Skip to contentSkip to conversation

Improve VCHIQ cache line size handling

PATCH Superseded Phil Elwell 4 participants 12 messages Sep 17, 2018 – Nov 6, 2018

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v3 cover Improve VCHIQ cache line size handling

Both sides of the VCHIQ communications mechanism need to agree on the cache line size. Using an incorrect value can lead to data corruption, but having the two sides using different values is usually worse.

In the absence of an obvious convenient run-time method to determine the correct value in the ARCH=arm world, the downstream Raspberry Pi trees used a Device Tree property, written by the firmware, to configure the kernel driver. This method was vetoed during the upstreaming process, so a fixed value of 32 was used instead, and some corruptions ensued. This is take 2 at arriving at the correct value.

Add a new compatible string - "brcm,bcm2836-vchiq" - to indicate an SoC with a 64-byte cache line. Document the new string in the binding, and use it on the appropriate platforms.

The final patch is a (seemingly cosmetic) correction of the Device Tree "reg" declaration for the device node, but it doubles as an indication to the Raspberry Pi firmware that the kernel driver is running a recent kernel driver that chooses the correct value. As such it would help if the DT patches are not merged before the driver patch.

v3: Builds without errors, tested on multiple Raspberry Pi models.
v2: Replaced ARM-specific logic used to determine cache line size with
a new compatible string for BCM2836 and BCM2837.

Phil Elwell (4):
staging/vc04_services: Use correct cache line size
dt-bindings: soc: Document "brcm,bcm2836-vchiq"
ARM: dts: bcm283x: Correct vchiq compatible string
ARM: dts: bcm283x: Correct mailbox register sizes

 .../bindings/soc/bcm/brcm,bcm2835-vchiq.txt        |  3 +-
 arch/arm/boot/dts/bcm2835-rpi.dtsi                 |  4 +--
 arch/arm/boot/dts/bcm2836-rpi-2-b.dts              |  2 +-
 arch/arm/boot/dts/bcm2836-rpi.dtsi                 |  6 ++++
 arch/arm/boot/dts/bcm2837-rpi-3-b-plus.dts         |  2 +-
 arch/arm/boot/dts/bcm2837-rpi-3-b.dts              |  2 +-
 arch/arm/boot/dts/bcm2837-rpi-cm3.dtsi             |  2 +-
 .../interface/vchiq_arm/vchiq_2835_arm.c           |  4 ++-
 .../vc04_services/interface/vchiq_arm/vchiq_arm.c  | 35 +++++++++++++++-------
 .../vc04_services/interface/vchiq_arm/vchiq_arm.h  |  5 ++++
 10 files changed, 47 insertions(+), 18 deletions(-)
 create mode 100644 arch/arm/boot/dts/bcm2836-rpi.dtsi
Signature · 1 line
2.7.4
#1 lore raw

PATCH v3 1/4 staging/vc04_services: Use correct cache line size

Use the compatible string in the DTB to select the correct cache line size for the SoC - 32 for BCM2835, and 64 for BCM2836 and BCM2837.

---
 .../interface/vchiq_arm/vchiq_2835_arm.c           |  4 ++-
 .../vc04_services/interface/vchiq_arm/vchiq_arm.c  | 35 +++++++++++++++-------
 .../vc04_services/interface/vchiq_arm/vchiq_arm.h  |  5 ++++
 3 files changed, 33 insertions(+), 11 deletions(-)
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c +3 -1
index e767209..83d740f 100644
@@ -109,7 +109,8 @@ free_pagelist(struct vchiq_pagelist_info *pagelistinfo,
 int vchiq_platform_init(struct platform_device *pdev, VCHIQ_STATE_T *state)
 {
 	struct device *dev = &pdev->dev;
-	struct rpi_firmware *fw = platform_get_drvdata(pdev);
+	struct vchiq_drvdata *drvdata = platform_get_drvdata(pdev);
+	struct rpi_firmware *fw = drvdata->fw;
 	VCHIQ_SLOT_ZERO_T *vchiq_slot_zero;
 	struct resource *res;
 	void *slot_mem;
@@ -127,6 +128,7 @@ int vchiq_platform_init(struct platform_device *pdev, VCHIQ_STATE_T *state)
 	if (err < 0)
 		return err;
 
+	g_cache_line_size = drvdata->cache_line_size;
 	g_fragments_size = 2 * g_cache_line_size;
 
 	/* Allocate space for the channels in coherent memory */
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.c +25 -10
index bc05c69..ea78937 100644
@@ -170,6 +170,14 @@ static struct device *vchiq_dev;
 static DEFINE_SPINLOCK(msg_queue_spinlock);
 static struct platform_device *bcm2835_camera;
 
+static struct vchiq_drvdata bcm2835_drvdata = {
+	.cache_line_size = 32,
+};
+
+static struct vchiq_drvdata bcm2836_drvdata = {
+	.cache_line_size = 64,
+};
+
 static const char *const ioctl_names[] = {
 	"CONNECT",
 	"SHUTDOWN",
@@ -3573,12 +3581,25 @@ void vchiq_platform_conn_state_changed(VCHIQ_STATE_T *state,
 	}
 }
 
+static const struct of_device_id vchiq_of_match[] = {
+	{ .compatible = "brcm,bcm2835-vchiq", .data = &bcm2835_drvdata },
+	{ .compatible = "brcm,bcm2836-vchiq", .data = &bcm2836_drvdata },
+	{},
+};
+MODULE_DEVICE_TABLE(of, vchiq_of_match);
+
 static int vchiq_probe(struct platform_device *pdev)
 {
 	struct device_node *fw_node;
-	struct rpi_firmware *fw;
+	const struct of_device_id *of_id;
+	struct vchiq_drvdata *drvdata;
 	int err;
 
+	of_id = of_match_node(vchiq_of_match, pdev->dev.of_node);
+	drvdata = (struct vchiq_drvdata *)of_id->data;
+	if (!drvdata)
+		return -EINVAL;
+
 	fw_node = of_find_compatible_node(NULL, NULL,
 					  "raspberrypi,bcm2835-firmware");
 	if (!fw_node) {
@@ -3586,12 +3607,12 @@ static int vchiq_probe(struct platform_device *pdev)
 		return -ENOENT;
 	}
 
-	fw = rpi_firmware_get(fw_node);
+	drvdata->fw = rpi_firmware_get(fw_node);
 	of_node_put(fw_node);
-	if (!fw)
+	if (!drvdata->fw)
 		return -EPROBE_DEFER;
 
-	platform_set_drvdata(pdev, fw);
+	platform_set_drvdata(pdev, drvdata);
 
 	err = vchiq_platform_init(pdev, &g_state);
 	if (err != 0)
@@ -3661,12 +3682,6 @@ static int vchiq_remove(struct platform_device *pdev)
 	return 0;
 }
 
-static const struct of_device_id vchiq_of_match[] = {
-	{ .compatible = "brcm,bcm2835-vchiq", },
-	{},
-};
-MODULE_DEVICE_TABLE(of, vchiq_of_match);
-
 static struct platform_driver vchiq_driver = {
 	.driver = {
 		.name = "bcm2835_vchiq",
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_arm.h +5 -0
index 40bb0c6..2f3ebc9 100644
@@ -123,6 +123,11 @@ typedef struct vchiq_arm_state_struct {
 
 } VCHIQ_ARM_STATE_T;
 
+struct vchiq_drvdata {
+	const unsigned int cache_line_size;
+	struct rpi_firmware *fw;
+};
+
 extern int vchiq_arm_log_level;
 extern int vchiq_susp_log_level;
 
Signature · 1 line
2.7.4
#3 lore raw

PATCH v3 2/4 dt-bindings: soc: Document "brcm,bcm2836-vchiq"

In reply to Phil Elwell

"brcm,bcm2836-vchiq" should be used on BCM2836 and BCM2837 to ensure correct operation.

---
 Documentation/devicetree/bindings/soc/bcm/brcm,bcm2835-vchiq.txt | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
Documentation/devicetree/bindings/soc/bcm/brcm,bcm2835-vchiq.txt +2 -1
index 8dd7b3a..f331316 100644
@@ -2,7 +2,8 @@ Broadcom VCHIQ firmware services
 
 Required properties:
 
-- compatible:	Should be "brcm,bcm2835-vchiq"
+- compatible:	Should be "brcm,bcm2835-vchiq" on BCM2835, otherwise
+		"brcm,bcm2836-vchiq".
 - reg:		Physical base address and length of the doorbell register pair
 - interrupts:	The interrupt number
 		  See bindings/interrupt-controller/brcm,bcm2835-armctrl-ic.txt
Signature · 1 line
2.7.4
#5 lore raw

PATCH v3 3/4 ARM: dts: bcm283x: Correct vchiq compatible string

In reply to Phil Elwell

To allow VCHIQ to determine the correct cache line size, use the new "brcm,bcm2836-vchiq" compatible string on BCM2836 and BCM2837.

---
 arch/arm/boot/dts/bcm2835-rpi.dtsi         | 2 +-
 arch/arm/boot/dts/bcm2836-rpi-2-b.dts      | 2 +-
 arch/arm/boot/dts/bcm2836-rpi.dtsi         | 6 ++++++
 arch/arm/boot/dts/bcm2837-rpi-3-b-plus.dts | 2 +-
 arch/arm/boot/dts/bcm2837-rpi-3-b.dts      | 2 +-
 arch/arm/boot/dts/bcm2837-rpi-cm3.dtsi     | 2 +-
 6 files changed, 11 insertions(+), 5 deletions(-)
 create mode 100644 arch/arm/boot/dts/bcm2836-rpi.dtsi
arch/arm/boot/dts/bcm2835-rpi.dtsi +1 -1
index cb2d6d7..215d8cc 100644
@@ -30,7 +30,7 @@
 			#power-domain-cells = <1>;
 		};
 
-		mailbox at 7e00b840 {
+		vchiq: mailbox at 7e00b840 {
 			compatible = "brcm,bcm2835-vchiq";
 			reg = <0x7e00b840 0xf>;
 			interrupts = <0 2>;
arch/arm/boot/dts/bcm2836-rpi-2-b.dts +1 -1
index 2fef70a..ac4408b 100644
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 /dts-v1/;
 #include "bcm2836.dtsi"
-#include "bcm2835-rpi.dtsi"
+#include "bcm2836-rpi.dtsi"
 #include "bcm283x-rpi-smsc9514.dtsi"
 #include "bcm283x-rpi-usb-host.dtsi"
 
arch/arm/boot/dts/bcm2836-rpi.dtsi new file +6 -0
new file mode 100644
index 0000000..c4c858b
@@ -0,0 +1,6 @@
+// SPDX-License-Identifier: GPL-2.0
+#include "bcm2835-rpi.dtsi"
+
+&vchiq {
+	compatible = "brcm,bcm2836-vchiq", "brcm,bcm2835-vchiq";
+};
arch/arm/boot/dts/bcm2837-rpi-3-b-plus.dts +1 -1
index 4adb85e..eca36e3 100644
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 /dts-v1/;
 #include "bcm2837.dtsi"
-#include "bcm2835-rpi.dtsi"
+#include "bcm2836-rpi.dtsi"
 #include "bcm283x-rpi-lan7515.dtsi"
 #include "bcm283x-rpi-usb-host.dtsi"
 
arch/arm/boot/dts/bcm2837-rpi-3-b.dts +1 -1
index c318bcb..a0ba0f6 100644
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 /dts-v1/;
 #include "bcm2837.dtsi"
-#include "bcm2835-rpi.dtsi"
+#include "bcm2836-rpi.dtsi"
 #include "bcm283x-rpi-smsc9514.dtsi"
 #include "bcm283x-rpi-usb-host.dtsi"
 
arch/arm/boot/dts/bcm2837-rpi-cm3.dtsi +1 -1
index 7b7ab6a..4a89a18 100644
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0
 /dts-v1/;
 #include "bcm2837.dtsi"
-#include "bcm2835-rpi.dtsi"
+#include "bcm2836-rpi.dtsi"
 
 / {
 	memory {
Signature · 1 line
2.7.4
#6 lore raw

PATCH v3 4/4 ARM: dts: bcm283x: Correct mailbox register sizes

In reply to Phil Elwell

The size field in a Device Tree "reg" property is encoded in bytes, not words.

Fixes 614fa22119d6 ("ARM: dts: bcm2835: Add VCHIQ node to the Raspberry Pi boards. (v3)")
---
 arch/arm/boot/dts/bcm2835-rpi.dtsi | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/boot/dts/bcm2835-rpi.dtsi +1 -1
index 215d8cc..29f970f 100644
@@ -32,7 +32,7 @@
 
 		vchiq: mailbox at 7e00b840 {
 			compatible = "brcm,bcm2835-vchiq";
-			reg = <0x7e00b840 0xf>;
+			reg = <0x7e00b840 0x3c>;
 			interrupts = <0 2>;
 		};
 	};
Signature · 1 line
2.7.4
#7 lore raw

v3 cover Improve VCHIQ cache line size handling

In reply to Phil Elwell

Hi Phil,

Am 17.09.2018 um 10:22 schrieb Phil Elwell:

Quoted 25 lines

Both sides of the VCHIQ communications mechanism need to agree on the cache line size. Using an incorrect value can lead to data corruption, but having the two sides using different values is usually worse.

In the absence of an obvious convenient run-time method to determine the correct value in the ARCH=arm world, the downstream Raspberry Pi trees used a Device Tree property, written by the firmware, to configure the kernel driver. This method was vetoed during the upstreaming process, so a fixed value of 32 was used instead, and some corruptions ensued. This is take 2 at arriving at the correct value.

Add a new compatible string - "brcm,bcm2836-vchiq" - to indicate an SoC with a 64-byte cache line. Document the new string in the binding, and use it on the appropriate platforms.

The final patch is a (seemingly cosmetic) correction of the Device Tree "reg" declaration for the device node, but it doubles as an indication to the Raspberry Pi firmware that the kernel driver is running a recent kernel driver that chooses the correct value. As such it would help if the DT patches are not merged before the driver patch.

v3: Builds without errors, tested on multiple Raspberry Pi models.
v2: Replaced ARM-specific logic used to determine cache line size with
a new compatible string for BCM2836 and BCM2837.

Quoted 5 lines

Phil Elwell (4):
staging/vc04_services: Use correct cache line size
dt-bindings: soc: Document "brcm,bcm2836-vchiq"
ARM: dts: bcm283x: Correct vchiq compatible string
ARM: dts: bcm283x: Correct mailbox register sizes

since my pull requests are out, would it be okay to apply patch #1 for 4.20 and the DT stuff for 4.21 (with the assumption Rob is okay with these patches)?

#8 lore raw

v3 cover Improve VCHIQ cache line size handling

Hi Stefan,

Quoted 34 lines · Stefan Wahren

Hi Phil,

Am 17.09.2018 um 10:22 schrieb Phil Elwell:

Quoted 25 lines

Both sides of the VCHIQ communications mechanism need to agree on the cache line size. Using an incorrect value can lead to data corruption, but having the two sides using different values is usually worse.

In the absence of an obvious convenient run-time method to determine the correct value in the ARCH=arm world, the downstream Raspberry Pi trees used a Device Tree property, written by the firmware, to configure the kernel driver. This method was vetoed during the upstreaming process, so a fixed value of 32 was used instead, and some corruptions ensued. This is take 2 at arriving at the correct value.

Add a new compatible string - "brcm,bcm2836-vchiq" - to indicate an SoC with a 64-byte cache line. Document the new string in the binding, and use it on the appropriate platforms.

The final patch is a (seemingly cosmetic) correction of the Device Tree "reg" declaration for the device node, but it doubles as an indication to the Raspberry Pi firmware that the kernel driver is running a recent kernel driver that chooses the correct value. As such it would help if the DT patches are not merged before the driver patch.

v3: Builds without errors, tested on multiple Raspberry Pi models.
v2: Replaced ARM-specific logic used to determine cache line size with
a new compatible string for BCM2836 and BCM2837.

Quoted 5 lines

Phil Elwell (4):
staging/vc04_services: Use correct cache line size
dt-bindings: soc: Document "brcm,bcm2836-vchiq"
ARM: dts: bcm283x: Correct vchiq compatible string
ARM: dts: bcm283x: Correct mailbox register sizes

Quoted 3 lines

since my pull requests are out, would it be okay to apply patch #1 for 4.20 and the DT stuff for 4.21 (with the assumption Rob is okay with these patches)?

Patch 4 is the only one I'd like to be delayed, but delaying 2-4 is fine with me.

Phil

#9 lore raw

v3 cover Improve VCHIQ cache line size handling

Quoted 37 lines · Phil Elwell

Hi Stefan,

Quoted 34 lines · Stefan Wahren

Hi Phil,

Am 17.09.2018 um 10:22 schrieb Phil Elwell:

Quoted 30 lines

Both sides of the VCHIQ communications mechanism need to agree on the cache line size. Using an incorrect value can lead to data corruption, but having the two sides using different values is usually worse.

In the absence of an obvious convenient run-time method to determine the correct value in the ARCH=arm world, the downstream Raspberry Pi trees used a Device Tree property, written by the firmware, to configure the kernel driver. This method was vetoed during the upstreaming process, so a fixed value of 32 was used instead, and some corruptions ensued. This is take 2 at arriving at the correct value.

Add a new compatible string - "brcm,bcm2836-vchiq" - to indicate an SoC with a 64-byte cache line. Document the new string in the binding, and use it on the appropriate platforms.

The final patch is a (seemingly cosmetic) correction of the Device Tree "reg" declaration for the device node, but it doubles as an indication to the Raspberry Pi firmware that the kernel driver is running a recent kernel driver that chooses the correct value. As such it would help if the DT patches are not merged before the driver patch.

v3: Builds without errors, tested on multiple Raspberry Pi models.
v2: Replaced ARM-specific logic used to determine cache line size with
a new compatible string for BCM2836 and BCM2837.

Phil Elwell (4):
staging/vc04_services: Use correct cache line size
dt-bindings: soc: Document "brcm,bcm2836-vchiq"
ARM: dts: bcm283x: Correct vchiq compatible string
ARM: dts: bcm283x: Correct mailbox register sizes

Quoted 5 lines
Quoted 3 lines

since my pull requests are out, would it be okay to apply patch #1 for 4.20 and the DT stuff for 4.21 (with the assumption Rob is okay with these patches)?

Patch 4 is the only one I'd like to be delayed, but delaying 2-4 is fine with me.

Humm, did you mean you would like not to be delayed? In any case Stefan, you can send an additional pull request, and I will merge it and send a second pull request towards ARM SoC maintainers, that's not a problem.

Signature · 1 line
Florian
#10 lore raw

v3 cover Improve VCHIQ cache line size handling

Quoted 42 lines · Florian Fainelli
Quoted 41 lines · Phil Elwell

Hi Stefan,

Quoted 37 lines · Stefan Wahren

Hi Phil,

Am 17.09.2018 um 10:22 schrieb Phil Elwell:

Quoted 25 lines

Both sides of the VCHIQ communications mechanism need to agree on the cache line size. Using an incorrect value can lead to data corruption, but having the two sides using different values is usually worse.

In the absence of an obvious convenient run-time method to determine the correct value in the ARCH=arm world, the downstream Raspberry Pi trees used a Device Tree property, written by the firmware, to configure the kernel driver. This method was vetoed during the upstreaming process, so a fixed value of 32 was used instead, and some corruptions ensued. This is take 2 at arriving at the correct value.

Add a new compatible string - "brcm,bcm2836-vchiq" - to indicate an SoC with a 64-byte cache line. Document the new string in the binding, and use it on the appropriate platforms.

The final patch is a (seemingly cosmetic) correction of the Device Tree "reg" declaration for the device node, but it doubles as an indication to the Raspberry Pi firmware that the kernel driver is running a recent kernel driver that chooses the correct value. As such it would help if the DT patches are not merged before the driver patch.

v3: Builds without errors, tested on multiple Raspberry Pi models.
v2: Replaced ARM-specific logic used to determine cache line size with
a new compatible string for BCM2836 and BCM2837.

Quoted 5 lines

Phil Elwell (4):
staging/vc04_services: Use correct cache line size
dt-bindings: soc: Document "brcm,bcm2836-vchiq"
ARM: dts: bcm283x: Correct vchiq compatible string
ARM: dts: bcm283x: Correct mailbox register sizes

since my pull requests are out, would it be okay to apply patch #1 for 4.20 and the DT stuff for 4.21 (with the assumption Rob is okay with these patches)?

Quoted 5 lines
Quoted 1 line

Patch 4 is the only one I'd like to be delayed, but delaying 2-4 is fine with me.

Humm, did you mean you would like not to be delayed? In any case Stefan, you can send an additional pull request, and I will merge it and send a second pull request towards ARM SoC maintainers, that's not a problem.

No, I meant what I wrote - I would prefer patch 1 to be merged before patch 4 (or at least in the same release) to avoid the need for another firmware change, hence delaying patch 4 is good. It makes sense for the other commits to be merged in that order, but the normal compatible-string fallback mechanism means there is no hard dependency there.

Phil

#11 lore raw

v3 cover Improve VCHIQ cache line size handling

Quoted 52 lines

Phil Elwell <[email protected]> hat am 17. September 2018 um 20:01 geschrieben:

Quoted 42 lines · Florian Fainelli
Quoted 41 lines · Phil Elwell

Hi Stefan,

Quoted 37 lines · Stefan Wahren

Hi Phil,

Am 17.09.2018 um 10:22 schrieb Phil Elwell:

Quoted 25 lines

Both sides of the VCHIQ communications mechanism need to agree on the cache line size. Using an incorrect value can lead to data corruption, but having the two sides using different values is usually worse.

In the absence of an obvious convenient run-time method to determine the correct value in the ARCH=arm world, the downstream Raspberry Pi trees used a Device Tree property, written by the firmware, to configure the kernel driver. This method was vetoed during the upstreaming process, so a fixed value of 32 was used instead, and some corruptions ensued. This is take 2 at arriving at the correct value.

Add a new compatible string - "brcm,bcm2836-vchiq" - to indicate an SoC with a 64-byte cache line. Document the new string in the binding, and use it on the appropriate platforms.

The final patch is a (seemingly cosmetic) correction of the Device Tree "reg" declaration for the device node, but it doubles as an indication to the Raspberry Pi firmware that the kernel driver is running a recent kernel driver that chooses the correct value. As such it would help if the DT patches are not merged before the driver patch.

v3: Builds without errors, tested on multiple Raspberry Pi models.
v2: Replaced ARM-specific logic used to determine cache line size with
a new compatible string for BCM2836 and BCM2837.

Quoted 5 lines

Phil Elwell (4):
staging/vc04_services: Use correct cache line size
dt-bindings: soc: Document "brcm,bcm2836-vchiq"
ARM: dts: bcm283x: Correct vchiq compatible string
ARM: dts: bcm283x: Correct mailbox register sizes

since my pull requests are out, would it be okay to apply patch #1 for 4.20 and the DT stuff for 4.21 (with the assumption Rob is okay with these patches)?

Quoted 5 lines
Quoted 1 line

Patch 4 is the only one I'd like to be delayed, but delaying 2-4 is fine with me.

Humm, did you mean you would like not to be delayed? In any case Stefan, you can send an additional pull request, and I will merge it and send a second pull request towards ARM SoC maintainers, that's not a problem.

Quoted 6 lines

No, I meant what I wrote - I would prefer patch 1 to be merged before patch 4 (or at least in the same release) to avoid the need for another firmware change, hence delaying patch 4 is good. It makes sense for the other commits to be merged in that order, but the normal compatible-string fallback mechanism means there is no hard dependency there.

Phil

Patches #2 - #4 applied to bcm2835-dt-next

Thanks