Dear Marek Vasut,
On Monday, November 5, 2012 11:54:12 PM, Marek Vasut wrote:
Dear Benoît Thébaudeau,
Hi Marek,
Thanks to Lucas' series coming with commits c7e3b2b and 676ae06, I'd like to use the multi-controller feature on MXC since most of these SoCs come with a USB IP supporting an OTG controller and multiple host-only controllers.
Currently the MXC code in ehci-mx{c|5|6}.c just ignores the index passed to ehci_hcd_init() and the like, and there are 3 port-specific configs (CONFIG_MXC_USB_PORT, CONFIG_MXC_USB_FLAGS and CONFIG_MXC_USB_PORTSC).
Not all USB ports from the USB IP will be available on each board for a given SoC, so we need a logical to physical USB port mapping.
I would suggest something like the following.
board.h: #define CONFIG_MXC_USB { \ { \ 0, \ MXC_EHCI_INTERNAL_PHY, \ MXC_EHCI_UTMI_16BIT | MXC_EHCI_MODE_UTMI \ }, { \ 1, \ MXC_EHCI_POWER_PINS_ENABLED | MXC_EHCI_PWR_PIN_ACTIVE_HIGH | \ MXC_EHCI_OC_PIN_ACTIVE_LOW, \ MXC_EHCI_MODE_ULPI \ }, \ }
ehci-fsl.h: struct mxc_ehci_cfg { int port; u32 flags; u32 portsc; };
ehci-mx{c|5|6}.c: static const struct mxc_ehci_cfg cfg[CONFIG_USB_MAX_CONTROLLER_COUNT] = CONFIG_MXC_USB;
Then, in ehci_hcd_init(), there would be the following replacements:
- CONFIG_MXC_USB_PORT -> cfg[index].port,
- CONFIG_MXC_USB_FLAGS -> cfg[index].flags,
- CONFIG_MXC_USB_PORTSC -> cfg[index].portsc.
What do you think?
What about passing port private / platform data instead of ID ?
The ID is already passed to ehci_hcd_init(), so we have to live with it if we don't want to change the newly introduced multi-controller infrastructure.
Or, perhaps this is what you meant, we could have some: int ehci_mxc_register(int index, const struct mxc_ehci_cfg *cfg); This function would simply fill an entry in the cfg array in ehci-mx{c|5|6}.c, this array becoming an array of pointers to struct mxc_ehci_cfg. This looks nicer, but it needs more code to do just the same thing as the CONFIG_MXC_USB would do, without adding any feature. The only benefit would be if index were actually the same as port here, but ehci_hcd_init() would still be called for all indexes, so it would have to fail e.g. if port 0 is unused but port 1 is used, which would probably generate some error noise for the user.
Or did you mean something else?
Best regards, Benoît