media: venus: introduce core selection

Presently the core (vcodec pipelines) assignment is static. Here we
introduce dynamic load balancing across the cores depending on the
current session load. The load on earch core is calculated and core
with minimum load is assigned to given instance. This will be
applicable on Venus v4 with more than one vcodec cores.

Signed-off-by: Aniket Masule <amasule@codeaurora.org>
Co-developed-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>

1 files changed, 171 insertions, 12 deletions

diff --git a/drivers/media/platform/qcom/venus/pm_helpers.c b/drivers/media/platform/qcom/venus/pm_helpers.c
index 455797e16ffa..abf93158857b 100644
--- a/drivers/media/platform/qcom/venus/pm_helpers.c
+++ b/drivers/media/platform/qcom/venus/pm_helpers.c
@@ -14,6 +14,7 @@
 #include <media/v4l2-mem2mem.h>
 
 #include "core.h"
+#include "hfi_parser.h"
 #include "hfi_venus_io.h"
 #include "pm_helpers.h"
 
@@ -480,6 +481,173 @@ static int poweron_coreid(struct venus_core *core, unsigned int coreid_mask)
 	return 0;
 }
 
+static void
+min_loaded_core(struct venus_inst *inst, u32 *min_coreid, u32 *min_load)
+{
+	u32 mbs_per_sec, load, core1_load = 0, core2_load = 0;
+	u32 cores_max = core_num_max(inst);
+	struct venus_core *core = inst->core;
+	struct venus_inst *inst_pos;
+	unsigned long vpp_freq;
+	u32 coreid;
+
+	mutex_lock(&core->lock);
+
+	list_for_each_entry(inst_pos, &core->instances, list) {
+		if (inst_pos == inst)
+			continue;
+		vpp_freq = inst_pos->clk_data.codec_freq_data->vpp_freq;
+		coreid = inst_pos->clk_data.core_id;
+
+		mbs_per_sec = load_per_instance(inst_pos);
+		load = mbs_per_sec * vpp_freq;
+
+		if ((coreid & VIDC_CORE_ID_3) == VIDC_CORE_ID_3) {
+			core1_load += load / 2;
+			core2_load += load / 2;
+		} else if (coreid & VIDC_CORE_ID_1) {
+			core1_load += load;
+		} else if (coreid & VIDC_CORE_ID_2) {
+			core2_load += load;
+		}
+	}
+
+	*min_coreid = core1_load <= core2_load ?
+			VIDC_CORE_ID_1 : VIDC_CORE_ID_2;
+	*min_load = min(core1_load, core2_load);
+
+	if (cores_max < VIDC_CORE_ID_2 || core->res->vcodec_num < 2) {
+		*min_coreid = VIDC_CORE_ID_1;
+		*min_load = core1_load;
+	}
+
+	mutex_unlock(&core->lock);
+}
+
+static int decide_core(struct venus_inst *inst)
+{
+	const u32 ptype = HFI_PROPERTY_CONFIG_VIDEOCORES_USAGE;
+	struct venus_core *core = inst->core;
+	u32 min_coreid, min_load, inst_load;
+	struct hfi_videocores_usage_type cu;
+	unsigned long max_freq;
+
+	if (legacy_binding) {
+		if (inst->session_type == VIDC_SESSION_TYPE_DEC)
+			cu.video_core_enable_mask = VIDC_CORE_ID_1;
+		else
+			cu.video_core_enable_mask = VIDC_CORE_ID_2;
+
+		goto done;
+	}
+
+	if (inst->clk_data.core_id != VIDC_CORE_ID_DEFAULT)
+		return 0;
+
+	inst_load = load_per_instance(inst);
+	inst_load *= inst->clk_data.codec_freq_data->vpp_freq;
+	max_freq = core->res->freq_tbl[0].freq;
+
+	min_loaded_core(inst, &min_coreid, &min_load);
+
+	if ((inst_load + min_load) > max_freq) {
+		dev_warn(core->dev, "HW is overloaded, needed: %u max: %lu\n",
+			 inst_load, max_freq);
+		return -EINVAL;
+	}
+
+	inst->clk_data.core_id = min_coreid;
+	cu.video_core_enable_mask = min_coreid;
+
+done:
+	return hfi_session_set_property(inst, ptype, &cu);
+}
+
+static int acquire_core(struct venus_inst *inst)
+{
+	struct venus_core *core = inst->core;
+	unsigned int coreid_mask = 0;
+
+	if (inst->core_acquired)
+		return 0;
+
+	inst->core_acquired = true;
+
+	if (inst->clk_data.core_id & VIDC_CORE_ID_1) {
+		if (core->core0_usage_count++)
+			return 0;
+
+		coreid_mask = VIDC_CORE_ID_1;
+	}
+
+	if (inst->clk_data.core_id & VIDC_CORE_ID_2) {
+		if (core->core1_usage_count++)
+			return 0;
+
+		coreid_mask |= VIDC_CORE_ID_2;
+	}
+
+	return poweron_coreid(core, coreid_mask);
+}
+
+static int release_core(struct venus_inst *inst)
+{
+	struct venus_core *core = inst->core;
+	unsigned int coreid_mask = 0;
+	int ret;
+
+	if (!inst->core_acquired)
+		return 0;
+
+	if (inst->clk_data.core_id & VIDC_CORE_ID_1) {
+		if (--core->core0_usage_count)
+			goto done;
+
+		coreid_mask = VIDC_CORE_ID_1;
+	}
+
+	if (inst->clk_data.core_id & VIDC_CORE_ID_2) {
+		if (--core->core1_usage_count)
+			goto done;
+
+		coreid_mask |= VIDC_CORE_ID_2;
+	}
+
+	ret = poweroff_coreid(core, coreid_mask);
+	if (ret)
+		return ret;
+
+done:
+	inst->clk_data.core_id = VIDC_CORE_ID_DEFAULT;
+	inst->core_acquired = false;
+	return 0;
+}
+
+static int coreid_power_v4(struct venus_inst *inst, int on)
+{
+	struct venus_core *core = inst->core;
+	int ret;
+
+	if (legacy_binding)
+		return 0;
+
+	if (on == POWER_ON) {
+		ret = decide_core(inst);
+		if (ret)
+			return ret;
+
+		mutex_lock(&core->lock);
+		ret = acquire_core(inst);
+		mutex_unlock(&core->lock);
+	} else {
+		mutex_lock(&core->lock);
+		ret = release_core(inst);
+		mutex_unlock(&core->lock);
+	}
+
+	return ret;
+}
+
 static int vdec_get_v4(struct device *dev)
 {
 	struct venus_core *core = dev_get_drvdata(dev);
@@ -661,22 +829,12 @@ static void core_put_v4(struct device *dev)
 static int core_power_v4(struct device *dev, int on)
 {
 	struct venus_core *core = dev_get_drvdata(dev);
-	const unsigned int coreid_mask = VIDC_CORE_ID_1 | VIDC_CORE_ID_2;
 	int ret = 0;
 
-	if (on == POWER_ON) {
+	if (on == POWER_ON)
 		ret = core_clks_enable(core);
-		if (ret)
-			return ret;
-
-		if (!legacy_binding)
-			ret = poweron_coreid(core, coreid_mask);
-	} else {
-		if (!legacy_binding)
-			ret = poweroff_coreid(core, coreid_mask);
-
+	else
 		core_clks_disable(core);
-	}
 
 	return ret;
 }
@@ -781,6 +939,7 @@ static const struct venus_pm_ops pm_ops_v4 = {
 	.venc_get = venc_get_v4,
 	.venc_put = venc_put_v4,
 	.venc_power = venc_power_v4,
+	.coreid_power = coreid_power_v4,
 	.load_scale = load_scale_v4,
 };