Support omap_pm_get_dev_context_loss_count() for OMAP4

Now that OMAP4 has hwmod/module level context lost counters, make omap_hwmod_get_context_loss_count() return them on platforms where they exist, else fall back on the pwrdm level counters for older platforms.

Signed-off-by Rajendra Nayak<[email protected]>

---
  arch/arm/mach-omap2/omap_hwmod.c |   11 ++++++++---
  1 files changed, 8 insertions(+), 3 deletions(-)

arch/arm/mach-omap2/omap_hwmod.c +0 -0

index ac17748..9d140c4 100644

Hi Rajendra,

Sorry for the delay, I kind of forgot your series :-(

Quoted 19 lines · Rajendra

Now that OMAP4 has hwmod/module level context lost counters, make omap_hwmod_get_context_loss_count() return them on platforms where they exist, else fall back on the pwrdm level counters for older platforms.

Signed-off-by Rajendra Nayak<[email protected]>

---
arch/arm/mach-omap2/omap_hwmod.c | 11 ++++++++---

1 files changed, 8 insertions(+), 3 deletions(-)

arch/arm/mach-omap2/omap_hwmod.c +0 -0

index ac17748..9d140c4 100644
@@ -2658,9 +2658,14 @@ u32 omap_hwmod_get_context_loss_count(struct
 omap_hwmod *oh)
 struct powerdomain *pwrdm;
 int ret = 0;
 

arch/arm/mach-omap2/omap_hwmod.c u32 omap_hwmod_get_context_loss_count(struct omap_hwmod *oh) · line 2661

Quoted 4 lines

arch/arm/mach-omap2/omap_hwmod.c +1 -3

- pwrdm = omap_hwmod_get_pwrdm(oh);
- if (pwrdm)
- ret = pwrdm_get_context_loss_count(pwrdm);
+ if (oh->prcm.omap4.context_offs) {

I don't think it is safe to use that, since this is an union. You do know what OMAP2 & 3 will put in that location. It might be zero, but who knows?

Quoted 11 lines

arch/arm/mach-omap2/omap_hwmod.c +7 -0

+ /* Support for per-hwmod context register */
+ ret = oh->prcm.omap4.context_lost_counter;
+ } else {
+ pwrdm = omap_hwmod_get_pwrdm(oh);
+ if (pwrdm)
+ ret = pwrdm_get_context_loss_count(pwrdm);
+ }
 
 return ret;

}

Regards,
Benoit

+ * XXX Will not work correctly if the RM_*_CONTEXT register
+ * offset is 0 -- probably a flag should be used to avoid this
+ * situation, rather than testing @oh->prcm.omap4.context_offs. No
+ * return value.

Hi

On Fri, 16 Dec 2011, Paul Walmsley wrote:

Quoted 4 lines

+ * XXX Will not work correctly if the RM_*_CONTEXT register
+ * offset is 0 -- probably a flag should be used to avoid this
+ * situation, rather than testing @oh->prcm.omap4.context_offs. No
+ * return value.

By the way, I'm also worried about this case. Am wondering if we should add a flag here just to be sure, rather than testing the offset. Thoughts?

- Paul

Quoted 10 lines · Paul Walmsley

Quoted 4 lines · Paul Walmsley

+ * XXX Will not work correctly if the RM_*_CONTEXT register
+ * offset is 0 -- probably a flag should be used to avoid this
+ * situation, rather than testing @oh->prcm.omap4.context_offs. No
+ * return value.

By the way, I'm also worried about this case. Am wondering if we should add a flag here just to be sure, rather than testing the offset. Thoughts?

I am not sure if we even have any modules on OMAP4 which *do not* support module level context loss status and for which we need to fall back on pwrdm level status. So maybe we could even get rid of the check for a valid context_offset altogether and expect it to be populated for all modules as we do with the clkctrl offset.

Hi Rajendra,

I've combined your patches down to one patch and changed the logic a little bit. It looks to me that the update and clear context registers can be done in one step, since an IP block shouldn't lose context when it's enabled. But maybe I'm missing something?

Anyway, I'd appreciate any comments you might have, and if you have an environment where you can test this patch, that would also be very welcome.

- Paul

From: Rajendra Nayak<[email protected]>
Date: Fri, 16 Dec 2011 04:12:48 -0700
Subject: [PATCH] ARM: OMAP: hwmod: Add support for per hwmod/module context
lost count

OMAP4 has module specific context lost registers which makes it now possible to have module level context loss count, instead of relying on the powerdomain level context count.

Add 2 private hwmod api's to update/clear the hwmod/module specific context lost counters/register.

Update the module specific context_lost_counter and clear the hardware bits just after enabling the module.

omap_hwmod_get_context_loss_count() now returns the hwmod context loss count them on platforms where they exist (OMAP4), else fall back on the pwrdm level counters for older platforms.

Signed-off-by Rajendra Nayak<[email protected]>

[paul at pwsan.com: added function kerneldoc, fixed structure kerneldoc,
rearranged structure to avoid memory waste, marked fns as OMAP4-specific, prevent fn entry on non-OMAP4 chips, reduced indentation, merged update and clear, merged patches]

Signed-off-by Paul Walmsley<[email protected]>

---
  arch/arm/mach-omap2/omap_hwmod.c             |   54 ++++++++++++++++++++++++--
  arch/arm/plat-omap/include/plat/omap_hwmod.h |    8 +++-
  2 files changed, 56 insertions(+), 6 deletions(-)

arch/arm/mach-omap2/omap_hwmod.c +50 -4

index 0a89335..5702462 100644
@@ -1435,6 +1435,42 @@ static int _reset(struct omap_hwmod *oh)
  }
 
  /**
+ * _omap4_update_context_lost - increment hwmod context loss counter if
+ * hwmod context was lost, and clear hardware context loss reg
+ * @oh: hwmod to check for context loss
+ *
+ * If the PRCM indicates that the hwmod @oh lost context, increment
+ * our in-memory context loss counter, and clear the RM_*_CONTEXT
+ * bits.  XXX Will not work correctly if the RM_*_CONTEXT register
+ * offset is 0 -- probably a flag should be used to avoid this
+ * situation, rather than testing @oh->prcm.omap4.context_offs.  No
+ * return value.
+ */
+static void _omap4_update_context_lost(struct omap_hwmod *oh)
+{
+	u32 r;
+
+	if (!cpu_is_omap44xx())
+		return;
+
+	if (!oh->prcm.omap4.context_offs)
+		return;
+
+	r = omap4_prminst_read_inst_reg(oh->clkdm->pwrdm.ptr->prcm_partition,
+					oh->clkdm->pwrdm.ptr->prcm_offs,
+					oh->prcm.omap4.context_offs);
+
+	if (!r)
+		return;
+
+	oh->prcm.omap4.context_lost_counter++;
+
+	omap4_prminst_write_inst_reg(r, oh->clkdm->pwrdm.ptr->prcm_partition,
+				     oh->clkdm->pwrdm.ptr->prcm_offs,
+				     oh->prcm.omap4.context_offs);
+}
+
+/**
   * _enable - enable an omap_hwmod
   * @oh: struct omap_hwmod *
   *

@@ -1512,6 +1548,8 @@ static int _enable(struct omap_hwmod *oh)
  	_enable_clocks(oh);
  	_enable_module(oh);
 
+	_omap4_update_context_lost(oh);
+
  	r = _wait_target_ready(oh);
  	if (!r) {
  		/*

@@ -2640,17 +2678,25 @@ ohsps_unlock:
   * omap_hwmod_get_context_loss_count - get lost context count
   * @oh: struct omap_hwmod *
   *
- * Query the powerdomain of of @oh to get the context loss
- * count for this device.
+ * Returns the context loss count of associated with @oh
+ * upon success, or zero if no context loss data is available.
   *
- * Returns the context loss count of the powerdomain assocated with @oh
- * upon success, or zero if no powerdomain exists for @oh.
+ * On OMAP4, this queries the per-hwmod context loss register,
+ * assuming one exists.  If not, or on OMAP2/3, this queries the
+ * enclosing powerdomain context loss count.
   */
  int omap_hwmod_get_context_loss_count(struct omap_hwmod *oh)
  {
  	struct powerdomain *pwrdm;
  	int ret = 0;
 
+	/*
+	 * XXX This should use a flag to determine whether to check
+	 * context_offs, as context_offs == 0 is potentially valid
+	 */
+	if (cpu_is_omap44xx()&&  oh->prcm.omap4.context_offs)
+		return oh->prcm.omap4.context_lost_counter;
+
  	pwrdm = omap_hwmod_get_pwrdm(oh);
  	if (pwrdm)
  		ret = pwrdm_get_context_loss_count(pwrdm);

arch/arm/plat-omap/include/plat/omap_hwmod.h +6 -2

index 1a13c02..3aea95e 100644
@@ -361,9 +361,12 @@ struct omap_hwmod_omap2_prcm {
 
  /**
   * struct omap_hwmod_omap4_prcm - OMAP4-specific PRCM data
- * @clkctrl_reg: PRCM address of the clock control register
- * @rstctrl_reg: address of the XXX_RSTCTRL register located in the PRM
+ * @clkctrl_offs: offset of the PRCM clock control register
+ * @rstctrl_offs: offset of the XXX_RSTCTRL register located in the PRM
+ * @context_offs: offset of the RM_*_CONTEXT register
   * @submodule_wkdep_bit: bit shift of the WKDEP range
+ * @modulemode: allowable modulemodes
+ * @context_lost_counter: Count of module level context lost
   */
  struct omap_hwmod_omap4_prcm {
  	u16		clkctrl_offs;

@@ -371,6 +374,7 @@ struct omap_hwmod_omap4_prcm {
  	u16		context_offs;
  	u8		submodule_wkdep_bit;
  	u8		modulemode;
+	unsigned	context_lost_counter;
  };
 
 

Quoted 3 lines · Paul Walmsley

Hi Rajendra,

Quoted 4 lines

I've combined your patches down to one patch and changed the logic a little bit. It looks to me that the update and clear context registers can be done in one step, since an IP block shouldn't lose context when it's enabled. But maybe I'm missing something?

Thanks for the patch. The changes look fine and I also did a quick test on an internal tree which supports core OSWR and it seems to work fine too.

Quoted 9 lines · Rajendra Nayak

Quoted 7 lines · Paul Walmsley

Hi Rajendra,

I've combined your patches down to one patch and changed the logic a little bit. It looks to me that the update and clear context registers can be done in one step, since an IP block shouldn't lose context when it's enabled. But maybe I'm missing something?

Quoted 3 lines

Thanks for the patch. The changes look fine and I also did a quick test on an internal tree which supports core OSWR and it seems to work fine too.

Great, thanks. Will wait on this one until you have a chance to tinker with it during the 3.4 time frame.

- Paul