gpu: nvgpu: add QoS notifier for common clk framework

Define specific QoS notifier for common clk framework
and protect it with CONFIG_COMMON_CLK

This new API will first get min/max requirements from
pm_qos and set min/max freq values in devfreq

A call to update_devfreq() will then ensure that
new estimated frequency is clipped appropriately
between min and max values
This also ensures that frequency is set along with
all the book-keeping

Add below platform specific notifier callback and use it
with pm_qos_add_notifier()
int (*qos_notify)()
If qos_notify is set, then only register the callback

We currently support only one qos_id which is treated
as notifier for min frequency
Remove dependency on qos_id, and use appropriate QoS
APIs like pm_qos_read_min/max_bound()

Store devfreq's min/max frequency in struct gk20a
for reference

Bug 1772462

Change-Id: I63d6d17451d19c9d376b67df7db775b38929287d
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/1161161
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>

5 files changed, 69 insertions, 19 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.h b/drivers/gpu/nvgpu/gk20a/gk20a.h
index 997da125..9acaa007 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.h
@@ -815,6 +815,8 @@ struct gk20a {
         wait_queue_head_t sw_irq_nonstall_last_handled_wq;
 
         struct devfreq *devfreq;
+        u32 devfreq_max_freq;
+        u32 devfreq_min_freq;
 
         struct gk20a_scale_profile *scale_profile;
 
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a_scale.c b/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
index 6d60d899..eeabd919 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a_scale.c
@@ -41,8 +41,44 @@
  * has changed. The function calls postscaling callback if it is defined.
  */
 
-static int gk20a_scale_qos_notify(struct notifier_block *nb,
-                                  unsigned long n, void *p)
+#if defined(CONFIG_COMMON_CLK)
+int gk20a_scale_qos_notify(struct notifier_block *nb,
+                          unsigned long n, void *p)
+{
+        struct gk20a_scale_profile *profile =
+                        container_of(nb, struct gk20a_scale_profile,
+                        qos_notify_block);
+        struct gk20a *g = get_gk20a(profile->dev);
+        struct devfreq *devfreq = g->devfreq;
+        s32 min_qos_freq, max_qos_freq;
+
+        if (!devfreq)
+                return NOTIFY_OK;
+
+        /* check for pm_qos min and max frequency requirement */
+        min_qos_freq = pm_qos_read_min_bound(PM_QOS_GPU_FREQ_BOUNDS) * 1000;
+        max_qos_freq = pm_qos_read_max_bound(PM_QOS_GPU_FREQ_BOUNDS) * 1000;
+
+        mutex_lock(&devfreq->lock);
+
+        devfreq->min_freq = max_t(u32, g->devfreq_min_freq, min_qos_freq);
+        devfreq->max_freq = min_t(u32, g->devfreq_max_freq, max_qos_freq);
+
+        WARN_ON(devfreq->min_freq > devfreq->max_freq);
+
+        /*
+         * update_devfreq() will adjust the current (or newly estimated)
+         * frequency based on devfreq->min_freq/max_freq
+         */
+        update_devfreq(devfreq);
+
+        mutex_unlock(&devfreq->lock);
+
+        return NOTIFY_OK;
+}
+#else
+int gk20a_scale_qos_notify(struct notifier_block *nb,
+                          unsigned long n, void *p)
 {
         struct gk20a_scale_profile *profile =
                 container_of(nb, struct gk20a_scale_profile,
@@ -57,7 +93,7 @@ static int gk20a_scale_qos_notify(struct notifier_block *nb,
         /* get the frequency requirement. if devfreq is enabled, check if it
          * has higher demand than qos */
         freq = platform->clk_round_rate(profile->dev,
-                                pm_qos_request(platform->qos_id));
+                        (u32)pm_qos_read_min_bound(PM_QOS_GPU_FREQ_BOUNDS));
         if (g->devfreq)
                 freq = max(g->devfreq->previous_freq, freq);
 
@@ -68,6 +104,7 @@ static int gk20a_scale_qos_notify(struct notifier_block *nb,
 
         return NOTIFY_OK;
 }
+#endif
 
 /*
  * gk20a_scale_make_freq_table(profile)
@@ -311,16 +348,19 @@ void gk20a_scale_init(struct device *dev)
                         devfreq = NULL;
 
                 g->devfreq = devfreq;
+                g->devfreq_max_freq = devfreq->max_freq;
+                g->devfreq_min_freq = devfreq->min_freq;
         }
 
         /* Should we register QoS callback for this device? */
-        if (platform->qos_id < PM_QOS_NUM_CLASSES &&
-            platform->qos_id != PM_QOS_RESERVED &&
-            platform->postscale) {
+        if (platform->qos_notify) {
                 profile->qos_notify_block.notifier_call =
-                        &gk20a_scale_qos_notify;
-                pm_qos_add_notifier(platform->qos_id,
-                                    &profile->qos_notify_block);
+                                        platform->qos_notify;
+
+                pm_qos_add_min_notifier(PM_QOS_GPU_FREQ_BOUNDS,
+                                        &profile->qos_notify_block);
+                pm_qos_add_max_notifier(PM_QOS_GPU_FREQ_BOUNDS,
+                                        &profile->qos_notify_block);
         }
 
         return;
@@ -335,10 +375,10 @@ void gk20a_scale_exit(struct device *dev)
         struct gk20a *g = platform->g;
         int err;
 
-        if (platform->qos_id < PM_QOS_NUM_CLASSES &&
-            platform->qos_id != PM_QOS_RESERVED &&
-            platform->postscale) {
-                pm_qos_remove_notifier(platform->qos_id,
+        if (platform->qos_notify) {
+                pm_qos_remove_min_notifier(PM_QOS_GPU_FREQ_BOUNDS,
+                                &g->scale_profile->qos_notify_block);
+                pm_qos_remove_max_notifier(PM_QOS_GPU_FREQ_BOUNDS,
                                 &g->scale_profile->qos_notify_block);
         }
 
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a_scale.h b/drivers/gpu/nvgpu/gk20a/gk20a_scale.h
index 5c8618eb..025c2070 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a_scale.h
+++ b/drivers/gpu/nvgpu/gk20a/gk20a_scale.h
@@ -47,11 +47,18 @@ void gk20a_scale_notify_idle(struct device *);
 
 void gk20a_scale_suspend(struct device *);
 void gk20a_scale_resume(struct device *);
+int gk20a_scale_qos_notify(struct notifier_block *nb,
+                        unsigned long n, void *p);
 #else
 static inline void gk20a_scale_notify_busy(struct device *dev) {}
 static inline void gk20a_scale_notify_idle(struct device *dev) {}
 static inline void gk20a_scale_suspend(struct device *dev) {}
 static inline void gk20a_scale_resume(struct device *dev) {}
+static inline int gk20a_scale_qos_notify(struct notifier_block *nb,
+                        unsigned long n, void *p)
+{
+        return -ENOSYS;
+}
 #endif
 
 #endif
diff --git a/drivers/gpu/nvgpu/gk20a/platform_gk20a.h b/drivers/gpu/nvgpu/gk20a/platform_gk20a.h
index 66d97915..a753201b 100644
--- a/drivers/gpu/nvgpu/gk20a/platform_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/platform_gk20a.h
@@ -181,10 +181,11 @@ struct gk20a_platform {
          * this governor to be used in scaling */
         const char *devfreq_governor;
 
-        /* Quality of service id. If this is set, the scaling routines
-         * will register a callback to id. Each time we receive a new value,
-         * the postscale callback gets called.  */
-        int qos_id;
+        /* Quality of service notifier callback. If this is set, the scaling
+         * routines will register a callback to Qos. Each time we receive
+         * a new value, this callback gets called.  */
+        int (*qos_notify)(struct notifier_block *nb,
+                          unsigned long n, void *p);
 
         /* Called as part of debug dump. If the gpu gets hung, this function
          * is responsible for delivering all necessary debug data of other
diff --git a/drivers/gpu/nvgpu/gk20a/platform_gk20a_tegra.c b/drivers/gpu/nvgpu/gk20a/platform_gk20a_tegra.c
index 618b1716..b8f70ab3 100644
--- a/drivers/gpu/nvgpu/gk20a/platform_gk20a_tegra.c
+++ b/drivers/gpu/nvgpu/gk20a/platform_gk20a_tegra.c
@@ -896,7 +896,7 @@ struct gk20a_platform gk20a_tegra_platform = {
         .prescale = gk20a_tegra_prescale,
         .postscale = gk20a_tegra_postscale,
         .devfreq_governor = "nvhost_podgov",
-        .qos_id = PM_QOS_GPU_FREQ_MIN,
+        .qos_notify = gk20a_scale_qos_notify,
 
         .secure_alloc = gk20a_tegra_secure_alloc,
         .secure_page_alloc = gk20a_tegra_secure_page_alloc,
@@ -956,7 +956,7 @@ struct gk20a_platform gm20b_tegra_platform = {
         .prescale = gk20a_tegra_prescale,
         .postscale = gk20a_tegra_postscale,
         .devfreq_governor = "nvhost_podgov",
-        .qos_id = PM_QOS_GPU_FREQ_MIN,
+        .qos_notify = gk20a_scale_qos_notify,
 
         .secure_alloc = gk20a_tegra_secure_alloc,
         .secure_page_alloc = gk20a_tegra_secure_page_alloc,