RE: [PATCH 0/5] rapidio: adding memory mapping IO support and misc fixes

RE: [PATCH 0/5] rapidio: adding memory mapping IO support and misc fixes

[Jan Neskudla]

Hallo, 

I tested your patches, and the MMIO is working when enabled in the
rionet driver. Only the compilation of rionet as modules was a problem.
I had to add following lines into rio.c to export missing symbols.

EXPORT_SYMBOL_GPL(rio_unmap_inb_region);
EXPORT_SYMBOL_GPL(rio_map_inb_region);
EXPORT_SYMBOL_GPL(rio_unmap_outb_region);
EXPORT_SYMBOL_GPL(rio_map_outb_region);

And one more think, when I enabled usage of DMA, rionet does not compile too, 
but in this case I do not have a fix. I tested this on kernel 2.6.29.1 and 
EP8548 as target board. 

                        Jan 




> 
> The patch series add MMIO support to Linux rapidio and fix a few nits.
> The patches cross rapidio, netdev, powerpc sub-systems.  It will be
> good for them to go through one tree.  Probably Andrew?  As the
> previous rapidio patches are mostly merged by you.
> 
> - Leo

Re: [PATCH 0/5] rapidio: adding memory mapping IO support and misc fixes

[Kumar Gala]

On May 7, 2009, at 9:10 AM, Jan Neskudla wrote:

> And one more think, when I enabled usage of DMA, rionet does not  
> compile too,
> but in this case I do not have a fix. I tested this on kernel  
> 2.6.29.1 and
> EP8548 as target board.

What exactly do you mean by that?  What CONFIG options cause compile  
failure?  Can you post the compiler error.

- k

Re: [PATCH 0/5] rapidio: adding memory mapping IO support and misc fixes

[Jan Neskudla]

On Thu, 2009-05-07 at 10:21 -0500, ext Kumar Gala wrote:
> On May 7, 2009, at 9:10 AM, Jan Neskudla wrote:
> 
> > And one more think, when I enabled usage of DMA, rionet does not  
> > compile too,
> > but in this case I do not have a fix. I tested this on kernel  
> > 2.6.29.1 and
> > EP8548 as target board.
> 
> What exactly do you mean by that?  What CONFIG options cause compile  
> failure?  Can you post the compiler error.
> 
> - k

The problem is with the missing stucture dma_client in the kernel tree
sources. It looks to me that a dma model changed after 2.6
.28. Here are the details. 

I used a pristine kernel 2.6.29 + patch 2.6.29.1 than Leo's patches in
this order. 

 rio-warn_unused_result-warnings-fix.patch
 rionet-add-memory-access-to-simulated-Ethernet-over-rapidio.patch
 powerpc-add-memory-map-support-to-Freescale-RapioIO-block.patch
 powerpc-fsl_rio-use-LAW-address-from-device-tree.patch
 rapidio-add-common-mapping-APIs-for-RapidIO-memory-access.patch

Important CONFIG options are: 
PPC_86xx=y
HPC8641_HPCN=y
RAPIDIO=y
DMADEVICES=y
FSL_DMA=y !!
NETDEVICES=y
RIONET=y/m
RIONET_MEMMAP=y
RIONET_DMA=y !!

And the error during compilation:


  CC      drivers/net/rionet.o
drivers/net/rionet.c:110: error: field `rio_dma_client' has incomplete
type
drivers/net/rionet.c: In function `rio_send_mem':
drivers/net/rionet.c:239: error: parse error before "rnet"
drivers/net/rionet.c: At top level:
drivers/net/rionet.c:514: warning: "enum dma_state" declared inside
parameter list
drivers/net/rionet.c:514: warning: its scope is only this definition or
declaration, which is probably not what you want
drivers/net/rionet.c:515: error: parameter `state' has incomplete type
drivers/net/rionet.c:515: error: return type is an incomplete type
drivers/net/rionet.c: In function `rionet_dma_event':
drivers/net/rionet.c:516: warning: type defaults to `int' in declaration
of `__mptr'
drivers/net/rionet.c:516: warning: initialization from incompatible
pointer type
drivers/net/rionet.c:518: error: variable `ack' has initializer but
incomplete type
drivers/net/rionet.c:518: error: `DMA_DUP' undeclared (first use in this
function)
drivers/net/rionet.c:518: error: (Each undeclared identifier is reported
only once
drivers/net/rionet.c:518: error: for each function it appears in.)
drivers/net/rionet.c:518: error: storage size of 'ack' isn't known
drivers/net/rionet.c:522: error: `DMA_RESOURCE_AVAILABLE' undeclared
(first use in this function)
drivers/net/rionet.c:524: error: `DMA_ACK' undeclared (first use in this
function)
drivers/net/rionet.c:531: error: `DMA_RESOURCE_REMOVED' undeclared
(first use in this function)
drivers/net/rionet.c:544: warning: `return' with a value, in function
returning void
drivers/net/rionet.c:518: warning: unused variable `ack'
drivers/net/rionet.c: In function `rionet_dma_register':
drivers/net/rionet.c:553: error: implicit declaration of function
`dma_async_client_register'
drivers/net/rionet.c:554: error: implicit declaration of function
`dma_async_client_chan_request'
drivers/net/rionet.c: In function `rionet_close':
drivers/net/rionet.c:731: error: implicit declaration of function
`dma_async_client_unregister'
make[2]: *** [drivers/net/rionet.o] Error 1


                            Jan 

Re: [PATCH 0/5] rapidio: adding memory mapping IO support and misc fixes

[Gerhard Jaeger]

Hi Jan,

On Friday 08 May 2009 12:06:35 Jan Neskudla wrote:
[SNIPSNAP]
> Important CONFIG options are: 
> PPC_86xx=y
> HPC8641_HPCN=y

you're using a e500 board (EP8548A), but the options above
will be used when building a kernel for a e600 machine (MPC8641).
Are you sure that is okay?

- Gerhard
-- 
Gerhard Jaeger <gjaeger@sysgo.com>            
SYSGO AG                                Embedding Innovations
www.sysgo.com | www.elinos.com | www.pikeos.com | www.osek.de 

Re: [PATCH 0/5] rapidio: adding memory mapping IO support and misc fixes

[Jan Neskudla]

Hi Gerhard, 

Yes I am sure, I know I chosen a different board than I use. I wanted to
show that this compilation problem is not influenced by our e500
patches. So I did the test on the pristine 2.6.29.1 kernel without any
external patches applied, and the problem is exactly the same when the
rionet is compiled for E500 with our patches. 

And anyway the dma_client structure is defined in the 2.6.28, but not in
the 2.6.29, so it looks to me that a rionet dma support is written for
older kernel. 

Here is the Linus tree and async_tx tree merge tree months ago
and info is:
 dmaengine: kill struct dma_client and supporting infrastructure

http://git.kernel.org/?
p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=d9e8a3a5b8298a3c814ed37ac5756e6f67b6be41

                     Jan 

On Fri, 2009-05-08 at 12:28 +0200, ext Gerhard Jaeger wrote:
> Hi Jan,
> 
> On Friday 08 May 2009 12:06:35 Jan Neskudla wrote:
> [SNIPSNAP]
> > Important CONFIG options are: 
> > PPC_86xx=y
> > HPC8641_HPCN=y
> 
> you're using a e500 board (EP8548A), but the options above
> will be used when building a kernel for a e600 machine (MPC8641).
> Are you sure that is okay?
> 
> - Gerhard

[PATCH 0/5] rapidio: adding memory mapping IO support and misc fixes

[Li Yang]

The patch series add MMIO support to Linux rapidio and fix a few nits.
 The patches cross rapidio, netdev, powerpc sub-systems.  It will be
good for them to go through one tree.  Probably Andrew?  As the
previous rapidio patches are mostly merged by you.

- Leo

Re: [PATCH 0/5] rapidio: adding memory mapping IO support and misc fixes

[Kumar Gala]

On Apr 28, 2009, at 5:16 AM, Li Yang wrote:

> The patch series add MMIO support to Linux rapidio and fix a few nits.
> The patches cross rapidio, netdev, powerpc sub-systems.  It will be
> good for them to go through one tree.  Probably Andrew?  As the
> previous rapidio patches are mostly merged by you.
>
> - Leo

When you repost use the -n option in git-format-patch so we get a  
sequence # for the patches.

Either Andrew or I can possible pick them up as long as we get the  
right maintainer ACKs.

- k