diff mbox

[dpdk-dev] bus/pci: fix wrong intr_handle.type with uio_pci_generic

Message ID 20171228061210.64767-1-zhiyong.yang@intel.com (mailing list archive)
State Superseded, archived
Delegated to: Thomas Monjalon
Headers

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK

Commit Message

Yang, Zhiyong Dec. 28, 2017, 6:12 a.m. UTC 
  In the function rte_pci_ioport_map, if uio_pci_generic is used on X86
platform, pci_ioport_map() is invoked, the operation
ev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; is execused directly,
it causes the wrong assignment for uio_pci_generic, the patch fixes it.

Fixes: 756ce64b1ecd ("eal: introduce PCI ioport API")
Cc: stable@dpdk.org
Signed-off-by: Zhiyong Yang <zhiyong.yang@intel.com>
---
 drivers/bus/pci/linux/pci.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

Comments

Thomas Monjalon Dec. 28, 2017, 9:05 a.m. UTC | #1 
28/12/2017 07:12, Zhiyong Yang:
> In the function rte_pci_ioport_map, if uio_pci_generic is used on X86
> platform, pci_ioport_map() is invoked, the operation
> ev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; is execused directly,
> it causes the wrong assignment for uio_pci_generic, the patch fixes it.
[...]
> --- a/drivers/bus/pci/linux/pci.c
> +++ b/drivers/bus/pci/linux/pci.c
> @@ -723,7 +723,9 @@ pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
>  	if (!found)
>  		return -1;
>  
> -	dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
> +	if (dev->kdrv == RTE_KDRV_NONE)
> +		dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;

I don't understand the logic.
NONE is different of UNKNOWN.

Your are talking about uio_pci_generic. In this case, it should be
RTE_KDRV_UIO_GENERIC.
Yang, Zhiyong Dec. 28, 2017, 9:37 a.m. UTC | #2 
Hi Thomas,

> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas@monjalon.net]
> Sent: Thursday, December 28, 2017 5:05 PM
> To: Yang, Zhiyong <zhiyong.yang@intel.com>
> Cc: dev@dpdk.org; Yigit, Ferruh <ferruh.yigit@intel.com>; stable@dpdk.org
> Subject: Re: [PATCH] bus/pci: fix wrong intr_handle.type with uio_pci_generic
> 
> 28/12/2017 07:12, Zhiyong Yang:
> > In the function rte_pci_ioport_map, if uio_pci_generic is used on X86
> > platform, pci_ioport_map() is invoked, the operation
> > ev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; is execused directly,
> > it causes the wrong assignment for uio_pci_generic, the patch fixes it.
> [...]
> > --- a/drivers/bus/pci/linux/pci.c
> > +++ b/drivers/bus/pci/linux/pci.c
> > @@ -723,7 +723,9 @@ pci_ioport_map(struct rte_pci_device *dev, int bar
> __rte_unused,
> >  	if (!found)
> >  		return -1;
> >
> > -	dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
> > +	if (dev->kdrv == RTE_KDRV_NONE)
> > +		dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
> 
> I don't understand the logic.
> NONE is different of UNKNOWN.
> 
> Your are talking about uio_pci_generic. In this case, it should be
> RTE_KDRV_UIO_GENERIC.

If we use uio_pci_generic,  dev->intr_handle.type has already been assigned to
The right value RTE_INTR_HANDLE_UIO_INTX before it, but in the function pci_ioport_map
the wrong value is assigned to dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
Two cases both call pci_ioport_map  on x86 platform.
one is RTE_KDRV_UIO_GENERIC
the other is RTE_KDRV_NONE 
if I understand right, for uio_generic, it should not be assigned to RTE_INTR_HANDLE_UNKNOWN;
This case has already the right value, don't need to assign again. 
The original code should be considered to handle RTE_KDRV_NONE case only.

thanks
Zhiyong
Thomas Monjalon Dec. 28, 2017, 10:49 a.m. UTC | #3 
28/12/2017 10:37, Yang, Zhiyong:
> Hi Thomas,
> 
> > -----Original Message-----
> > From: Thomas Monjalon [mailto:thomas@monjalon.net]
> > Sent: Thursday, December 28, 2017 5:05 PM
> > To: Yang, Zhiyong <zhiyong.yang@intel.com>
> > Cc: dev@dpdk.org; Yigit, Ferruh <ferruh.yigit@intel.com>; stable@dpdk.org
> > Subject: Re: [PATCH] bus/pci: fix wrong intr_handle.type with uio_pci_generic
> > 
> > 28/12/2017 07:12, Zhiyong Yang:
> > > In the function rte_pci_ioport_map, if uio_pci_generic is used on X86
> > > platform, pci_ioport_map() is invoked, the operation
> > > ev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; is execused directly,
> > > it causes the wrong assignment for uio_pci_generic, the patch fixes it.
> > [...]
> > > --- a/drivers/bus/pci/linux/pci.c
> > > +++ b/drivers/bus/pci/linux/pci.c
> > > @@ -723,7 +723,9 @@ pci_ioport_map(struct rte_pci_device *dev, int bar
> > __rte_unused,
> > >  	if (!found)
> > >  		return -1;
> > >
> > > -	dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
> > > +	if (dev->kdrv == RTE_KDRV_NONE)
> > > +		dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
> > 
> > I don't understand the logic.
> > NONE is different of UNKNOWN.
> > 
> > Your are talking about uio_pci_generic. In this case, it should be
> > RTE_KDRV_UIO_GENERIC.
> 
> If we use uio_pci_generic,  dev->intr_handle.type has already been assigned to
> The right value RTE_INTR_HANDLE_UIO_INTX before it, but in the function pci_ioport_map
> the wrong value is assigned to dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
> Two cases both call pci_ioport_map  on x86 platform.
> one is RTE_KDRV_UIO_GENERIC
> the other is RTE_KDRV_NONE 
> if I understand right, for uio_generic, it should not be assigned to RTE_INTR_HANDLE_UNKNOWN;
> This case has already the right value, don't need to assign again. 
> The original code should be considered to handle RTE_KDRV_NONE case only.

OK thanks for the explanation.
I had overlooked your patch.

Please, could add some of this text in v2?
I think it is good to explain that the right value is already set,
and we must avoid overwriting the default "UNKNOWN" value.

One more question, why is it needed to set the value to "UNKNOWN" in this
function? It should already be set to this value (0) at init.
Yang, Zhiyong Dec. 29, 2017, 2:10 a.m. UTC | #4 
> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas@monjalon.net]
> Sent: Thursday, December 28, 2017 6:49 PM
> To: Yang, Zhiyong <zhiyong.yang@intel.com>
> Cc: dev@dpdk.org; Yigit, Ferruh <ferruh.yigit@intel.com>; stable@dpdk.org
> Subject: Re: [PATCH] bus/pci: fix wrong intr_handle.type with uio_pci_generic
> 
> 28/12/2017 10:37, Yang, Zhiyong:
> > Hi Thomas,
> >
> > > -----Original Message-----
> > > From: Thomas Monjalon [mailto:thomas@monjalon.net]
> > > Sent: Thursday, December 28, 2017 5:05 PM
> > > To: Yang, Zhiyong <zhiyong.yang@intel.com>
> > > Cc: dev@dpdk.org; Yigit, Ferruh <ferruh.yigit@intel.com>;
> > > stable@dpdk.org
> > > Subject: Re: [PATCH] bus/pci: fix wrong intr_handle.type with
> > > uio_pci_generic
> > >
> > > 28/12/2017 07:12, Zhiyong Yang:
> > > > In the function rte_pci_ioport_map, if uio_pci_generic is used on
> > > > X86 platform, pci_ioport_map() is invoked, the operation
> > > > ev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; is execused
> > > > ev->directly,
> > > > it causes the wrong assignment for uio_pci_generic, the patch fixes it.
> > > [...]
> > > > --- a/drivers/bus/pci/linux/pci.c
> > > > +++ b/drivers/bus/pci/linux/pci.c
> > > > @@ -723,7 +723,9 @@ pci_ioport_map(struct rte_pci_device *dev, int
> > > > bar
> > > __rte_unused,
> > > >  	if (!found)
> > > >  		return -1;
> > > >
> > > > -	dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
> > > > +	if (dev->kdrv == RTE_KDRV_NONE)
> > > > +		dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
> > >
> > > I don't understand the logic.
> > > NONE is different of UNKNOWN.
> > >
> > > Your are talking about uio_pci_generic. In this case, it should be
> > > RTE_KDRV_UIO_GENERIC.
> >
> > If we use uio_pci_generic,  dev->intr_handle.type has already been
> > assigned to The right value RTE_INTR_HANDLE_UIO_INTX before it, but in
> > the function pci_ioport_map the wrong value is assigned to
> > dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN; Two cases both call
> pci_ioport_map  on x86 platform.
> > one is RTE_KDRV_UIO_GENERIC
> > the other is RTE_KDRV_NONE
> > if I understand right, for uio_generic, it should not be assigned to
> > RTE_INTR_HANDLE_UNKNOWN; This case has already the right value, don't
> need to assign again.
> > The original code should be considered to handle RTE_KDRV_NONE case only.
> 
> OK thanks for the explanation.
> I had overlooked your patch.
> 
> Please, could add some of this text in v2?
> I think it is good to explain that the right value is already set, and we must avoid
> overwriting the default "UNKNOWN" value.

Ok.  add the text in V2 later.

> 
> One more question, why is it needed to set the value to "UNKNOWN" in this
> function? It should already be set to this value (0) at init.

Good suggestion, we should remove it to fix issue . Any different opinion?  Everyone.

BTW,  The bug can be seen if run the following test case.

Testpmd startup fails  using uio_pci_generic running on virtio legacy device.

Thanks
Zhiyong
diff mbox

Patch

diff --git a/drivers/bus/pci/linux/pci.c b/drivers/bus/pci/linux/pci.c
index 5da6728fb..8ff7dbfd7 100644
--- a/drivers/bus/pci/linux/pci.c
+++ b/drivers/bus/pci/linux/pci.c
@@ -723,7 +723,9 @@  pci_ioport_map(struct rte_pci_device *dev, int bar __rte_unused,
 	if (!found)
 		return -1;
 
-	dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
+	if (dev->kdrv == RTE_KDRV_NONE)
+		dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
+
 	p->base = start;
 	RTE_LOG(DEBUG, EAL, "PCI Port IO found start=0x%x\n", start);