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

[Andrew Jeffery <andrew@codeconstruct.com.au>]

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