On Tue, Jul 23, 2019 at 6:18 PM Pierre-Louis Bossart
<pierre-louis.bossart@xxxxxxxxxxxxxxx> wrote:
>
>
> > diff --git a/sound/soc/sof/imx/Makefile b/sound/soc/sof/imx/Makefile
> > new file mode 100644
> > index 000000000000..c69237971da5
> > --- /dev/null
> > +++ b/sound/soc/sof/imx/Makefile
> > @@ -0,0 +1,7 @@
> > +# SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
> > +
> > +ccflags-y += -DDEBUG
>
> this should be removed or in a separate patch.
All right.
>
>
> > +struct imx8_priv {
> > + struct device *dev;
> > + struct snd_sof_dev *sdev;
> > + struct imx_dsp_ipc *dsp_ipc;
> > + struct imx_sc_ipc *sc_ipc;
>
> maybe a comment to explain what 'sc' stands for?
Sure.
>
> > +};
> > +
> > +static void imx8_get_windows(struct snd_sof_dev *sdev)
> > +{
> > + struct sof_ipc_window_elem *elem;
> > + u32 outbox_offset = 0;
> > + u32 stream_offset = 0;
> > + u32 inbox_offset = 0;
> > + u32 outbox_size = 0;
> > + u32 stream_size = 0;
> > + u32 inbox_size = 0;
> > + int i;
> > +
> > + if (!sdev->info_window) {
> > + dev_err(sdev->dev, "error: have no window info\n");
> > + return;
> > + }
> > +
> > + for (i = 0; i < sdev->info_window->num_windows; i++) {
> > + elem = &sdev->info_window->window[i];
> > +
> > + switch (elem->type) {
> > + case SOF_IPC_REGION_UPBOX:
> > + inbox_offset = elem->offset + MBOX_OFFSET;
> > + inbox_size = elem->size;
> > + snd_sof_debugfs_io_item(sdev,
> > + sdev->bar[SOF_FW_BLK_TYPE_SRAM]
> > + + inbox_offset,
> > + elem->size, "inbox",
> > + SOF_DEBUGFS_ACCESS_D0_ONLY);
> > + break;
> > + case SOF_IPC_REGION_DOWNBOX:
> > + outbox_offset = elem->offset + MBOX_OFFSET;
> > + outbox_size = elem->size;
> > + snd_sof_debugfs_io_item(sdev,
> > + sdev->bar[SOF_FW_BLK_TYPE_SRAM]
> > + + outbox_offset,
> > + elem->size, "outbox",
> > + SOF_DEBUGFS_ACCESS_D0_ONLY);
> > + break;
> > + case SOF_IPC_REGION_TRACE:
> > + snd_sof_debugfs_io_item(sdev,
> > + sdev->bar[SOF_FW_BLK_TYPE_SRAM]
> > + + elem->offset + MBOX_OFFSET,
> > + elem->size, "etrace",
> > + SOF_DEBUGFS_ACCESS_D0_ONLY);
> > + break;
> > + case SOF_IPC_REGION_DEBUG:
> > + snd_sof_debugfs_io_item(sdev,
> > + sdev->bar[SOF_FW_BLK_TYPE_SRAM]
> > + + elem->offset + MBOX_OFFSET,
> > + elem->size, "debug",
> > + SOF_DEBUGFS_ACCESS_D0_ONLY);
> > + break;
> > + case SOF_IPC_REGION_STREAM:
> > + stream_offset = elem->offset + MBOX_OFFSET;
> > + stream_size = elem->size;
> > + snd_sof_debugfs_io_item(sdev,
> > + sdev->bar[SOF_FW_BLK_TYPE_SRAM]
> > + + stream_offset,
> > + elem->size, "stream",
> > + SOF_DEBUGFS_ACCESS_D0_ONLY);
> > + break;
> > + case SOF_IPC_REGION_REGS:
> > + snd_sof_debugfs_io_item(sdev,
> > + sdev->bar[SOF_FW_BLK_TYPE_SRAM]
> > + + elem->offset + MBOX_OFFSET,
> > + elem->size, "regs",
> > + SOF_DEBUGFS_ACCESS_D0_ONLY);
> > + break;
> > + case SOF_IPC_REGION_EXCEPTION:
> > + sdev->dsp_oops_offset = elem->offset + MBOX_OFFSET;
> > + snd_sof_debugfs_io_item(sdev,
> > + sdev->bar[SOF_FW_BLK_TYPE_SRAM]
> > + + elem->offset + MBOX_OFFSET,
> > + elem->size, "exception",
> > + SOF_DEBUGFS_ACCESS_D0_ONLY);
> > + break;
> > + default:
> > + dev_err(sdev->dev, "error: get illegal window info\n");
> > + return;
> > + }
> > + }
> > +
> > + if (outbox_size == 0 || inbox_size == 0) {
> > + dev_err(sdev->dev, "error: get illegal mailbox window\n");
> > + return;
> > + }
> > +
> > + snd_sof_dsp_mailbox_init(sdev, inbox_offset, inbox_size,
> > + outbox_offset, outbox_size);
> > + sdev->stream_box.offset = stream_offset;
> > + sdev->stream_box.size = stream_size;
> > +
> > + dev_dbg(sdev->dev, " mailbox upstream 0x%x - size 0x%x\n",
> > + inbox_offset, inbox_size);
> > + dev_dbg(sdev->dev, " mailbox downstream 0x%x - size 0x%x\n",
> > + outbox_offset, outbox_size);
> > + dev_dbg(sdev->dev, " stream region 0x%x - size 0x%x\n",
> > + stream_offset, stream_size);
> > +}
>
> This looks 100% similar to Baytrail?
Yes!
>
> > +
> > +/*
> > + * IPC Firmware ready.
> > + */
> > +static int imx8_fw_ready(struct snd_sof_dev *sdev, u32 msg_id)
> > +{
> > + struct sof_ipc_fw_ready *fw_ready = &sdev->fw_ready;
> > + u32 offset;
> > + int ret;
> > +
> > + /* mailbox must be on 4k boundary */
> > + offset = MBOX_OFFSET;
> > +
> > + dev_dbg(sdev->dev, "ipc: DSP is ready 0x%8.8x offset 0x%x\n",
> > + msg_id, offset);
> > +
> > + /* no need to re-check version/ABI for subsequent boots */
> > + if (!sdev->first_boot)
> > + return 0;
> > +
> > + /* copy data from the DSP FW ready offset */
> > + sof_block_read(sdev, sdev->mailbox_bar, offset, fw_ready,
> > + sizeof(*fw_ready));
> > + snd_sof_dsp_mailbox_init(sdev, fw_ready->dspbox_offset,
> > + fw_ready->dspbox_size,
> > + fw_ready->hostbox_offset,
> > + fw_ready->hostbox_size);
> > +
> > + /* make sure ABI version is compatible */
> > + ret = snd_sof_ipc_valid(sdev);
> > + if (ret < 0)
> > + return ret;
> > +
> > + /* now check for extended data */
> > + snd_sof_fw_parse_ext_data(sdev, SOF_FW_BLK_TYPE_SRAM, MBOX_OFFSET +
> > + sizeof(struct sof_ipc_fw_ready));
> > +
> > + imx8_get_windows(sdev);
> > +
> > + return 0;
> > +}
>
> That code looks nearly similar to the baytrail one except for the last
> line, we should look into factoring this.
Yes, I got most of my inspiration from intel code.
>
> > +
> > +static void imx8_get_reply(struct snd_sof_dev *sdev)
> > +{
> > + struct snd_sof_ipc_msg *msg = sdev->msg;
> > + struct sof_ipc_reply reply;
> > + unsigned long flags;
> > + int ret = 0;
> > +
> > + if (!msg) {
> > + dev_warn(sdev->dev, "unexpected ipc interrupt\n");
> > + return;
> > + }
> > +
> > + /* get reply */
> > + sof_mailbox_read(sdev, sdev->host_box.offset, &reply, sizeof(reply));
> > +
> > + spin_lock_irqsave(&sdev->ipc_lock, flags);
> > +
> > + if (reply.error < 0) {
> > + memcpy(msg->reply_data, &reply, sizeof(reply));
> > + ret = reply.error;
> > + } else {
> > + /* reply has correct size? */
> > + if (reply.hdr.size != msg->reply_size) {
> > + dev_err(sdev->dev, "error: reply expected %zu got %u bytes\n",
> > + msg->reply_size, reply.hdr.size);
> > + ret = -EINVAL;
> > + }
> > +
> > + /* read the message */
> > + if (msg->reply_size > 0)
> > + sof_mailbox_read(sdev, sdev->host_box.offset,
> > + msg->reply_data, msg->reply_size);
> > + }
> > +
> > + msg->reply_error = ret;
> > +
> > + spin_unlock_irqrestore(&sdev->ipc_lock, flags);
>
> I don't see a spin_lock/unlock for the get_reply in the Intel code, is
> this necessary?
Hmm, you are right. I think I've used an older version of the intel code
where there a lock.
>
> > +}
> > +
> > +void imx_dsp_handle_reply(struct imx_dsp_ipc *ipc)
> > +{
> > + struct imx8_priv *priv = imx_dsp_get_data(ipc);
> > +
> > + imx8_get_reply(priv->sdev);
> > + snd_sof_ipc_reply(priv->sdev, 0);
> > +}
> > +
> > +void imx_dsp_handle_request(struct imx_dsp_ipc *ipc)
> > +{
> > + struct imx8_priv *priv = imx_dsp_get_data(ipc);
> > +
> > + snd_sof_ipc_msgs_rx(priv->sdev);
> > +}
> > +
> > +struct imx_dsp_ops dsp_ops = {
> > + .handle_reply = imx_dsp_handle_reply,
> > + .handle_request = imx_dsp_handle_request,
> > +};
> > +
> > +static int imx8_send_msg(struct snd_sof_dev *sdev, struct snd_sof_ipc_msg *msg)
> > +{
> > + struct imx8_priv *priv = (struct imx8_priv *)sdev->private;
> > +
> > + sof_mailbox_write(sdev, sdev->host_box.offset, msg->msg_data,
> > + msg->msg_size);
> > + imx_dsp_ring_doorbell(priv->dsp_ipc, 0);
> > +
> > + return 0;
> > +}
> > +
> > +/*
> > + * DSP control.
> > + */
> > +static int imx8_run(struct snd_sof_dev *sdev)
> > +{
> > + int ret;
> > + struct imx8_priv *dsp_priv = (struct imx8_priv *)sdev->private;
> > +
> > + ret = imx_sc_misc_set_control(dsp_priv->sc_ipc, IMX_SC_R_DSP,
> > + IMX_SC_C_OFS_SEL, 1);
> > + if (ret < 0) {
> > + dev_err(sdev->dev, "Error system address offset source select\n");
> > + return ret;
> > + }
> > +
> > + ret = imx_sc_misc_set_control(dsp_priv->sc_ipc, IMX_SC_R_DSP,
> > + IMX_SC_C_OFS_AUDIO, 0x80);
> > + if (ret < 0) {
> > + dev_err(sdev->dev, "Error system address offset of AUDIO\n");
> > + return ret;
> > + }
> > +
> > + ret = imx_sc_misc_set_control(dsp_priv->sc_ipc, IMX_SC_R_DSP,
> > + IMX_SC_C_OFS_PERIPH, 0x5A);
> > + if (ret < 0) {
> > + dev_err(sdev->dev, "Error system address offset of PERIPH %d\n",
> > + ret);
> > + return ret;
> > + }
> > +
> > + ret = imx_sc_misc_set_control(dsp_priv->sc_ipc, IMX_SC_R_DSP,
> > + IMX_SC_C_OFS_IRQ, 0x51);
> > + if (ret < 0) {
> > + dev_err(sdev->dev, "Error system address offset of IRQ\n");
> > + return ret;
> > + }
> > +
> > + imx_sc_pm_cpu_start(dsp_priv->sc_ipc, IMX_SC_R_DSP, true,
> > + RESET_VECTOR_VADDR);
> > +
> > + return 0;
> > +}
> > +
> > +static int imx8_probe(struct snd_sof_dev *sdev)
> > +{
> > + struct imx8_priv *priv;
> > + int i;
> > + struct platform_device *pdev =
> > + container_of(sdev->dev, struct platform_device, dev);
> > + struct platform_device *ipc_dev;
> > + struct resource *mmio;
> > + int num_domains = 0;
> > + u32 base, size;
> > + int ret = 0;
> > + struct device_node *np = pdev->dev.of_node;
> > + struct device_node *res_node;
> > + struct resource res;
>
> nit-pick: can we reorder so that we have all counters last and a nice
> xmas-tree shape.
Ack.
>
> > +
> > + priv = devm_kzalloc(&pdev->dev, sizeof(*priv), GFP_KERNEL);
> > + if (!priv)
> > + return -ENOMEM;
> > +
> > + sdev->private = priv;
> > + priv->dev = sdev->dev;
> > + priv->sdev = sdev;
> > +
> > + ret = imx_scu_get_handle(&priv->sc_ipc);
> > + if (ret) {
> > + dev_err(sdev->dev, "Cannot obtain SCU handle (err = %d)\n",
> > + ret);
> > + return ret;
> > + }
> > +
> > + ipc_dev = platform_device_register_data(sdev->dev, "imx-dsp",
> > + PLATFORM_DEVID_NONE,
> > + pdev, sizeof(*pdev));
> > + if (IS_ERR(ipc_dev)) {
> > + dev_err(sdev->dev, "Failed to register platform device\n");
> > + return PTR_ERR(ipc_dev);
> > + }
> > +
> > + priv->dsp_ipc = dev_get_drvdata(&ipc_dev->dev);
> > + if (!priv->dsp_ipc)
> > + return -EPROBE_DEFER;
> > +
> > + imx_dsp_set_data(priv->dsp_ipc, priv);
> > + priv->dsp_ipc->ops = &dsp_ops;
> > +
> > + num_domains = of_count_phandle_with_args(np, "power-domains",
> > + "#power-domain-cells");
> > + for (i = 0; i < num_domains; i++) {
> > + struct device *pd_dev;
> > + struct device_link *link;
> > +
> > + pd_dev = dev_pm_domain_attach_by_id(&pdev->dev, i);
> > + if (IS_ERR(pd_dev))
> > + return PTR_ERR(pd_dev);
> > +
> > + link = device_link_add(&pdev->dev, pd_dev,
> > + DL_FLAG_STATELESS |
> > + DL_FLAG_PM_RUNTIME |
> > + DL_FLAG_RPM_ACTIVE);
> > + if (IS_ERR(link))
> > + return PTR_ERR(link);
>
> Question: is the error flow final? Wondering if we release all the
> resources/memory/devices on errors?
Will check again. It seemed no need for resource freeing.
>
> Also are all the resources device-managed, I don't see a remove()?
Good catch for pm stuff. We mostly didn't care about remove because
drivers are always Y in our distribution.
Thanks Pierre for review, I will let some time for others to have a look and
send a new version.
thanks,
Daniel.
