[PATCH v7 0/2] Resolve MPM register space situation

[PATCH v7 0/2] Resolve MPM register space situation

[Konrad Dybcio]

v6 -> v7:
- Actually check the rvalue of of_address_to_resource
- Pick up rb
- Link to v6: https://lore.kernel.org/r/20230328-topic-msgram_mpm-v6-0-682e4855b7e2@linaro.org

v5 -> v6:
- Resending due to no responses.
- Change of base to next-20231124 (no changes to the patch)

Link to v5: https://lore.kernel.org/r/20230328-topic-msgram_mpm-v5-0-6e06278896b5@linaro.org

v4 -> v5:
- Pick up tags
- Rebase on Rob's of_ header untanglement

Link to v4: https://lore.kernel.org/r/20230328-topic-msgram_mpm-v4-0-bae382dc0f92@linaro.org

v3 -> v4:
- Fix up indentation in the bindings patch
- Add an example glink-edge subnode to remoteproc-rpm (its bindings
  require that..)

Link to v3: https://lore.kernel.org/r/20230328-topic-msgram_mpm-v3-0-2c72f27b4706@linaro.org

v2 -> v3:
- Fix the example
- Pick up tags
- remove the outdated example from the cover letter, check bindings
  should you want to see one

The bindings for the wrapper node used in the yaml example are merged
in qcom/for-next

Link to v2: https://lore.kernel.org/r/20230328-topic-msgram_mpm-v2-0-e24a48e57f0d@linaro.org

v1 -> v2:
- deprecate 'reg', make qcom,rpm-msg-ram required [1/2]
- Use devm_ioremap() [2/2]

Link to v1: https://lore.kernel.org/r/20230328-topic-msgram_mpm-v1-0-1b788a5f5a33@linaro.org

Depends on resolution of https://github.com/devicetree-org/dt-schema/issues/104

The MPM (and some other things, irrelevant to this patchset) resides
(as far as the ARM cores are concerned, anyway) in a MMIO-mapped region
that's a portion of the RPM (low-power management core)'s RAM, known
as the RPM Message RAM. Representing this relation in the Device Tree
creates some challenges, as one would either have to treat a memory
region as a bus, map nodes in a way such that their reg-s would be
overlapping, or supply the nodes with a slice of that region.

This series implements the third option, by adding a qcom,rpm-msg-ram
property, which has been used for some drivers poking into this region
before. Bindings ABI compatibility is preserved through keeping the
"normal" (a.k.a read the reg property and map that region) way of
passing the register space.

Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
---
Konrad Dybcio (2):
      dt-bindings: interrupt-controller: mpm: Pass MSG RAM slice through phandle
      irqchip: irq-qcom-mpm: Support passing a slice of SRAM as reg space

 .../bindings/interrupt-controller/qcom,mpm.yaml    | 52 +++++++++++++++-------
 drivers/irqchip/irq-qcom-mpm.c                     | 26 +++++++++--
 2 files changed, 58 insertions(+), 20 deletions(-)
---
base-commit: 48bbaf8b793e0770798519f8ee1ea2908ff0943a
change-id: 20230328-topic-msgram_mpm-c688be3bc294

Best regards,
-- 
Konrad Dybcio <konrad.dybcio@linaro.org>

[PATCH v7 1/2] dt-bindings: interrupt-controller: mpm: Pass MSG RAM slice through phandle

[Konrad Dybcio]

Due to the wild nature of the Qualcomm RPM Message RAM, we can't really
use 'reg' to point to the MPM's slice of Message RAM without cutting into
an already-defined RPM MSG RAM node used for GLINK and SMEM.

Document passing the register space as a slice of SRAM through the
qcom,rpm-msg-ram property. This also makes 'reg' deprecated.

Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
---
 .../bindings/interrupt-controller/qcom,mpm.yaml    | 52 +++++++++++++++-------
 1 file changed, 35 insertions(+), 17 deletions(-)

diff --git a/Documentation/devicetree/bindings/interrupt-controller/qcom,mpm.yaml b/Documentation/devicetree/bindings/interrupt-controller/qcom,mpm.yaml
index 509d20c091af..4ce7912d8047 100644
--- a/Documentation/devicetree/bindings/interrupt-controller/qcom,mpm.yaml
+++ b/Documentation/devicetree/bindings/interrupt-controller/qcom,mpm.yaml
@@ -29,6 +29,12 @@ properties:
     maxItems: 1
     description:
       Specifies the base address and size of vMPM registers in RPM MSG RAM.
+    deprecated: true
+
+  qcom,rpm-msg-ram:
+    $ref: /schemas/types.yaml#/definitions/phandle
+    description:
+      Phandle to the APSS MPM slice of the RPM Message RAM
 
   interrupts:
     maxItems: 1
@@ -64,33 +70,45 @@ properties:
 
 required:
   - compatible
-  - reg
   - interrupts
   - mboxes
   - interrupt-controller
   - '#interrupt-cells'
   - qcom,mpm-pin-count
   - qcom,mpm-pin-map
+  - qcom,rpm-msg-ram
 
 additionalProperties: false
 
 examples:
   - |
     #include <dt-bindings/interrupt-controller/arm-gic.h>
-    mpm: interrupt-controller@45f01b8 {
-        compatible = "qcom,mpm";
-        interrupts = <GIC_SPI 197 IRQ_TYPE_EDGE_RISING>;
-        reg = <0x45f01b8 0x1000>;
-        mboxes = <&apcs_glb 1>;
-        interrupt-controller;
-        #interrupt-cells = <2>;
-        interrupt-parent = <&intc>;
-        qcom,mpm-pin-count = <96>;
-        qcom,mpm-pin-map = <2 275>,
-                           <5 296>,
-                           <12 422>,
-                           <24 79>,
-                           <86 183>,
-                           <90 260>,
-                           <91 260>;
+
+    remoteproc-rpm {
+        compatible = "qcom,msm8998-rpm-proc", "qcom,rpm-proc";
+
+        glink-edge {
+            compatible = "qcom,glink-rpm";
+
+            interrupts = <GIC_SPI 168 IRQ_TYPE_EDGE_RISING>;
+            qcom,rpm-msg-ram = <&rpm_msg_ram>;
+            mboxes = <&apcs_glb 0>;
+        };
+
+        mpm: interrupt-controller {
+            compatible = "qcom,mpm";
+            qcom,rpm-msg-ram = <&apss_mpm>;
+            interrupts = <GIC_SPI 197 IRQ_TYPE_EDGE_RISING>;
+            mboxes = <&apcs_glb 1>;
+            interrupt-controller;
+            #interrupt-cells = <2>;
+            interrupt-parent = <&intc>;
+            qcom,mpm-pin-count = <96>;
+            qcom,mpm-pin-map = <2 275>,
+                               <5 296>,
+                               <12 422>,
+                               <24 79>,
+                               <86 183>,
+                               <91 260>;
+        };
     };

-- 
2.43.0

[PATCH v7 2/2] irqchip: irq-qcom-mpm: Support passing a slice of SRAM as reg space

[Konrad Dybcio]

The MPM hardware is accessible to us from the ARM CPUs through a shared
memory region (RPM MSG RAM) that's also concurrently accessed by other
kinds of cores on the system (like modem, ADSP etc.). Modeling this
relation in a (somewhat) sane manner in the device tree basically
requires us to either present the MPM as a child of said memory region
(which makes little sense, as a mapped memory carveout is not a bus),
define nodes which bleed their register spaces into one another, or
passing their slice of the MSG RAM through some kind of a property.

Go with the third option and add a way to map a region passed through
the "qcom,rpm-msg-ram" property as our register space.

The current way of using 'reg' is preserved for ABI reasons.

Acked-by: Shawn Guo <shawn.guo@linaro.org>
Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
---
 drivers/irqchip/irq-qcom-mpm.c | 26 +++++++++++++++++++++++---
 1 file changed, 23 insertions(+), 3 deletions(-)

diff --git a/drivers/irqchip/irq-qcom-mpm.c b/drivers/irqchip/irq-qcom-mpm.c
index 7124565234a5..cda5838d2232 100644
--- a/drivers/irqchip/irq-qcom-mpm.c
+++ b/drivers/irqchip/irq-qcom-mpm.c
@@ -14,6 +14,7 @@
 #include <linux/mailbox_client.h>
 #include <linux/module.h>
 #include <linux/of.h>
+#include <linux/of_address.h>
 #include <linux/of_platform.h>
 #include <linux/platform_device.h>
 #include <linux/pm_domain.h>
@@ -322,8 +323,10 @@ static int qcom_mpm_init(struct device_node *np, struct device_node *parent)
 	struct device *dev = &pdev->dev;
 	struct irq_domain *parent_domain;
 	struct generic_pm_domain *genpd;
+	struct device_node *msgram_np;
 	struct qcom_mpm_priv *priv;
 	unsigned int pin_cnt;
+	struct resource res;
 	int i, irq;
 	int ret;
 
@@ -374,9 +377,26 @@ static int qcom_mpm_init(struct device_node *np, struct device_node *parent)
 
 	raw_spin_lock_init(&priv->lock);
 
-	priv->base = devm_platform_ioremap_resource(pdev, 0);
-	if (IS_ERR(priv->base))
-		return PTR_ERR(priv->base);
+	/* If we have a handle to an RPM message ram partition, use it. */
+	msgram_np = of_parse_phandle(np, "qcom,rpm-msg-ram", 0);
+	if (msgram_np) {
+		ret = of_address_to_resource(msgram_np, 0, &res);
+		if (ret) {
+			of_node_put(msgram_np);
+			return ret;
+		}
+
+		/* Don't use devm_ioremap_resource, as we're accessing a shared region. */
+		priv->base = devm_ioremap(dev, res.start, resource_size(&res));
+		of_node_put(msgram_np);
+		if (IS_ERR(priv->base))
+			return PTR_ERR(priv->base);
+	} else {
+		/* Otherwise, fall back to simple MMIO. */
+		priv->base = devm_platform_ioremap_resource(pdev, 0);
+		if (IS_ERR(priv->base))
+			return PTR_ERR(priv->base);
+	}
 
 	for (i = 0; i < priv->reg_stride; i++) {
 		qcom_mpm_write(priv, MPM_REG_ENABLE, i, 0);

-- 
2.43.0

Re: [PATCH v7 0/2] Resolve MPM register space situation

[Thomas Gleixner]

Konrad!

On Mon, Nov 27 2023 at 16:52, Konrad Dybcio wrote:

This list of version changes is useful, but seriously not on top of the
cover letter. It makes me scroll down 60 lines to find the real meat.

<SNIP>

> Depends on resolution of https://github.com/devicetree-org/dt-schema/issues/104

Is this resolved? I can't figure out from this non-format whether this
belongs to V1 or to the actual cover letter...

> The MPM (and some other things, irrelevant to this patchset) resides
> (as far as the ARM cores are concerned, anyway) in a MMIO-mapped region
> that's a portion of the RPM (low-power management core)'s RAM, known
> as the RPM Message RAM. Representing this relation in the Device Tree
> creates some challenges, as one would either have to treat a memory
> region as a bus, map nodes in a way such that their reg-s would be
> overlapping, or supply the nodes with a slice of that region.
>
> This series implements the third option, by adding a qcom,rpm-msg-ram
> property, which has been used for some drivers poking into this region
> before. Bindings ABI compatibility is preserved through keeping the
> "normal" (a.k.a read the reg property and map that region) way of
> passing the register space.
>
> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>

Signed-off-by on the cover letter is completely meaningless.

Thanks,

        tglx

Re: [PATCH v7 2/2] irqchip: irq-qcom-mpm: Support passing a slice of SRAM as reg space

[Thomas Gleixner]

On Mon, Nov 27 2023 at 16:52, Konrad Dybcio wrote:

The prefix in the subject is wrong. Also please write out register. This
is not Xitter.

> The MPM hardware is accessible to us from the ARM CPUs through a shared

to us? Can you access that hardware? I doubt it.

Please use neutral tone as documented in Documentation/process/

> memory region (RPM MSG RAM) that's also concurrently accessed by other
> kinds of cores on the system (like modem, ADSP etc.). Modeling this
> relation in a (somewhat) sane manner in the device tree basically
> requires us to either present the MPM as a child of said memory region
> (which makes little sense, as a mapped memory carveout is not a bus),
> define nodes which bleed their register spaces into one another, or
> passing their slice of the MSG RAM through some kind of a property.
>
> Go with the third option and add a way to map a region passed through
> the "qcom,rpm-msg-ram" property as our register space.
>
> The current way of using 'reg' is preserved for ABI reasons.

It's not an ABI reason. It's backwards compatibility with old device
trees, right?

I'll fix it up for you this time. No need to resend.

Thanks,

        tglx

Re: [PATCH v7 0/2] Resolve MPM register space situation

[Konrad Dybcio]

On 8.12.2023 15:33, Thomas Gleixner wrote:
> Konrad!
> 
> On Mon, Nov 27 2023 at 16:52, Konrad Dybcio wrote:
> 
> This list of version changes is useful, but seriously not on top of the
> cover letter. It makes me scroll down 60 lines to find the real meat.
Ack

> 
> <SNIP>
> 
>> Depends on resolution of https://github.com/devicetree-org/dt-schema/issues/104
> 
> Is this resolved? I can't figure out from this non-format whether this
> belongs to V1 or to the actual cover letter...
Yes it has been resolved since

> 
>> The MPM (and some other things, irrelevant to this patchset) resides
>> (as far as the ARM cores are concerned, anyway) in a MMIO-mapped region
>> that's a portion of the RPM (low-power management core)'s RAM, known
>> as the RPM Message RAM. Representing this relation in the Device Tree
>> creates some challenges, as one would either have to treat a memory
>> region as a bus, map nodes in a way such that their reg-s would be
>> overlapping, or supply the nodes with a slice of that region.
>>
>> This series implements the third option, by adding a qcom,rpm-msg-ram
>> property, which has been used for some drivers poking into this region
>> before. Bindings ABI compatibility is preserved through keeping the
>> "normal" (a.k.a read the reg property and map that region) way of
>> passing the register space.
>>
>> Signed-off-by: Konrad Dybcio <konrad.dybcio@linaro.org>
> 
> Signed-off-by on the cover letter is completely meaningless.
Yes, I realize that. Blame b4.

Konrad

Re: [PATCH v7 2/2] irqchip: irq-qcom-mpm: Support passing a slice of SRAM as reg space

[Konrad Dybcio]

On 8.12.2023 15:37, Thomas Gleixner wrote:
> On Mon, Nov 27 2023 at 16:52, Konrad Dybcio wrote:
> 
> The prefix in the subject is wrong. Also please write out register. This
> is not Xitter.
Had a feeling it would be too long, but actually it'd be perfect
72 chars :)

> 
>> The MPM hardware is accessible to us from the ARM CPUs through a shared
> 
> to us? Can you access that hardware? I doubt it.
With a small enough "stick".. but I get your point

> 
> Please use neutral tone as documented in Documentation/process/
> 
>> memory region (RPM MSG RAM) that's also concurrently accessed by other
>> kinds of cores on the system (like modem, ADSP etc.). Modeling this
>> relation in a (somewhat) sane manner in the device tree basically
>> requires us to either present the MPM as a child of said memory region
>> (which makes little sense, as a mapped memory carveout is not a bus),
>> define nodes which bleed their register spaces into one another, or
>> passing their slice of the MSG RAM through some kind of a property.
>>
>> Go with the third option and add a way to map a region passed through
>> the "qcom,rpm-msg-ram" property as our register space.
>>
>> The current way of using 'reg' is preserved for ABI reasons.
> 
> It's not an ABI reason. It's backwards compatibility with old device
> trees, right?
Yes, I thought of something else.

> 
> I'll fix it up for you this time. No need to resend.
Thanks!

Konrad

Re: [PATCH v7 0/2] Resolve MPM register space situation

[Konstantin Ryabitsev]

December 9, 2023 at 9:00 AM, "Konrad Dybcio" <konrad.dybcio@linaro.org> wrote:
> >  Signed-off-by on the cover letter is completely meaningless.
> 
> Yes, I realize that. Blame b4.

It's not -- some subsystems use cover letters as merge commits.

-K