RE: [PATCH V6 3/4] mailbox: imx: add SCU MU support

From: Peng Fan
Date: Wed Mar 18 2020 - 08:17:14 EST

Next message: kajoljain: "Re: [PATCH v5 08/11] perf/tools: Refactoring metricgroup__add_metric function"
Previous message: Jason Gunthorpe: "Re: [PATCHv7 2/3] mm/gup: fix omission of check on FOLL_LONGTERM in gup fast path"
In reply to: Leonard Crestez: "Re: [PATCH V6 3/4] mailbox: imx: add SCU MU support"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

> Subject: Re: [PATCH V6 3/4] mailbox: imx: add SCU MU support
>
> On 2020-03-04 7:56 AM, Peng Fan wrote:
> > From: Peng Fan <peng.fan@xxxxxxx>
> >
> > i.MX8/8X SCU MU is dedicated for communication between SCU and
> > Cortex-A cores from hardware design, and could not be reused for other
> purpose.
> >
> > Per i.MX8/8X Reference mannual, Chapter "12.9.2.3.2 Messaging
> Examples",
> > Passing short messages: Transmit register(s) can be used to pass
> > short messages from one to four words in length. For example, when
> > a four-word message is desired, only one of the registers needs to
> > have its corresponding interrupt enable bit set at the receiver side;
> > the message's first three words are written to the registers whose
> > interrupt is masked, and the fourth word is written to the other
> > register (which triggers an interrupt at the receiver side).
> >
> > i.MX8/8X SCU firmware IPC is an implementation of passing short
> > messages. But current imx-mailbox driver only support one word
> > message, i.MX8/8X linux side firmware has to request four TX and four
> > RX to support IPC to SCU firmware. This is low efficent and more
> > interrupts triggered compared with one TX and one RX.
> >
> > To make SCU MU work,
> > - parse the size of msg.
> > - Only enable TR0/RR0 interrupt for transmit/receive message.
> > - For TX/RX, only support one TX channel and one RX channel
> > - For RX, support receive msg larger than 4 u32 words.
> > - Support 6 channels, TX0/RX0/RXDB[0-3], not support TXDB.
> >
> > Reviewed-by: Oleksij Rempel <o.rempel@xxxxxxxxxxxxxx>
> > Signed-off-by: Peng Fan <peng.fan@xxxxxxx>
> > ---
> >
> > V6:
> > Add R-b tag
> > Use %zu for printk sizeof
> >
> > V5:
> > Code style cleanup
> > Add more debug msg
> > Drop __packed aligned
> > idx santity check in scu xlate
> >
> > V4:
> > Added separate chans init and xlate function for SCU MU
> > Limit chans to TX0/RX0/RXDB[0-3], max 6 chans.
> > Santity check to msg size
> >
> > V3:
> > Added scu type tx/rx and SCU MU type
> >
> > drivers/mailbox/imx-mailbox.c | 134
> ++++++++++++++++++++++++++++++++++++++++++
> > 1 file changed, 134 insertions(+)
> >
> > diff --git a/drivers/mailbox/imx-mailbox.c
> > b/drivers/mailbox/imx-mailbox.c index df6c4ecd913c..0895ccd23d17
> > 100644
> > --- a/drivers/mailbox/imx-mailbox.c
> > +++ b/drivers/mailbox/imx-mailbox.c
> > @@ -4,6 +4,7 @@
> > */
> >
> > #include <linux/clk.h>
> > +#include <linux/firmware/imx/ipc.h>
> > #include <linux/interrupt.h>
> > #include <linux/io.h>
> > #include <linux/kernel.h>
> > @@ -27,6 +28,8 @@
> > #define IMX_MU_xCR_GIRn(x) BIT(16 + (3 - (x)))
> >
> > #define IMX_MU_CHANS 16
> > +/* TX0/RX0/RXDB[0-3] */
> > +#define IMX_MU_SCU_CHANS 6
> > #define IMX_MU_CHAN_NAME_SIZE 20
> >
> > enum imx_mu_chan_type {
> > @@ -36,6 +39,11 @@ enum imx_mu_chan_type {
> > IMX_MU_TYPE_RXDB, /* Rx doorbell */
> > };
> >
> > +struct imx_sc_rpc_msg_max {
> > + struct imx_sc_rpc_msg hdr;
> > + u32 data[7];
> > +};
> > +
> > struct imx_mu_con_priv {
> > unsigned int idx;
> > char irq_desc[IMX_MU_CHAN_NAME_SIZE];
> > @@ -134,6 +142,63 @@ static int imx_mu_generic_rx(struct imx_mu_priv
> *priv,
> > return 0;
> > }
> >
> > +static int imx_mu_scu_tx(struct imx_mu_priv *priv,
> > + struct imx_mu_con_priv *cp,
> > + void *data)
> > +{
> > + struct imx_sc_rpc_msg_max *msg = data;
> > + u32 *arg = data;
> > + int i;
> > +
> > + switch (cp->type) {
> > + case IMX_MU_TYPE_TX:
> > + if (msg->hdr.size > sizeof(*msg)) {
> > + /*
> > + * The real message size can be different to
> > + * struct imx_sc_rpc_msg_max size
> > + */
> > + dev_err(priv->dev, "Exceed max msg size (%zu) on TX,
> got: %i\n", sizeof(*msg), msg->hdr.size);
> > + return -EINVAL;
> > + }
> > +
> > + for (i = 0; i < msg->hdr.size; i++) > + imx_mu_write(priv,
> *arg++, priv->dcfg->xTR[i % 4]);
>
> Need to poll for TE, otherwise for long messages we could overwrite TR0
> before SCU reads it.

I planned to add a follow up patch for long word messages.
Since you raised it, I'll add check for TX for send, RX for recv.

>
> > +
> > + imx_mu_xcr_rmw(priv, IMX_MU_xCR_TIEn(cp->idx), 0);
>
> Shouldn't TIE be armed ahead of writing? In theory SCU could read first
> register before loop above is over.

This is to avoid TX interrupt be triggered early.

>
> > + break;
> > + default:
> > + dev_warn_ratelimited(priv->dev, "Send data on wrong channel
> type: %d\n", cp->type);
> > + return -EINVAL;
> > + }
> > +
> > + return 0;
> > +}
> > +
> > +static int imx_mu_scu_rx(struct imx_mu_priv *priv,
> > + struct imx_mu_con_priv *cp)
> > +{
> > + struct imx_sc_rpc_msg_max msg;
> > + u32 *data = (u32 *)&msg;
> > + int i;
> > +
> > + imx_mu_xcr_rmw(priv, 0, IMX_MU_xCR_RIEn(0));
> > + *data++ = imx_mu_read(priv, priv->dcfg->xRR[0]);
> > +
> > + if (msg.hdr.size > sizeof(msg)) {
> > + dev_err(priv->dev, "Exceed max msg size (%zu) on RX, got: %i\n",
> > + sizeof(msg), msg.hdr.size);
> > + return -EINVAL;
> > + }
> > +
> > + for (i = 1; i < msg.hdr.size; i++)
> > + *data++ = imx_mu_read(priv, priv->dcfg->xRR[i % 4]);
>
> Shouldn't you poll for !RE? It's possible to receive RX interrupt and start
> reading before other side writes the second word.

Will add RE check.

Thanks,
Peng.

>
> > +
> > + imx_mu_xcr_rmw(priv, IMX_MU_xCR_RIEn(0), 0);
> > + mbox_chan_received_data(cp->chan, (void *)&msg);
> > +
> > + return 0;
> > +}
> > +
> > static void imx_mu_txdb_tasklet(unsigned long data)
> > {
> > struct imx_mu_con_priv *cp = (struct imx_mu_con_priv *)data; @@
> > -263,6 +328,42 @@ static const struct mbox_chan_ops imx_mu_ops = {
> > .shutdown = imx_mu_shutdown,
> > };
> >
> > +static struct mbox_chan *imx_mu_scu_xlate(struct mbox_controller
> *mbox,
> > + const struct of_phandle_args *sp) {
> > + u32 type, idx, chan;
> > +
> > + if (sp->args_count != 2) {
> > + dev_err(mbox->dev, "Invalid argument count %d\n",
> sp->args_count);
> > + return ERR_PTR(-EINVAL);
> > + }
> > +
> > + type = sp->args[0]; /* channel type */
> > + idx = sp->args[1]; /* index */
> > +
> > + switch (type) {
> > + case IMX_MU_TYPE_TX:
> > + case IMX_MU_TYPE_RX:
> > + if (idx != 0)
> > + dev_err(mbox->dev, "Invalid chan idx: %d\n", idx);
> > + chan = type;
> > + break;
> > + case IMX_MU_TYPE_RXDB:
> > + chan = 2 + idx;
> > + break;
> > + default:
> > + dev_err(mbox->dev, "Invalid chan type: %d\n", type);
> > + return NULL;
> > + }
> > +
> > + if (chan >= mbox->num_chans) {
> > + dev_err(mbox->dev, "Not supported channel number: %d. (type: %d,
> idx: %d)\n", chan, type, idx);
> > + return ERR_PTR(-EINVAL);
> > + }
> > +
> > + return &mbox->chans[chan];
> > +}
> > +
> > static struct mbox_chan * imx_mu_xlate(struct mbox_controller *mbox,
> > const struct of_phandle_args *sp)
> > {
> > @@ -310,6 +411,28 @@ static void imx_mu_init_generic(struct
> imx_mu_priv *priv)
> > imx_mu_write(priv, 0, priv->dcfg->xCR);
> > }
> >
> > +static void imx_mu_init_scu(struct imx_mu_priv *priv) {
> > + unsigned int i;
> > +
> > + for (i = 0; i < IMX_MU_SCU_CHANS; i++) {
> > + struct imx_mu_con_priv *cp = &priv->con_priv[i];
> > +
> > + cp->idx = i < 2 ? 0 : i - 2;
> > + cp->type = i < 2 ? i : IMX_MU_TYPE_RXDB;
> > + cp->chan = &priv->mbox_chans[i];
> > + priv->mbox_chans[i].con_priv = cp;
> > + snprintf(cp->irq_desc, sizeof(cp->irq_desc),
> > + "imx_mu_chan[%i-%i]", cp->type, cp->idx);
> > + }
> > +
> > + priv->mbox.num_chans = IMX_MU_SCU_CHANS;
> > + priv->mbox.of_xlate = imx_mu_scu_xlate;
> > +
> > + /* Set default MU configuration */
> > + imx_mu_write(priv, 0, priv->dcfg->xCR); }
> > +
> > static int imx_mu_probe(struct platform_device *pdev)
> > {
> > struct device *dev = &pdev->dev;
> > @@ -396,9 +519,20 @@ static const struct imx_mu_dcfg
> imx_mu_cfg_imx7ulp = {
> > .xCR = 0x64,
> > };
> >
> > +static const struct imx_mu_dcfg imx_mu_cfg_imx8_scu = {
> > + .tx = imx_mu_scu_tx,
> > + .rx = imx_mu_scu_rx,
> > + .init = imx_mu_init_scu,
> > + .xTR = {0x0, 0x4, 0x8, 0xc},
> > + .xRR = {0x10, 0x14, 0x18, 0x1c},
> > + .xSR = 0x20,
> > + .xCR = 0x24,
> > +};
> > +
> > static const struct of_device_id imx_mu_dt_ids[] = {
> > { .compatible = "fsl,imx7ulp-mu", .data = &imx_mu_cfg_imx7ulp },
> > { .compatible = "fsl,imx6sx-mu", .data = &imx_mu_cfg_imx6sx },
> > + { .compatible = "fsl,imx8-mu-scu", .data = &imx_mu_cfg_imx8_scu },
> > { },
> > };
> > MODULE_DEVICE_TABLE(of, imx_mu_dt_ids);
> >

Next message: kajoljain: "Re: [PATCH v5 08/11] perf/tools: Refactoring metricgroup__add_metric function"
Previous message: Jason Gunthorpe: "Re: [PATCHv7 2/3] mm/gup: fix omission of check on FOLL_LONGTERM in gup fast path"
In reply to: Leonard Crestez: "Re: [PATCH V6 3/4] mailbox: imx: add SCU MU support"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]