Re: [PATCH 2/2] soc: qcom: rpmhpd: Make mx as a parent of cx only for sdm845

From: Rajendra Nayak
Date: Thu Dec 02 2021 - 02:49:09 EST



On 11/23/2021 12:45 PM, Rajendra Nayak wrote:

On 11/19/2021 8:49 AM, Bjorn Andersson wrote:
On Mon 15 Nov 23:26 CST 2021, Rajendra Nayak wrote:

The requirement to specify the active + sleep and active-only MX power
domains as the parents of the corresponding CX power domains is applicable
only on the sdm845 SoC. With the same struct definition reused for all the
SoCs this condition was wrongly applied to all those SoCs as well, which
isn't needed. Define new sdm845 specific structures to manage this
dependency and remove the parent assignements from the common structure.


Looking at the downstream sm8150 dts I see that both cx and mmcx
specifies mx as parent "supply".

Is this not needed or should we instead name these resources
"cx_with_mx_parent" and have sm8150 opt in as well?

Right, looks like these are needed, after talking to some more folks
I was told RPMh does not really enforce any dependencies on any of the
SoCs, so my earlier statement was wrong that this was managed by RPMh.
Some SoCs just have some digital domain requirements which need these
dependencies to be managed (not all SoCs) and when we end up with such
a situation its almost always expected to be managed by the RPMh masters
(APPS running hlos in this case)
This is not just across cx/mx but others as well like mmcx/mxc/gfx etc.

Unfortunately I could not find this very well documented at an SoC level,
so perhaps the best way to go about is to look at downstream dependencies
and try to match them upstream :/
I will respin this to add the 8150 dependencies back (and if I see any more
for the others)

Looking through this more in downstream files, I see atleast the mx being a
parent of cx is needed on pretty much all upstream supported SoCs, except the
sc7280. There seem to be more complex dependencies that downstream models
across other rails, mainly for 8150/8250/8350 and the most recent 8450
but looks like we have been able to live without those upstream so I plan
to leave them for now and just re-post this with an additional cx_with_mx_parent
and a cx to distinguish between these 2 cases.



Regards,
Bjorn

Signed-off-by: Rajendra Nayak <rnayak@xxxxxxxxxxxxxx>
---
  drivers/soc/qcom/rpmhpd.c | 22 ++++++++++++++++++----
  1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/drivers/soc/qcom/rpmhpd.c b/drivers/soc/qcom/rpmhpd.c
index c71481d..12d8ce9 100644
--- a/drivers/soc/qcom/rpmhpd.c
+++ b/drivers/soc/qcom/rpmhpd.c
@@ -108,7 +108,6 @@ static struct rpmhpd cx_ao;
  static struct rpmhpd cx = {
      .pd = { .name = "cx", },
      .peer = &cx_ao,
-    .parent = &mx.pd,
      .res_name = "cx.lvl",
  };
@@ -116,7 +115,6 @@ static struct rpmhpd cx_ao = {
      .pd = { .name = "cx_ao", },
      .active_only = true,
      .peer = &cx,
-    .parent = &mx_ao.pd,
      .res_name = "cx.lvl",
  };
@@ -149,12 +147,28 @@ static struct rpmhpd mxc_ao = {
  };
  /* SDM845 RPMH powerdomains */
+static struct rpmhpd sdm845_cx_ao;
+static struct rpmhpd sdm845_cx = {
+    .pd = { .name = "cx", },
+    .peer = &sdm845_cx_ao,
+    .parent = &mx.pd,
+    .res_name = "cx.lvl",
+};
+
+static struct rpmhpd sdm845_cx_ao = {
+    .pd = { .name = "cx_ao", },
+    .active_only = true,
+    .peer = &sdm845_cx,
+    .parent = &mx_ao.pd,
+    .res_name = "cx.lvl",
+};
+
  static struct rpmhpd *sdm845_rpmhpds[] = {
      [SDM845_EBI] = &ebi,
      [SDM845_MX] = &mx,
      [SDM845_MX_AO] = &mx_ao,
-    [SDM845_CX] = &cx,
-    [SDM845_CX_AO] = &cx_ao,
+    [SDM845_CX] = &sdm845_cx,
+    [SDM845_CX_AO] = &sdm845_cx_ao,
      [SDM845_LMX] = &lmx,
      [SDM845_LCX] = &lcx,
      [SDM845_GFX] = &gfx,
--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation



--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation