Skip to contentSkip to conversation

Venus cleanups

PATCH Superseded Konrad Dybcio 1 participant 7 messages Feb 9, 2024

Subsystems (4)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover Venus cleanups

With the driver supporting multiple generations of hardware, some mold has definitely grown over the code..

This series attempts to amend this situation a bit by commonizing some code paths and fixing some bugs while at it.

Only tested on SM8250.

Definitely needs testing on:

- SDM845 with old bindings
- SDM845 with new bindings or 7180
- MSM8916
- MSM8996

---
Changes in v2:
- Fix "set but unused" warning in "Drop cache properties in resource struct"
- Fix modular build with "Commonize vdec_get()"
- Rebase
- Test again on 8250, since nobody else tested other platforms since the last submission (or at least hasn't reported that), I'm assuming nobody cares
- Needs to be tested atop [1] and similar, it's in latest -next already
- Link to v1: https://lore.kernel.org/r/[email protected]

[1] https://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux.git/commit/?h=for-next&id=d2cd22c9c384aa50c0b4530e842bd078427e6279

---
Konrad Dybcio (20):
media: venus: pm_helpers: Only set rate of the core clock in core_clks_enable
media: venus: pm_helpers: Rename core_clks_get to venus_clks_get
media: venus: pm_helpers: Add kerneldoc to venus_clks_get()
media: venus: core: Set OPP clkname in a common code path
media: venus: pm_helpers: Kill dead code
media: venus: pm_helpers: Move reset acquisition to common code
media: venus: core: Constify all members of the resource struct
media: venus: core: Deduplicate OPP genpd names
media: venus: core: Get rid of vcodec_num
media: venus: core: Drop cache properties in resource struct
media: venus: core: Use GENMASK for dma_mask
media: venus: core: Remove cp_start
media: venus: pm_helpers: Commonize core_power
media: venus: pm_helpers: Remove pm_ops->core_put
media: venus: core: Define a pointer to core->res
media: venus: pm_helpers: Simplify vcodec clock handling
media: venus: pm_helpers: Commonize getting clocks and GenPDs
media: venus: pm_helpers: Commonize vdec_get()
media: venus: pm_helpers: Commonize venc_get()
media: venus: pm_helpers: Use reset_bulk API

 drivers/media/platform/qcom/venus/core.c       | 139 ++++-------
 drivers/media/platform/qcom/venus/core.h       |  66 +++--
 drivers/media/platform/qcom/venus/firmware.c   |   3 +-
 drivers/media/platform/qcom/venus/hfi_venus.c  |  10 +-
 drivers/media/platform/qcom/venus/pm_helpers.c | 323 +++++++++----------------
 drivers/media/platform/qcom/venus/pm_helpers.h |  10 +-
 drivers/media/platform/qcom/venus/vdec.c       |   9 +-
 drivers/media/platform/qcom/venus/venc.c       |   9 +-
 8 files changed, 213 insertions(+), 356 deletions(-)

---
base-commit: 445a555e0623387fa9b94e68e61681717e70200a
change-id: 20230911-topic-mars-e60bb2269411

Best regards,

Signature · 1 line
Konrad Dybcio <[email protected]>
#1 lore raw

PATCH v2 1/20 media: venus: pm_helpers: Only set rate of the core clock in core_clks_enable

Commit c22b1a29497c ("media: venus: core,pm: Vote for min clk freq during venus boot") intended to up the rate of the Venus core clock from the XO minimum to something more reasonable, based on the per- SoC frequency table.

Unfortunately, it ended up calling set_rate with that same argument on all clocks in res->clks. Fix that using the OPP API.

Fixes c22b1a29497c ("media: venus: core,pm: Vote for min clk freq during venus boot")
---
 drivers/media/platform/qcom/venus/pm_helpers.c | 23 +++++++++++------------
 1 file changed, 11 insertions(+), 12 deletions(-)
drivers/media/platform/qcom/venus/pm_helpers.c +11 -12
index 502822059498..8bd0ce4ce69d 100644
@@ -41,24 +41,23 @@ static int core_clks_get(struct venus_core *core)
 static int core_clks_enable(struct venus_core *core)
 {
 	const struct venus_resources *res = core->res;
-	const struct freq_tbl *freq_tbl = core->res->freq_tbl;
-	unsigned int freq_tbl_size = core->res->freq_tbl_size;
-	unsigned long freq;
+	struct dev_pm_opp *opp;
+	unsigned long freq = 0;
 	unsigned int i;
 	int ret;
 
-	if (!freq_tbl)
-		return -EINVAL;
+	if (core->has_opp_table) {
+		opp = dev_pm_opp_find_freq_ceil(core->dev, &freq);
+		if (IS_ERR(opp))
+			return PTR_ERR(opp);
+		dev_pm_opp_put(opp);
 
-	freq = freq_tbl[freq_tbl_size - 1].freq;
+		ret = dev_pm_opp_set_rate(core->dev, freq);
+		if (ret)
+			return ret;
+	}
 
 	for (i = 0; i < res->clks_num; i++) {
-		if (IS_V6(core)) {
-			ret = clk_set_rate(core->clks[i], freq);
-			if (ret)
-				goto err;
-		}
-
 		ret = clk_prepare_enable(core->clks[i]);
 		if (ret)
 			goto err;
Signature · 1 line
2.43.0
#2 lore raw

PATCH v2 2/20 media: venus: pm_helpers: Rename core_clks_get to venus_clks_get

In reply to Konrad Dybcio

"core" is used in multiple contexts when talking about Venus, rename the function to save on confusion.

---
 drivers/media/platform/qcom/venus/pm_helpers.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/media/platform/qcom/venus/pm_helpers.c +3 -3
index 8bd0ce4ce69d..ac7c83404c6e 100644
@@ -23,7 +23,7 @@
 
 static bool legacy_binding;
 
-static int core_clks_get(struct venus_core *core)
+static int venus_clks_get(struct venus_core *core)
 {
 	const struct venus_resources *res = core->res;
 	struct device *dev = core->dev;
@@ -294,7 +294,7 @@ static int core_get_v1(struct venus_core *core)
 {
 	int ret;
 
-	ret = core_clks_get(core);
+	ret = venus_clks_get(core);
 	if (ret)
 		return ret;
 
@@ -961,7 +961,7 @@ static int core_get_v4(struct venus_core *core)
 	const struct venus_resources *res = core->res;
 	int ret;
 
-	ret = core_clks_get(core);
+	ret = venus_clks_get(core);
 	if (ret)
 		return ret;
 
Signature · 1 line
2.43.0
#3 lore raw

PATCH v2 3/20 media: venus: pm_helpers: Add kerneldoc to venus_clks_get()

In reply to Konrad Dybcio

To make it easier to understand the various clock requirements within this driver, add kerneldoc to venus_clk_get() explaining the fluff.

---
 drivers/media/platform/qcom/venus/pm_helpers.c | 28 ++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)
drivers/media/platform/qcom/venus/pm_helpers.c +28 -0
index ac7c83404c6e..ea0a7d4601e2 100644
@@ -23,6 +23,34 @@
 
 static bool legacy_binding;
 
+/**
+ * venus_clks_get() - Get Venus clocks that are not bound to a vcodec
+ * @core: A pointer to the venus core resource
+ *
+ * The Venus block (depending on the generation) can be split into a couple
+ * of clock domains: one for "main logic" and one for each video core (0-2pcs).
+ *
+ * MSM8916 (and possibly other HFIv1 users) only feature the "main logic"
+ * domain, so this function is the only kind if clk_get necessary there.
+ *
+ * MSM8996 (and other HFIv3 users) feature two video cores, with core0 being
+ * statically proclaimed a decoder and core1 an encoder, with both having
+ * their own clock domains.
+ *
+ * SDM845 features two video cores, each one of which may or may not be
+ * subdivided into 2 enc/dec threads.
+ *
+ * Other SoCs either feature a single video core (with its own clock domain)
+ * or 1 video core and 1 CVP (Computer Vision Processor) core. In both cases
+ * we treat it the same (CVP only happens to live near-by Venus on the SoC).
+ *
+ * Due to unfortunate developments in the past, we have to support bindings
+ * (MSM8996, SDM660, SDM845) that require specifying the clocks and
+ * power-domains associated with a video core domain in a bogus subnode,
+ * which means that additional fluff is necessary..
+ *
+ * Return: 0 on success, negative errno on failure.
+ */
 static int venus_clks_get(struct venus_core *core)
 {
 	const struct venus_resources *res = core->res;
Signature · 1 line
2.43.0
#4 lore raw

PATCH v2 4/20 media: venus: core: Set OPP clkname in a common code path

In reply to Konrad Dybcio

Calling devm_pm_opp_set_clkname() is repeated for all HFI versions in pm_ops->core_power.

Move it to the common codepath.

This also lets us get rid of core_get_v1.

---
 drivers/media/platform/qcom/venus/core.c       |  5 +++++
 drivers/media/platform/qcom/venus/pm_helpers.c | 23 ++---------------------
 2 files changed, 7 insertions(+), 21 deletions(-)
drivers/media/platform/qcom/venus/core.c +5 -0
index ce206b709754..5ab3c414ec0f 100644
@@ -14,6 +14,7 @@
 #include <linux/of.h>
 #include <linux/of_platform.h>
 #include <linux/platform_device.h>
+#include <linux/pm_opp.h>
 #include <linux/slab.h>
 #include <linux/types.h>
 #include <linux/pm_domain.h>
@@ -319,6 +320,10 @@ static int venus_probe(struct platform_device *pdev)
 	if (!core->pm_ops)
 		return -ENODEV;
 
+	ret = devm_pm_opp_set_clkname(dev, "core");
+	if (ret)
+		return ret;
+
 	if (core->pm_ops->core_get) {
 		ret = core->pm_ops->core_get(core);
 		if (ret)
drivers/media/platform/qcom/venus/pm_helpers.c +2 -21
index ea0a7d4601e2..1ba65345a5e2 100644
@@ -318,21 +318,6 @@ static int load_scale_v1(struct venus_inst *inst)
 	return ret;
 }
 
-static int core_get_v1(struct venus_core *core)
-{
-	int ret;
-
-	ret = venus_clks_get(core);
-	if (ret)
-		return ret;
-
-	ret = devm_pm_opp_set_clkname(core->dev, "core");
-	if (ret)
-		return ret;
-
-	return 0;
-}
-
 static void core_put_v1(struct venus_core *core)
 {
 }
@@ -350,7 +335,7 @@ static int core_power_v1(struct venus_core *core, int on)
 }
 
 static const struct venus_pm_ops pm_ops_v1 = {
-	.core_get = core_get_v1,
+	.core_get = venus_clks_get,
 	.core_put = core_put_v1,
 	.core_power = core_power_v1,
 	.load_scale = load_scale_v1,
@@ -423,7 +408,7 @@ static int venc_power_v3(struct device *dev, int on)
 }
 
 static const struct venus_pm_ops pm_ops_v3 = {
-	.core_get = core_get_v1,
+	.core_get = venus_clks_get,
 	.core_put = core_put_v1,
 	.core_power = core_power_v1,
 	.vdec_get = vdec_get_v3,
@@ -1013,10 +998,6 @@ static int core_get_v4(struct venus_core *core)
 	if (legacy_binding)
 		return 0;
 
-	ret = devm_pm_opp_set_clkname(dev, "core");
-	if (ret)
-		return ret;
-
 	ret = vcodec_domains_get(core);
 	if (ret)
 		return ret;
Signature · 1 line
2.43.0
#5 lore raw

PATCH v2 5/20 media: venus: pm_helpers: Kill dead code

In reply to Konrad Dybcio

A situation like:

if (!foo)

	goto bar;

for (i = 0; i < foo; i++)

	...1...

bar:

	...2...

is totally identical to:

for (i = 0; i < 0; i++) // === if (0)

	...1...

...2...

Get rid of such boilerplate.

---
 drivers/media/platform/qcom/venus/pm_helpers.c | 10 ----------
 1 file changed, 10 deletions(-)
drivers/media/platform/qcom/venus/pm_helpers.c +0 -10
index 1ba65345a5e2..7193075e8c04 100644
@@ -878,14 +878,10 @@ static int vcodec_domains_get(struct venus_core *core)
 		.pd_flags = PD_FLAG_NO_DEV_LINK,
 	};
 
-	if (!res->vcodec_pmdomains_num)
-		goto skip_pmdomains;
-
 	ret = dev_pm_domain_attach_list(dev, &vcodec_data, &core->pmdomains);
 	if (ret < 0)
 		return ret;
 
-skip_pmdomains:
 	if (!core->res->opp_pmdomain)
 		return 0;
 
@@ -928,9 +924,6 @@ static int core_resets_reset(struct venus_core *core)
 	unsigned int i;
 	int ret;
 
-	if (!res->resets_num)
-		return 0;
-
 	for (i = 0; i < res->resets_num; i++) {
 		ret = reset_control_assert(core->resets[i]);
 		if (ret)
@@ -953,9 +946,6 @@ static int core_resets_get(struct venus_core *core)
 	unsigned int i;
 	int ret;
 
-	if (!res->resets_num)
-		return 0;
-
 	for (i = 0; i < res->resets_num; i++) {
 		core->resets[i] =
 			devm_reset_control_get_exclusive(dev, res->resets[i]);
Signature · 1 line
2.43.0
#6 lore raw
In reply to Konrad Dybcio
Quoted 10 lines · Konrad Dybcio

With the driver supporting multiple generations of hardware, some mold has definitely grown over the code..

This series attempts to amend this situation a bit by commonizing some code paths and fixing some bugs while at it.

Only tested on SM8250.

Definitely needs testing on:

Quoted 7 lines

- SDM845 with old bindings
- SDM845 with new bindings or 7180
- MSM8916
- MSM8996

---

Apologies for sending this twice. The other submission should be looked at.

Konrad