Skip to contentSkip to conversation

Should I worry about "omap-mcbsp.2: alias fck already exists" or is it expected?

PATCH NeilBrown 3 participants 3 messages Dec 11, 2011 – Dec 16, 2011

Subsystems (2)
lore
Share & tools
lore

NeilBrown

author
#0 lore raw

In 3.2-rc5 (and some earlier kernels) I'm getting the boot-time warning:

[    0.186828]  omap-mcbsp.2: alias fck already exists

and

[    0.188476]  omap-mcbsp.3: alias fck already exists

This happens because omap_alloc_device() contains:

	for (i = 0; i < oh_cnt; i++) {
		hwmods[i]->od = od;
		_add_hwmod_clocks_clkdev(od, hwmods[i]);
	}

so if oh_cnt is ever > 1 (which is is for mcbsp.2 and .3 in omap_hwmod_3xxx_data.c as they both declare a '.devattr'), then _add_hwmod_clocks_clkdev will be called twice with the one 'od', and both calls will try to add an alias 'fsck' to that same device, and the second will always give a warning.

Is this actually a bug, or should we silence the warning on second and subsequent calls to omap_alloc_device()?

Like this?

Thanks,
NeilBrown

omap: don't try to register the main clock twice.

If omap_device_alloc is given 2 or more "struct omap_hwmod" it will try to register the 'main_clk' of each of them with the same alias - "fck" - against the same device. This fails.
So to avoid a warning, don't even try.

arch/arm/plat-omap/omap_device.c +5 -3
index e8d9869..11012bd 100644
@@ -294,6 +294,7 @@ static void _add_clkdev(struct omap_device *od, const char *clk_alias,
  * and main clock
  * @od: struct omap_device *od
  * @oh: struct omap_hwmod *oh
+ * @sub: this is a subordinate device, so don't try to register fck
  *
  * For the main clock and every optional clock present per hwmod per
  * omap_device, this function adds an entry in the clkdev table of the
@@ -309,11 +310,12 @@ static void _add_clkdev(struct omap_device *od, const char *clk_alias,
  * No return value.
  */
 static void _add_hwmod_clocks_clkdev(struct omap_device *od,
-				     struct omap_hwmod *oh)
+				     struct omap_hwmod *oh, int sub)
 {
 	int i;
 
-	_add_clkdev(od, "fck", oh->main_clk);
+	if (!sub)
+		_add_clkdev(od, "fck", oh->main_clk);
 
 	for (i = 0; i < oh->opt_clks_cnt; i++)
 		_add_clkdev(od, oh->opt_clks[i].role, oh->opt_clks[i].clk);
@@ -576,7 +578,7 @@ static struct omap_device *omap_device_alloc(struct platform_device *pdev,
 
 	for (i = 0; i < oh_cnt; i++) {
 		hwmods[i]->od = od;
-		_add_hwmod_clocks_clkdev(od, hwmods[i]);
+		_add_hwmod_clocks_clkdev(od, hwmods[i], i);
 	}
 
 	return od;
#1 lore raw

+ Paul and Peter

Hi Neil,

Quoted 17 lines · NeilBrown

In 3.2-rc5 (and some earlier kernels) I'm getting the boot-time warning:

[    0.186828]  omap-mcbsp.2: alias fck already exists

and

[    0.188476]  omap-mcbsp.3: alias fck already exists

This happens because omap_alloc_device() contains:

	for (i = 0; i<  oh_cnt; i++) {
		hwmods[i]->od = od;
		_add_hwmod_clocks_clkdev(od, hwmods[i]);
	}
Quoted 7 lines

so if oh_cnt is ever> 1 (which is is for mcbsp.2 and .3 in omap_hwmod_3xxx_data.c as they both declare a '.devattr'), then _add_hwmod_clocks_clkdev will be called twice with the one 'od', and both calls will try to add an alias 'fsck' to that same device, and the second will always give a warning.

Is this actually a bug,

It is indeed a bug. The point is that in theory most hwmod should have an unique device to represent them. But in the case of the Mcbsp2 and 3, due to the tightly coupled sidetone, and because each one has a dedicated OCP port, it was decided to have 2 hwmods but only one omap_device.

I'm wondering now if another device should not be considered, but I'll let Peter decide for the McBSP driver organization.

Quoted 2 lines

or should we silence the warning on second and
subsequent calls to omap_alloc_device()?

We can have only one "fck" alias per device, so preventing to register any subsequent fck seems the good approach to me. The warning was added to detect a potential static version of the clkdev and thus eliminate it, so we should keep it.

Quoted 6 lines

Like this?

Thanks,
NeilBrown

omap: don't try to register the main clock twice.

ARM: OMAP2+: Don't try to register the main clock twice

Quoted 6 lines

If omap_device_alloc is given 2 or more "struct omap_hwmod" it will try to register the 'main_clk' of each of them with the same alias - "fck" - against the same device. This fails.
So to avoid a warning, don't even try.

Quoted 28 lines
arch/arm/plat-omap/omap_device.c +4 -2
index e8d9869..11012bd 100644
@@ -294,6 +294,7 @@ static void _add_clkdev(struct omap_device *od, const char *clk_alias,
   * and main clock
   * @od: struct omap_device *od
   * @oh: struct omap_hwmod *oh
+ * @sub: this is a subordinate device, so don't try to register fck
   *
   * For the main clock and every optional clock present per hwmod per
   * omap_device, this function adds an entry in the clkdev table of the
@@ -309,11 +310,12 @@ static void _add_clkdev(struct omap_device *od, const char *clk_alias,
   * No return value.
   */
  static void _add_hwmod_clocks_clkdev(struct omap_device *od,
-				     struct omap_hwmod *oh)
+				     struct omap_hwmod *oh, int sub)
  {
  	int i;
 
-	_add_clkdev(od, "fck", oh->main_clk);
+	if (!sub)
+		_add_clkdev(od, "fck", oh->main_clk);
 
  	for (i = 0; i<  oh->opt_clks_cnt; i++)
  		_add_clkdev(od, oh->opt_clks[i].role, oh->opt_clks[i].clk);

arch/arm/plat-omap/omap_device.c static struct omap_device *omap_device_alloc(struct platform_device *pdev, · line 584

Quoted 8 lines
arch/arm/plat-omap/omap_device.c +1 -1
@@ -576,7 +578,7 @@ static struct omap_device *omap_device_alloc(struct platform_device *pdev,
 
  	for (i = 0; i<  oh_cnt; i++) {
  		hwmods[i]->od = od;
-		_add_hwmod_clocks_clkdev(od, hwmods[i]);
+		_add_hwmod_clocks_clkdev(od, hwmods[i], i);
  	}
 
  	return od;

Thanks for the fix.

Regards,
Benoit

#2 lore raw

(cc Jarkko)

Quoted 26 lines · Benoit
Quoted 17 lines · NeilBrown

In 3.2-rc5 (and some earlier kernels) I'm getting the boot-time warning:

[    0.186828]  omap-mcbsp.2: alias fck already exists

and

[    0.188476]  omap-mcbsp.3: alias fck already exists

This happens because omap_alloc_device() contains:

	for (i = 0; i<  oh_cnt; i++) {
		hwmods[i]->od = od;
		_add_hwmod_clocks_clkdev(od, hwmods[i]);
	}
Quoted 7 lines

so if oh_cnt is ever> 1 (which is is for mcbsp.2 and .3 in omap_hwmod_3xxx_data.c as they both declare a '.devattr'), then _add_hwmod_clocks_clkdev will be called twice with the one 'od', and both calls will try to add an alias 'fsck' to that same device, and the second will always give a warning.

Is this actually a bug,

Quoted 7 lines

It is indeed a bug. The point is that in theory most hwmod should have an unique device to represent them. But in the case of the Mcbsp2 and 3, due to the tightly coupled sidetone, and because each one has a dedicated OCP port, it was decided to have 2 hwmods but only one omap_device.

I'm wondering now if another device should not be considered, but I'll let Peter decide for the McBSP driver organization.

The sidetone blocks should definitely have separate devices & drivers, but AFAIK from Jarkko, this turns out to be not so easy due to an OMAP3 hardware bug:

http://www.spinics.net/lists/linux-omap/msg58101.html

So basically the interface clock autoidle handling needs to be fixed in the omap_hwmod code, so IP blocks with buggy autoidle handling can be marked in the hwmod data, and the hwmod code can control the interface clock autoidle bits along with the module's PM runtime status.

Patches welcome :-)

- Paul