System ACE MPU interface to CF card

System ACE MPU interface to CF card

[Sharad Gupta]

Hi All,

I would like to know if anyone in the group or otherwise is/was using
Xilinx's System ACE MPU interface to read/write to Compact Flash card on
their hardware ? I see that there is a device driver provided by Xilinx
for system ace mpu bus in the linux kernel tree (2.4.23). But, I dont
see anyone actually implementing it in their design.

It will be nice to follow someone's footsteps if it has already been
implemented before. Else, please direct me to appropriate resources
about how to go about using it in my design.

Thanks
Sharad.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Stephen Williams]

Sharad Gupta sharad-at-buckeye.ece.drexel.edu |PPC Linux Embedded| wrote:
>
> Hi All,
>
> I would like to know if anyone in the group or otherwise is/was using
> Xilinx's System ACE MPU interface to read/write to Compact Flash card on
> their hardware ? I see that there is a device driver provided by Xilinx
> for system ace mpu bus in the linux kernel tree (2.4.23). But, I dont
> see anyone actually implementing it in their design.

That would be me. I haven't yet submitted patches for my JSE
board, but if you look back to the last few weeks in this list,
you'll see plenty of messages with SystemACE in their subject.

Including a patch on 1 April 2004 to generalize the driver a
bit. MonteVista folks have approved the patch, so I hope it
turns up inofficial trees in due course.

--
Steve Williams                "The woods are lovely, dark and deep.
steve at XXXXXXXXXX           But I have promises to keep,
http://www.XXXXXXXXXX         and lines to code before I sleep,
http://www.picturel.com       And lines to code before I sleep."


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Tom Rini]

On Wed, Apr 07, 2004 at 04:09:33PM -0700, Stephen Williams wrote:
>
> Sharad Gupta sharad-at-buckeye.ece.drexel.edu |PPC Linux Embedded| wrote:
> >
> >Hi All,
> >
> >I would like to know if anyone in the group or otherwise is/was using
> >Xilinx's System ACE MPU interface to read/write to Compact Flash card on
> >their hardware ? I see that there is a device driver provided by Xilinx
> >for system ace mpu bus in the linux kernel tree (2.4.23). But, I dont
> >see anyone actually implementing it in their design.
>
> That would be me. I haven't yet submitted patches for my JSE
> board, but if you look back to the last few weeks in this list,
> you'll see plenty of messages with SystemACE in their subject.
>
> Including a patch on 1 April 2004 to generalize the driver a
> bit. MonteVista folks have approved the patch, so I hope it
> turns up inofficial trees in due course.

The previous message reminded me that I still hadn't pushed that patch,
so I did that a few hours ago (and it should show up on the commit list
in due course).

--
Tom Rini
http://gate.crashing.org/~trini/

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Stephen Williams]

[-- Attachment #1: Type: text/plain, Size: 961 bytes --]

Sharad Gupta sharad-at-buckeye.ece.drexel.edu |PPC Linux Embedded| wrote:

> Can you send me your board patch so that I can configure and build ACE
> driver for my board ? I want some help in knowing how and where to
> customize the driver for one's board.

The attached patch demonstrates how I enabled the SYSACE driver
for my (JSE) board. I'm not so sure I'm happy with it, because
each board will have to do something similar. I'm thinking that
the CONFIG_XILINX_SYSACE define should lose its dependence on
CONFIG_XILINX_OCP, but I'm not clear on the best way to account
for this in the xilinx_ocp/Makefile.

NOTE: This patch is meant to demonstrate my thinking on this
matter. It's not complete.

--
Steve Williams                "The woods are lovely, dark and deep.
steve at XXXXXXXXXX           But I have promises to keep,
http://www.XXXXXXXXXX         and lines to code before I sleep,
http://www.picturel.com       And lines to code before I sleep."

[-- Attachment #2: foo.txt --]
[-- Type: text/plain, Size: 1919 bytes --]

Index: arch/ppc/platforms/xilinx_ocp/Makefile
===================================================================
RCS file: /cvsroot/linuxppc_2_4_devel/arch/ppc/platforms/xilinx_ocp/Makefile,v
retrieving revision 1.1
diff -u -r1.1 Makefile
--- arch/ppc/platforms/xilinx_ocp/Makefile	1 Dec 2002 22:42:56 -0000	1.1
+++ arch/ppc/platforms/xilinx_ocp/Makefile	13 Apr 2004 17:48:37 -0000
@@ -13,6 +13,7 @@
 			   xipif_v1_23_b.o xpacket_fifo_v1_00_b.o xversion.o

 obj-$(CONFIG_XILINX_OCP) := xilinx_ocp.o
+obj-$(CONFIG_JSE) := xilinx_ocp.o

 xilinx_ocp.o: $(xilinx_ocp-objs)
 	$(LD) -r -o $@ $(xilinx_ocp-objs)
Index: arch/ppc/platforms/xilinx_ocp/xparameters.h
===================================================================
RCS file: /cvsroot/linuxppc_2_4_devel/arch/ppc/platforms/xilinx_ocp/xparameters.h,v
retrieving revision 1.2
diff -u -r1.2 xparameters.h
--- arch/ppc/platforms/xilinx_ocp/xparameters.h	30 Oct 2003 00:32:20 -0000	1.2
+++ arch/ppc/platforms/xilinx_ocp/xparameters.h	13 Apr 2004 17:48:37 -0000
@@ -16,3 +16,7 @@
 #if defined(CONFIG_XILINX_ML300)
 #include "xparameters_ml300.h"
 #endif
+
+#if defined(CONFIG_JSE)
+#include "xparameters_jse.h"
+#endif
Index: drivers/block/Config.in
===================================================================
RCS file: /cvsroot/linuxppc_2_4_devel/drivers/block/Config.in,v
retrieving revision 1.5
diff -u -r1.5 Config.in
--- drivers/block/Config.in	30 Oct 2003 00:32:48 -0000	1.5
+++ drivers/block/Config.in	13 Apr 2004 17:48:37 -0000
@@ -40,7 +40,7 @@
 dep_tristate 'Mylex DAC960/DAC1100 PCI RAID Controller support' CONFIG_BLK_DEV_DAC960 $CONFIG_PCI
 dep_tristate 'Micro Memory MM5415 Battery Backed RAM support (EXPERIMENTAL)' CONFIG_BLK_DEV_UMEM $CONFIG_PCI $CONFIG_EXPERIMENTAL

-if [ "$CONFIG_XILINX_OCP" = "y" ]; then
+if [ "$CONFIG_XILINX_OCP" = "y" -o "$CONFIG_JSE" = "y" ]; then
    tristate 'Xilinx on-chip System ACE' CONFIG_XILINX_SYSACE
 fi

Re: System ACE MPU interface to CF card

[Sharad Gupta]

Stephen Williams wrote:

> The attached patch demonstrates how I enabled the SYSACE driver
> for my (JSE) board. I'm not so sure I'm happy with it, because
> each board will have to do something similar. I'm thinking that
> the CONFIG_XILINX_SYSACE define should lose its dependence on
> CONFIG_XILINX_OCP, but I'm not clear on the best way to account
> for this in the xilinx_ocp/Makefile.

Thanks Stephen,

Your patch helped me a lot in compiling and configuring my kernel. Just
on a side note, have you been able to talk to compact flash (mount fs on
it, fsck it ?) through system ace ?

My kernel hangs while doing partition check on the drive. All I ever see is

Partition check:
xsysacea:

(then nothing happens)

I am using SanDisk's 128MB CF card and system ace is configured in the
byte mode. Any pointers as to why it is not going ahead will be highly
appreciated.

Thanks
Sharad.


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Stephen Williams]

Sharad Gupta sharad-at-buckeye.ece.drexel.edu |PPC Linux Embedded| wrote:

 > Thanks Stephen,
 >
 > Your patch helped me a lot in compiling and configuring my kernel. Just
 > on a side note, have you been able to talk to compact flash (mount fs on
 > it, fsck it ?) through system ace ?
 >
 > My kernel hangs while doing partition check on the drive. All I ever see is
 >
 > Partition check:
 > xsysacea:
 >
 > (then nothing happens)
 >
 > I am using SanDisk's 128MB CF card and system ace is configured in the
 > byte mode. Any pointers as to why it is not going ahead will be highly
 > appreciated.

Works just peachy for me, with 32M and 256M disks installed. I
have most of my embedded system living on the disk (a little
bit of ramdisk for root, and tmpfs for /tmp) with not a spot
of trouble.

I have noticed that if the disk does not have a .ace file on
it, and a Xilinx chip to program, the driver fails to get any
partition info from the disk and it all fails from there. So
I suspect the driver still to some degree relies on the state
of the chip after a successful Xilinx load. If you are trying
to use the chip without an FPGA attached, you may be running
into that.
--
Steve Williams                "The woods are lovely, dark and deep.
steve at XXXXXXXXXX           But I have promises to keep,
http://www.XXXXXXXXXX         and lines to code before I sleep,
http://www.picturel.com       And lines to code before I sleep."


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Jon Masters]

Stephen Williams wrote:

| I have noticed that if the disk does not have a .ace file on
| it, and a Xilinx chip to program, the driver fails to get any
| partition info from the disk and it all fails from there. So
| I suspect the driver still to some degree relies on the state
| of the chip after a successful Xilinx load. If you are trying
| to use the chip without an FPGA attached, you may be running
| into that.

I am having problems with xilinx_sysace at my end. This is on a
non-ML300 which is not running at anything like the same clock as the
rest of you at this moment in time (for various hardware development
reasons unrelated to the Linux side). I have a classic case of debugging
printk code altering the rate of execution and making things work.

Otherwise we get hard lockups and corrupted CURRENT data which has all
the hallmarks of a failure to re-aquire io_request_lock somewhere.

Obviously I will fix this problem soon but if anyone has already caught
this nasty and fixed it then do drop me a line. The driver will need
plenty of cleanup for kernel 2.6. Even then the HAL approach clearly
needs to be replaced also (but I will send an update on that later).

Jon.


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Jon Masters]

Jon Masters wrote:

| I am having problems with xilinx_sysace at my end. This is on a
| non-ML300 which is not running at anything like the same clock as the
| rest of you at this moment in time (for various hardware development
| reasons unrelated to the Linux side). I have a classic case of debugging
| printk code altering the rate of execution and making things work.

I received a mail offlist about this concerning the issues with an
earlier revision of the xilinx_sysace driver brokenly not following the
datasheet and failing to hold the device configuration controller in
reset while doing transfers. I have now upgraded to a new verson.

This has fixed the ``hardware seems fairly screwed, it's not giving me
interrupts when I want it! arrrgh!'' type problems but I still get
random lockups and corruption of the request queue, which I think is
related to a missing spinlock aquisition somewhere else.

To be honest I might soon get annoyed enough with it to rewrite it.
Clearly Xilinx think it is a really good idea to have this HAL concept
(and I think it is a really really bad one) - and I am one of the people
that ends up using this code so really want something that works :-).

Jon.


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Jeff Angielski]

On Thu, 2004-04-15 at 15:05, Sharad Gupta wrote:
> Your patch helped me a lot in compiling and configuring my kernel. Just
> on a side note, have you been able to talk to compact flash (mount fs on
> it, fsck it ?) through system ace ?
>
> My kernel hangs while doing partition check on the drive. All I ever see is
>
> Partition check:
> xsysacea:
>
> (then nothing happens)

Sounds to me like you do not have your interrupts configured properly
between the SystemACE and the CPU or your xsa_thread is dying
prematurely.

Either way, the kernel is getting no response when it actually tries to
access the CF thru the SystemACE.


Jeff Angielski
The PTR Group


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Sharad Gupta]

Jeff Angielski wrote:
> Sounds to me like you do not have your interrupts configured properly
> between the SystemACE and the CPU or your xsa_thread is dying
> prematurely.
>
> Either way, the kernel is getting no response when it actually tries to
> access the CF thru the SystemACE.
>
>
> Jeff Angielski
> The PTR Group

I do not enable interrupts between SystemACE and the CPU. Following the
debug cycle:

xsysace_init()
->register_disk()
-->grok_partitions()
--->check_partition()
---->msdos_partition()
----->read_dev_sector()
------>read_cache_page()
------->lock_page()
-------->__lock_page()
--------->schedule()

And it never resumes after that. The code just goes to power_save mode.
Any help would be appreciated.


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: System ACE MPU interface to CF card

[Jeff Angielski]

On Mon, 2004-04-19 at 14:06, Sharad Gupta wrote:
> I do not enable interrupts between SystemACE and the CPU.

Doh.  It is in your best interest to get the interrupts working...

> Following the
> debug cycle:
>
> xsysace_init()
> ->register_disk()
> -->grok_partitions()
> --->check_partition()
> ---->msdos_partition()
> ----->read_dev_sector()
> ------>read_cache_page()
> ------->lock_page()
> -------->__lock_page()
> --------->schedule()
>
> And it never resumes after that. The code just goes to power_save mode.
> Any help would be appreciated.

I would suggest instrumenting the code in the driver that actually
handles the block request from the kernel.  For instance, is the request
being received by the xsa_thread?  Is the driver actually making a
request to the device.  And most importantly and where I suspect your
problem lies, is the xsa_complete_request() ever getting run?

Jeff Angielski
The PTR Group


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

more SystemACE problems

[Sharad Gupta]

Jeff Angielski wrote:
> On Mon, 2004-04-19 at 14:06, Sharad Gupta wrote:
> I would suggest instrumenting the code in the driver that actually
> handles the block request from the kernel.  For instance, is the request
> being received by the xsa_thread?  Is the driver actually making a
> request to the device.  And most importantly and where I suspect your
> problem lies, is the xsa_complete_request() ever getting run?
>
> Jeff Angielski
> The PTR Group

Hi Jeff,

Thanks very much for your help. I have enabled the interrupts for
SystemACE now. And in fact xsa_complete_request() was never called. I
fixed that in xsa_thread.

I was wondering if you could send in your modifications to the
xilinx_systemace/* files so that I can know better where else the
potential problems may lie.

Do I need to have a device entry for xsystemacea (0/1) in my /dev
directory ? If yes, then what should it be like ?

Thanks
Sharad.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/