Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Thadeu Lima de Souza Cascardo]

> ----- Original Message -----
> From: cascardo@linux.vnet.ibm.com
> To: venkat.x.venkatsubra@oracle.com
> Cc: davem@davemloft.net, netdev@vger.kernel.org, linux-kernel@vger.kernel.org, cascardo@linux.vnet.ibm.com, dledford@redhat.com, Jes.Sorensen@redhat.com, rds-devel@oss.oracle.com
> Sent: Wednesday, January 4, 2012 4:03:42 PM GMT -06:00 US/Canada Central
> Subject: [PATCH] rds_rdma: don't assume infiniband device is PCI
> 
> RDS code assumes that the struct ib_device dma_device member, which is a
> pointer, points to a struct device embedded in a struct pci_dev.
> 
> This is not the case for ehca, for example, which is a OF driver, and
> makes dma_device point to a struct device embedded in a struct
> platform_device.
> 
> This will make the system crash when rds_rdma is loaded in a system
> with ehca, since it will try to access the bus member of a non-existent
> struct pci_dev.
> 
> The only reason rds_rdma uses the struct pci_dev is to get the NUMA node
> the device is attached to. Using dev_to_node for that is much better,
> since it won't assume which bus the infiniband is attached to.
> 
> Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
> Cc: dledford@redhat.com
> Cc: Jes.Sorensen@redhat.com
> Cc: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
> Cc: rds-devel@oss.oracle.com
> ---
>  net/rds/ib.h |    3 +--
>  1 files changed, 1 insertions(+), 2 deletions(-)
> 
> diff --git a/net/rds/ib.h b/net/rds/ib.h
> index edfaaaf..8d2b3d5 100644
> --- a/net/rds/ib.h
> +++ b/net/rds/ib.h
> @@ -186,8 +186,7 @@ struct rds_ib_device {
>  	struct work_struct	free_work;
>  };
>  
> -#define pcidev_to_node(pcidev) pcibus_to_node(pcidev->bus)
> -#define ibdev_to_node(ibdev) pcidev_to_node(to_pci_dev(ibdev->dma_device))
> +#define ibdev_to_node(ibdev) dev_to_node(ibdev->dma_device)
>  #define rdsibdev_to_node(rdsibdev) ibdev_to_node(rdsibdev->dev)
>  
>  /* bits for i_ack_flags */
> -- 
> 1.7.4.4
> 

On Thu, Jan 05, 2012 at 08:56:34AM -0800, Venkat Venkatsubra wrote:
> Hi Cascardo,
> 
> Your changes look good to me.
> But our latest code doesn't use this rdsibdev_to_node macro anywhere.
> Checking with the people in my group who know the history of the NUMA feature.
> Trying to find out if the call to kzalloc_node() can be replaced by kzalloc().
> In which case this macro can be removed.
> 
> I will keep you posted.
> 
> Venkat
> 

Hi, Venkat.

Do you have any public tree where we can track the last changes in RDS?
Note that I have changed ibsdev_to_node, which rdsibdev_to_node makes
use of. Anyway, replacing kzalloc_node with kzalloc has crossed my mind,
but since I was not sure if this would affect latency of RDS in any use
cases, I kept that and used a better function to get the node from the
device. And we have dev_to_node since 2.6.20, so it should not be a
problem to use it.

If possible, keep everyone copied and avoid top posting.

Regards.
Cascardo.

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[David Miller]

From: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Date: Thu, 5 Jan 2012 15:05:24 -0200

> On Thu, Jan 05, 2012 at 08:56:34AM -0800, Venkat Venkatsubra wrote:
>> Hi Cascardo,
>> 
>> Your changes look good to me.
>> But our latest code doesn't use this rdsibdev_to_node macro anywhere.
>> Checking with the people in my group who know the history of the NUMA feature.
>> Trying to find out if the call to kzalloc_node() can be replaced by kzalloc().
>> In which case this macro can be removed.
>> 
>> I will keep you posted.
>> 
>> Venkat
>> 
> 
> Hi, Venkat.
> 
> Do you have any public tree where we can track the last changes in RDS?
> Note that I have changed ibsdev_to_node, which rdsibdev_to_node makes
> use of. Anyway, replacing kzalloc_node with kzalloc has crossed my mind,
> but since I was not sure if this would affect latency of RDS in any use
> cases, I kept that and used a better function to get the node from the
> device. And we have dev_to_node since 2.6.20, so it should not be a
> problem to use it.
> 
> If possible, keep everyone copied and avoid top posting.

Indeed, otherwise it's impossible for anyone to follow the progress on
this patch.  If anything, you should never remove netdev from the CC:
list when discussing a patch.  Otherwise the followups don't make into
our patch tracking system at:

http://patchwork.ozlabs.org/project/netdev/list/

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Thadeu Lima de Souza Cascardo]

On Thu, Jan 05, 2012 at 03:05:24PM -0200, Thadeu Lima de Souza Cascardo wrote:
> > ----- Original Message -----
> > From: cascardo@linux.vnet.ibm.com
> > To: venkat.x.venkatsubra@oracle.com
> > Cc: davem@davemloft.net, netdev@vger.kernel.org, linux-kernel@vger.kernel.org, cascardo@linux.vnet.ibm.com, dledford@redhat.com, Jes.Sorensen@redhat.com, rds-devel@oss.oracle.com
> > Sent: Wednesday, January 4, 2012 4:03:42 PM GMT -06:00 US/Canada Central
> > Subject: [PATCH] rds_rdma: don't assume infiniband device is PCI
> > 
> > RDS code assumes that the struct ib_device dma_device member, which is a
> > pointer, points to a struct device embedded in a struct pci_dev.
> > 
> > This is not the case for ehca, for example, which is a OF driver, and
> > makes dma_device point to a struct device embedded in a struct
> > platform_device.
> > 
> > This will make the system crash when rds_rdma is loaded in a system
> > with ehca, since it will try to access the bus member of a non-existent
> > struct pci_dev.
> > 
> > The only reason rds_rdma uses the struct pci_dev is to get the NUMA node
> > the device is attached to. Using dev_to_node for that is much better,
> > since it won't assume which bus the infiniband is attached to.
> > 
> > Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
> > Cc: dledford@redhat.com
> > Cc: Jes.Sorensen@redhat.com
> > Cc: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
> > Cc: rds-devel@oss.oracle.com
> > ---
> >  net/rds/ib.h |    3 +--
> >  1 files changed, 1 insertions(+), 2 deletions(-)
> > 
> > diff --git a/net/rds/ib.h b/net/rds/ib.h
> > index edfaaaf..8d2b3d5 100644
> > --- a/net/rds/ib.h
> > +++ b/net/rds/ib.h
> > @@ -186,8 +186,7 @@ struct rds_ib_device {
> >  	struct work_struct	free_work;
> >  };
> >  
> > -#define pcidev_to_node(pcidev) pcibus_to_node(pcidev->bus)
> > -#define ibdev_to_node(ibdev) pcidev_to_node(to_pci_dev(ibdev->dma_device))
> > +#define ibdev_to_node(ibdev) dev_to_node(ibdev->dma_device)
> >  #define rdsibdev_to_node(rdsibdev) ibdev_to_node(rdsibdev->dev)
> >  
> >  /* bits for i_ack_flags */
> > -- 
> > 1.7.4.4
> > 
> 
> On Thu, Jan 05, 2012 at 08:56:34AM -0800, Venkat Venkatsubra wrote:
> > Hi Cascardo,
> > 
> > Your changes look good to me.
> > But our latest code doesn't use this rdsibdev_to_node macro anywhere.
> > Checking with the people in my group who know the history of the NUMA feature.
> > Trying to find out if the call to kzalloc_node() can be replaced by kzalloc().
> > In which case this macro can be removed.
> > 
> > I will keep you posted.
> > 
> > Venkat
> > 
> 
> Hi, Venkat.
> 
> Do you have any public tree where we can track the last changes in RDS?
> Note that I have changed ibsdev_to_node, which rdsibdev_to_node makes
> use of. Anyway, replacing kzalloc_node with kzalloc has crossed my mind,
> but since I was not sure if this would affect latency of RDS in any use
> cases, I kept that and used a better function to get the node from the
> device. And we have dev_to_node since 2.6.20, so it should not be a
> problem to use it.
> 
> If possible, keep everyone copied and avoid top posting.
> 
> Regards.
> Cascardo.

Just to make it clean, ibdev_to_node is used in Linus' tree and
next/master as of today and this patch fixes a real oops. I should have
probably added the stack trace in the log message. For completion, here
it is, running 3.2.0-rc7-00039-g89307ba:

NET: Registered protocol family 21
Registered RDS/iwarp transport
Unable to handle kernel paging request for data at address 0x00000088
Faulting instruction address: 0xc0000000000336d0
cpu 0x6: Vector: 300 (Data Access) at [c0000000ddd837e0]
    pc: c0000000000336d0: .pcibus_to_node+0x0/0x10
    lr: d000000004590de0: .rds_ib_add_one+0x100/0x360 [rds_rdma]
    sp: c0000000ddd83a60
   msr: 8000000000009032
   dar: 88
 dsisr: 40000000
  current = 0xc0000000ee6c0000
  paca    = 0xc000000006c32d00
    pid   = 8574, comm = modprobe
enter ? for help
[link register   ] d000000004590de0 .rds_ib_add_one+0x100/0x360
[rds_rdma]
[c0000000ddd83a60] d000000004590d88 .rds_ib_add_one+0xa8/0x360
[rds_rdma] (unreliable)
[c0000000ddd83b10] d000000003195240 .ib_register_client+0xc0/0x120
[ib_core]
[c0000000ddd83bb0] d00000000459074c .rds_ib_init+0x2c/0xf0 [rds_rdma]
[c0000000ddd83c30] d000000004590144 .rds_rdma_init+0xe4/0x180 [rds_rdma]
[c0000000ddd83cd0] c00000000000aa54 .do_one_initcall+0x64/0x1e0
[c0000000ddd83d90] c0000000000d359c .SyS_init_module+0xcc/0x220
[c0000000ddd83e30] c0000000000097e4 syscall_exit+0x0/0x40
--- Exception: c00 (System Call) at 00000fff7dc3d3d8
SP (fffe87eeeb0) is in userspace

Regards.
Cascardo.

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Thadeu Lima de Souza Cascardo]

On Fri, Jan 06, 2012 at 01:57:50PM -0200, Thadeu Lima de Souza Cascardo wrote:
> On Thu, Jan 05, 2012 at 03:05:24PM -0200, Thadeu Lima de Souza Cascardo wrote:
> > > ----- Original Message -----
> > > From: cascardo@linux.vnet.ibm.com
> > > To: venkat.x.venkatsubra@oracle.com
> > > Cc: davem@davemloft.net, netdev@vger.kernel.org, linux-kernel@vger.kernel.org, cascardo@linux.vnet.ibm.com, dledford@redhat.com, Jes.Sorensen@redhat.com, rds-devel@oss.oracle.com
> > > Sent: Wednesday, January 4, 2012 4:03:42 PM GMT -06:00 US/Canada Central
> > > Subject: [PATCH] rds_rdma: don't assume infiniband device is PCI
> > > 
> > > RDS code assumes that the struct ib_device dma_device member, which is a
> > > pointer, points to a struct device embedded in a struct pci_dev.
> > > 
> > > This is not the case for ehca, for example, which is a OF driver, and
> > > makes dma_device point to a struct device embedded in a struct
> > > platform_device.
> > > 
> > > This will make the system crash when rds_rdma is loaded in a system
> > > with ehca, since it will try to access the bus member of a non-existent
> > > struct pci_dev.
> > > 
> > > The only reason rds_rdma uses the struct pci_dev is to get the NUMA node
> > > the device is attached to. Using dev_to_node for that is much better,
> > > since it won't assume which bus the infiniband is attached to.
> > > 
> > > Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
> > > Cc: dledford@redhat.com
> > > Cc: Jes.Sorensen@redhat.com
> > > Cc: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
> > > Cc: rds-devel@oss.oracle.com
> > > ---
> > >  net/rds/ib.h |    3 +--
> > >  1 files changed, 1 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/net/rds/ib.h b/net/rds/ib.h
> > > index edfaaaf..8d2b3d5 100644
> > > --- a/net/rds/ib.h
> > > +++ b/net/rds/ib.h
> > > @@ -186,8 +186,7 @@ struct rds_ib_device {
> > >  	struct work_struct	free_work;
> > >  };
> > >  
> > > -#define pcidev_to_node(pcidev) pcibus_to_node(pcidev->bus)
> > > -#define ibdev_to_node(ibdev) pcidev_to_node(to_pci_dev(ibdev->dma_device))
> > > +#define ibdev_to_node(ibdev) dev_to_node(ibdev->dma_device)
> > >  #define rdsibdev_to_node(rdsibdev) ibdev_to_node(rdsibdev->dev)
> > >  
> > >  /* bits for i_ack_flags */
> > > -- 
> > > 1.7.4.4
> > > 
> > 
> > On Thu, Jan 05, 2012 at 08:56:34AM -0800, Venkat Venkatsubra wrote:
> > > Hi Cascardo,
> > > 
> > > Your changes look good to me.
> > > But our latest code doesn't use this rdsibdev_to_node macro anywhere.
> > > Checking with the people in my group who know the history of the NUMA feature.
> > > Trying to find out if the call to kzalloc_node() can be replaced by kzalloc().
> > > In which case this macro can be removed.
> > > 
> > > I will keep you posted.
> > > 
> > > Venkat
> > > 
> > 
> > Hi, Venkat.
> > 
> > Do you have any public tree where we can track the last changes in RDS?
> > Note that I have changed ibsdev_to_node, which rdsibdev_to_node makes
> > use of. Anyway, replacing kzalloc_node with kzalloc has crossed my mind,
> > but since I was not sure if this would affect latency of RDS in any use
> > cases, I kept that and used a better function to get the node from the
> > device. And we have dev_to_node since 2.6.20, so it should not be a
> > problem to use it.
> > 
> > If possible, keep everyone copied and avoid top posting.
> > 
> > Regards.
> > Cascardo.
> 
> Just to make it clean, ibdev_to_node is used in Linus' tree and
> next/master as of today and this patch fixes a real oops. I should have
> probably added the stack trace in the log message. For completion, here
> it is, running 3.2.0-rc7-00039-g89307ba:
> 
> NET: Registered protocol family 21
> Registered RDS/iwarp transport
> Unable to handle kernel paging request for data at address 0x00000088
> Faulting instruction address: 0xc0000000000336d0
> cpu 0x6: Vector: 300 (Data Access) at [c0000000ddd837e0]
>     pc: c0000000000336d0: .pcibus_to_node+0x0/0x10
>     lr: d000000004590de0: .rds_ib_add_one+0x100/0x360 [rds_rdma]
>     sp: c0000000ddd83a60
>    msr: 8000000000009032
>    dar: 88
>  dsisr: 40000000
>   current = 0xc0000000ee6c0000
>   paca    = 0xc000000006c32d00
>     pid   = 8574, comm = modprobe
> enter ? for help
> [link register   ] d000000004590de0 .rds_ib_add_one+0x100/0x360
> [rds_rdma]
> [c0000000ddd83a60] d000000004590d88 .rds_ib_add_one+0xa8/0x360
> [rds_rdma] (unreliable)
> [c0000000ddd83b10] d000000003195240 .ib_register_client+0xc0/0x120
> [ib_core]
> [c0000000ddd83bb0] d00000000459074c .rds_ib_init+0x2c/0xf0 [rds_rdma]
> [c0000000ddd83c30] d000000004590144 .rds_rdma_init+0xe4/0x180 [rds_rdma]
> [c0000000ddd83cd0] c00000000000aa54 .do_one_initcall+0x64/0x1e0
> [c0000000ddd83d90] c0000000000d359c .SyS_init_module+0xcc/0x220
> [c0000000ddd83e30] c0000000000097e4 syscall_exit+0x0/0x40
> --- Exception: c00 (System Call) at 00000fff7dc3d3d8
> SP (fffe87eeeb0) is in userspace
> 
> Regards.
> Cascardo.
> 

David,

The References was broken by Venkat email, so patchwork is not catching
up the new messages in the thread. At least, that's what I expect it
uses. So, I am sending this message with a fixed References. Sorry for
that.

Regards.
Cascardo.

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Venkat Venkatsubra]

Looks great to me:

Reviewed-by: Roland Dreier <roland@purestorage.com>
Acked-by: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>

[PATCH] rds_rdma: don't assume infiniband device is PCI

[Thadeu Lima de Souza Cascardo]

RDS code assumes that the struct ib_device dma_device member, which is a
pointer, points to a struct device embedded in a struct pci_dev.

This is not the case for ehca, for example, which is a OF driver, and
makes dma_device point to a struct device embedded in a struct
platform_device.

This will make the system crash when rds_rdma is loaded in a system
with ehca, since it will try to access the bus member of a non-existent
struct pci_dev.

The only reason rds_rdma uses the struct pci_dev is to get the NUMA node
the device is attached to. Using dev_to_node for that is much better,
since it won't assume which bus the infiniband is attached to.

Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Cc: dledford@redhat.com
Cc: Jes.Sorensen@redhat.com
Cc: Venkat Venkatsubra <venkat.x.venkatsubra@oracle.com>
Cc: rds-devel@oss.oracle.com
---
 net/rds/ib.h |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

diff --git a/net/rds/ib.h b/net/rds/ib.h
index edfaaaf..8d2b3d5 100644
--- a/net/rds/ib.h
+++ b/net/rds/ib.h
@@ -186,8 +186,7 @@ struct rds_ib_device {
 	struct work_struct	free_work;
 };
 
-#define pcidev_to_node(pcidev) pcibus_to_node(pcidev->bus)
-#define ibdev_to_node(ibdev) pcidev_to_node(to_pci_dev(ibdev->dma_device))
+#define ibdev_to_node(ibdev) dev_to_node(ibdev->dma_device)
 #define rdsibdev_to_node(rdsibdev) ibdev_to_node(rdsibdev->dev)
 
 /* bits for i_ack_flags */
-- 
1.7.4.4

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Roland Dreier]

Looks great to me:

Reviewed-by: Roland Dreier <roland@purestorage.com>

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Roland Dreier]

On Fri, Jan 6, 2012 at 9:47 AM, Roland Dreier <roland@purestorage.com> wrote:
> Looks great to me:

And BTW if Oracle has some non-published work they should probably just
rebase it on top of this fix.  We should merge this rather than shipping a
kernel with an obvious know crash like this.

 - R.

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Thadeu Lima de Souza Cascardo]

On Fri, Jan 06, 2012 at 09:49:36AM -0800, Roland Dreier wrote:
> On Fri, Jan 6, 2012 at 9:47 AM, Roland Dreier <roland@purestorage.com> wrote:
> > Looks great to me:
> 
> And BTW if Oracle has some non-published work they should probably just
> rebase it on top of this fix.  We should merge this rather than shipping a
> kernel with an obvious know crash like this.
> 
>  - R.
> 

Hi, Roland.

Sorry that I didn't include you. Should I include the infiniband users
under net/ in the files that Infiniband maintainers take care of?

David,

Is this going through your tree? In patchwork, its state is "Not
Applicable".

Thanks.
Cascardo.

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[David Miller]

From: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Date: Fri, 13 Jan 2012 16:17:06 -0200

> Is this going through your tree? In patchwork, its state is "Not
> Applicable".

There was an active RDS maintainer, though I haven't seen patches
from that person in a while.

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Thadeu Lima de Souza Cascardo]

On Fri, Jan 13, 2012 at 10:25:18AM -0800, David Miller wrote:
> From: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
> Date: Fri, 13 Jan 2012 16:17:06 -0200
> 
> > Is this going through your tree? In patchwork, its state is "Not
> > Applicable".
> 
> There was an active RDS maintainer, though I haven't seen patches
> from that person in a while.
> 

Hi, David.

Venkat is the current RDS maintainer and he has Acked the patch.

Thanks.
Cascardo.

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[David Miller]

From: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
Date: Fri, 13 Jan 2012 16:52:58 -0200

> On Fri, Jan 13, 2012 at 10:25:18AM -0800, David Miller wrote:
>> From: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
>> Date: Fri, 13 Jan 2012 16:17:06 -0200
>> 
>> > Is this going through your tree? In patchwork, its state is "Not
>> > Applicable".
>> 
>> There was an active RDS maintainer, though I haven't seen patches
>> from that person in a while.
>> 
> 
> Hi, David.
> 
> Venkat is the current RDS maintainer and he has Acked the patch.

Not according to patchwork.  I don't see him reply at all to the
patch posting and give his explicit "Acked-by: "

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Thadeu Lima de Souza Cascardo]

On Fri, Jan 13, 2012 at 01:12:48PM -0800, David Miller wrote:
> From: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
> Date: Fri, 13 Jan 2012 16:52:58 -0200
> 
> > On Fri, Jan 13, 2012 at 10:25:18AM -0800, David Miller wrote:
> >> From: Thadeu Lima de Souza Cascardo <cascardo@linux.vnet.ibm.com>
> >> Date: Fri, 13 Jan 2012 16:17:06 -0200
> >> 
> >> > Is this going through your tree? In patchwork, its state is "Not
> >> > Applicable".
> >> 
> >> There was an active RDS maintainer, though I haven't seen patches
> >> from that person in a while.
> >> 
> > 
> > Hi, David.
> > 
> > Venkat is the current RDS maintainer and he has Acked the patch.
> 
> Not according to patchwork.  I don't see him reply at all to the
> patch posting and give his explicit "Acked-by: "
> 

Venkat,

Care to send an Acked-by using a different MUA? Yours sends no
In-Reply-To or References header field.

David,

This is the message Venkat has sent.

http://marc.info/?l=linux-netdev&m=132587819932339&w=2

Thanks.
Cascardo.

Re: [PATCH] rds_rdma: don't assume infiniband device is PCI

[Roland Dreier]

On Fri, Jan 13, 2012 at 10:17 AM, Thadeu Lima de Souza Cascardo
<cascardo@linux.vnet.ibm.com> wrote:
>
> Hi, Roland.
>
> Sorry that I didn't include you. Should I include the infiniband users
> under net/ in the files that Infiniband maintainers take care of?

I don't maintain those things but I may have an opinion about them :)

 - R.