usb: gadget: renesas: add support for RZ/G2L SoC
Subsystems (2)
Quoted 13 lines · Michele Bisogno
Support the Renesas RZ/G2L (r9a07g044) USBHS IP by addressing platform-specific reset, glue logic, and FIFO access requirements.
- regulator: rzg2l: Implement
CON_CTRLmanual connectivity notification. By settingSEL_CONNECTandCONNECT_1, the UTM+ core is notified of connection status in peripheral mode.
- gadget: common: Add "renesas,rzg2l-usbhs" compatible and support reset deassertion. Fix a re-entrancy bug by callingusb_del_gadget_udc()in the remove callback to prevent stale UDC list entries.
- gadget: fifo: Refactor PIO push/pop logic to handle trailing bytes using iowrite8/ioread8. This ensures correct data alignment on RZ/G2L and simplifies the data transfer loop.
Quoted 3 lines
Signed-off-by Michele Bisogno <[email protected]>--- .../power/regulator/rzg2l-usbphy-regulator.c | 10 ++++++--
Regulator change should be a separate patch please.
Quoted 2 lines
drivers/usb/gadget/rcar/common.c | 15 ++++++++++++ drivers/usb/gadget/rcar/fifo.c | 23 +++++++++++--------
drivers/usb/gadget/rcar is pulled from Linux 1b4861e32e46 ("Linux 6.9.3") , but none of these changes in this patch are even in Linux next. Is there at least a submitted Linux kernel patch which adds these matching changes ? I would very much like to keep the USBHS drivers synchronized between U-Boot and Linux.
The addition of RZ/G2L into USBHS driver in Linux was much simpler:
PATCH 1/1 usb: gadget: renesas: add support for RZ/G2L SoC
Support the Renesas RZ/G2L (r9a07g044) USBHS IP by addressing platform-specific reset, glue logic, and FIFO access requirements.
- regulator: rzg2l: Implement CON_CTRL manual connectivity notification. By setting SEL_CONNECT and CONNECT_1, the UTM+ core is notified of connection status in peripheral mode.
- gadget: common: Add "renesas,rzg2l-usbhs" compatible and support reset deassertion. Fix a re-entrancy bug by calling usb_del_gadget_udc() in the remove callback to prevent stale UDC list entries.
- gadget: fifo: Refactor PIO push/pop logic to handle trailing bytes using iowrite8/ioread8. This ensures correct data alignment on RZ/G2L and simplifies the data transfer loop.
--- .../power/regulator/rzg2l-usbphy-regulator.c | 10 ++++++-- drivers/usb/gadget/rcar/common.c | 15 ++++++++++++ drivers/usb/gadget/rcar/fifo.c | 23 +++++++++++-------- 3 files changed, 36 insertions(+), 12 deletions(-)
drivers/power/regulator/rzg2l-usbphy-regulator.c +8 -2
@@ -9,16 +9,22 @@#include <renesas/rzg2l-usbphy.h> #define VBENCTL 0x03c +#define CON_CTRL 0x020#define VBENCTL_VBUS_SEL BIT(0) +#define SEL_CONNECT BIT(4) +#define CONNECT_1 BIT(0)static int rzg2l_usbphy_regulator_set_enable (struct udevice *dev,bool enable) {struct rzg2l_usbphy_ctrl_priv *priv =dev_get_priv (dev->parent); -if (enable) +if (enable) {clrbits_le32 (priv->regs +VBENCTL ,VBENCTL_VBUS_SEL ); -else +setbits_le32 (priv->regs +CON_CTRL ,SEL_CONNECT |CONNECT_1 ); + }else {setbits_le32 (priv->regs +VBENCTL ,VBENCTL_VBUS_SEL ); +clrbits_le32 (priv->regs +CON_CTRL ,SEL_CONNECT |CONNECT_1 ); + }return 0 ; }
drivers/usb/gadget/rcar/common.c +15 -0
@@ -16,6 +16,7 @@#include <linux/usb/ch9.h> #include <linux/usb/gadget.h> #include <usb.h> +#include <reset.h> #include "common.h"
@@ -397,6 +398,7 @@ static int usbhs_udc_otg_probe(struct udevice *dev)struct usbhs_priv_otg_data *priv =dev_get_priv (dev);struct usb_gadget *gadget;struct clk_bulk clk_bulk; +struct reset_ctl_bulk reset_bulk;int ret = -EINVAL ; priv->base =dev_read_addr_ptr (dev);
@@ -411,6 +413,14 @@ static int usbhs_udc_otg_probe(struct udevice *dev)if (ret)return ret; + ret =reset_get_bulk (dev, &reset_bulk); +if (ret) +return ret; + + ret =reset_deassert_bulk (&reset_bulk); +if (ret) +return ret; +clrsetbits_le32 (priv->base +UGCTRL2 ,UGCTRL2_USB0SEL_MASK ,UGCTRL2_USB0SEL_EHCI );clrsetbits_le16 (priv->base +LPSTS ,LPSTS_SUSPM ,LPSTS_SUSPM );
@@ -447,7 +457,11 @@ err_clk:static int usbhs_udc_otg_remove (struct udevice *dev) {struct usbhs_priv_otg_data *priv =dev_get_priv (dev); +struct usb_gadget *gadget; + gadget =usbhsg_get_gadget (&priv->usbhs_priv); +if (gadget) +usb_del_gadget_udc (gadget);usbhs_rcar3_power_ctrl (&priv->usbhs_priv, false);usbhs_mod_remove (&priv->usbhs_priv);usbhs_fifo_remove (&priv->usbhs_priv);
@@ -460,6 +474,7 @@ static int usbhs_udc_otg_remove(struct udevice *dev)static const struct udevice_id usbhs_udc_otg_ids[] = { { .compatible ="renesas,rcar-gen3-usbhs" }, + { .compatible ="renesas,rzg2l-usbhs" }, {}, };
drivers/usb/gadget/rcar/fifo.c +13 -10
@@ -535,18 +535,21 @@ static int usbhsf_pio_try_push(struct usbhs_pkt *pkt, int *is_done)* 32-bit access only */ if (len >=4 && !((unsigned long )buf &0x03 )) { -iowrite32_rep (addr, buf, len /4 ); - len %=4 ; - buf += total_len - len; +int words = len /4 ; + +iowrite32_rep (addr, buf, words); + buf += (words *4 ); + len -= (words *4 ); }/* the rest operation */ -if (usbhs_get_dparam (priv, cfifo_byte_addr)) { -for (i =0 ; i < len; i++) -iowrite8 (buf[i], addr + (i &0x03 )); - }else { -for (i =0 ; i < len; i++) -iowrite8 (buf[i], addr + (0x03 - (i &0x03 ))); +if (len >0 ) { +if (len ==2 && !((unsigned long )buf &0x01 )) { +iowrite16 (*(u16 *)buf, addr); + }else { +for (i =0 ; i < len; i++) +iowrite8 (buf[i], addr); + } }/*
@@ -716,7 +719,7 @@ static int usbhsf_pio_try_pop(struct usbhs_pkt *pkt, int *is_done)if (!(i &0x03 )) data =ioread32 (addr); - buf[i] = (data >> ((i &0x03 ) *8 )) &0xff ; + buf[i] =ioread8 (addr); } usbhs_fifo_read_end:
Signature · 1 line
2.34.1
Hi Marek,
Thank you for the review and the link to the Linux implementation.
Quoted 1 line
Regulator change should be a separate patch please.
I agree. I will split the regulator changes into a separate patch for the v2 series to keep the subsystem changes distinct.
Quoted 2 lines
drivers/usb/gadget/rcaris pulled from Linux1b4861e32e46... but none of these changes in this patch are even in Linux next.
The RZ/G2L support in Linux relies heavily on DMA for data transfers. In U-Boot, we are primarily using the PIO path. The refactoring in fifo.c (switching to ioread8/iowrite8) was found to be necessary to handle trailing bytes and alignment specifically for the RZ/G2L bus when DMA is not active.
Additionally, the usb_del_gadget_udc() change in common.c addresses a re-entrancy bug unique to U-Boot's gadget lifecycle. Without this, the UDC list becomes stale, causing the board to crash if a command like 'ums' is executed more than once.
Quoted 1 line
The addition of RZ/G2L into USBHS driver in Linux was much simpler
I agree that moving the RZ/G2L specifics in common.c to a platform dependent file (similar to Linux) would be the cleaner long-term architecture. However, doing so would require a much more invasive refactoring of the existing driver structure. I opted for these minimal changes to maintain stability for existing R-Car Gen3 users while enabling RZ/G2L support.
I will prepare a v2 series with the patches split as requested.
Kind regards,
Michele
On 4/16/26 4:14 PM, Michele Bisogno wrote:
Hello Michele,
Quoted 7 lines
Thank you for the review and the link to the Linux implementation.
Quoted 1 line
Regulator change should be a separate patch please.
I agree. I will split the regulator changes into a separate patch for the v2 series to keep the subsystem changes distinct.
Quoted 8 lines
Quoted 2 lines
drivers/usb/gadget/rcaris pulled from Linux1b4861e32e46... but none of these changes in this patch are even in Linux next.The RZ/G2L support in Linux relies heavily on DMA for data transfers. In U-Boot, we are primarily using the PIO path. The refactoring in
fifo.c(switching to ioread8/iowrite8) was found to be necessary to handle trailing bytes and alignment specifically for the RZ/G2L bus when DMA is not active.
If Linux also contains the PIO code, it would also benefit from matching improvement, right ?
Quoted 4 lines
Additionally, the
usb_del_gadget_udc()change incommon.caddresses a re-entrancy bug unique to U-Boot's gadget lifecycle. Without this, the UDC list becomes stale, causing the board to crash if a command like 'ums' is executed more than once.
This should be a separate bugfix patch.
Quoted 10 lines
Quoted 1 line
The addition of RZ/G2L into USBHS driver in Linux was much simpler
I agree that moving the RZ/G2L specifics in
common.cto a platform dependent file (similar to Linux) would be the cleaner long-term architecture. However, doing so would require a much more invasive refactoring of the existing driver structure. I opted for these minimal changes to maintain stability for existing R-Car Gen3 users while enabling RZ/G2L support.I will prepare a v2 series with the patches split as requested.
Ideally, let's try to keep the Linux and U-Boot driver code base synchronized as much as possible.
Thank you !
Hi Marek,
thank you for your comments and suggestions.
Quoted 2 lines
If Linux also contains the PIO code, it would also benefit from matching improvement, right ?
Yes, though I have found the fix is not strictly required. We can maintain synchronization between both repositories by utilizing the cfifo_byte_addr parameter for the RZ/G2L and similar devices, allowing us to leave the existing PIO code untouched.
Quoted 1 line
This should be a separate bugfix patch.
Agreed.
Quoted 2 lines
Ideally, let's try to keep the Linux and U-Boot driver code base synchronized as much as possible.
Understood. I am currently preparing a revised patch set that aligns with this approach to ensure consistency between the codebases.
Kind regards,
Michele
On 4/21/26 9:32 AM, Michele Bisogno wrote:
Hello Michele,
Quoted 12 lines
thank you for your comments and suggestions.
Quoted 2 lines
If Linux also contains the PIO code, it would also benefit from matching improvement, right ?
Yes, though I have found the fix is not strictly required. We can maintain synchronization between both repositories by utilizing the
cfifo_byte_addrparameter for the RZ/G2L and similar devices, allowing us to leave the existing PIO code untouched.Quoted 1 line
This should be a separate bugfix patch.
Quoted 7 lines
Agreed.
Quoted 2 lines
Ideally, let's try to keep the Linux and U-Boot driver code base synchronized as much as possible.
Understood. I am currently preparing a revised patch set that aligns with this approach to ensure consistency between the codebases.
Thank you !