diff mbox series

[RFC,2/3] pci/doe: Use devm_xa_init()

Message ID 20220705232159.2218958-3-ira.weiny@intel.com
State Rejected
Headers show
Series Introduce devm_xa_init | expand

Commit Message

Ira Weiny July 5, 2022, 11:21 p.m. UTC
From: Ira Weiny <ira.weiny@intel.com>

The XArray being used to store the protocols does not even store
allocated objects.

Use devm_xa_init() to automatically destroy the XArray when the PCI
device goes away.

Signed-off-by: Ira Weiny <ira.weiny@intel.com>
---
 drivers/pci/doe.c | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)

Comments

Bjorn Helgaas July 7, 2022, 4:06 p.m. UTC | #1
On Tue, Jul 05, 2022 at 04:21:58PM -0700, ira.weiny@intel.com wrote:
> From: Ira Weiny <ira.weiny@intel.com>
> 
> The XArray being used to store the protocols does not even store
> allocated objects.

I guess the point is that the doe_mb->prots XArray doesn't reference
any other objects that would need to be freed when destroying
doe_mb->prots?  A few more words here would make the commit log more
useful to non-XArray experts.

s|pci/doe|PCI/DOE| in subject to match the drivers/pci convention.

> Use devm_xa_init() to automatically destroy the XArray when the PCI
> device goes away.
> 
> Signed-off-by: Ira Weiny <ira.weiny@intel.com>
> ---
>  drivers/pci/doe.c | 14 ++------------
>  1 file changed, 2 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/pci/doe.c b/drivers/pci/doe.c
> index 0b02f33ef994..aa36f459d375 100644
> --- a/drivers/pci/doe.c
> +++ b/drivers/pci/doe.c
> @@ -386,13 +386,6 @@ static int pci_doe_cache_protocols(struct pci_doe_mb *doe_mb)
>  	return 0;
>  }
>  
> -static void pci_doe_xa_destroy(void *mb)
> -{
> -	struct pci_doe_mb *doe_mb = mb;
> -
> -	xa_destroy(&doe_mb->prots);
> -}
> -
>  static void pci_doe_destroy_workqueue(void *mb)
>  {
>  	struct pci_doe_mb *doe_mb = mb;
> @@ -440,11 +433,8 @@ struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset)
>  	doe_mb->pdev = pdev;
>  	doe_mb->cap_offset = cap_offset;
>  	init_waitqueue_head(&doe_mb->wq);
> -
> -	xa_init(&doe_mb->prots);
> -	rc = devm_add_action(dev, pci_doe_xa_destroy, doe_mb);
> -	if (rc)
> -		return ERR_PTR(rc);
> +	if (devm_xa_init(dev, &doe_mb->prots))
> +		return ERR_PTR(-ENOMEM);
>  
>  	doe_mb->work_queue = alloc_ordered_workqueue("DOE: [%x]", 0,
>  						     doe_mb->cap_offset);
> -- 
> 2.35.3
>
Ira Weiny July 8, 2022, 2:45 p.m. UTC | #2
On Thu, Jul 07, 2022 at 11:06:46AM -0500, Bjorn Helgaas wrote:
> On Tue, Jul 05, 2022 at 04:21:58PM -0700, ira.weiny@intel.com wrote:
> > From: Ira Weiny <ira.weiny@intel.com>
> > 
> > The XArray being used to store the protocols does not even store
> > allocated objects.
> 
> I guess the point is that the doe_mb->prots XArray doesn't reference
> any other objects that would need to be freed when destroying
> doe_mb->prots?

Yes.

> A few more words here would make the commit log more
> useful to non-XArray experts.

I'll update this to be more clear in a V1 if it goes that far.  But to clarify
here; the protocol information is a u16 vendor id and u8 protocol number.  So
we are able to store that in the unsigned long value that would normally be a
pointer to something in the XArray.

> 
> s|pci/doe|PCI/DOE| in subject to match the drivers/pci convention.

Yes. Sorry,

Thanks for the review,
Ira

> 
> > Use devm_xa_init() to automatically destroy the XArray when the PCI
> > device goes away.
> > 
> > Signed-off-by: Ira Weiny <ira.weiny@intel.com>
> > ---
> >  drivers/pci/doe.c | 14 ++------------
> >  1 file changed, 2 insertions(+), 12 deletions(-)
> > 
> > diff --git a/drivers/pci/doe.c b/drivers/pci/doe.c
> > index 0b02f33ef994..aa36f459d375 100644
> > --- a/drivers/pci/doe.c
> > +++ b/drivers/pci/doe.c
> > @@ -386,13 +386,6 @@ static int pci_doe_cache_protocols(struct pci_doe_mb *doe_mb)
> >  	return 0;
> >  }
> >  
> > -static void pci_doe_xa_destroy(void *mb)
> > -{
> > -	struct pci_doe_mb *doe_mb = mb;
> > -
> > -	xa_destroy(&doe_mb->prots);
> > -}
> > -
> >  static void pci_doe_destroy_workqueue(void *mb)
> >  {
> >  	struct pci_doe_mb *doe_mb = mb;
> > @@ -440,11 +433,8 @@ struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset)
> >  	doe_mb->pdev = pdev;
> >  	doe_mb->cap_offset = cap_offset;
> >  	init_waitqueue_head(&doe_mb->wq);
> > -
> > -	xa_init(&doe_mb->prots);
> > -	rc = devm_add_action(dev, pci_doe_xa_destroy, doe_mb);
> > -	if (rc)
> > -		return ERR_PTR(rc);
> > +	if (devm_xa_init(dev, &doe_mb->prots))
> > +		return ERR_PTR(-ENOMEM);
> >  
> >  	doe_mb->work_queue = alloc_ordered_workqueue("DOE: [%x]", 0,
> >  						     doe_mb->cap_offset);
> > -- 
> > 2.35.3
> >
Matthew Wilcox July 8, 2022, 2:49 p.m. UTC | #3
On Fri, Jul 08, 2022 at 07:45:12AM -0700, Ira Weiny wrote:
> On Thu, Jul 07, 2022 at 11:06:46AM -0500, Bjorn Helgaas wrote:
> > On Tue, Jul 05, 2022 at 04:21:58PM -0700, ira.weiny@intel.com wrote:
> > > From: Ira Weiny <ira.weiny@intel.com>
> > > 
> > > The XArray being used to store the protocols does not even store
> > > allocated objects.
> > 
> > I guess the point is that the doe_mb->prots XArray doesn't reference
> > any other objects that would need to be freed when destroying
> > doe_mb->prots?
> 
> Yes.
> 
> > A few more words here would make the commit log more
> > useful to non-XArray experts.
> 
> I'll update this to be more clear in a V1 if it goes that far.  But to clarify
> here; the protocol information is a u16 vendor id and u8 protocol number.  So
> we are able to store that in the unsigned long value that would normally be a
> pointer to something in the XArray.

Er.  Signed long.  I can't find drivers/pci/doe.c in linux-next, so
I have no idea if you're doing something wrong.  But what you said here
sounds wrong.
Ira Weiny July 8, 2022, 2:57 p.m. UTC | #4
On Fri, Jul 08, 2022 at 03:49:51PM +0100, Matthew Wilcox wrote:
> On Fri, Jul 08, 2022 at 07:45:12AM -0700, Ira Weiny wrote:
> > On Thu, Jul 07, 2022 at 11:06:46AM -0500, Bjorn Helgaas wrote:
> > > On Tue, Jul 05, 2022 at 04:21:58PM -0700, ira.weiny@intel.com wrote:
> > > > From: Ira Weiny <ira.weiny@intel.com>
> > > > 
> > > > The XArray being used to store the protocols does not even store
> > > > allocated objects.
> > > 
> > > I guess the point is that the doe_mb->prots XArray doesn't reference
> > > any other objects that would need to be freed when destroying
> > > doe_mb->prots?
> > 
> > Yes.
> > 
> > > A few more words here would make the commit log more
> > > useful to non-XArray experts.
> > 
> > I'll update this to be more clear in a V1 if it goes that far.  But to clarify
> > here; the protocol information is a u16 vendor id and u8 protocol number.  So
> > we are able to store that in the unsigned long value that would normally be a
> > pointer to something in the XArray.
> 
> Er.  Signed long.

Sorry I misspoke, xa_mk_value() takes an unsigned long.

> I can't find drivers/pci/doe.c in linux-next, so
> I have no idea if you're doing something wrong.

Sorry doe.c does not exist yet.  I came up with this idea while developing a
CXL series which is still in review.[0]

> But what you said here
> sounds wrong.

:-/

Can't I use xa_mk_value() to store data directly in the entry "pointer"?

From patch 3/9 in that series.[1]

+static void *pci_doe_xa_prot_entry(u16 vid, u8 prot)
+{
+	return xa_mk_value(((unsigned long)vid << 16) | prot);
+}

Both Dan and I thought this was acceptable in XArray?

Ira

[0] https://lore.kernel.org/linux-cxl/20220705154932.2141021-1-ira.weiny@intel.com/
[1] https://lore.kernel.org/linux-cxl/20220705154932.2141021-4-ira.weiny@intel.com/
Matthew Wilcox July 8, 2022, 3:04 p.m. UTC | #5
On Fri, Jul 08, 2022 at 07:57:10AM -0700, Ira Weiny wrote:
> > > I'll update this to be more clear in a V1 if it goes that far.  But to clarify
> > > here; the protocol information is a u16 vendor id and u8 protocol number.  So
> > > we are able to store that in the unsigned long value that would normally be a
> > > pointer to something in the XArray.
> > 
> > Er.  Signed long.
> 
> Sorry I misspoke, xa_mk_value() takes an unsigned long.

It does, *but* ...

static inline void *xa_mk_value(unsigned long v)
{
        WARN_ON((long)v < 0);
        return (void *)((v << 1) | 1);
}

... you can't pass an integer that has the top bit set to it.

> Can't I use xa_mk_value() to store data directly in the entry "pointer"?

Yes, that's the purpose of xa_mk_value().  From what you said, it sounded
like you were just storing the integer directly, which won't work.

> +static void *pci_doe_xa_prot_entry(u16 vid, u8 prot)
> +{
> +	return xa_mk_value(((unsigned long)vid << 16) | prot);
> +}
> 
> Both Dan and I thought this was acceptable in XArray?

You haven't tested that on 32-bit, have you?  Shift vid by 8 instead of
16, and it'll be fine.

(Oh, and you don't need to cast vid; the standard C integer promotions
will promote vid to int before shifting, and you won't lose any bits)
Ira Weiny July 8, 2022, 3:49 p.m. UTC | #6
On Fri, Jul 08, 2022 at 04:04:05PM +0100, Matthew Wilcox wrote:
> On Fri, Jul 08, 2022 at 07:57:10AM -0700, Ira Weiny wrote:
> > > > I'll update this to be more clear in a V1 if it goes that far.  But to clarify
> > > > here; the protocol information is a u16 vendor id and u8 protocol number.  So
> > > > we are able to store that in the unsigned long value that would normally be a
> > > > pointer to something in the XArray.
> > > 
> > > Er.  Signed long.
> > 
> > Sorry I misspoke, xa_mk_value() takes an unsigned long.
> 
> It does, *but* ...
> 
> static inline void *xa_mk_value(unsigned long v)
> {
>         WARN_ON((long)v < 0);
>         return (void *)((v << 1) | 1);
> }
> 
> ... you can't pass an integer that has the top bit set to it.
> 
> > Can't I use xa_mk_value() to store data directly in the entry "pointer"?
> 
> Yes, that's the purpose of xa_mk_value().  From what you said, it sounded
> like you were just storing the integer directly, which won't work.
> 
> > +static void *pci_doe_xa_prot_entry(u16 vid, u8 prot)
> > +{
> > +	return xa_mk_value(((unsigned long)vid << 16) | prot);
> > +}
> > 
> > Both Dan and I thought this was acceptable in XArray?
> 
> You haven't tested that on 32-bit, have you?  Shift vid by 8 instead of
> 16, and it'll be fine.

Ah ok.

> 
> (Oh, and you don't need to cast vid; the standard C integer promotions
> will promote vid to int before shifting, and you won't lose any bits)

Will do, thanks!
Ira
diff mbox series

Patch

diff --git a/drivers/pci/doe.c b/drivers/pci/doe.c
index 0b02f33ef994..aa36f459d375 100644
--- a/drivers/pci/doe.c
+++ b/drivers/pci/doe.c
@@ -386,13 +386,6 @@  static int pci_doe_cache_protocols(struct pci_doe_mb *doe_mb)
 	return 0;
 }
 
-static void pci_doe_xa_destroy(void *mb)
-{
-	struct pci_doe_mb *doe_mb = mb;
-
-	xa_destroy(&doe_mb->prots);
-}
-
 static void pci_doe_destroy_workqueue(void *mb)
 {
 	struct pci_doe_mb *doe_mb = mb;
@@ -440,11 +433,8 @@  struct pci_doe_mb *pcim_doe_create_mb(struct pci_dev *pdev, u16 cap_offset)
 	doe_mb->pdev = pdev;
 	doe_mb->cap_offset = cap_offset;
 	init_waitqueue_head(&doe_mb->wq);
-
-	xa_init(&doe_mb->prots);
-	rc = devm_add_action(dev, pci_doe_xa_destroy, doe_mb);
-	if (rc)
-		return ERR_PTR(rc);
+	if (devm_xa_init(dev, &doe_mb->prots))
+		return ERR_PTR(-ENOMEM);
 
 	doe_mb->work_queue = alloc_ordered_workqueue("DOE: [%x]", 0,
 						     doe_mb->cap_offset);