On 04/01/2019 14:54, Jon Nettleton wrote:
On Fri, Jan 4, 2019 at 12:22 PM Jon Nettleton jon@solid-run.com wrote:
On Fri, Jan 4, 2019 at 11:25 AM Jean-Jacques Hiblot jjhiblot@ti.com wrote:
On 04/01/2019 11:19, Lukasz Majewski wrote:
Hi Jean-Jacques,
On 04/01/2019 08:05, Jon Nettleton wrote:
On Fri, Dec 7, 2018 at 8:46 AM Jean-Jacques Hiblot jjhiblot@ti.com wrote: > On 06/12/2018 21:56, Loic Devulder wrote: >> Hi, >> >> I re-tested this series on Khadas VIM1 and it still fix the USB >> issue I had with 'usb reset' (I already tested v2 patches). >> >> I just have a message and I'm not sure that I had it last time: >> "Error disabling PHY supply >> Can't shutdown USB PHY1 for dwc3@c9000000" >> >> But USB stack looks ok, I can plug/unplug USB key and do the >> reset, all works as expected. >> >> Since my last test lot of changes has been done in Amlogic SoCs, >> maybe it's because of this? >> >> So if you want you can add my tested-by flag. >> Tested-by: Loic Devulder ldevulder@suse.de > Thank you for testing >> On 11/29/18 10:52 AM, Jean-Jacques Hiblot wrote: >>> This series aims at bringing improvements to the dwc3_generic >>> driver so that it can be used by most of the platforms using the >>> dwc3 controller. >>> >>> I tested this on with DRA7 and AM57x platforms for both >>> Peripheral and Host operations. The code to enable DM USB host & >>> dev support for those platforms will be submitted in a separate >>> series. >>> >>> Michal Simek has tested this series: >>> " I have tested it on zcu100 with usb stick, usb to ethernet >>> converter and also dfu. >>> Tested-by: Michal Simek michal.simek@xilinx.com" >>> >>> Enhancements: >>> - use separate Kconfig option for DM USB Periphal and DM USB >>> Host. This allow platforms to keep their non-DM USB peripheral >>> code and use the DM USB host. >>> - fixes the bind/probe confusion in dwc3_generic. The probe is >>> done when the USB device is first needed. >>> - handles PHYs when in the peripheral mode. The code to handle >>> the PHYs is shared with the host side >>> - handles clock and reset >>> - bind host controller to the more generic driver 'xhci-dwc3' >>> >>> >>> Changes in v4: >>> - rebased on latest U-Boot >>> - renamed DM_USB_DEV as DM_USB_GADGET >>> - Add a new commit to create a new UCLASS for USB gadget drivers >>> >>> Changes in v3: >>> - fixes bug dwc3_setup_phy(): the phy arrays wasn't returned. >>> This was visible only when the device is removed. >>> - Stub the DWC3 PHY operations if CONFIG_IS_ENABLED(PHY) is >>> false. This fixes all build issues but one (evb-rk3328). >>> - Fix build issue with evb-rk3328 by enabling CONFIG_USB_DWC3. >>> This has little impact on the footprint and should not break the >>> runtime as the xhci-rockchip driver has its own probe function. >>> Nevertheless this was !!! NOT TESTED !!! by lack of hw >>> >>> Changes in v2: >>> - Updated commit log >>> - Fixed typo in thordown.c >>> - select DM_USB_DEV by default for zynqmp platforms >>> >>> Jean-Jacques Hiblot (10): >>> usb: gadget: Do not call board_usb_xxx() directly in USB >>> gadget drivers >>> usb: introduce a separate config option for DM USB device >>> usb: udc: implement DM versions of >>> usb_gadget_initialize()/_release()/_handle_interrupt() >>> dwc3_generic: do not probe the USB device driver when it's >>> bound dwc3: move phy operation to core.c >>> dm: usb: create a new UCLASS ID for USB gadget devices >>> configs: evb-rk3328: Enable CONFIG_USB_DWC3 >>> dwc3-generic: Handle the PHYs, the clocks and the reset lines >>> dwc3-generic: Add select_dr_mode operation >>> usb: dwc3: Fix a compilation error with the edison defconfig >>> >>> arch/arm/Kconfig | 2 + >>> board/sunxi/board.c | 2 +- >>> cmd/fastboot.c | 4 +- >>> cmd/rockusb.c | 4 +- >>> cmd/thordown.c | 4 +- >>> cmd/usb_gadget_sdp.c | 4 +- >>> cmd/usb_mass_storage.c | 4 +- >>> common/dfu.c | 6 +- >>> configs/evb-rk3328_defconfig | 1 + >>> drivers/usb/Kconfig | 14 +++ >>> drivers/usb/dwc3/Kconfig | 7 +- >>> drivers/usb/dwc3/core.c | 89 ++++++++++++++- >>> drivers/usb/dwc3/dwc3-generic.c | 208 >>> ++++++++++++++++++++++++++++-------- >>> drivers/usb/dwc3/ep0.c | 2 +- >>> drivers/usb/gadget/ether.c | 40 ++----- >>> drivers/usb/gadget/udc/Makefile | 4 + >>> drivers/usb/gadget/udc/udc-core.c | 3 +- >>> drivers/usb/gadget/udc/udc-uclass.c | 58 ++++++++++ >>> drivers/usb/host/xhci-dwc3.c | 95 ++-------------- >>> drivers/usb/musb-new/omap2430.c | 2 +- >>> drivers/usb/musb-new/sunxi.c | 2 +- >>> include/dm/uclass-id.h | 1 + >>> include/dwc3-uboot.h | 19 ++++ >>> include/linux/usb/gadget.h | 18 ++++ 24 files changed, >>> 401 insertions(+), 192 deletions(-) create mode 100644 >>> drivers/usb/gadget/udc/udc-uclass.c > _______________________________________________ > U-Boot mailing list > U-Boot@lists.denx.de > https://lists.denx.de/listinfo/u-boot First off thanks for the work. I have imported it into my u-boot tree for the iMX8M and now have both usb host and gadget mode working simultaneously. There is one catch. Currently, the iMX8M codebase does not support clock and power device initialization from DM. I have patched the driver to punt back to the board to do these initializations in the following patch. Yes I know not ideal, but functional. I am wondering if we should support architecture specific implementation quirks for boards that don't fully support DM initialization yet.
That is not fitting the DM model. I had the same issue and finally enabled the clocks in the board_late_init(): board/ti/dra7xx/evm.c:662:int board_late_init(void)
I am open to writing something more complete ifwe decide it is something worth pursuing, otherwise I will carry this patch in my tree for time being.
The true solution would be to implement the support for clk.
Many other people on the ML (including me) see the urgent need to implement the DM_CLK for e.g. imx6/5/7/8, etc.
I think that it would be best to add it clock by clock. The stubs in board code/files is a dead end, IMHO.
I agree 100%.
We had a similar discussion for the TI platforms, but due to the specific design of the TI power/clk in Linux (hwmod), we chose to keep the clocks handling in the board file at the moment.
-Jon
diff --git a/drivers/usb/dwc3/dwc3-generic.c b/drivers/usb/dwc3/dwc3-generic.c index 6d9be9b245..90d4ff2ea7 100644 --- a/drivers/usb/dwc3/dwc3-generic.c +++ b/drivers/usb/dwc3/dwc3-generic.c @@ -119,6 +119,15 @@ struct dwc3_glue_ops { enum usb_dr_mode mode); };
+__weak int board_usb_of_init(struct udevice *dev) +{
return 0;+}
+__weak void board_usb_of_cleanup(struct udevice *dev) +{ +}
- static int dwc3_glue_bind(struct udevice *parent) { const void *fdt = gd->fdt_blob;
@@ -242,6 +251,8 @@ static int dwc3_glue_probe(struct udevice *dev) index++; }
ret = board_usb_of_init(dev);return 0;@@ -249,6 +260,8 @@ static int dwc3_glue_remove(struct udevice *dev) { struct dwc3_glue_data *glue = dev_get_platdata(dev);
board_usb_of_cleanup(dev);reset_release_bulk(&glue->resets); clk_release_bulk(&glue->clks);@@ -257,6 +270,7 @@ static int dwc3_glue_remove(struct udevice *dev) }
Best regards,
Lukasz Majewski
--
DENX Software Engineering GmbH, Managing Director: Wolfgang Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: (+49)-8142-66989-59 Fax: (+49)-8142-66989-80 Email: lukma@denx.de
Obviously the proper solution is implementing a proper DM clk driver. I generally refrain from just enabling a bunch of clocks and code in our board file, since our devices are SOM + carrier, and many customers design their own boards and then continue to use our bootloader. I am fine carrying the patch until we have the time to build the proper infrastructure in mainline.
Quick followup.
Has anyone tried to build this into SPL? I am getting lots of undefined symbols. undefined reference to `usb_get_dr_mode' , undefined reference to `device_bind_driver_to_node' etc.
I've used it in SPL on am33xx based platforms. You may have to enable a couple of options, including SPL_DM.
-Jon