[PATCH v5 0/2] ASPEED: Add mailbox driver for AST2700 series

[PATCH v5 0/2] ASPEED: Add mailbox driver for AST2700 series

[Jammy Huang]

Add mailbox controller driver for AST27XX SoCs, which provides
independent tx/rx mailbox between different processors. There are 4
channels for each tx/rx mailbox and each channel has an 32-byte FIFO.

 v5 changes:
  - Update document
     1. Separate reg from 1 to 2. 1st is tx controller; 2nd is rx.
     2. Remove 'Reviewed-by' since the patch has changed.
  - Update driver, no functional changes.
     1. Update since there is 2 reg base now.
     2. Refine reg definitions
     3. Add spinlock to protect registers
     4. Use bool as return value for ast2700_mbox_tx_done
     5. Rename variable from drv_data to dev_data.

 v4 changes:
  - Update driver, no functional changes.
     1. Remove unused variable, rx_buff, in struct ast2700_mbox.
     2. Remove unneeded cast on device_get_match_data.
     3. Remove the usage of writel/readl_relaxed.
     4. Improve readability.
 v3 changes:
  - Correct document
     1. Use 32-bit addressing in dts example property, reg.
 v2 changes:
  - Update document
     1. Correct error in dts example.
     2. Drop description for mbox-cell per suggestion previously.

Jammy Huang (2):
  dt-bindings: mailbox: Add ASPEED AST2700 series SoC
  mailbox: aspeed: add mailbox driver for AST27XX series SoC

 .../mailbox/aspeed,ast2700-mailbox.yaml       |  60 +++++
 drivers/mailbox/Kconfig                       |   8 +
 drivers/mailbox/Makefile                      |   2 +
 drivers/mailbox/ast2700-mailbox.c             | 240 ++++++++++++++++++
 4 files changed, 310 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml
 create mode 100644 drivers/mailbox/ast2700-mailbox.c


base-commit: ec7714e4947909190ffb3041a03311a975350fe0
-- 
2.25.1

[PATCH v5 1/2] dt-bindings: mailbox: Add ASPEED AST2700 series SoC

[Jammy Huang]

Introduce the mailbox module for AST27XX series SoC, which is responsible
for interchanging messages between asymmetric processors.

Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com>
---
 .../mailbox/aspeed,ast2700-mailbox.yaml       | 60 +++++++++++++++++++
 1 file changed, 60 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml

diff --git a/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml b/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml
new file mode 100644
index 000000000000..0a5f43de5f28
--- /dev/null
+++ b/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml
@@ -0,0 +1,60 @@
+# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/mailbox/aspeed,ast2700-mailbox.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: ASPEED AST2700 mailbox controller
+
+maintainers:
+  - Jammy Huang <jammy_huang@aspeedtech.com>
+
+description:
+  ASPEED AST2700 has multiple processors that need to communicate with each
+  other. The mailbox controller provides a way for these processors to send
+  messages to each other. It is a hardware-based inter-processor communication
+  mechanism that allows processors to send and receive messages through
+  dedicated channels.
+  The mailbox's tx/rx are independent, meaning that one processor can send a
+  message while another processor is receiving a message simultaneously.
+  There are 4 channels available for both tx and rx operations. Each channel
+  has a FIFO buffer that can hold messages of a fixed size (32 bytes in this
+  case).
+  The mailbox controller also supports interrupt generation, allowing
+  processors to notify each other when a message is available or when an event
+  occurs.
+
+properties:
+  compatible:
+    const: aspeed,ast2700-mailbox
+
+  reg:
+    maxItems: 2
+    description:
+      Contains the base addresses and sizes of the mailbox controller. 1st one
+      is for TX control register; 2nd one is for RX control register.
+
+  interrupts:
+    maxItems: 1
+
+  "#mbox-cells":
+    const: 1
+
+required:
+  - compatible
+  - reg
+  - interrupts
+  - "#mbox-cells"
+
+additionalProperties: false
+
+examples:
+  - |
+    #include <dt-bindings/interrupt-controller/arm-gic.h>
+
+    mailbox@12c1c200 {
+        compatible = "aspeed,ast2700-mailbox";
+        reg = <0x12c1c200 0x100>, <0x12c1c300 0x100>;
+        interrupts = <GIC_SPI 103 IRQ_TYPE_LEVEL_HIGH>;
+        #mbox-cells = <1>;
+    };
-- 
2.25.1

Re: [PATCH v5 1/2] dt-bindings: mailbox: Add ASPEED AST2700 series SoC

[Rob Herring]

On Wed, Jun 25, 2025 at 03:34:16PM +0800, Jammy Huang wrote:
> Introduce the mailbox module for AST27XX series SoC, which is responsible
> for interchanging messages between asymmetric processors.
> 
> Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com>

You didn't add Krzysztof's Reviewed-by...

> ---
>  .../mailbox/aspeed,ast2700-mailbox.yaml       | 60 +++++++++++++++++++
>  1 file changed, 60 insertions(+)
>  create mode 100644 Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml
> 
> diff --git a/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml b/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml
> new file mode 100644
> index 000000000000..0a5f43de5f28
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml
> @@ -0,0 +1,60 @@
> +# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause)
> +%YAML 1.2
> +---
> +$id: http://devicetree.org/schemas/mailbox/aspeed,ast2700-mailbox.yaml#
> +$schema: http://devicetree.org/meta-schemas/core.yaml#
> +
> +title: ASPEED AST2700 mailbox controller
> +
> +maintainers:
> +  - Jammy Huang <jammy_huang@aspeedtech.com>
> +
> +description:

You need '>' to preserve paragraphs.

> +  ASPEED AST2700 has multiple processors that need to communicate with each
> +  other. The mailbox controller provides a way for these processors to send
> +  messages to each other. It is a hardware-based inter-processor communication
> +  mechanism that allows processors to send and receive messages through
> +  dedicated channels.

And a blank line between paragraphs.

> +  The mailbox's tx/rx are independent, meaning that one processor can send a
> +  message while another processor is receiving a message simultaneously.
> +  There are 4 channels available for both tx and rx operations. Each channel
> +  has a FIFO buffer that can hold messages of a fixed size (32 bytes in this
> +  case).

And here.

> +  The mailbox controller also supports interrupt generation, allowing
> +  processors to notify each other when a message is available or when an event
> +  occurs.
> +
> +properties:
> +  compatible:
> +    const: aspeed,ast2700-mailbox
> +
> +  reg:
> +    maxItems: 2
> +    description:
> +      Contains the base addresses and sizes of the mailbox controller. 1st one
> +      is for TX control register; 2nd one is for RX control register.

Instead, just:

items:
  - description: TX control register
  - description: RX control register

> +
> +  interrupts:
> +    maxItems: 1
> +
> +  "#mbox-cells":
> +    const: 1
> +
> +required:
> +  - compatible
> +  - reg
> +  - interrupts
> +  - "#mbox-cells"
> +
> +additionalProperties: false
> +
> +examples:
> +  - |
> +    #include <dt-bindings/interrupt-controller/arm-gic.h>
> +
> +    mailbox@12c1c200 {
> +        compatible = "aspeed,ast2700-mailbox";
> +        reg = <0x12c1c200 0x100>, <0x12c1c300 0x100>;
> +        interrupts = <GIC_SPI 103 IRQ_TYPE_LEVEL_HIGH>;
> +        #mbox-cells = <1>;
> +    };
> -- 
> 2.25.1
> 

RE: [PATCH v5 1/2] dt-bindings: mailbox: Add ASPEED AST2700 series SoC

[Jammy Huang]

> On Wed, Jun 25, 2025 at 03:34:16PM +0800, Jammy Huang wrote:
> > Introduce the mailbox module for AST27XX series SoC, which is
> > responsible for interchanging messages between asymmetric processors.
> >
> > Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com>
> 
> You didn't add Krzysztof's Reviewed-by...
Hi Rob,

I checked submitting patches' guide, and it says that if the patch has changed
substantially in following version, these tags might not be applicable anymore
and thus should be removed. I also mentioned this in changelog on the cover.

I am not sure about this, so may I keep Krzk's Reviewed-by tag?
If it's OK, I will add it back on next patch.

> 
> > ---
> >  .../mailbox/aspeed,ast2700-mailbox.yaml       | 60
> +++++++++++++++++++
> >  1 file changed, 60 insertions(+)
> >  create mode 100644
> > Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yaml
> >
> > diff --git
> > a/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yam
> > l
> > b/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox.yam
> > l
> > new file mode 100644
> > index 000000000000..0a5f43de5f28
> > --- /dev/null
> > +++ b/Documentation/devicetree/bindings/mailbox/aspeed,ast2700-mailbox
> > +++ .yaml
> > @@ -0,0 +1,60 @@
> > +# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause) %YAML 1.2
> > +---
> > +$id:
> > +http://devicetree.org/schemas/mailbox/aspeed,ast2700-mailbox.yaml#
> > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > +
> > +title: ASPEED AST2700 mailbox controller
> > +
> > +maintainers:
> > +  - Jammy Huang <jammy_huang@aspeedtech.com>
> > +
> > +description:
> 
> You need '>' to preserve paragraphs.
OK

> 
> > +  ASPEED AST2700 has multiple processors that need to communicate
> > + with each  other. The mailbox controller provides a way for these
> > + processors to send  messages to each other. It is a hardware-based
> > + inter-processor communication  mechanism that allows processors to
> > + send and receive messages through  dedicated channels.
> 
> And a blank line between paragraphs.
> 
OK

> > +  The mailbox's tx/rx are independent, meaning that one processor can
> > + send a  message while another processor is receiving a message
> simultaneously.
> > +  There are 4 channels available for both tx and rx operations. Each
> > + channel  has a FIFO buffer that can hold messages of a fixed size
> > + (32 bytes in this  case).
> 
> And here.
> 
Sure.

> > +  The mailbox controller also supports interrupt generation, allowing
> > + processors to notify each other when a message is available or when
> > + an event  occurs.
> > +
> > +properties:
> > +  compatible:
> > +    const: aspeed,ast2700-mailbox
> > +
> > +  reg:
> > +    maxItems: 2
> > +    description:
> > +      Contains the base addresses and sizes of the mailbox controller. 1st
> one
> > +      is for TX control register; 2nd one is for RX control register.
> 
> Instead, just:
> 
> items:
>   - description: TX control register
>   - description: RX control register
> 
Agree.

> > +
> > +  interrupts:
> > +    maxItems: 1
> > +
> > +  "#mbox-cells":
> > +    const: 1
> > +
> > +required:
> > +  - compatible
> > +  - reg
> > +  - interrupts
> > +  - "#mbox-cells"
> > +
> > +additionalProperties: false
> > +
> > +examples:
> > +  - |
> > +    #include <dt-bindings/interrupt-controller/arm-gic.h>
> > +
> > +    mailbox@12c1c200 {
> > +        compatible = "aspeed,ast2700-mailbox";
> > +        reg = <0x12c1c200 0x100>, <0x12c1c300 0x100>;
> > +        interrupts = <GIC_SPI 103 IRQ_TYPE_LEVEL_HIGH>;
> > +        #mbox-cells = <1>;
> > +    };
> > --
> > 2.25.1
> >
Thanks for your help.

Regards
Jammy

Re: [PATCH v5 1/2] dt-bindings: mailbox: Add ASPEED AST2700 series SoC

[Andrew Jeffery]

On Wed, 2025-06-25 at 10:50 -0500, Rob Herring wrote:
> On Wed, Jun 25, 2025 at 03:34:16PM +0800, Jammy Huang wrote:
> > Introduce the mailbox module for AST27XX series SoC, which is responsible
> > for interchanging messages between asymmetric processors.
> > 
> > Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com>
> 

*snip*

> > +properties:
> > +  compatible:
> > +    const: aspeed,ast2700-mailbox
> > +
> > +  reg:
> > +    maxItems: 2
> > +    description:
> > +      Contains the base addresses and sizes of the mailbox controller. 1st one
> > +      is for TX control register; 2nd one is for RX control register.
> 
> Instead, just:
> 
> items:
>   - description: TX control register
>   - description: RX control register

Maybe also specify reg-names with "tx" and "rx"? That way we can use
devm_platform_ioremap_resource_byname() in the linux driver
implementation, which should look tidier.

Andrew

RE: [PATCH v5 1/2] dt-bindings: mailbox: Add ASPEED AST2700 series SoC

[Jammy Huang]

> On Wed, 2025-06-25 at 10:50 -0500, Rob Herring wrote:
> > On Wed, Jun 25, 2025 at 03:34:16PM +0800, Jammy Huang wrote:
> > > Introduce the mailbox module for AST27XX series SoC, which is
> > > responsible for interchanging messages between asymmetric processors.
> > >
> > > Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com>
> >
> 
> *snip*
> 
> > > +properties:
> > > +  compatible:
> > > +    const: aspeed,ast2700-mailbox
> > > +
> > > +  reg:
> > > +    maxItems: 2
> > > +    description:
> > > +      Contains the base addresses and sizes of the mailbox
> > > +controller. 1st one
> > > +      is for TX control register; 2nd one is for RX control register.
> >
> > Instead, just:
> >
> > items:
> >   - description: TX control register
> >   - description: RX control register
> 
> Maybe also specify reg-names with "tx" and "rx"? That way we can use
> devm_platform_ioremap_resource_byname() in the linux driver
> implementation, which should look tidier.
Sounds good. I will do it.
> 
> Andrew

Regards
Jammy

[PATCH v5 2/2] mailbox: aspeed: add mailbox driver for AST27XX series SoC

[Jammy Huang]

Add mailbox controller driver for AST27XX SoCs, which provides
independent tx/rx mailbox between different processors. There are 4
channels for each tx/rx mailbox and each channel has an 32-byte FIFO.

Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com>
---
 drivers/mailbox/Kconfig           |   8 +
 drivers/mailbox/Makefile          |   2 +
 drivers/mailbox/ast2700-mailbox.c | 240 ++++++++++++++++++++++++++++++
 3 files changed, 250 insertions(+)
 create mode 100644 drivers/mailbox/ast2700-mailbox.c

diff --git a/drivers/mailbox/Kconfig b/drivers/mailbox/Kconfig
index 68eeed660a4a..1c38cd570091 100644
--- a/drivers/mailbox/Kconfig
+++ b/drivers/mailbox/Kconfig
@@ -340,4 +340,12 @@ config THEAD_TH1520_MBOX
 	  kernel is running, and E902 core used for power management among other
 	  things.
 
+config AST2700_MBOX
+	tristate "ASPEED AST2700 IPC driver"
+	depends on ARCH_ASPEED || COMPILE_TEST
+	help
+	  Mailbox driver implementation for ASPEED AST27XX SoCs. This driver
+	  can be used to send message between different processors in SoC.
+	  The driver provides mailbox support for sending interrupts to the
+	  clients. Say Y here if you want to build this driver.
 endif
diff --git a/drivers/mailbox/Makefile b/drivers/mailbox/Makefile
index 13a3448b3271..9a9add9a7548 100644
--- a/drivers/mailbox/Makefile
+++ b/drivers/mailbox/Makefile
@@ -72,3 +72,5 @@ obj-$(CONFIG_QCOM_CPUCP_MBOX)	+= qcom-cpucp-mbox.o
 obj-$(CONFIG_QCOM_IPCC)		+= qcom-ipcc.o
 
 obj-$(CONFIG_THEAD_TH1520_MBOX)	+= mailbox-th1520.o
+
+obj-$(CONFIG_AST2700_MBOX)	+= ast2700-mailbox.o
diff --git a/drivers/mailbox/ast2700-mailbox.c b/drivers/mailbox/ast2700-mailbox.c
new file mode 100644
index 000000000000..5470053f8139
--- /dev/null
+++ b/drivers/mailbox/ast2700-mailbox.c
@@ -0,0 +1,240 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright Aspeed Technology Inc. (C) 2025. All rights reserved
+ */
+
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/iopoll.h>
+#include <linux/kernel.h>
+#include <linux/mailbox_controller.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+
+/* Each bit in the register represents an IPC ID */
+#define IPCR_TX_TRIG		0x00
+#define IPCR_ENABLE		0x04
+#define IPCR_STATUS		0x08
+#define  RX_IRQ(n)		BIT(n)
+#define  RX_IRQ_MASK		0xf
+#define IPCR_DATA		0x10
+
+struct ast2700_mbox_data {
+	u8 num_chans;
+	u8 msg_size;
+};
+
+struct ast2700_mbox {
+	struct mbox_controller mbox;
+	u8 msg_size;
+	void __iomem *tx_regs;
+	void __iomem *rx_regs;
+	spinlock_t lock;
+};
+
+static inline int ch_num(struct mbox_chan *chan)
+{
+	return chan - chan->mbox->chans;
+}
+
+static inline bool ast2700_mbox_tx_done(struct ast2700_mbox *mb, int idx)
+{
+	return !(readl(mb->tx_regs + IPCR_STATUS) & BIT(idx));
+}
+
+static irqreturn_t ast2700_mbox_irq(int irq, void *p)
+{
+	struct ast2700_mbox *mb = p;
+	void __iomem *data_reg;
+	int num_words;
+	u32 *word_data;
+	u32 status;
+	int n;
+
+	/* Only examine channels that are currently enabled. */
+	status = readl(mb->rx_regs + IPCR_ENABLE) &
+		 readl(mb->rx_regs + IPCR_STATUS);
+
+	if (!(status & RX_IRQ_MASK))
+		return IRQ_NONE;
+
+	for (n = 0; n < mb->mbox.num_chans; ++n) {
+		struct mbox_chan *chan = &mb->mbox.chans[n];
+
+		if (!(status & RX_IRQ(n)))
+			continue;
+
+		/* Read the message data */
+		for (data_reg = mb->rx_regs + IPCR_DATA + mb->msg_size * n,
+		     word_data = chan->con_priv,
+		     num_words = (mb->msg_size / sizeof(u32));
+		     num_words;
+		     num_words--, data_reg += sizeof(u32), word_data++)
+			*word_data = readl(data_reg);
+
+		mbox_chan_received_data(chan, chan->con_priv);
+
+		/* The IRQ can be cleared only once the FIFO is empty. */
+		writel(RX_IRQ(n), mb->rx_regs + IPCR_STATUS);
+	}
+
+	return IRQ_HANDLED;
+}
+
+static int ast2700_mbox_send_data(struct mbox_chan *chan, void *data)
+{
+	struct ast2700_mbox *mb = dev_get_drvdata(chan->mbox->dev);
+	void __iomem *data_reg;
+	u32 *word_data;
+	int num_words;
+	int idx = ch_num(chan);
+
+	if (!(readl(mb->tx_regs + IPCR_ENABLE) & BIT(idx))) {
+		dev_warn(mb->mbox.dev, "%s: Ch-%d not enabled yet\n", __func__, idx);
+		return -EBUSY;
+	}
+
+	if (!(ast2700_mbox_tx_done(mb, idx))) {
+		dev_warn(mb->mbox.dev, "%s: Ch-%d last data has not finished\n", __func__, idx);
+		return -EBUSY;
+	}
+
+	/* Write the message data */
+	for (data_reg = mb->tx_regs + IPCR_DATA + mb->msg_size * idx,
+	     word_data = (u32 *)data,
+	     num_words = (mb->msg_size / sizeof(u32));
+	     num_words;
+	     num_words--, data_reg += sizeof(u32), word_data++)
+		writel(*word_data, data_reg);
+
+	writel(BIT(idx), mb->tx_regs + IPCR_TX_TRIG);
+	dev_dbg(mb->mbox.dev, "%s: Ch-%d sent\n", __func__, idx);
+
+	return 0;
+}
+
+static int ast2700_mbox_startup(struct mbox_chan *chan)
+{
+	struct ast2700_mbox *mb = dev_get_drvdata(chan->mbox->dev);
+	int idx = ch_num(chan);
+	void __iomem *reg = mb->rx_regs + IPCR_ENABLE;
+	unsigned long flags;
+
+	spin_lock_irqsave(&mb->lock, flags);
+	writel(readl(reg) | BIT(idx), reg);
+	spin_unlock_irqrestore(&mb->lock, flags);
+
+	return 0;
+}
+
+static void ast2700_mbox_shutdown(struct mbox_chan *chan)
+{
+	struct ast2700_mbox *mb = dev_get_drvdata(chan->mbox->dev);
+	int idx = ch_num(chan);
+	void __iomem *reg = mb->rx_regs + IPCR_ENABLE;
+	unsigned long flags;
+
+	spin_lock_irqsave(&mb->lock, flags);
+	writel(readl(reg) & ~BIT(idx), reg);
+	spin_unlock_irqrestore(&mb->lock, flags);
+}
+
+static bool ast2700_mbox_last_tx_done(struct mbox_chan *chan)
+{
+	struct ast2700_mbox *mb = dev_get_drvdata(chan->mbox->dev);
+	int idx = ch_num(chan);
+
+	return ast2700_mbox_tx_done(mb, idx);
+}
+
+static const struct mbox_chan_ops ast2700_mbox_chan_ops = {
+	.send_data	= ast2700_mbox_send_data,
+	.startup	= ast2700_mbox_startup,
+	.shutdown	= ast2700_mbox_shutdown,
+	.last_tx_done	= ast2700_mbox_last_tx_done,
+};
+
+static int ast2700_mbox_probe(struct platform_device *pdev)
+{
+	struct ast2700_mbox *mb;
+	const struct ast2700_mbox_data *dev_data;
+	struct device *dev = &pdev->dev;
+	int irq, ret;
+
+	if (!pdev->dev.of_node)
+		return -ENODEV;
+
+	dev_data = device_get_match_data(&pdev->dev);
+
+	mb = devm_kzalloc(dev, sizeof(*mb), GFP_KERNEL);
+	if (!mb)
+		return -ENOMEM;
+
+	mb->mbox.chans = devm_kcalloc(&pdev->dev, dev_data->num_chans,
+				      sizeof(*mb->mbox.chans), GFP_KERNEL);
+	if (!mb->mbox.chans)
+		return -ENOMEM;
+
+	/* con_priv of each channel is used to store the message received */
+	for (int i = 0; i < dev_data->num_chans; i++) {
+		mb->mbox.chans[i].con_priv = devm_kcalloc(dev, dev_data->msg_size,
+							  sizeof(u8), GFP_KERNEL);
+		if (!mb->mbox.chans[i].con_priv)
+			return -ENOMEM;
+	}
+
+	platform_set_drvdata(pdev, mb);
+
+	mb->tx_regs = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(mb->tx_regs))
+		return PTR_ERR(mb->tx_regs);
+
+	mb->rx_regs = devm_platform_ioremap_resource(pdev, 1);
+	if (IS_ERR(mb->rx_regs))
+		return PTR_ERR(mb->rx_regs);
+
+	mb->msg_size = dev_data->msg_size;
+	mb->mbox.dev = dev;
+	mb->mbox.num_chans = dev_data->num_chans;
+	mb->mbox.ops = &ast2700_mbox_chan_ops;
+	mb->mbox.txdone_irq = false;
+	mb->mbox.txdone_poll = true;
+	mb->mbox.txpoll_period = 5;
+	spin_lock_init(&mb->lock);
+
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return irq;
+
+	ret = devm_request_irq(dev, irq, ast2700_mbox_irq, 0, dev_name(dev), mb);
+	if (ret)
+		return ret;
+
+	return devm_mbox_controller_register(dev, &mb->mbox);
+}
+
+static const struct ast2700_mbox_data ast2700_dev_data = {
+	.num_chans = 4,
+	.msg_size = 0x20,
+};
+
+static const struct of_device_id ast2700_mbox_of_match[] = {
+	{ .compatible = "aspeed,ast2700-mailbox", .data = &ast2700_dev_data },
+	{}
+};
+MODULE_DEVICE_TABLE(of, ast2700_mbox_of_match);
+
+static struct platform_driver ast2700_mbox_driver = {
+	.driver = {
+		.name = "ast2700-mailbox",
+		.of_match_table = ast2700_mbox_of_match,
+	},
+	.probe = ast2700_mbox_probe,
+};
+module_platform_driver(ast2700_mbox_driver);
+
+MODULE_AUTHOR("Jammy Huang <jammy_huang@aspeedtech.com>");
+MODULE_DESCRIPTION("ASPEED AST2700 IPC driver");
+MODULE_LICENSE("GPL");
-- 
2.25.1

Re: [PATCH v5 2/2] mailbox: aspeed: add mailbox driver for AST27XX series SoC

[Andrew Jeffery]

On Wed, 2025-06-25 at 15:34 +0800, Jammy Huang wrote:
> Add mailbox controller driver for AST27XX SoCs, which provides
> independent tx/rx mailbox between different processors. There are 4
> channels for each tx/rx mailbox and each channel has an 32-byte FIFO.
> 
> Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com>
> ---
>  drivers/mailbox/Kconfig           |   8 +
>  drivers/mailbox/Makefile          |   2 +
>  drivers/mailbox/ast2700-mailbox.c | 240 ++++++++++++++++++++++++++++++
>  3 files changed, 250 insertions(+)
>  create mode 100644 drivers/mailbox/ast2700-mailbox.c
> 
> diff --git a/drivers/mailbox/Kconfig b/drivers/mailbox/Kconfig
> index 68eeed660a4a..1c38cd570091 100644
> --- a/drivers/mailbox/Kconfig
> +++ b/drivers/mailbox/Kconfig
> @@ -340,4 +340,12 @@ config THEAD_TH1520_MBOX
>           kernel is running, and E902 core used for power management among other
>           things.
>  
> +config AST2700_MBOX
> +       tristate "ASPEED AST2700 IPC driver"
> +       depends on ARCH_ASPEED || COMPILE_TEST
> +       help
> +         Mailbox driver implementation for ASPEED AST27XX SoCs. This driver
> +         can be used to send message between different processors in SoC.
> +         The driver provides mailbox support for sending interrupts to the
> +         clients. Say Y here if you want to build this driver.
>  endif
> diff --git a/drivers/mailbox/Makefile b/drivers/mailbox/Makefile
> index 13a3448b3271..9a9add9a7548 100644
> --- a/drivers/mailbox/Makefile
> +++ b/drivers/mailbox/Makefile
> @@ -72,3 +72,5 @@ obj-$(CONFIG_QCOM_CPUCP_MBOX) += qcom-cpucp-mbox.o
>  obj-$(CONFIG_QCOM_IPCC)                += qcom-ipcc.o
>  
>  obj-$(CONFIG_THEAD_TH1520_MBOX)        += mailbox-th1520.o
> +
> +obj-$(CONFIG_AST2700_MBOX)     += ast2700-mailbox.o
> diff --git a/drivers/mailbox/ast2700-mailbox.c b/drivers/mailbox/ast2700-mailbox.c
> new file mode 100644
> index 000000000000..5470053f8139
> --- /dev/null
> +++ b/drivers/mailbox/ast2700-mailbox.c
> @@ -0,0 +1,240 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Copyright Aspeed Technology Inc. (C) 2025. All rights reserved
> + */
> +
> +#include <linux/interrupt.h>
> +#include <linux/io.h>
> +#include <linux/iopoll.h>
> +#include <linux/kernel.h>
> +#include <linux/mailbox_controller.h>
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/platform_device.h>
> +#include <linux/slab.h>
> +
> +/* Each bit in the register represents an IPC ID */
> +#define IPCR_TX_TRIG           0x00
> +#define IPCR_ENABLE            0x04
> +#define IPCR_STATUS            0x08
> +#define  RX_IRQ(n)             BIT(n)
> +#define  RX_IRQ_MASK           0xf
> +#define IPCR_DATA              0x10
> +
> +struct ast2700_mbox_data {
> +       u8 num_chans;
> +       u8 msg_size;
> +};
> +
> +struct ast2700_mbox {
> +       struct mbox_controller mbox;
> +       u8 msg_size;
> +       void __iomem *tx_regs;
> +       void __iomem *rx_regs;
> +       spinlock_t lock;
> +};
> +
> +static inline int ch_num(struct mbox_chan *chan)
> +{
> +       return chan - chan->mbox->chans;
> +}
> +
> +static inline bool ast2700_mbox_tx_done(struct ast2700_mbox *mb, int idx)
> +{
> +       return !(readl(mb->tx_regs + IPCR_STATUS) & BIT(idx));
> +}
> +
> +static irqreturn_t ast2700_mbox_irq(int irq, void *p)
> +{
> +       struct ast2700_mbox *mb = p;
> +       void __iomem *data_reg;
> +       int num_words;
> +       u32 *word_data;
> +       u32 status;
> +       int n;
> +
> +       /* Only examine channels that are currently enabled. */
> +       status = readl(mb->rx_regs + IPCR_ENABLE) &
> +                readl(mb->rx_regs + IPCR_STATUS);

Also need to lock over these reads to handle concurrent
startup/shutdown?

> +
> +       if (!(status & RX_IRQ_MASK))
> +               return IRQ_NONE;
> +
> +       for (n = 0; n < mb->mbox.num_chans; ++n) {
> +               struct mbox_chan *chan = &mb->mbox.chans[n];
> +
> +               if (!(status & RX_IRQ(n)))
> +                       continue;
> +
> +               /* Read the message data */
> +               for (data_reg = mb->rx_regs + IPCR_DATA + mb->msg_size * n,
> +                    word_data = chan->con_priv,
> +                    num_words = (mb->msg_size / sizeof(u32));
> +                    num_words;
> +                    num_words--, data_reg += sizeof(u32), word_data++)
> +                       *word_data = readl(data_reg);
> +
> +               mbox_chan_received_data(chan, chan->con_priv);
> +
> +               /* The IRQ can be cleared only once the FIFO is empty. */
> +               writel(RX_IRQ(n), mb->rx_regs + IPCR_STATUS);

Can we rather clear the status once outside the loop immediately before
return?

writel(status, mb->rx_regs + IPCR_STATUS);

> +       }
> +
> +       return IRQ_HANDLED;
> +}
> +
> +static int ast2700_mbox_send_data(struct mbox_chan *chan, void *data)
> +{
> +       struct ast2700_mbox *mb = dev_get_drvdata(chan->mbox->dev);
> +       void __iomem *data_reg;
> +       u32 *word_data;
> +       int num_words;
> +       int idx = ch_num(chan);
> +
> +       if (!(readl(mb->tx_regs + IPCR_ENABLE) & BIT(idx))) {
> +               dev_warn(mb->mbox.dev, "%s: Ch-%d not enabled yet\n", __func__, idx);
> +               return -EBUSY;

-ENODEV? I wouldn't say it's busy here :)

> +       }
> +
> +       if (!(ast2700_mbox_tx_done(mb, idx))) {
> +               dev_warn(mb->mbox.dev, "%s: Ch-%d last data has not finished\n", __func__, idx);
> +               return -EBUSY;
> +       }
> +
> +       /* Write the message data */
> +       for (data_reg = mb->tx_regs + IPCR_DATA + mb->msg_size * idx,
> +            word_data = (u32 *)data,
> +            num_words = (mb->msg_size / sizeof(u32));
> +            num_words;
> +            num_words--, data_reg += sizeof(u32), word_data++)
> +               writel(*word_data, data_reg);
> +
> +       writel(BIT(idx), mb->tx_regs + IPCR_TX_TRIG);
> +       dev_dbg(mb->mbox.dev, "%s: Ch-%d sent\n", __func__, idx);
> +
> +       return 0;
> +}
> +

*snip*

> +
> +static int ast2700_mbox_probe(struct platform_device *pdev)
> +{
> +       struct ast2700_mbox *mb;
> +       const struct ast2700_mbox_data *dev_data;
> +       struct device *dev = &pdev->dev;
> +       int irq, ret;
> +
> +       if (!pdev->dev.of_node)
> +               return -ENODEV;
> +
> +       dev_data = device_get_match_data(&pdev->dev);
> +
> +       mb = devm_kzalloc(dev, sizeof(*mb), GFP_KERNEL);
> +       if (!mb)
> +               return -ENOMEM;
> +
> +       mb->mbox.chans = devm_kcalloc(&pdev->dev, dev_data->num_chans,
> +                                     sizeof(*mb->mbox.chans), GFP_KERNEL);
> +       if (!mb->mbox.chans)
> +               return -ENOMEM;
> +
> +       /* con_priv of each channel is used to store the message received */
> +       for (int i = 0; i < dev_data->num_chans; i++) {
> +               mb->mbox.chans[i].con_priv = devm_kcalloc(dev, dev_data->msg_size,
> +                                                         sizeof(u8), GFP_KERNEL);
> +               if (!mb->mbox.chans[i].con_priv)
> +                       return -ENOMEM;
> +       }
> +
> +       platform_set_drvdata(pdev, mb);
> +
> +       mb->tx_regs = devm_platform_ioremap_resource(pdev, 0);

See the thought about reg-names on the binding.

Andrew

RE: [PATCH v5 2/2] mailbox: aspeed: add mailbox driver for AST27XX series SoC

[Jammy Huang]

> On Wed, 2025-06-25 at 15:34 +0800, Jammy Huang wrote:
> > Add mailbox controller driver for AST27XX SoCs, which provides
> > independent tx/rx mailbox between different processors. There are 4
> > channels for each tx/rx mailbox and each channel has an 32-byte FIFO.
> >
> > Signed-off-by: Jammy Huang <jammy_huang@aspeedtech.com>
> > ---
> >  drivers/mailbox/Kconfig           |   8 +
> >  drivers/mailbox/Makefile          |   2 +
> >  drivers/mailbox/ast2700-mailbox.c | 240
> > ++++++++++++++++++++++++++++++
> >  3 files changed, 250 insertions(+)
> >  create mode 100644 drivers/mailbox/ast2700-mailbox.c
> >
> > diff --git a/drivers/mailbox/Kconfig b/drivers/mailbox/Kconfig index
> > 68eeed660a4a..1c38cd570091 100644
> > --- a/drivers/mailbox/Kconfig
> > +++ b/drivers/mailbox/Kconfig
> > @@ -340,4 +340,12 @@ config THEAD_TH1520_MBOX
> >           kernel is running, and E902 core used for power
> management
> > among other
> >           things.
> >
> > +config AST2700_MBOX
> > +       tristate "ASPEED AST2700 IPC driver"
> > +       depends on ARCH_ASPEED || COMPILE_TEST
> > +       help
> > +         Mailbox driver implementation for ASPEED AST27XX SoCs.
> This
> > +driver
> > +         can be used to send message between different processors in
> SoC.
> > +         The driver provides mailbox support for sending interrupts
> > +to the
> > +         clients. Say Y here if you want to build this driver.
> >  endif
> > diff --git a/drivers/mailbox/Makefile b/drivers/mailbox/Makefile index
> > 13a3448b3271..9a9add9a7548 100644
> > --- a/drivers/mailbox/Makefile
> > +++ b/drivers/mailbox/Makefile
> > @@ -72,3 +72,5 @@ obj-$(CONFIG_QCOM_CPUCP_MBOX) +=
> qcom-cpucp-mbox.o
> >  obj-$(CONFIG_QCOM_IPCC)                += qcom-ipcc.o
> >
> >  obj-$(CONFIG_THEAD_TH1520_MBOX)        += mailbox-th1520.o
> > +
> > +obj-$(CONFIG_AST2700_MBOX)     += ast2700-mailbox.o
> > diff --git a/drivers/mailbox/ast2700-mailbox.c
> > b/drivers/mailbox/ast2700-mailbox.c
> > new file mode 100644
> > index 000000000000..5470053f8139
> > --- /dev/null
> > +++ b/drivers/mailbox/ast2700-mailbox.c
> > @@ -0,0 +1,240 @@
> > +// SPDX-License-Identifier: GPL-2.0-only
> > +/*
> > + * Copyright Aspeed Technology Inc. (C) 2025. All rights reserved  */
> > +
> > +#include <linux/interrupt.h>
> > +#include <linux/io.h>
> > +#include <linux/iopoll.h>
> > +#include <linux/kernel.h>
> > +#include <linux/mailbox_controller.h> #include <linux/module.h>
> > +#include <linux/of.h> #include <linux/platform_device.h> #include
> > +<linux/slab.h>
> > +
> > +/* Each bit in the register represents an IPC ID */ #define
> > +IPCR_TX_TRIG           0x00 #define
> IPCR_ENABLE            0x04
> > +#define IPCR_STATUS            0x08
> #define  RX_IRQ(n)
> > +BIT(n) #define  RX_IRQ_MASK           0xf #define
> IPCR_DATA
> > +0x10
> > +
> > +struct ast2700_mbox_data {
> > +       u8 num_chans;
> > +       u8 msg_size;
> > +};
> > +
> > +struct ast2700_mbox {
> > +       struct mbox_controller mbox;
> > +       u8 msg_size;
> > +       void __iomem *tx_regs;
> > +       void __iomem *rx_regs;
> > +       spinlock_t lock;
> > +};
> > +
> > +static inline int ch_num(struct mbox_chan *chan) {
> > +       return chan - chan->mbox->chans; }
> > +
> > +static inline bool ast2700_mbox_tx_done(struct ast2700_mbox *mb, int
> > +idx) {
> > +       return !(readl(mb->tx_regs + IPCR_STATUS) & BIT(idx)); }
> > +
> > +static irqreturn_t ast2700_mbox_irq(int irq, void *p) {
> > +       struct ast2700_mbox *mb = p;
> > +       void __iomem *data_reg;
> > +       int num_words;
> > +       u32 *word_data;
> > +       u32 status;
> > +       int n;
> > +
> > +       /* Only examine channels that are currently enabled. */
> > +       status = readl(mb->rx_regs + IPCR_ENABLE) &
> > +                readl(mb->rx_regs + IPCR_STATUS);
> 
> Also need to lock over these reads to handle concurrent startup/shutdown?
It read-only here. I think it's fine.

> 
> > +
> > +       if (!(status & RX_IRQ_MASK))
> > +               return IRQ_NONE;
> > +
> > +       for (n = 0; n < mb->mbox.num_chans; ++n) {
> > +               struct mbox_chan *chan = &mb->mbox.chans[n];
> > +
> > +               if (!(status & RX_IRQ(n)))
> > +                       continue;
> > +
> > +               /* Read the message data */
> > +               for (data_reg = mb->rx_regs + IPCR_DATA +
> mb->msg_size
> > +* n,
> > +                    word_data = chan->con_priv,
> > +                    num_words = (mb->msg_size / sizeof(u32));
> > +                    num_words;
> > +                    num_words--, data_reg += sizeof(u32),
> > +word_data++)
> > +                       *word_data = readl(data_reg);
> > +
> > +               mbox_chan_received_data(chan, chan->con_priv);
> > +
> > +               /* The IRQ can be cleared only once the FIFO is
> empty.
> > +*/
> > +               writel(RX_IRQ(n), mb->rx_regs + IPCR_STATUS);
> 
> Can we rather clear the status once outside the loop immediately before
> return?
> 
> writel(status, mb->rx_regs + IPCR_STATUS);
I think it's better to keep current way. Because tx side could polling check rx-status
to know if tx is done. mbox_chan_received_data() here will invoke the callback registered
by mailbox-client. It could take times depending on how the callback is implemented.
Since we have 4 channels now, clear it earlier can have better performance.

> 
> > +       }
> > +
> > +       return IRQ_HANDLED;
> > +}
> > +
> > +static int ast2700_mbox_send_data(struct mbox_chan *chan, void *data)
> > +{
> > +       struct ast2700_mbox *mb =
> dev_get_drvdata(chan->mbox->dev);
> > +       void __iomem *data_reg;
> > +       u32 *word_data;
> > +       int num_words;
> > +       int idx = ch_num(chan);
> > +
> > +       if (!(readl(mb->tx_regs + IPCR_ENABLE) & BIT(idx))) {
> > +               dev_warn(mb->mbox.dev, "%s: Ch-%d not enabled
> yet\n",
> > +__func__, idx);
> > +               return -EBUSY;
> 
> -ENODEV? I wouldn't say it's busy here :)
> 
> > +       }
> > +
> > +       if (!(ast2700_mbox_tx_done(mb, idx))) {
> > +               dev_warn(mb->mbox.dev, "%s: Ch-%d last data has
> not
> > +finished\n", __func__, idx);
> > +               return -EBUSY;
> > +       }
> > +
> > +       /* Write the message data */
> > +       for (data_reg = mb->tx_regs + IPCR_DATA + mb->msg_size * idx,
> > +            word_data = (u32 *)data,
> > +            num_words = (mb->msg_size / sizeof(u32));
> > +            num_words;
> > +            num_words--, data_reg += sizeof(u32), word_data++)
> > +               writel(*word_data, data_reg);
> > +
> > +       writel(BIT(idx), mb->tx_regs + IPCR_TX_TRIG);
> > +       dev_dbg(mb->mbox.dev, "%s: Ch-%d sent\n", __func__, idx);
> > +
> > +       return 0;
> > +}
> > +
> 
> *snip*
> 
> > +
> > +static int ast2700_mbox_probe(struct platform_device *pdev) {
> > +       struct ast2700_mbox *mb;
> > +       const struct ast2700_mbox_data *dev_data;
> > +       struct device *dev = &pdev->dev;
> > +       int irq, ret;
> > +
> > +       if (!pdev->dev.of_node)
> > +               return -ENODEV;
> > +
> > +       dev_data = device_get_match_data(&pdev->dev);
> > +
> > +       mb = devm_kzalloc(dev, sizeof(*mb), GFP_KERNEL);
> > +       if (!mb)
> > +               return -ENOMEM;
> > +
> > +       mb->mbox.chans = devm_kcalloc(&pdev->dev,
> dev_data->num_chans,
> >
> +                                     sizeof(*mb->mbox.c
> hans),
> > +GFP_KERNEL);
> > +       if (!mb->mbox.chans)
> > +               return -ENOMEM;
> > +
> > +       /* con_priv of each channel is used to store the message
> > +received */
> > +       for (int i = 0; i < dev_data->num_chans; i++) {
> > +               mb->mbox.chans[i].con_priv = devm_kcalloc(dev,
> > +dev_data->msg_size,
> >
> +
>        sizeof(u8),
> > +GFP_KERNEL);
> > +               if (!mb->mbox.chans[i].con_priv)
> > +                       return -ENOMEM;
> > +       }
> > +
> > +       platform_set_drvdata(pdev, mb);
> > +
> > +       mb->tx_regs = devm_platform_ioremap_resource(pdev, 0);
> 
> See the thought about reg-names on the binding.
Agree.

> 
> Andrew

Regards
Jammy