On Tue, Sep 24, 2024 at 10:39:40AM GMT, Shivnandan Kumar wrote: > The SC7280 CPUCP mailbox controller is compatible with legacy mailbox > hardware. > Implement support for this functionality which enable HLOS to s/HLOS/Linux/ > CPUCP communication. This enables Linux to do this and that. > > Signed-off-by: Shivnandan Kumar <quic_kshivnan@xxxxxxxxxxx> > --- > drivers/mailbox/qcom-cpucp-mbox.c | 156 +++++++++++++++++++++++------- > 1 file changed, 122 insertions(+), 34 deletions(-) > > diff --git a/drivers/mailbox/qcom-cpucp-mbox.c b/drivers/mailbox/qcom-cpucp-mbox.c > index e5437c294803..faae6e069ea1 100644 > --- a/drivers/mailbox/qcom-cpucp-mbox.c > +++ b/drivers/mailbox/qcom-cpucp-mbox.c > @@ -13,18 +13,24 @@ > #include <linux/platform_device.h> > > #define APSS_CPUCP_IPC_CHAN_SUPPORTED 3 > -#define APSS_CPUCP_MBOX_CMD_OFF 0x4 > - > -/* Tx Registers */ > -#define APSS_CPUCP_TX_MBOX_CMD(i) (0x100 + ((i) * 8)) > > /* Rx Registers */ > -#define APSS_CPUCP_RX_MBOX_CMD(i) (0x100 + ((i) * 8)) > -#define APSS_CPUCP_RX_MBOX_MAP 0x4000 > -#define APSS_CPUCP_RX_MBOX_STAT 0x4400 > -#define APSS_CPUCP_RX_MBOX_CLEAR 0x4800 > -#define APSS_CPUCP_RX_MBOX_EN 0x4c00 > -#define APSS_CPUCP_RX_MBOX_CMD_MASK GENMASK_ULL(63, 0) > +#define APSS_CPUCP_V2_RX_MBOX_CMD_MASK GENMASK_ULL(63, 0) > +#define APSS_CPUCP_V1_SEND_IRQ_VAL BIT(28) > +#define APSS_CPUCP_V1_CLEAR_IRQ_VAL BIT(3) > +#define APSS_CPUCP_V1_STATUS_IRQ_VAL BIT(3) > + > +struct qcom_cpucp_mbox_desc { > + u32 enable_reg; > + u32 map_reg; > + u32 rx_reg; > + u32 tx_reg; > + u32 status_reg; > + u32 clear_reg; > + u32 chan_stride; > + bool v2_mbox; > + u32 num_chans; > +}; > > /** > * struct qcom_cpucp_mbox - Holder for the mailbox driver > @@ -35,6 +41,7 @@ > */ > struct qcom_cpucp_mbox { > struct mbox_chan chans[APSS_CPUCP_IPC_CHAN_SUPPORTED]; > + const struct qcom_cpucp_mbox_desc *desc; > struct mbox_controller mbox; > void __iomem *tx_base; > void __iomem *rx_base; > @@ -48,13 +55,40 @@ static inline int channel_number(struct mbox_chan *chan) > static irqreturn_t qcom_cpucp_mbox_irq_fn(int irq, void *data) qcom_cpucp_v1_mbox_irq_fn() > { > struct qcom_cpucp_mbox *cpucp = data; > + const struct qcom_cpucp_mbox_desc *desc = cpucp->desc; > + int i; > + > + for (i = 0; i < desc->num_chans; i++) { > + u32 val = readl(cpucp->rx_base + desc->status_reg + (i * desc->chan_stride)); > + struct mbox_chan *chan = &cpucp->chans[i]; > + unsigned long flags; > + > + if (val & APSS_CPUCP_V1_STATUS_IRQ_VAL) { > + writel(APSS_CPUCP_V1_CLEAR_IRQ_VAL, > + cpucp->rx_base + desc->clear_reg + (i * desc->chan_stride)); > + /* Make sure reg write is complete before proceeding */ > + mb(); > + spin_lock_irqsave(&chan->lock, flags); > + if (chan->cl) > + mbox_chan_received_data(chan, NULL); v1 clears IRQ before processing, v2 does it after pinging mbox subsys. Is that expected? What warrants such a difference? > + spin_unlock_irqrestore(&chan->lock, flags); > + } > + } > + > + return IRQ_HANDLED; > +} > + > +static irqreturn_t qcom_cpucp_v2_mbox_irq_fn(int irq, void *data) > +{ > + struct qcom_cpucp_mbox *cpucp = data; > + const struct qcom_cpucp_mbox_desc *desc = cpucp->desc; > u64 status; > int i; > > - status = readq(cpucp->rx_base + APSS_CPUCP_RX_MBOX_STAT); > + status = readq(cpucp->rx_base + desc->status_reg); > > - for_each_set_bit(i, (unsigned long *)&status, APSS_CPUCP_IPC_CHAN_SUPPORTED) { > - u32 val = readl(cpucp->rx_base + APSS_CPUCP_RX_MBOX_CMD(i) + APSS_CPUCP_MBOX_CMD_OFF); > + for_each_set_bit(i, (unsigned long *)&status, desc->num_chans) { > + u32 val = readl(cpucp->rx_base + desc->rx_reg + (i * desc->chan_stride)); > struct mbox_chan *chan = &cpucp->chans[i]; > unsigned long flags; > > @@ -62,7 +96,7 @@ static irqreturn_t qcom_cpucp_mbox_irq_fn(int irq, void *data) > spin_lock_irqsave(&chan->lock, flags); > if (chan->cl) > mbox_chan_received_data(chan, &val); > - writeq(BIT(i), cpucp->rx_base + APSS_CPUCP_RX_MBOX_CLEAR); > + writeq(BIT(i), cpucp->rx_base + desc->clear_reg); > spin_unlock_irqrestore(&chan->lock, flags); > } > > @@ -72,12 +106,15 @@ static irqreturn_t qcom_cpucp_mbox_irq_fn(int irq, void *data) > static int qcom_cpucp_mbox_startup(struct mbox_chan *chan) > { > struct qcom_cpucp_mbox *cpucp = container_of(chan->mbox, struct qcom_cpucp_mbox, mbox); > + const struct qcom_cpucp_mbox_desc *desc = cpucp->desc; > unsigned long chan_id = channel_number(chan); > u64 val; > > - val = readq(cpucp->rx_base + APSS_CPUCP_RX_MBOX_EN); > - val |= BIT(chan_id); > - writeq(val, cpucp->rx_base + APSS_CPUCP_RX_MBOX_EN); > + if (desc->v2_mbox) { > + val = readq(cpucp->rx_base + desc->enable_reg); > + val |= BIT(chan_id); > + writeq(val, cpucp->rx_base + desc->enable_reg); > + } So, startup and shutdown are effectively empty for v1 CPUCP chips? Could you please add two separate families of functions and two separate mbox_chan_ops instances? Which probably will mean one register less in the qcom_cpucp_mbox_desc structure. > > return 0; > } > @@ -85,22 +122,26 @@ static int qcom_cpucp_mbox_startup(struct mbox_chan *chan) > static void qcom_cpucp_mbox_shutdown(struct mbox_chan *chan) > { > struct qcom_cpucp_mbox *cpucp = container_of(chan->mbox, struct qcom_cpucp_mbox, mbox); > + const struct qcom_cpucp_mbox_desc *desc = cpucp->desc; > unsigned long chan_id = channel_number(chan); > u64 val; > > - val = readq(cpucp->rx_base + APSS_CPUCP_RX_MBOX_EN); > - val &= ~BIT(chan_id); > - writeq(val, cpucp->rx_base + APSS_CPUCP_RX_MBOX_EN); > + if (desc->v2_mbox) { > + val = readq(cpucp->rx_base + desc->enable_reg); > + val &= ~BIT(chan_id); > + writeq(val, cpucp->rx_base + desc->enable_reg); > + } > } > > static int qcom_cpucp_mbox_send_data(struct mbox_chan *chan, void *data) > { > struct qcom_cpucp_mbox *cpucp = container_of(chan->mbox, struct qcom_cpucp_mbox, mbox); > + const struct qcom_cpucp_mbox_desc *desc = cpucp->desc; > + u32 val = desc->v2_mbox ? *(u32 *)data : APSS_CPUCP_V1_SEND_IRQ_VAL; > unsigned long chan_id = channel_number(chan); > - u32 *val = data; > - > - writel(*val, cpucp->tx_base + APSS_CPUCP_TX_MBOX_CMD(chan_id) + APSS_CPUCP_MBOX_CMD_OFF); > + u32 offset = desc->v2_mbox ? (chan_id * desc->chan_stride) : 0; > > + writel(val, cpucp->tx_base + desc->tx_reg + offset); > return 0; > } > > @@ -112,41 +153,66 @@ static const struct mbox_chan_ops qcom_cpucp_mbox_chan_ops = { > > static int qcom_cpucp_mbox_probe(struct platform_device *pdev) > { > + const struct qcom_cpucp_mbox_desc *desc; > struct device *dev = &pdev->dev; > struct qcom_cpucp_mbox *cpucp; > struct mbox_controller *mbox; > + struct resource *res; > int irq, ret; > > + desc = device_get_match_data(&pdev->dev); > + if (!desc) > + return -EINVAL; > + > cpucp = devm_kzalloc(dev, sizeof(*cpucp), GFP_KERNEL); > if (!cpucp) > return -ENOMEM; > > - cpucp->rx_base = devm_of_iomap(dev, dev->of_node, 0, NULL); > - if (IS_ERR(cpucp->rx_base)) > - return PTR_ERR(cpucp->rx_base); > + cpucp->desc = desc; > + > + if (desc->v2_mbox) { > + cpucp->rx_base = devm_of_iomap(dev, dev->of_node, 0, NULL); > + if (IS_ERR(cpucp->rx_base)) > + return PTR_ERR(cpucp->rx_base); > + /* Legacy mailbox quirks due to shared region with EPSS register space */ Does that mean that on these platforms cpucp should be a child node of the EPSS? Also it might be a high time when the drivers should be switched to use regmaps. > + } else { > + res = platform_get_resource(pdev, IORESOURCE_MEM, 0); > + if (!res) { > + dev_err(&pdev->dev, "Failed to get the device base address\n"); > + return -ENODEV; > + } > + cpucp->rx_base = devm_ioremap(dev, res->start, resource_size(res)); > + if (!cpucp->rx_base) { > + dev_err(dev, "Failed to ioremap the cpucp rx irq addr\n"); > + return -ENOMEM; > + } > + } > > cpucp->tx_base = devm_of_iomap(dev, dev->of_node, 1, NULL); > if (IS_ERR(cpucp->tx_base)) > return PTR_ERR(cpucp->tx_base); > > - writeq(0, cpucp->rx_base + APSS_CPUCP_RX_MBOX_EN); > - writeq(0, cpucp->rx_base + APSS_CPUCP_RX_MBOX_CLEAR); > - writeq(0, cpucp->rx_base + APSS_CPUCP_RX_MBOX_MAP); > + if (desc->v2_mbox) { > + writeq(0, cpucp->rx_base + desc->enable_reg); > + writeq(0, cpucp->rx_base + desc->clear_reg); > + writeq(0, cpucp->rx_base + desc->map_reg); > + } If these registers are only present on V2, there is no need to have them in the qcom_cpucp_mbox_desc. > > irq = platform_get_irq(pdev, 0); > if (irq < 0) > return irq; > > - ret = devm_request_irq(dev, irq, qcom_cpucp_mbox_irq_fn, > - IRQF_TRIGGER_HIGH, "apss_cpucp_mbox", cpucp); > + ret = devm_request_irq(dev, irq, desc->v2_mbox ? qcom_cpucp_v2_mbox_irq_fn : > + qcom_cpucp_mbox_irq_fn, IRQF_TRIGGER_HIGH, "apss_cpucp_mbox", cpucp); > if (ret < 0) > return dev_err_probe(dev, ret, "Failed to register irq: %d\n", irq); > > - writeq(APSS_CPUCP_RX_MBOX_CMD_MASK, cpucp->rx_base + APSS_CPUCP_RX_MBOX_MAP); > + if (desc->v2_mbox) > + writeq(APSS_CPUCP_V2_RX_MBOX_CMD_MASK, cpucp->rx_base + desc->map_reg); > > mbox = &cpucp->mbox; > mbox->dev = dev; > - mbox->num_chans = APSS_CPUCP_IPC_CHAN_SUPPORTED; > + mbox->num_chans = desc->num_chans; > mbox->chans = cpucp->chans; > mbox->ops = &qcom_cpucp_mbox_chan_ops; > > @@ -157,8 +223,30 @@ static int qcom_cpucp_mbox_probe(struct platform_device *pdev) > return 0; > } > > +static const struct qcom_cpucp_mbox_desc sc7280_cpucp_mbox = { > + .tx_reg = 0xC, > + .chan_stride = 0x1000, > + .status_reg = 0x30C, > + .clear_reg = 0x308, > + .v2_mbox = false, > + .num_chans = 2, > +}; > + > +static const struct qcom_cpucp_mbox_desc x1e80100_cpucp_mbox = { > + .rx_reg = 0x104, > + .tx_reg = 0x104, > + .chan_stride = 0x8, > + .map_reg = 0x4000, > + .status_reg = 0x4400, > + .clear_reg = 0x4800, > + .enable_reg = 0x4C00, > + .v2_mbox = true, > + .num_chans = 3, > +}; > + > static const struct of_device_id qcom_cpucp_mbox_of_match[] = { > - { .compatible = "qcom,x1e80100-cpucp-mbox" }, > + { .compatible = "qcom,x1e80100-cpucp-mbox", .data = &x1e80100_cpucp_mbox}, > + { .compatible = "qcom,sc7280-cpucp-mbox", .data = &sc7280_cpucp_mbox}, Please keep the table sorted. > {} > }; > MODULE_DEVICE_TABLE(of, qcom_cpucp_mbox_of_match); > -- > 2.25.1 > -- With best wishes Dmitry