,
> -					  &pdev->p2pdma->devmap_ref);
> -	if (error)
> -		goto pgmap_free;
> -
>  	pci_info(pdev, "added peer-to-peer DMA memory %pR\n",
>  		 &pgmap->res);
> 
> --
> 2.19.0
Logan Gunthorpe - Dec. 6, 2018, 10:17 p.m.
On 2018-12-06 1:46 p.m., Bjorn Helgaas wrote:
> On Fri, Nov 30, 2018 at 03:59:11PM -0700, Logan Gunthorpe wrote:
>> "mm-hmm-mark-hmm_devmem_add-add_resource-export_symbol_gpl.patch" in the
>> mm tree breaks p2pdma. The patch was written and reviewed before p2pdma
>> was merged so the necessary changes were not done to the call site in
>> that code.
>>
>> Without this patch, all drivers will fail to register P2P resources
>> because devm_memremap_pages() will return -EINVAL due to the 'kill'
>> member of the pagemap structure not yet being set.
>>
>> Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
>> Cc: Andrew Morton <akpm@linux-foundation.org>
>> Cc: Dan Williams <dan.j.williams@intel.com>
>> Cc: Bjorn Helgaas <bhelgaas@google.com>
> 
> Applied with Dan's reviewed-by to pci/peer-to-peer for v4.21, thanks!
> 
> If the mm patch you mention gets merged for v4.20, let me know and I can
> promote this to for-linus so v4.20 doesn't end up broken.

Thanks Bjorn, but I think Andrew has also picked it up in the mm tree
with the patch in question. My hope is that he squashes the two but I'm
not sure what his intentions are. I mostly copied you for  information
purposes as this patch shouldn't even compile without Dan's patch.

Though, I guess we'll find out what goes in after the merge window --
you may not need to do anything.

Logan
Bjorn Helgaas - Dec. 11, 2018, 1:56 p.m.
On Thu, Dec 06, 2018 at 03:17:00PM -0700, Logan Gunthorpe wrote:
> 
> 
> On 2018-12-06 1:46 p.m., Bjorn Helgaas wrote:
> > On Fri, Nov 30, 2018 at 03:59:11PM -0700, Logan Gunthorpe wrote:
> >> "mm-hmm-mark-hmm_devmem_add-add_resource-export_symbol_gpl.patch" in the
> >> mm tree breaks p2pdma. The patch was written and reviewed before p2pdma
> >> was merged so the necessary changes were not done to the call site in
> >> that code.
> >>
> >> Without this patch, all drivers will fail to register P2P resources
> >> because devm_memremap_pages() will return -EINVAL due to the 'kill'
> >> member of the pagemap structure not yet being set.
> >>
> >> Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
> >> Cc: Andrew Morton <akpm@linux-foundation.org>
> >> Cc: Dan Williams <dan.j.williams@intel.com>
> >> Cc: Bjorn Helgaas <bhelgaas@google.com>
> > 
> > Applied with Dan's reviewed-by to pci/peer-to-peer for v4.21, thanks!
> > 
> > If the mm patch you mention gets merged for v4.20, let me know and I can
> > promote this to for-linus so v4.20 doesn't end up broken.
> 
> Thanks Bjorn, but I think Andrew has also picked it up in the mm tree
> with the patch in question. My hope is that he squashes the two but I'm
> not sure what his intentions are. I mostly copied you for  information
> purposes as this patch shouldn't even compile without Dan's patch.
> 
> Though, I guess we'll find out what goes in after the merge window --
> you may not need to do anything.

OK, I dropped this because I don't have the matching mm patch.  I agree,
it should be squashed so we avoid a bisection hole.  Let me know if I
need to do anything else with this.

Bjorn

Patch

diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c
index ae3c5b25dcc7..a2eb25271c96 100644
--- a/drivers/pci/p2pdma.c
+++ b/drivers/pci/p2pdma.c
@@ -82,10 +82,8 @@  static void pci_p2pdma_percpu_release(struct percpu_ref *ref)
 	complete_all(&p2p->devmap_ref_done);
 }

-static void pci_p2pdma_percpu_kill(void *data)
+static void pci_p2pdma_percpu_kill(struct percpu_ref *ref)
 {
-	struct percpu_ref *ref = data;
-
 	/*
 	 * pci_p2pdma_add_resource() may be called multiple times
 	 * by a driver and may register the percpu_kill devm action multiple
@@ -198,6 +196,7 @@  int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size,
 	pgmap->type = MEMORY_DEVICE_PCI_P2PDMA;
 	pgmap->pci_p2pdma_bus_offset = pci_bus_address(pdev, bar) -
 		pci_resource_start(pdev, bar);
+	pgmap->kill = pci_p2pdma_percpu_kill;

 	addr = devm_memremap_pages(&pdev->dev, pgmap);
 	if (IS_ERR(addr)) {
@@ -211,11 +210,6 @@  int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size,
 	if (error)
 		goto pgmap_free;

-	error = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_percpu_kill,
-					  &pdev->p2pdma->devmap_ref);
-	if (error)
-		goto pgmap_free;
-
 	pci_info(pdev, "added peer-to-peer DMA memory %pR\n",
 		 &pgmap->res);