[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20230303-topic-rpmcc_sleep-v2-13-ae80a325fe94@linaro.org>
Date: Wed, 08 Mar 2023 22:35:29 +0100
From: Konrad Dybcio <konrad.dybcio@...aro.org>
To: Andy Gross <agross@...nel.org>,
Bjorn Andersson <andersson@...nel.org>,
Michael Turquette <mturquette@...libre.com>,
Stephen Boyd <sboyd@...nel.org>,
Rob Herring <robh+dt@...nel.org>,
Krzysztof Kozlowski <krzysztof.kozlowski+dt@...aro.org>
Cc: linux-arm-msm@...r.kernel.org, linux-clk@...r.kernel.org,
linux-kernel@...r.kernel.org,
Krzysztof Kozlowski <krzysztof.kozlowski@...aro.org>,
devicetree@...r.kernel.org,
Konrad Dybcio <konrad.dybcio@...aro.org>,
Shawn Guo <shawn.guo@...aro.org>
Subject: [PATCH RFT v2 13/14] clk: qcom: smd-rpm: Mark clock enabled in
clk_smd_rpm_handoff()
From: Shawn Guo <shawn.guo@...aro.org>
The result of clock handoff is that the clock is voted by APSS and
enabled by RPM. So it should be marked as enabled.
This, combined with .is_enabled/prepared will ultimately cause RPM
clocks that were enabled by the bootloader to actually be shut down
if unused. We can only afford to do so if we have a functioning ICC
driver. An immediate thought to test for that can be to provide an
ICC path to the RPMCC node, however that would create a couple of
circular dependencies between GCC/DISPCC/.../RPMCC and ICC.
The next best thing to do without breaking older device trees is to
add an opt-in DTS property. Do just that.
Signed-off-by: Shawn Guo <shawn.guo@...aro.org>
[Konrad: make conditional, explain the consequences]
Signed-off-by: Konrad Dybcio <konrad.dybcio@...aro.org>
---
drivers/clk/qcom/clk-smd-rpm.c | 17 +++++++++++++++--
1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/drivers/clk/qcom/clk-smd-rpm.c b/drivers/clk/qcom/clk-smd-rpm.c
index ef3157fd29d5..6736e53e607a 100644
--- a/drivers/clk/qcom/clk-smd-rpm.c
+++ b/drivers/clk/qcom/clk-smd-rpm.c
@@ -197,7 +197,7 @@ struct rpm_smd_clk_desc {
static DEFINE_MUTEX(rpm_smd_clk_lock);
-static int clk_smd_rpm_handoff(struct clk_smd_rpm *r)
+static int clk_smd_rpm_handoff(struct clk_smd_rpm *r, bool disable_unused_clks)
{
int ret;
struct clk_smd_rpm_req req = {
@@ -222,6 +222,10 @@ static int clk_smd_rpm_handoff(struct clk_smd_rpm *r)
if (ret)
return ret;
+ /* Marking clocks enabled here will trigger unused cleanup */
+ if (disable_unused_clks)
+ r->enabled = true;
+
return 0;
}
@@ -1319,6 +1323,7 @@ static int rpm_smd_clk_probe(struct platform_device *pdev)
struct qcom_smd_rpm *rpm;
struct clk_smd_rpm **rpm_smd_clks;
const struct rpm_smd_clk_desc *desc;
+ bool disable_unused_clks;
rpm = dev_get_drvdata(pdev->dev.parent);
if (!rpm) {
@@ -1326,6 +1331,14 @@ static int rpm_smd_clk_probe(struct platform_device *pdev)
return -ENODEV;
}
+ /*
+ * We can only really park unused clocks if we have a sane interconnect
+ * driver. Otherwise, the platform may (and probably will) try accessing
+ * IPs that are hosted on unclocked buses. In an effort not to break
+ * older DTs, make this an opt-in through a DT property.
+ */
+ disable_unused_clks = of_property_read_bool(pdev->dev.of_node, "qcom,clk-disable-unused");
+
desc = of_device_get_match_data(&pdev->dev);
if (!desc)
return -EINVAL;
@@ -1339,7 +1352,7 @@ static int rpm_smd_clk_probe(struct platform_device *pdev)
rpm_smd_clks[i]->rpm = rpm;
- ret = clk_smd_rpm_handoff(rpm_smd_clks[i]);
+ ret = clk_smd_rpm_handoff(rpm_smd_clks[i], disable_unused_clks);
if (ret)
goto err;
}
--
2.39.2
Powered by blists - more mailing lists