On Tue, May 5, 2020 at 7:00 AM Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Thu, Apr 30, 2020 at 07:12:57AM -0400, Al Cooper wrote:
> > Add a new EHCI driver for Broadcom STB SoC's. A new EHCI driver
> > was created instead of adding support to the existing ehci platform
> > driver because of the code required to workaround bugs in the EHCI
> > controller.
> >
> > Signed-off-by: Al Cooper <alcooperx@xxxxxxxxx>
> > Reviewed-by: Andy Shevchenko <andy.shevchenko@xxxxxxxxx>
> > ---
> > drivers/usb/host/ehci-brcm.c | 290 +++++++++++++++++++++++++++++++++++
> > 1 file changed, 290 insertions(+)
> > create mode 100644 drivers/usb/host/ehci-brcm.c
>
> I need an ack from the EHCI maintainer to agree that this needs a whole
> new driver file...
>
> >
> > diff --git a/drivers/usb/host/ehci-brcm.c b/drivers/usb/host/ehci-brcm.c
> > new file mode 100644
> > index 000000000000..381bed5fdab0
> > --- /dev/null
> > +++ b/drivers/usb/host/ehci-brcm.c
> > @@ -0,0 +1,290 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/* Copyright (c) 2020, Broadcom */
> > +
> > +#include <linux/clk.h>
> > +#include <linux/dma-mapping.h>
> > +#include <linux/err.h>
> > +#include <linux/kernel.h>
> > +#include <linux/io.h>
> > +#include <linux/module.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/usb.h>
> > +#include <linux/usb/hcd.h>
> > +#include <linux/iopoll.h>
> > +
> > +#include "ehci.h"
> > +
> > +#define hcd_to_ehci_priv(h) ((struct brcm_priv *)hcd_to_ehci(h)->priv)
> > +
> > +struct brcm_priv {
> > + struct clk *clk;
> > +};
> > +
> > +static const char brcm_hcd_name[] = "ehci-brcm";
>
> You don't use this anywhere? Are you sure this didn't cause compile
> warnings?
I'll remove it. I'm not getting a warning because it's a "static
const" which doesn't give unused warnings.
>
> > +
> > +static int (*org_hub_control)(struct usb_hcd *hcd,
> > + u16 typeReq, u16 wValue, u16 wIndex,
> > + char *buf, u16 wLength);
>
> So you only support one device per system? That feels bad...
When this driver was originally written, the ehci_hub_control()
function was a static function in ehci-hub.c and couldn't be called
directly. Instead, the function pointer was taken out of "struct
hc_driver" and since it couldn't change for multiple devices, only one
pointer was needed. The ehci_hub_control function is now global so it
can be called directly and this can be removed. It's nice to get rid
of this hack, thanks.
>
>
> > +
> > +/*
> > + * ehci_brcm_wait_for_sof
> > + * Wait for start of next microframe, then wait extra delay microseconds
> > + */
> > +static inline void ehci_brcm_wait_for_sof(struct ehci_hcd *ehci, u32 delay)
> > +{
> > + u32 frame_idx = ehci_readl(ehci, &ehci->regs->frame_index);
> > + u32 val;
> > + int res;
> > +
> > + /* Wait for next microframe (every 125 usecs) */
> > + res = readl_relaxed_poll_timeout(&ehci->regs->frame_index, val,
> > + val != frame_idx, 1, 130);
> > + if (res)
> > + dev_err(ehci_to_hcd(ehci)->self.controller,
> > + "Error waiting for SOF\n");
> > + udelay(delay);
> > +}
> > +
> > +/*
> > + * ehci_brcm_hub_control
> > + * Intercept echi-hcd request to complete RESUME and align it to the start
> > + * of the next microframe.
> > + * If RESUME is complete too late in the microframe, host controller
> > + * detects babble on suspended port and resets the port afterwards.
> > + * This s/w workaround allows to avoid this problem.
> > + * See SWLINUX-1909 for more details
> > + */
> > +static int ehci_brcm_hub_control(
> > + struct usb_hcd *hcd,
> > + u16 typeReq,
> > + u16 wValue,
> > + u16 wIndex,
> > + char *buf,
> > + u16 wLength)
> > +{
> > + struct ehci_hcd *ehci = hcd_to_ehci(hcd);
> > + int ports = HCS_N_PORTS(ehci->hcs_params);
> > + u32 __iomem *status_reg = &ehci->regs->port_status[
> > + (wIndex & 0xff) - 1];
>
> Horrid line-wrapping, put this assignment below so it can be read.
>
> And wIndex is little endian? Or native?
native
>
> > + unsigned long flags;
> > + int retval, irq_disabled = 0;
> > +
> > + /*
> > + * RESUME is cleared when GetPortStatus() is called 20ms after start
> > + * of RESUME
> > + */
> > + if ((typeReq == GetPortStatus) &&
> > + (wIndex && wIndex <= ports) &&
> > + ehci->reset_done[wIndex-1] &&
> > + time_after_eq(jiffies, ehci->reset_done[wIndex-1]) &&
> > + (ehci_readl(ehci, status_reg) & PORT_RESUME)) {
> > +
> > + /*
> > + * to make sure we are not interrupted until RESUME bit
> > + * is cleared, disable interrupts on current CPU
> > + */
> > + ehci_dbg(ehci, "SOF alignment workaround\n");
> > + irq_disabled = 1;
> > + local_irq_save(flags);
> > + ehci_brcm_wait_for_sof(ehci, 5);
> > + }
> > + retval = (*org_hub_control)(hcd, typeReq, wValue, wIndex, buf, wLength);
>
> But this might not be set, did you just crash?
>
> If it is always set, then why does it need to be a function pointer at
> all?
Explained above (and removed).
>
> > + if (irq_disabled)
> > + local_irq_restore(flags);
> > + return retval;
> > +}
> > +
> > +static int ehci_brcm_reset(struct usb_hcd *hcd)
> > +{
> > + struct ehci_hcd *ehci = hcd_to_ehci(hcd);
> > +
> > + ehci->big_endian_mmio = 1;
> > +
> > + ehci->caps = (struct ehci_caps *) hcd->regs;
> > + ehci->regs = (struct ehci_regs *) (hcd->regs +
>
> coding style, did you run this through checkpatch.pl?
Checkpatch did not complain, but I'll clean it up.
>
> > + HC_LENGTH(ehci, ehci_readl(ehci, &ehci->caps->hc_capbase)));
> > +
> > + /* This fixes the lockup during reboot due to prior interrupts */
> > + ehci_writel(ehci, CMD_RESET, &ehci->regs->command);
> > + mdelay(10);
> > +
> > + /*
> > + * SWLINUX-1705: Avoid OUT packet underflows during high memory
> > + * bus usage
> > + * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00 @ 0x90
> > + */
> > + ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
> > + ehci_writel(ehci, 0x00000001, &ehci->regs->port_status[0x12]);
> > +
> > + return ehci_setup(hcd);
> > +}
> > +
> > +static struct hc_driver __read_mostly ehci_brcm_hc_driver;
> > +
> > +static const struct ehci_driver_overrides brcm_overrides __initconst = {
> > +
>
> No blank line.
Fixed.
>
> > + .reset = ehci_brcm_reset,
> > + .extra_priv_size = sizeof(struct brcm_priv),
> > +};
> > +
> > +static int ehci_brcm_probe(struct platform_device *pdev)
> > +{
> > + struct device *dev = &pdev->dev;
> > + struct resource *res_mem;
> > + struct brcm_priv *priv;
> > + struct usb_hcd *hcd;
> > + int irq;
> > + int err;
> > +
> > + if (usb_disabled())
> > + return -ENODEV;
> > +
> > + err = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
> > + if (err)
> > + return err;
> > +
> > + irq = platform_get_irq(pdev, 0);
> > + if (irq < 0)
>
> "<=" right?
The example in the comment above platform_get_irq in platform.c says
"if (irq < 0)"
>
> > + return irq;
> > +
> > + /* Hook the hub control routine to work around a bug */
>
> What bug? This feels wrong.
The bug is explained in a comment above the ehci_brcm_hub_control()
routine as follows:
/*
* ehci_brcm_hub_control
* Intercept echi-hcd request to complete RESUME and align it to the start
* of the next microframe.
* If RESUME is complete too late in the microframe, host controller
* detects babble on suspended port and resets the port afterwards.
* This s/w workaround allows to avoid this problem.
* See SWLINUX-1909 for more details
*/
I'll remove the internal bug tracking reference from the comment.
>
> > + if (!org_hub_control)
> > + org_hub_control = ehci_brcm_hc_driver.hub_control;
> > + ehci_brcm_hc_driver.hub_control = ehci_brcm_hub_control;
> > +
> > + /* initialize hcd */
> > + hcd = usb_create_hcd(&ehci_brcm_hc_driver, dev, dev_name(dev));
> > + if (!hcd)
> > + return -ENOMEM;
> > +
> > + platform_set_drvdata(pdev, hcd);
> > + priv = hcd_to_ehci_priv(hcd);
> > +
> > + priv->clk = devm_clk_get_optional(dev, NULL);
> > + if (IS_ERR(priv->clk)) {
> > + err = PTR_ERR(priv->clk);
> > + goto err_hcd;
> > + }
> > +
> > + err = clk_prepare_enable(priv->clk);
>
> But clk was optional, will this break?
devm_clk_get_optional() will set clk to NULL if it's not found and all
the clk_... routines handle a NULL clk by just returning.
>
> > + if (err)
> > + goto err_hcd;
> > +
> > + hcd->regs = devm_platform_get_and_ioremap_resource(pdev, 0, &res_mem);
> > + if (IS_ERR(hcd->regs)) {
> > + err = PTR_ERR(hcd->regs);
> > + goto err_clk;
> > + }
> > + hcd->rsrc_start = res_mem->start;
> > + hcd->rsrc_len = resource_size(res_mem);
> > + err = usb_add_hcd(hcd, irq, IRQF_SHARED);
> > + if (err)
> > + goto err_clk;
> > +
> > + device_wakeup_enable(hcd->self.controller);
> > + device_enable_async_suspend(hcd->self.controller);
> > + platform_set_drvdata(pdev, hcd);
>
> Shouldn't that be set before you register the hcd?
Since usb_add_hcd(hcd,...) takes the hcd pointer and not a device
pointer it shouldn't require platform_set_drvdata() to be called
first.
Looking at the code made me notice that platform_set_drvdata() is also
called at the beginning of the function and it's not needed twice so
I'll remove this second call. This driver used the ehci and ohci
platform drivers as a starting point and they also do this call twice
in probe.
>
> > +
> > + return 0;
> > +
> > +err_clk:
> > + clk_disable_unprepare(priv->clk);
> > +err_hcd:
> > + usb_put_hcd(hcd);
> > +
> > + return err;
> > +}
> > +
> > +static int ehci_brcm_remove(struct platform_device *dev)
> > +{
> > + struct usb_hcd *hcd = platform_get_drvdata(dev);
> > + struct brcm_priv *priv = hcd_to_ehci_priv(hcd);
> > +
> > + usb_remove_hcd(hcd);
> > + clk_disable_unprepare(priv->clk);
> > + usb_put_hcd(hcd);
> > + return 0;
> > +}
> > +
> > +static int __maybe_unused ehci_brcm_suspend(struct device *dev)
> > +{
> > + int ret;
> > + struct usb_hcd *hcd = dev_get_drvdata(dev);
> > + struct brcm_priv *priv = hcd_to_ehci_priv(hcd);
> > + bool do_wakeup = device_may_wakeup(dev);
> > +
> > + ret = ehci_suspend(hcd, do_wakeup);
> > + if (ret)
> > + return ret;
> > + clk_disable_unprepare(priv->clk);
> > + return 0;
> > +}
> > +
> > +static int __maybe_unused ehci_brcm_resume(struct device *dev)
> > +{
> > + struct usb_hcd *hcd = dev_get_drvdata(dev);
> > + struct ehci_hcd *ehci = hcd_to_ehci(hcd);
> > + struct brcm_priv *priv = hcd_to_ehci_priv(hcd);
> > + int err;
> > +
> > + err = clk_prepare_enable(priv->clk);
> > + if (err)
> > + return err;
> > + /*
> > + * SWLINUX-1705: Avoid OUT packet underflows during high memory
> > + * bus usage
> > + * port_status[0x0f] = Broadcom-proprietary USB_EHCI_INSNREG00
> > + * @ 0x90
> > + */
> > + ehci_writel(ehci, 0x00800040, &ehci->regs->port_status[0x10]);
> > + ehci_writel(ehci, 0x00000001, &ehci->regs->port_status[0x12]);
> > +
> > + ehci_resume(hcd, false);
> > +
> > + pm_runtime_disable(dev);
> > + pm_runtime_set_active(dev);
> > + pm_runtime_enable(dev);
> > +
> > + return 0;
> > +}
> > +
> > +static SIMPLE_DEV_PM_OPS(ehci_brcm_pm_ops, ehci_brcm_suspend,
> > + ehci_brcm_resume);
> > +
> > +static const struct of_device_id brcm_ehci_of_match[] = {
> > + { .compatible = "brcm,ehci-brcm-v2", },
> > + { .compatible = "brcm,bcm7445-ehci", },
> > + {}
> > +};
> > +
> > +static struct platform_driver ehci_brcm_driver = {
> > + .probe = ehci_brcm_probe,
> > + .remove = ehci_brcm_remove,
> > + .shutdown = usb_hcd_platform_shutdown,
> > + .driver = {
> > + .name = "ehci-brcm",
> > + .pm = &ehci_brcm_pm_ops,
> > + .of_match_table = brcm_ehci_of_match,
> > + }
> > +};
> > +
> > +static int __init ehci_brcm_init(void)
> > +{
> > + if (usb_disabled())
> > + return -ENODEV;
>
> You check this here, so why are you also checking it in the probe
> function?
Good point. I'll remove the one in probe. This was another thing that
came from the ehci and ohci platform drivers. About a third of the
drivers in this directory do this, probably because they started with
the platform drivers. Do you think it's worth me submitting a patch to
clean this up at some point?
>
> thanks,
>
> greg k-h
I'll wait for your response before sending a V7.
Thanks for the review.
Al
