Skip to contentSkip to conversation

ppc4xx: Fix problem with board_eth_init() vs cpu_eth_init() on AMCC boards

PATCH New Stefan Roese 3 participants 7 messages Feb 11, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ppc4xx: Fix problem with board_eth_init() vs cpu_eth_init() on AMCC boards

Some AMCC eval boards do have a board_eth_init() function calling pci_eth_init(). These boards need to call cpu_eth_init() explicitly now with the new eth_init rework.

---
 board/amcc/katmai/katmai.c   |    8 +++++++-
 board/amcc/taihu/taihu.c     |    8 +++++++-
 board/amcc/taishan/taishan.c |    8 +++++++-
 board/amcc/yucca/yucca.c     |    8 +++++++-
 4 files changed, 28 insertions(+), 4 deletions(-)
board/amcc/katmai/katmai.c +7 -1
index b6c0c11..cace93f 100644
@@ -451,5 +451,11 @@ int post_hotkeys_pressed(void)
 
 int board_eth_init(bd_t *bis)
 {
-	return pci_eth_init(bis);
+	cpu_eth_init(bis);
+	pci_eth_init(bis);
+
+	/*
+	 * Return 0 so that cpu_eth_init() won't get executed again
+	 */
+	return 0;
 }
board/amcc/taihu/taihu.c +7 -1
index 5224378..8e74b3c 100644
@@ -195,5 +195,11 @@ int pci_pre_init(struct pci_controller *hose)
 
 int board_eth_init(bd_t *bis)
 {
-	return pci_eth_init(bis);
+	cpu_eth_init(bis);
+	pci_eth_init(bis);
+
+	/*
+	 * Return 0 so that cpu_eth_init() won't get executed again
+	 */
+	return 0;
 }
board/amcc/taishan/taishan.c +7 -1
index 28bdab5..0b5cdcc 100644
@@ -315,5 +315,11 @@ int post_hotkeys_pressed(void)
 
 int board_eth_init(bd_t *bis)
 {
-	return pci_eth_init(bis);
+	cpu_eth_init(bis);
+	pci_eth_init(bis);
+
+	/*
+	 * Return 0 so that cpu_eth_init() won't get executed again
+	 */
+	return 0;
 }
board/amcc/yucca/yucca.c +7 -1
index c805568..6036b29 100644
@@ -956,5 +956,11 @@ int onboard_pci_arbiter_selected(int core_pci)
 
 int board_eth_init(bd_t *bis)
 {
-	return pci_eth_init(bis);
+	cpu_eth_init(bis);
+	pci_eth_init(bis);
+
+	/*
+	 * Return 0 so that cpu_eth_init() won't get executed again
+	 */
+	return 0;
 }
Signature · 1 line
1.6.1.3
#1 lore raw

PATCH 1/1 ppc4xx: Fix problem with board_eth_init() vs cpu_eth_init() on AMCC boards

Dear Stefan Roese,

Quoted 18 lines · In message <[email protected]> you wrote:

Some AMCC eval boards do have a board_eth_init() function calling pci_eth_init(). These boards need to call cpu_eth_init() explicitly now with the new eth_init rework.

---
 board/amcc/katmai/katmai.c   |    8 +++++++-
 board/amcc/taihu/taihu.c     |    8 +++++++-
 board/amcc/taishan/taishan.c |    8 +++++++-
 board/amcc/yucca/yucca.c     |    8 +++++++-
 4 files changed, 28 insertions(+), 4 deletions(-)
board/amcc/katmai/katmai.c +0 -0
index b6c0c11..cace93f 100644
@@ -451,5 +451,11 @@ int post_hotkeys_pressed(void)
 

board/amcc/katmai/katmai.c int post_hotkeys_pressed(void) · line 460

Quoted 10 lines
board/amcc/katmai/katmai.c +7 -1
 int board_eth_init(bd_t *bis)
 {
-	return pci_eth_init(bis);
+	cpu_eth_init(bis);
+	pci_eth_init(bis);
+
+	/*
+	 * Return 0 so that cpu_eth_init() won't get executed again
+	 */
+	return 0;

What happens in case of errors? This looks broken to me, or I misinderstand the comment.

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Generally speaking, there are other ways to accomplish whatever it is
that you think you need ...                               - Doug Gwyn
#2 lore raw

PATCH 1/1 ppc4xx: Fix problem with board_eth_init() vs cpu_eth_init() on AMCC boards

Quoted 16 lines · Wolfgang Denk

board/amcc/katmai/katmai.c int post_hotkeys_pressed(void) · line 460

Quoted 13 lines

+++ b/board/amcc/katmai/katmai.c

board/amcc/katmai/katmai.c +7 -1
@@ -451,5 +451,11 @@ int post_hotkeys_pressed(void)
 
 int board_eth_init(bd_t *bis)
 {
-	return pci_eth_init(bis);
+	cpu_eth_init(bis);
+	pci_eth_init(bis);
+
+	/*
+	 * Return 0 so that cpu_eth_init() won't get executed again
+	 */
+	return 0;

What happens in case of errors? This looks broken to me, or I misinderstand the comment.

This is the code calling board_eth_init() from net/eth.c:

        /* Try board-specific initialization first.  If it fails or isn't
         * present, try the cpu-specific initialization */
        if (board_eth_init(bis) < 0)
                cpu_eth_init(bis);

So if we return with an error in board_eth_init(), cpu_eth_init() will get called again. Another way to fix this problem would be this implementation:

board_eth_init()

{
	pci_eth_init(bis);

	/*
	 * Return -1 so that cpu_eth_init() will get executed in net/eth.c
	 */
	return -1;
}

But I like the former implementation better, since the cpu internal ethernet interfaces are added first to the network devices list.

Best regards,
Stefan

=====================================================================
DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: [email protected]
=====================================================================

#3 lore raw

PATCH 1/1 ppc4xx: Fix problem with board_eth_init() vs cpu_eth_init() on AMCC boards

Dear Stefan,

Quoted 14 lines · In message <[email protected]> you wrote:
Quoted 11 lines

board/amcc/katmai/katmai.c int post_hotkeys_pressed(void) · line 460

Quoted 8 lines
board/amcc/katmai/katmai.c +7 -0
+	cpu_eth_init(bis);
+	pci_eth_init(bis);
+
+	/*
+	 * Return 0 so that cpu_eth_init() won't get executed again
+	 */
+	return 0;

What happens in case of errors? This looks broken to me, or I misinderstand the comment.

This is the code calling board_eth_init() from net/eth.c:

Quoted 7 lines
        /* Try board-specific initialization first.  If it fails or isn't
         * present, try the cpu-specific initialization */
        if (board_eth_init(bis) < 0)
                cpu_eth_init(bis);

So if we return with an error in board_eth_init(), cpu_eth_init() will get called again. Another way to fix this problem would be this implementation:

I consider this a buggy design that should be fixed. It should be possible to handle the situation that pci_eth_init() returns an error code.

Quoted 10 lines

board_eth_init()

{
	pci_eth_init(bis);

	/*
	 * Return -1 so that cpu_eth_init() will get executed in net/eth.c
	 */
	return -1;
}
Quoted 2 lines

But I like the former implementation better, since the cpu internal ethernet interfaces are added first to the network devices list.

That would be as bad as the previous solution, or actually worse as it looks as if board_eth_init() was always failing.

I think the key problems is here:

Quoted 4 lines
        /* Try board-specific initialization first.  If it fails or isn't
         * present, try the cpu-specific initialization */
        if (board_eth_init(bis) < 0)
                cpu_eth_init(bis);

I think we must differentiate between board_eth_init() not existing and a failure in board_eth_init(); these are two very different situations.

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
"Well I don't see why I have to make one man  miserable  when  I  can
make so many men happy."              - Ellyn Mustard, about marriage
#4 lore raw

PATCH 1/1 ppc4xx: Fix problem with board_eth_init() vs cpu_eth_init() on AMCC boards

(Added Ben to CC)

Quoted 18 lines · Wolfgang Denk
Quoted 18 lines
Quoted 10 lines

board/amcc/katmai/katmai.c int post_hotkeys_pressed(void) · line 460

Quoted 7 lines
board/amcc/katmai/katmai.c +7 -0
+	cpu_eth_init(bis);
+	pci_eth_init(bis);
+
+	/*
+	 * Return 0 so that cpu_eth_init() won't get executed again
+	 */
+	return 0;

What happens in case of errors? This looks broken to me, or I misinderstand the comment.

This is the code calling board_eth_init() from net/eth.c:

        /* Try board-specific initialization first.  If it fails or isn't
         * present, try the cpu-specific initialization */
        if (board_eth_init(bis) < 0)
                cpu_eth_init(bis);
Quoted 7 lines
Quoted 3 lines

So if we return with an error in board_eth_init(), cpu_eth_init() will get called again. Another way to fix this problem would be this implementation:

I consider this a buggy design that should be fixed. It should be possible to handle the situation that pci_eth_init() returns an error code.

pci_eth_init() is called to add additional *optional* network interfaces. Since PCI boards may or may not exist, I think that a non existant PCI ethernet device should not result in an error.

What sort of error handling do you have in mind here?

Quoted 10 lines
Quoted 10 lines

board_eth_init()

{
	pci_eth_init(bis);

	/*
	 * Return -1 so that cpu_eth_init() will get executed in net/eth.c
	 */
	return -1;
}
Quoted 5 lines
Quoted 2 lines

But I like the former implementation better, since the cpu internal ethernet interfaces are added first to the network devices list.

That would be as bad as the previous solution, or actually worse as it looks as if board_eth_init() was always failing.

Right. That's also one reason why I implemented the first version.

Quoted 9 lines

I think the key problems is here:

Quoted 4 lines
        /* Try board-specific initialization first.  If it fails or isn't
         * present, try the cpu-specific initialization */
        if (board_eth_init(bis) < 0)
                cpu_eth_init(bis);

I think we must differentiate between board_eth_init() not existing and a failure in board_eth_init(); these are two very different situations.

board_eth_init() not existing is the default. We have a weak implementation for board_eth_init() in eth.c:

/*
* CPU and board-specific Ethernet initializations. Aliased function
* signals caller to move on
*/
static int __def_eth_init(bd_t *bis)

{
        return -1;
}

int cpu_eth_init(bd_t *bis) __attribute((weak, alias("__def_eth_init"))); int board_eth_init(bd_t *bis) __attribute((weak, alias("__def_eth_init")));

What change do you have in mind here?

Best regards,
Stefan

=====================================================================
DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: office at denx.de
=====================================================================

#5 lore raw

PATCH 1/1 ppc4xx: Fix problem with board_eth_init() vs cpu_eth_init() on AMCC boards

Dear Stefan Roese,

Quoted 6 lines · In message <[email protected]> you wrote:

pci_eth_init() is called to add additional *optional* network interfaces. Since PCI boards may or may not exist, I think that a non existant PCI ethernet device should not result in an error.

What sort of error handling do you have in mind here?

I don;t know. I just see that pci_eth_init() is declared to return "int". If the return value is not used, we should make the function void.

Quoted 17 lines
Quoted 8 lines
Quoted 4 lines
        /* Try board-specific initialization first.  If it fails or isn't
         * present, try the cpu-specific initialization */
        if (board_eth_init(bis) < 0)
                cpu_eth_init(bis);

I think we must differentiate between board_eth_init() not existing and a failure in board_eth_init(); these are two very different situations.

board_eth_init() not existing is the default. We have a weak implementation for board_eth_init() in eth.c:

/*
* CPU and board-specific Ethernet initializations. Aliased function
* signals caller to move on
*/
static int __def_eth_init(bd_t *bis)

Quoted 8 lines
{
        return -1;
}

int cpu_eth_init(bd_t *bis) __attribute((weak, alias("__def_eth_init"))); int board_eth_init(bd_t *bis) __attribute((weak, alias("__def_eth_init")));

What change do you have in mind here?

I don't understand the intended logic. As far as I can see, pci_eth_init() will return the number of initialized boards, i. e. a non-negative value. Then having board_eth_init() reuturn the same return code should be no problem - you get -1 (i. e. < 0) if board_eth_init() does not exist, and >= 0 else.

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
No journaling file system can recover your data if the disk dies.
                                 - Steve Rago in <[email protected]>
#6 lore raw

PATCH 1/1 ppc4xx: Fix problem with board_eth_init() vs cpu_eth_init() on AMCC boards

In reply to Wolfgang Denk

Hi Stefan/Wolfgang,

Quoted 16 lines · Wolfgang Denk

Dear Stefan,

Quoted 12 lines · In message <[email protected]> you wrote:
Quoted 11 lines

board/amcc/katmai/katmai.c int post_hotkeys_pressed(void) · line 460

Quoted 8 lines

+ cpu_eth_init(bis);
+ pci_eth_init(bis);

board/amcc/katmai/katmai.c +5 -0
+
+ /*
+  * Return 0 so that cpu_eth_init() won't get executed again
+  */
+ return 0;

What happens in case of errors? This looks broken to me, or I misinderstand the comment.

I think this is right, although the undocumented return code of 0 may not be (more on this later in this e-mail)

Quoted 8 lines
Quoted 8 lines

This is the code calling board_eth_init() from net/eth.c:

        /* Try board-specific initialization first.  If it fails or isn't
         * present, try the cpu-specific initialization */
        if (board_eth_init(bis) < 0)
                cpu_eth_init(bis);
Quoted 9 lines
Quoted 1 line

So if we return with an error in board_eth_init(), cpu_eth_init() will

get

Quoted 1 line

called again. Another way to fix this problem would be this

implementation:

I consider this a buggy design that should be fixed. It should be possible to handle the situation that pci_eth_init() returns an error code.

What handling should there be in the case of a device initialization error? Drivers currently printf something and don't register the device. What else is appropriate?

Quoted 20 lines
Quoted 11 lines

board_eth_init()

{
      pci_eth_init(bis);

      /*
       * Return -1 so that cpu_eth_init() will get executed in net/eth.c
       */
      return -1;
}
Quoted 1 line

But I like the former implementation better, since the cpu internal

ethernet

Quoted 1 line

interfaces are added first to the network devices list.

That would be as bad as the previous solution, or actually worse as it looks as if board_eth_init() was always failing.

I think the key problems is here:

Quoted 9 lines
Quoted 4 lines
        /* Try board-specific initialization first.  If it fails or isn't
         * present, try the cpu-specific initialization */
        if (board_eth_init(bis) < 0)
                cpu_eth_init(bis);

I think we must differentiate between board_eth_init() not existing and a failure in board_eth_init(); these are two very different situations.

Sorry this is confusing. The -1 return code currently means "does not exist", although there are a few individual drivers that return -1 on failure, and should be cleaned up. All of the PCI drivers return the number of interfaces found (i.e. return 0 if nothing is found or there's an error). While this flaunts UNIX convention, it has the advantage of letting us count the interfaces at initialization without stepping through the list. I was thinking of taking advantage of this to get rid of the CONFIG_NET_MULTI thing some time in the future.

Quoted 5 lines

Best regards,

Wolfgang Denk

regards,
Ben