Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Message-ID: <1547037068.4433.122.camel@mhfsdcap03>
Subject: Re: [PATCH 4/4] usb: musb: Add support for MediaTek musb controller
From:   Min Guo <min.guo@mediatek.com>
To:     Bin Liu <b-liu@ti.com>
CC:     Rob Herring <robh+dt@kernel.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Mark Rutland <mark.rutland@arm.com>,
        "Matthias Brugger" <matthias.bgg@gmail.com>,
        Alan Stern <stern@rowland.harvard.edu>,
        <chunfeng.yun@mediatek.com>, <linux-usb@vger.kernel.org>,
        <devicetree@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
        <linux-arm-kernel@lists.infradead.org>,
        <linux-mediatek@lists.infradead.org>,
        Yonglong Wu <yonglong.wu@mediatek.com>
Date:   Wed, 9 Jan 2019 20:31:08 +0800
In-Reply-To: <20190108154441.GG25910@uda0271908>
References: <1545896066-897-1-git-send-email-min.guo@mediatek.com>
         <1545896066-897-5-git-send-email-min.guo@mediatek.com>
         <20190108154441.GG25910@uda0271908>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
MIME-Version: 1.0
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Bin,
On Tue, 2019-01-08 at 09:44 -0600, Bin Liu wrote:
> Hi,
> 
> On Thu, Dec 27, 2018 at 03:34:26PM +0800, min.guo@mediatek.com wrote:
> > From: Min Guo <min.guo@mediatek.com>
> > 
> > This adds support for MediaTek musb controller in
> > host, peripheral and otg mode
> > 
> > Signed-off-by: Min Guo <min.guo@mediatek.com>
> > Signed-off-by: Yonglong Wu <yonglong.wu@mediatek.com>
> > ---
> >  drivers/usb/musb/Kconfig     |   8 +-
> >  drivers/usb/musb/Makefile    |   1 +
> >  drivers/usb/musb/mediatek.c  | 562 +++++++++++++++++++++++++++++++++++++++++++
> >  drivers/usb/musb/musb_core.c |  10 +
> >  drivers/usb/musb/musb_core.h |   1 +
> >  drivers/usb/musb/musb_dma.h  |   1 +
> >  drivers/usb/musb/musb_host.c |  79 ++++--
> >  drivers/usb/musb/musb_regs.h |   6 +
> >  drivers/usb/musb/musbhsdma.c |  34 ++-
> >  9 files changed, 671 insertions(+), 31 deletions(-)
> >  create mode 100644 drivers/usb/musb/mediatek.c
> > 
> > diff --git a/drivers/usb/musb/Kconfig b/drivers/usb/musb/Kconfig
> > index ad08895..540fc9f 100644
> > --- a/drivers/usb/musb/Kconfig
> > +++ b/drivers/usb/musb/Kconfig
> > @@ -115,6 +115,12 @@ config USB_MUSB_JZ4740
> >  	depends on USB_MUSB_GADGET
> >  	depends on USB_OTG_BLACKLIST_HUB
> >  
> > +config USB_MUSB_MEDIATEK
> > +	tristate "MediaTek platforms"
> > +	depends on ARCH_MEDIATEK
> 
> Please also add '|| COMPILE_TEST' to make testing easier.

Ok

> > +	depends on NOP_USB_XCEIV
> > +	depends on GENERIC_PHY
> > +
> >  config USB_MUSB_AM335X_CHILD
> >  	tristate
> >  
> > @@ -141,7 +147,7 @@ config USB_UX500_DMA
> >  
> >  config USB_INVENTRA_DMA
> >  	bool 'Inventra'
> > -	depends on USB_MUSB_OMAP2PLUS
> > +	depends on USB_MUSB_OMAP2PLUS || USB_MUSB_MEDIATEK
> >  	help
> >  	  Enable DMA transfers using Mentor's engine.
> >  
> > diff --git a/drivers/usb/musb/Makefile b/drivers/usb/musb/Makefile
> > index 3a88c79..63d82d0 100644
> > --- a/drivers/usb/musb/Makefile
> > +++ b/drivers/usb/musb/Makefile
> > @@ -24,6 +24,7 @@ obj-$(CONFIG_USB_MUSB_DA8XX)			+= da8xx.o
> >  obj-$(CONFIG_USB_MUSB_UX500)			+= ux500.o
> >  obj-$(CONFIG_USB_MUSB_JZ4740)			+= jz4740.o
> >  obj-$(CONFIG_USB_MUSB_SUNXI)			+= sunxi.o
> > +obj-$(CONFIG_USB_MUSB_MEDIATEK)      		+= mediatek.o
> >  
> >  
> >  obj-$(CONFIG_USB_MUSB_AM335X_CHILD)		+= musb_am335x.o
> > diff --git a/drivers/usb/musb/mediatek.c b/drivers/usb/musb/mediatek.c
> > new file mode 100644
> > index 0000000..15a6460
> > --- /dev/null
> > +++ b/drivers/usb/musb/mediatek.c
> 
> [snip]
> I will review this section later after we sorted out other things.
> 
> > diff --git a/drivers/usb/musb/musb_core.c b/drivers/usb/musb/musb_core.c
> > index b7d5627..d60f76f 100644
> > --- a/drivers/usb/musb/musb_core.c
> > +++ b/drivers/usb/musb/musb_core.c
> > @@ -1028,6 +1028,16 @@ static void musb_disable_interrupts(struct musb *musb)
> >  	temp = musb_readb(mbase, MUSB_INTRUSB);
> >  	temp = musb_readw(mbase, MUSB_INTRTX);
> >  	temp = musb_readw(mbase, MUSB_INTRRX);
> > +
> > +	/*  MediaTek controller interrupt status is W1C */
> 
> This W1C doesn't match to the MUSB Programming Guide that I have. Those
> registers are read-only.
> Is the difference due to the IP intergration in the mtk platforms? or is
> it a new version of the MUSB controller? If latter, what is the version?

This is difference due to the IP intergration in mtk platforms. W1C is
easy for CpdeViser debug.

> > +	if (musb->ops->quirks & MUSB_MTK_QUIRKS) {
> 
> Basically we don't want to use this type of platform specific quirks if
> possible, so let's try to not use it.

I will try my best to avoid using it.

> > +		musb_writeb(mbase, MUSB_INTRUSB,
> > +			musb_readb(mbase, MUSB_INTRUSB));
> 
> For this clearing register bit operation, please create platform hooks
> musb_clearb() and musb_clearw() in struct musb_platform_ops instead,
> then follow how musb_readb() pointer is assigned in
> musb_init_controller() to use the W1C version for mtk platform.

I have tried implementing musb_readb(), musb_readw() interface with
interrupt status W1C function in struct musb_platform_ops. But this
interface will require a global variable to hold MAC basic address for
judgment, and then special handling of the interrupt state. A global
variable will make the driver work with only a single instance, so it
can't work on some MTK platforms which have two instances.
How about creating musb_clearb/w() as following:
void (*clearb)(void __iomem *addr, unsigned offset, u8 data);
void (*clearw)(void __iomem *addr, unsigned offset, u16 data);


> > +		musb_writew(mbase, MUSB_INTRRX,
> > +			musb_readw(mbase, MUSB_INTRRX));
> > +		musb_writew(mbase, MUSB_INTRTX,
> > +			musb_readw(mbase, MUSB_INTRTX));
> > +	}
> >  }
> >  
> >  static void musb_enable_interrupts(struct musb *musb)
> > diff --git a/drivers/usb/musb/musb_core.h b/drivers/usb/musb/musb_core.h
> > index 04203b7..1bf4e9a 100644
> > --- a/drivers/usb/musb/musb_core.h
> > +++ b/drivers/usb/musb/musb_core.h
> > @@ -138,6 +138,7 @@ enum musb_g_ep0_state {
> >   */
> >  struct musb_platform_ops {
> >  
> > +#define MUSB_MTK_QUIRKS	BIT(10)
> >  #define MUSB_G_NO_SKB_RESERVE	BIT(9)
> >  #define MUSB_DA8XX		BIT(8)
> >  #define MUSB_PRESERVE_SESSION	BIT(7)
> > diff --git a/drivers/usb/musb/musb_dma.h b/drivers/usb/musb/musb_dma.h
> > index 281e75d3..b218210 100644
> > --- a/drivers/usb/musb/musb_dma.h
> > +++ b/drivers/usb/musb/musb_dma.h
> > @@ -197,6 +197,7 @@ static inline void musb_dma_controller_destroy(struct dma_controller *d) { }
> >  extern struct dma_controller *
> >  musbhs_dma_controller_create(struct musb *musb, void __iomem *base);
> >  extern void musbhs_dma_controller_destroy(struct dma_controller *c);
> > +extern irqreturn_t dma_controller_irq(int irq, void *private_data);
> >  
> >  extern struct dma_controller *
> >  tusb_dma_controller_create(struct musb *musb, void __iomem *base);
> > diff --git a/drivers/usb/musb/musb_host.c b/drivers/usb/musb/musb_host.c
> > index b59ce9a..b1b0216 100644
> > --- a/drivers/usb/musb/musb_host.c
> > +++ b/drivers/usb/musb/musb_host.c
> > @@ -292,20 +292,73 @@ static inline void musb_save_toggle(struct musb_qh *qh, int is_in,
> >  {
> >  	void __iomem		*epio = qh->hw_ep->regs;
> >  	u16			csr;
> > +	struct musb *musb = qh->hw_ep->musb;
> >  
> >  	/*
> >  	 * FIXME: the current Mentor DMA code seems to have
> >  	 * problems getting toggle correct.
> >  	 */
> >  
> > -	if (is_in)
> > -		csr = musb_readw(epio, MUSB_RXCSR) & MUSB_RXCSR_H_DATATOGGLE;
> > -	else
> > -		csr = musb_readw(epio, MUSB_TXCSR) & MUSB_TXCSR_H_DATATOGGLE;
> > +	/* MediaTek controller has private toggle register */
> 
> only one toggle register for all endpoints? how does it handle
> difference toggle values for different endpoints?

MediaTek controller has separate registers to describe TX/RX toggle.

> > +	if (musb->ops->quirks & MUSB_MTK_QUIRKS) {
> > +		u16 toggle;
> > +		u8 epnum = qh->hw_ep->epnum;
> > +
> > +		if (is_in)
> > +			toggle = musb_readl(musb->mregs, MUSB_RXTOG);
> 
> should use musb_readw() instead? MUSB_RXTOG seems to be 16bit.

Ok

> > +		else
> > +			toggle = musb_readl(musb->mregs, MUSB_TXTOG);
> > +
> > +		csr = toggle & (1 << epnum);
> > +	} else {
> > +		if (is_in)
> > +			csr = musb_readw(epio, MUSB_RXCSR)
> > +				& MUSB_RXCSR_H_DATATOGGLE;
> > +		else
> > +			csr = musb_readw(epio, MUSB_TXCSR)
> > +				& MUSB_TXCSR_H_DATATOGGLE;
> 
> Does this logic still work for the mtk platform even if it has its own
> private toggle register? If so, we don't need to change here.

Sorry, this logic can not work on mtk platform, bit
MUSB_RXCSR_H_DATATOGGLE and MUSB_TXCSR_H_DATATOGGLE are used for other
function.

> If not, let's try to not use this quirk flag. Please create a hook
> musb_platform_get_toggle() in struct musb_platform_ops.

Does the method of implement musb_platform_get_toggle() is prepare
musb_default_get_toggle with common function, then follow how
musb_readb() pointer is assigned in musb_init_controller()?

How about creating musb_platform_get_toggle() as following:
u16 (*get_toggle)(struct musb* musb, struct musb_qh *qh, int is_in);

> > +	}
> >  
> >  	usb_settoggle(urb->dev, qh->epnum, !is_in, csr ? 1 : 0);
> >  }
> >  
> > +static inline u16 musb_set_toggle(struct musb_qh *qh, int is_in,
> > +					struct urb *urb)
> > +{
> > +	u16 csr = 0;
> > +	u16 toggle = 0;
> > +	struct musb *musb = qh->hw_ep->musb;
> > +	u8 epnum = qh->hw_ep->epnum;
> > +
> > +	toggle = usb_gettoggle(urb->dev, qh->epnum, !is_in);
> > +
> > +	/* MediaTek controller has private toggle register */
> > +	if (musb->ops->quirks & MUSB_MTK_QUIRKS) {
> > +		if (is_in) {
> > +			musb_writel(musb->mregs, MUSB_RXTOGEN, (1 << epnum));
> > +			musb_writel(musb->mregs, MUSB_RXTOG, (toggle << epnum));
> > +		} else {
> > +			musb_writel(musb->mregs, MUSB_TXTOGEN, (1 << epnum));
> > +			musb_writel(musb->mregs, MUSB_TXTOG, (toggle << epnum));
> > +		}
> > +	} else {
> > +		if (is_in) {
> > +			if (toggle)
> > +				csr = MUSB_RXCSR_H_WR_DATATOGGLE
> > +						| MUSB_RXCSR_H_DATATOGGLE;
> > +			else
> > +				csr = 0;
> > +		} else {
> > +			if (toggle)
> > +				csr |= MUSB_TXCSR_H_WR_DATATOGGLE
> > +						| MUSB_TXCSR_H_DATATOGGLE;
> > +			else
> > +				csr |= MUSB_TXCSR_CLRDATATOG;
> > +		}
> > +	}
> > +	return csr;
> 
> Please create a seperate patch for this musb_set_toggle() without adding
> the mtk logic. It is a nice cleanup.

Does this like get toggle implementation, create a hook
musb_platform_set_toggle() in struct musb_platform_ops?

> > +}
> > +
> >  /*
> >   * Advance this hardware endpoint's queue, completing the specified URB and
> >   * advancing to either the next URB queued to that qh, or else invalidating
> > @@ -772,13 +825,8 @@ static void musb_ep_program(struct musb *musb, u8 epnum,
> >  					);
> >  			csr |= MUSB_TXCSR_MODE;
> >  
> > -			if (!hw_ep->tx_double_buffered) {
> > -				if (usb_gettoggle(urb->dev, qh->epnum, 1))
> > -					csr |= MUSB_TXCSR_H_WR_DATATOGGLE
> > -						| MUSB_TXCSR_H_DATATOGGLE;
> > -				else
> > -					csr |= MUSB_TXCSR_CLRDATATOG;
> > -			}
> > +			if (!hw_ep->tx_double_buffered)
> > +				csr |= musb_set_toggle(qh, !is_out, urb);
> >  
> >  			musb_writew(epio, MUSB_TXCSR, csr);
> >  			/* REVISIT may need to clear FLUSHFIFO ... */
> > @@ -860,17 +908,12 @@ static void musb_ep_program(struct musb *musb, u8 epnum,
> >  
> >  	/* IN/receive */
> >  	} else {
> > -		u16	csr;
> > +		u16	csr = 0;
> >  
> >  		if (hw_ep->rx_reinit) {
> >  			musb_rx_reinit(musb, qh, epnum);
> > +			csr |= musb_set_toggle(qh, !is_out, urb);
> >  
> > -			/* init new state: toggle and NYET, maybe DMA later */
> > -			if (usb_gettoggle(urb->dev, qh->epnum, 0))
> > -				csr = MUSB_RXCSR_H_WR_DATATOGGLE
> > -					| MUSB_RXCSR_H_DATATOGGLE;
> > -			else
> > -				csr = 0;
> >  			if (qh->type == USB_ENDPOINT_XFER_INT)
> >  				csr |= MUSB_RXCSR_DISNYET;
> >  
> > diff --git a/drivers/usb/musb/musb_regs.h b/drivers/usb/musb/musb_regs.h
> > index 5cd7264..ffbe267 100644
> > --- a/drivers/usb/musb/musb_regs.h
> > +++ b/drivers/usb/musb/musb_regs.h
> > @@ -273,6 +273,12 @@
> >  #define MUSB_RXHUBADDR		0x06
> >  #define MUSB_RXHUBPORT		0x07
> >  
> > +/* MediaTek controller toggle enable and status reg */
> > +#define MUSB_RXTOG		0x80
> > +#define MUSB_RXTOGEN		0x82
> > +#define MUSB_TXTOG		0x84
> > +#define MUSB_TXTOGEN		0x86
> 
> Again, these offsets are for different registers in the MUSB version I
> have, please let me know if you have different version of the MUSB IP.

Sorry, these are MediaTek controller private registers used for control
toggle.

> > +
> >  static inline u8 musb_read_configdata(void __iomem *mbase)
> >  {
> >  	musb_writeb(mbase, MUSB_INDEX, 0);
> > diff --git a/drivers/usb/musb/musbhsdma.c b/drivers/usb/musb/musbhsdma.c
> > index 824adcb..c545475 100644
> > --- a/drivers/usb/musb/musbhsdma.c
> > +++ b/drivers/usb/musb/musbhsdma.c
> > @@ -263,7 +263,7 @@ static int dma_channel_abort(struct dma_channel *channel)
> >  	return 0;
> >  }
> >  
> > -static irqreturn_t dma_controller_irq(int irq, void *private_data)
> > +irqreturn_t dma_controller_irq(int irq, void *private_data)
> >  {
> >  	struct musb_dma_controller *controller = private_data;
> >  	struct musb *musb = controller->private_data;
> > @@ -285,6 +285,8 @@ static irqreturn_t dma_controller_irq(int irq, void *private_data)
> >  	spin_lock_irqsave(&musb->lock, flags);
> >  
> >  	int_hsdma = musb_readb(mbase, MUSB_HSDMA_INTR);
> > +	if (musb->ops->quirks & MUSB_MTK_QUIRKS)
> > +		musb_writeb(musb->mregs, MUSB_HSDMA_INTR, int_hsdma);
> 
> you can use musb_clearb() defined above to get rid of this quirk.

OK. 

> >  
> >  	if (!int_hsdma) {
> >  		musb_dbg(musb, "spurious DMA irq");
> > @@ -377,15 +379,17 @@ static irqreturn_t dma_controller_irq(int irq, void *private_data)
> >  	spin_unlock_irqrestore(&musb->lock, flags);
> >  	return retval;
> >  }
> > +EXPORT_SYMBOL_GPL(dma_controller_irq);
> >  
> >  void musbhs_dma_controller_destroy(struct dma_controller *c)
> >  {
> >  	struct musb_dma_controller *controller = container_of(c,
> >  			struct musb_dma_controller, controller);
> > +	struct musb *musb = controller->private_data;
> >  
> >  	dma_controller_stop(controller);
> >  
> > -	if (controller->irq)
> > +	if (!(musb->ops->quirks & MUSB_MTK_QUIRKS) && controller->irq)
> >  		free_irq(controller->irq, c);
> >  
> >  	kfree(controller);
> > @@ -398,11 +402,15 @@ struct dma_controller *musbhs_dma_controller_create(struct musb *musb,
> >  	struct musb_dma_controller *controller;
> >  	struct device *dev = musb->controller;
> >  	struct platform_device *pdev = to_platform_device(dev);
> > -	int irq = platform_get_irq_byname(pdev, "dma");
> > +	int irq = -1;
> >  
> > -	if (irq <= 0) {
> > -		dev_err(dev, "No DMA interrupt line!\n");
> > -		return NULL;
> > +	if (!(musb->ops->quirks & MUSB_MTK_QUIRKS)) {
> > +		irq = platform_get_irq_byname(pdev, "dma");
> > +
> > +		if (irq < 0) {
> > +			dev_err(dev, "No DMA interrupt line!\n");
> > +			return NULL;
> > +		}
> >  	}
> 
> Please create musbhs_dma_controller_destroy_noirq() for your platform to
> not use the quirk.

OK.

> >  
> >  	controller = kzalloc(sizeof(*controller), GFP_KERNEL);
> > @@ -418,15 +426,17 @@ struct dma_controller *musbhs_dma_controller_create(struct musb *musb,
> >  	controller->controller.channel_program = dma_channel_program;
> >  	controller->controller.channel_abort = dma_channel_abort;
> >  
> > -	if (request_irq(irq, dma_controller_irq, 0,
> > +	if (!(musb->ops->quirks & MUSB_MTK_QUIRKS)) {
> > +		if (request_irq(irq, dma_controller_irq, 0,
> >  			dev_name(musb->controller), &controller->controller)) {
> > -		dev_err(dev, "request_irq %d failed!\n", irq);
> > -		musb_dma_controller_destroy(&controller->controller);
> > +			dev_err(dev, "request_irq %d failed!\n", irq);
> > +			musb_dma_controller_destroy(&controller->controller);
> >  
> > -		return NULL;
> > -	}
> > +			return NULL;
> > +		}
> >  
> > -	controller->irq = irq;
> > +		controller->irq = irq;
> > +	}
> >  
> >  	return &controller->controller;
> >  }
> 
> Same here, create musbhs_dma_controller_create_noirq(). Then use both
> new API for the mtk glue driver.

OK.

> Regards,
> -Bin.