| Message ID | 1421789010-21900-3-git-send-email-sylvain.rochet@finsecur.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Sylvain, On Tue, 20 Jan 2015 22:23:29 +0100 Sylvain Rochet <sylvain.rochet@finsecur.com> wrote: > Vbus IRQ handler needs a started UDC driver to work because it uses > udc->driver, which is set by the UDC start handler. The previous way > chosen was to return from interrupt if udc->driver is NULL using a > spinlock. > > This patch now request the Vbus signal IRQ in UDC start instead of UDC > probe and release the IRQ in UDC stop before udc->driver is set back to > NULL. This way we don't need the check about udc->driver in interruption > handler, therefore we don't need the spinlock as well anymore. > > This was chosen against using set_irq_flags() to request a not auto > enabled IRQ (IRQF_NOAUTOEN flag) because set_irq_flags() can't change > just one flag, therefore it must be called with all flags, without > respecting what the AIC previously did. Naively copying IRQ flags > currently set by the AIC looked like error-prone if defaults flags > change at some point in the future. > > Signed-off-by: Sylvain Rochet <sylvain.rochet@finsecur.com> > Suggested-by: Boris Brezillon <boris.brezillon@free-electrons.com> > --- > drivers/usb/gadget/udc/atmel_usba_udc.c | 64 ++++++++++++++++----------------- > 1 file changed, 32 insertions(+), 32 deletions(-) > > diff --git a/drivers/usb/gadget/udc/atmel_usba_udc.c b/drivers/usb/gadget/udc/atmel_usba_udc.c > index e207d75..546da63 100644 > --- a/drivers/usb/gadget/udc/atmel_usba_udc.c > +++ b/drivers/usb/gadget/udc/atmel_usba_udc.c > @@ -1729,10 +1729,6 @@ static irqreturn_t usba_vbus_irq(int irq, void *devid) > > spin_lock(&udc->lock); > > - /* May happen if Vbus pin toggles during probe() */ > - if (!udc->driver) > - goto out; > - > vbus = vbus_is_present(udc); > if (vbus != udc->vbus_prev) { > if (vbus) { > @@ -1753,7 +1749,6 @@ static irqreturn_t usba_vbus_irq(int irq, void *devid) > udc->vbus_prev = vbus; > } > > -out: > spin_unlock(&udc->lock); > > return IRQ_HANDLED; > @@ -1767,23 +1762,27 @@ static int atmel_usba_start(struct usb_gadget *gadget, > unsigned long flags; > > spin_lock_irqsave(&udc->lock, flags); > - > udc->devstatus = 1 << USB_DEVICE_SELF_POWERED; > udc->driver = driver; > spin_unlock_irqrestore(&udc->lock, flags); > > ret = clk_prepare_enable(udc->pclk); > if (ret) > - return ret; > + goto err_pclk; > ret = clk_prepare_enable(udc->hclk); > - if (ret) { > - clk_disable_unprepare(udc->pclk); > - return ret; > - } > + if (ret) > + goto err_hclk; > > udc->vbus_prev = 0; > - if (gpio_is_valid(udc->vbus_pin)) > - enable_irq(gpio_to_irq(udc->vbus_pin)); > + if (gpio_is_valid(udc->vbus_pin)) { > + ret = request_irq(gpio_to_irq(udc->vbus_pin), > + usba_vbus_irq, 0, > + "atmel_usba_udc", udc); > + if (ret) { > + udc->vbus_pin = -ENODEV; I guess you're trying to protect against free_irq by changing the vbus_pin value (making it an invalid gpio id), but I think you should leave it unchanged for two reasons: 1) If the request_irq call temporary fails (an ENOMEM for example) then you should be able to retry later, and modifying the vbus_pin value will prevent that. 2) atmel_usba_stop will never be called if the atmel_usba_start failed, so there's no need to protect against this free_irq. Best Regards, Boris
Hi Boris, On Wed, Jan 21, 2015 at 10:20:16AM +0100, Boris Brezillon wrote: > > > > udc->vbus_prev = 0; > > - if (gpio_is_valid(udc->vbus_pin)) > > - enable_irq(gpio_to_irq(udc->vbus_pin)); > > + if (gpio_is_valid(udc->vbus_pin)) { > > + ret = request_irq(gpio_to_irq(udc->vbus_pin), > > + usba_vbus_irq, 0, > > + "atmel_usba_udc", udc); > > + if (ret) { > > + udc->vbus_pin = -ENODEV; > > I guess you're trying to protect against free_irq by changing the > vbus_pin value (making it an invalid gpio id), but I think you should > leave it unchanged for two reasons: > 1) If the request_irq call temporary fails (an ENOMEM for example) then > you should be able to retry later, and modifying the vbus_pin value > will prevent that. > 2) atmel_usba_stop will never be called if the atmel_usba_start failed, > so there's no need to protect against this free_irq. Indeed. By the way, I just discovered irq_set_status_flags(irq, IRQ_NOAUTOEN); … so I am going for a v5 and the previous part is not relevant anymore. Sylvain
diff --git a/drivers/usb/gadget/udc/atmel_usba_udc.c b/drivers/usb/gadget/udc/atmel_usba_udc.c index e207d75..546da63 100644 --- a/drivers/usb/gadget/udc/atmel_usba_udc.c +++ b/drivers/usb/gadget/udc/atmel_usba_udc.c @@ -1729,10 +1729,6 @@ static irqreturn_t usba_vbus_irq(int irq, void *devid) spin_lock(&udc->lock); - /* May happen if Vbus pin toggles during probe() */ - if (!udc->driver) - goto out; - vbus = vbus_is_present(udc); if (vbus != udc->vbus_prev) { if (vbus) { @@ -1753,7 +1749,6 @@ static irqreturn_t usba_vbus_irq(int irq, void *devid) udc->vbus_prev = vbus; } -out: spin_unlock(&udc->lock); return IRQ_HANDLED; @@ -1767,23 +1762,27 @@ static int atmel_usba_start(struct usb_gadget *gadget, unsigned long flags; spin_lock_irqsave(&udc->lock, flags); - udc->devstatus = 1 << USB_DEVICE_SELF_POWERED; udc->driver = driver; spin_unlock_irqrestore(&udc->lock, flags); ret = clk_prepare_enable(udc->pclk); if (ret) - return ret; + goto err_pclk; ret = clk_prepare_enable(udc->hclk); - if (ret) { - clk_disable_unprepare(udc->pclk); - return ret; - } + if (ret) + goto err_hclk; udc->vbus_prev = 0; - if (gpio_is_valid(udc->vbus_pin)) - enable_irq(gpio_to_irq(udc->vbus_pin)); + if (gpio_is_valid(udc->vbus_pin)) { + ret = request_irq(gpio_to_irq(udc->vbus_pin), + usba_vbus_irq, 0, + "atmel_usba_udc", udc); + if (ret) { + udc->vbus_pin = -ENODEV; + goto err; + } + } /* If Vbus is present, enable the controller and wait for reset */ spin_lock_irqsave(&udc->lock, flags); @@ -1797,6 +1796,17 @@ static int atmel_usba_start(struct usb_gadget *gadget, spin_unlock_irqrestore(&udc->lock, flags); return 0; + +err: + clk_disable_unprepare(udc->hclk); +err_hclk: + clk_disable_unprepare(udc->pclk); +err_pclk: + spin_lock_irqsave(&udc->lock, flags); + udc->devstatus &= ~(1 << USB_DEVICE_SELF_POWERED); + udc->driver = NULL; + spin_unlock_irqrestore(&udc->lock, flags); + return ret; } static int atmel_usba_stop(struct usb_gadget *gadget) @@ -1805,7 +1815,7 @@ static int atmel_usba_stop(struct usb_gadget *gadget) unsigned long flags; if (gpio_is_valid(udc->vbus_pin)) - disable_irq(gpio_to_irq(udc->vbus_pin)); + free_irq(gpio_to_irq(udc->vbus_pin), udc); spin_lock_irqsave(&udc->lock, flags); udc->gadget.speed = USB_SPEED_UNKNOWN; @@ -2047,24 +2057,14 @@ static int usba_udc_probe(struct platform_device *pdev) } udc->irq = irq; - if (gpio_is_valid(udc->vbus_pin)) { - if (!devm_gpio_request(&pdev->dev, udc->vbus_pin, "atmel_usba_udc")) { - ret = devm_request_irq(&pdev->dev, - gpio_to_irq(udc->vbus_pin), - usba_vbus_irq, 0, - "atmel_usba_udc", udc); - if (ret) { - udc->vbus_pin = -ENODEV; - dev_warn(&udc->pdev->dev, - "failed to request vbus irq; " - "assuming always on\n"); - } else { - disable_irq(gpio_to_irq(udc->vbus_pin)); - } - } else { - /* gpio_request fail so use -EINVAL for gpio_is_valid */ - udc->vbus_pin = -EINVAL; - } + if (gpio_is_valid(udc->vbus_pin) + && devm_gpio_request(&pdev->dev, udc->vbus_pin, + "atmel_usba_udc")) { + /* gpio_request fail so use -EINVAL for gpio_is_valid */ + udc->vbus_pin = -EINVAL; + dev_warn(&udc->pdev->dev, + "failed to request vbus gpio; " + "assuming always on\n"); } ret = usb_add_gadget_udc(&pdev->dev, &udc->gadget);
Vbus IRQ handler needs a started UDC driver to work because it uses udc->driver, which is set by the UDC start handler. The previous way chosen was to return from interrupt if udc->driver is NULL using a spinlock. This patch now request the Vbus signal IRQ in UDC start instead of UDC probe and release the IRQ in UDC stop before udc->driver is set back to NULL. This way we don't need the check about udc->driver in interruption handler, therefore we don't need the spinlock as well anymore. This was chosen against using set_irq_flags() to request a not auto enabled IRQ (IRQF_NOAUTOEN flag) because set_irq_flags() can't change just one flag, therefore it must be called with all flags, without respecting what the AIC previously did. Naively copying IRQ flags currently set by the AIC looked like error-prone if defaults flags change at some point in the future. Signed-off-by: Sylvain Rochet <sylvain.rochet@finsecur.com> Suggested-by: Boris Brezillon <boris.brezillon@free-electrons.com> --- drivers/usb/gadget/udc/atmel_usba_udc.c | 64 ++++++++++++++++----------------- 1 file changed, 32 insertions(+), 32 deletions(-)