msm: kgsl: Dynamic AB voting based on actual bus transactions

Right now AB vote is calulated as a percentage of total possible BW
which causes higher AB vote than actual GPU bus transactions in some
use cases. We can calculate actual GPU bus BW if we know bus_width
for the target. This change will derive AB vote based on bus_width
defined in dtsi and VBIF counters.

Define bus_width in number of bytes in target gpu dtsi file to use
this new AB vote calculation.

Change-Id: I75d19c18649d9a87d20e3dbf7b623b772265fb5b
Signed-off-by: Mihir Patel <mihirp@codeaurora.org>

Documentation/devicetree/bindings/gpu/adreno.txt

@@ -82,6 +82,8 @@ Optional Properties:
 - qcom,initial-powerlevel: This value indicates which qcom,gpu-pwrlevel should be used at start time
 			   and when coming back out of resume
 - qcom,bus-control:	   Boolean. Enables an independent bus vote from the gpu frequency
+- qcom,bus-width:	   Bus width in number of bytes. This enables dynamic AB bus voting based on
+			   bus width and actual bus transactions.
 - qcom,gpubw-dev:	   a phandle to a device representing bus bandwidth requirements
 			   (see devdw.txt)
 - qcom,idle-timeout:	   This property represents the time in microseconds for idle timeout.

drivers/devfreq/governor_gpubw_mon.c

@@ -1,4 +1,4 @@
-/* Copyright (c) 2014, The Linux Foundation. All rights reserved.
+/* Copyright (c) 2014-2015, The Linux Foundation. All rights reserved.
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License version 2 and
@@ -23,6 +23,8 @@
 #define HIST                    5
 #define TARGET                  80
 #define CAP                     75
+/* AB vote is in multiple of BW_STEP Mega bytes */
+#define BW_STEP                 160
 
 static void _update_cutoff(struct devfreq_msm_adreno_tz_data *priv,
 					unsigned int norm_max)
@@ -59,6 +61,7 @@ static int devfreq_gpubw_get_target(struct devfreq *df,
 	 */
 	static int norm_ab_max = 300;
 	int norm_ab;
+	unsigned long ab_mbytes = 0;
 
 	stats.private_data = &b;
 
@@ -103,8 +106,15 @@ static int devfreq_gpubw_get_target(struct devfreq *df,
 			bus_profile->flag = DEVFREQ_FLAG_SLOW_HINT;
 	}
 
-	/* Re-calculate the AB percentage for a new IB vote */
-	if (bus_profile->flag) {
+	/* Calculate the AB vote based on bus width if defined */
+	if (priv->bus.width) {
+		norm_ab =  (unsigned int)priv->bus.ram_time /
+			(unsigned int) priv->bus.total_time;
+		/* Calculate AB in Mega Bytes and roundup in BW_STEP */
+		ab_mbytes = (norm_ab * priv->bus.width * 1000000ULL) >> 20;
+		bus_profile->ab_mbytes = roundup(ab_mbytes, BW_STEP);
+	} else if (bus_profile->flag) {
+		/* Re-calculate the AB percentage for a new IB vote */
 		norm_ab =  (unsigned int)priv->bus.ram_time /
 			(unsigned int) priv->bus.total_time;
 		if (norm_ab > norm_ab_max)

drivers/gpu/msm/kgsl_pwrctrl.c

@@ -168,8 +168,10 @@ static void _ab_buslevel_update(struct kgsl_pwrctrl *pwr,
 		return;
 	if (ib == 0)
 		*ab = 0;
-	else if (!pwr->bus_percent_ab)
+	else if ((!pwr->bus_percent_ab) && (!pwr->bus_ab_mbytes))
 		*ab = DEFAULT_BUS_P * ib / 100;
+	else if (pwr->bus_width)
+		*ab = pwr->bus_ab_mbytes;
 	else
 		*ab = (pwr->bus_percent_ab * max_bw) / 100;
 
@@ -250,6 +252,8 @@ void kgsl_pwrctrl_buslevel_update(struct kgsl_device *device,
 	} else {
 		/* If the bus is being turned off, reset to default level */
 		pwr->bus_mod = 0;
+		pwr->bus_percent_ab = 0;
+		pwr->bus_ab_mbytes = 0;
 	}
 	trace_kgsl_buslevel(device, pwr->active_pwrlevel, buslevel);
 	last_vote_buslevel = buslevel;
@@ -1553,6 +1557,10 @@ int kgsl_pwrctrl_init(struct kgsl_device *device)
 
 	/* Set if independent bus BW voting is supported */
 	pwr->bus_control = pdata->bus_control;
+	/* Bus width in bytes, set it to zero if not found */
+	if (of_property_read_u32(pdev->dev.of_node, "qcom,bus-width",
+		&pwr->bus_width))
+		pwr->bus_width = 0;
 
 	/* Check if gpu bandwidth vote device is defined in dts */
 	gpubw_dev_node = of_parse_phandle(pdev->dev.of_node,

drivers/gpu/msm/kgsl_pwrctrl.h

@@ -108,6 +108,8 @@ struct kgsl_pwr_constraint {
  * @bus_control - true if the bus calculation is independent
  * @bus_mod - modifier from the current power level for the bus vote
  * @bus_percent_ab - current percent of total possible bus usage
+ * @bus_width - target specific bus width in number of bytes
+ * @bus_ab_mbytes - AB vote in Mbytes for current bus usage
  * @bus_index - default bus index into the bus_ib table
  * @bus_ib - the set of unique ib requests needed for the bus calculation
  * @constraint - currently active power constraint
@@ -152,6 +154,8 @@ struct kgsl_pwrctrl {
 	bool bus_control;
 	int bus_mod;
 	unsigned int bus_percent_ab;
+	unsigned int bus_width;
+	unsigned long bus_ab_mbytes;
 	struct device *devbw;
 	unsigned int bus_index[KGSL_MAX_PWRLEVELS];
 	uint64_t *bus_ib;

drivers/gpu/msm/kgsl_pwrscale.c

@@ -715,6 +715,7 @@ int kgsl_busmon_target(struct device *dev, unsigned long *freq, u32 flags)
 
 	if (pwr->bus_mod != b) {
 		pwr->bus_percent_ab = device->pwrscale.bus_profile.percent_ab;
+		pwr->bus_ab_mbytes =  device->pwrscale.bus_profile.ab_mbytes;
 		kgsl_pwrctrl_buslevel_update(device, true);
 	}
 
@@ -797,6 +798,7 @@ int kgsl_pwrscale_init(struct device *dev, const char *governor)
 		data->bus.num = out;
 		data->bus.ib = &pwr->bus_ib[0];
 		data->bus.index = &pwr->bus_index[0];
+		data->bus.width = pwr->bus_width;
 	} else
 		data->bus.num = 0;
 

include/linux/msm_adreno_devfreq.h

@@ -36,6 +36,7 @@ struct devfreq_msm_adreno_tz_data {
 		u64 gpu_time;
 		u32 num;
 		u32 max;
+		u32 width;
 		u32 up[MSM_ADRENO_MAX_PWRLEVELS];
 		u32 down[MSM_ADRENO_MAX_PWRLEVELS];
 		u32 p_up[MSM_ADRENO_MAX_PWRLEVELS];
@@ -61,6 +62,7 @@ struct msm_adreno_extended_profile {
 struct msm_busmon_extended_profile {
 	u32 flag;
 	unsigned long percent_ab;
+	unsigned long ab_mbytes;
 	struct devfreq_msm_adreno_tz_data *private_data;
 	struct devfreq_dev_profile profile;
 };