Hi,

On 06/14/2015 01:46 PM, Ian Campbell wrote:

On Sun, 2015-06-14 at 12:40 +0200, Hans de Goede wrote:

...

Fully resetting the controller is a too big hammer, and the musb_core will then afterwards fail to communicate with any endpoints other then 0 as too much state was cleared.

Instead report vbus low for 200ms which will effectively end the current session without needing to do a full reset.

This fixes usb mass-storage devices no longer working after a "usb reset"

Signed-off-by: Hans de Goede hdegoede@redhat.com

Acked-by: Ian Campbell ijc@hellion.org.uk

One question (which turned into two) (both more for the musb maint than you):

...

+/* musb_core does not call enable / disable in a balanced manner <sigh> */ +static bool enabled = false;

Is this sufficient, or should we be reference counting? Or should the core be fixed?

The u-boot musb-new code is a copy of the kernel code, and both are of, ah, interesting quality. One problem is that the core calls musb_platform_disable as part of musb_init_controller and thus before musb_platform_enable is ever called.

Ideally the core would not do this, but in u-boot it is used on 4 different SoCs and in the kernel even more, and chances are that that call was added to put the hw in a clean state on some SoC...

Basically my approach for these kind of problems in the musb core so far has been to work around them were possible, so as to not cause any regressions on other SoCs.

Regards,

Hans

On Sun, 2015-06-14 at 12:40 +0200, Hans de Goede wrote:

Remove the unused sunxi_musb_exit method, there is no code in u-boot calling the exit method, and our implementation was broken as it did not disable the clocks and asserted reset.

Seems to me the call should either be called (and the hooks fixed) or removed from the core but so far as sunxi goes:

Signed-off-by: Hans de Goede hdegoede@redhat.com

Acked-by: Ian Campbell ijc@hellion.org.uk

Hi,

On 15-06-15 21:21, Paul Kocialkowski wrote:

Le dimanche 14 juin 2015 à 12:40 +0200, Hans de Goede a écrit :

...

Hi Ian, Paul,

Here is a patch to fix the problems where most usb devices will no longer work after a "usb reset ", when connected to the otg controller in host mode + a related cleanup patch.

This works fine for an USB storage device and an USB2 keyboard but does not work with an USB1 keyboard, with error:

sunxi# usb reset resetting USB... USB0: scanning bus 0 for devices... USB device descriptor short read (expected 8, got 0) No USB Device found

Hmm, did you test my sunxi-wip branch perhaps? This bug does exist there, but it is the result of me refactoring things so that the musb code can use the device-model when build in host mode, which will allow enabling both the otg port in host mode and regular usb hosts in a single build, which is esp. useful for boards which have the otg hooked up in host-only mode (e.g. connected to an usb-a receptacle, or usb <-> sata bridge).

I've just retested current u-boot/master with just this patch and the problem you are describing does not happen then.

Sorry about my sunxi-wip being broken atm I try to always keep it 100% functional, but well it is a wip tree after all.

Regards,

Hans

On Friday, June 19, 2015 at 03:07:04 PM, Hans de Goede wrote:

Hi,

On 19-06-15 14:35, Marek Vasut wrote:

...

On Sunday, June 14, 2015 at 12:40:11 PM, Hans de Goede wrote:

...

Hi Ian, Paul,

Here is a patch to fix the problems where most usb devices will no longer work after a "usb reset ", when connected to the otg controller in host mode + a related cleanup patch.

Ian I would like to send out a PR with these 2 as fixed for v2015.07, can you review them please? Note I've not tested this with the otg in gadget mode, but we do not have gadget mode enabled by default anywhere atm, so I still consider this suitable as a bugfix for v2015.07.

Paul, can you test these with gadget mode? Specifically if they help the problem you were seeing when switching roles?

Also this bit from the kernel code for the sunxi glue may be relevant

to your problems: if ((musb->int_usb & MUSB_INTR_RESET) && !is_host_active(musb)) {

             /* ep0 FADDR must be 0 when (re)entering peripheral
             mode */ musb_ep_select(musb->mregs, 0);
             musb_writeb(musb->mregs, MUSB_FADDR, 0);

     }

This is from the interrupt handler in the sunxi-musb glue in the kernel,

maybe we can do the same, and/or maybe we need to do: /* ep0 FADDR must be 0 when (re)entering peripheral mode */ musb_ep_select(musb->mregs, 0); musb_writeb(musb->mregs, MUSB_FADDR, 0);

From sunxi_musb_disable?

From my experience sofar we should avoid doing a full reset from musb_stop

/ sunxi_musb_disable as musb_init_controller never gets re-run, so doing a full reset leaves things in a bad state where only ep0 still seems to work, this may be what you were seeing before.

Hi,

do you want me to pick this via u-boot-usb/master ?

No, I've already send a pull-req for them via u-boot-sunxi/master and they are already merged :)

OK, roger.

Best regards, Marek Vasut

Hi,

On 26-06-15 12:20, Maxime Ripard wrote:

Hi,

On Sun, Jun 14, 2015 at 12:40:11PM +0200, Hans de Goede wrote:

...

Paul, can you test these with gadget mode? Specifically if they help the problem you were seeing when switching roles?

Also this bit from the kernel code for the sunxi glue may be relevant to your problems:

     if ((musb->int_usb & MUSB_INTR_RESET) && !is_host_active(musb)) {
             /* ep0 FADDR must be 0 when (re)entering peripheral mode */
             musb_ep_select(musb->mregs, 0);
             musb_writeb(musb->mregs, MUSB_FADDR, 0);
     }

This is from the interrupt handler in the sunxi-musb glue in the kernel, maybe we can do the same, and/or maybe we need to do:

/* ep0 FADDR must be 0 when (re)entering peripheral mode */ musb_ep_select(musb->mregs, 0); musb_writeb(musb->mregs, MUSB_FADDR, 0);

From sunxi_musb_disable?

So, I was affected by this issue on a SinA33, even when not switching roles (since the SinA33 is peripheral only). The issue I was seeing what that when I was two subsequent use of any gadget (fastboot, USB ethernet traffic), the second would silently fail on U-boot side with the host failing to enumerate the device.

Paul pointed me to that thread, and your suggestion on EP0 seems like a good lead, since the ugly http://pastebin.com/9Y6S9Hpw makes it work.

Ah, that is not that ugly actually, I can take that as is if you want me too just submit it to the u-boot list with me in the Cc and I'll pick it up and put it in u-boot-sunxi/next

Regards,

Hans

Hi,

On 26-06-15 12:20, Maxime Ripard wrote:

Hi,

On Sun, Jun 14, 2015 at 12:40:11PM +0200, Hans de Goede wrote:

...

Paul, can you test these with gadget mode? Specifically if they help the problem you were seeing when switching roles?

Also this bit from the kernel code for the sunxi glue may be relevant to your problems:

     if ((musb->int_usb & MUSB_INTR_RESET) && !is_host_active(musb)) {
             /* ep0 FADDR must be 0 when (re)entering peripheral mode */
             musb_ep_select(musb->mregs, 0);
             musb_writeb(musb->mregs, MUSB_FADDR, 0);
     }

This is from the interrupt handler in the sunxi-musb glue in the kernel, maybe we can do the same, and/or maybe we need to do:

/* ep0 FADDR must be 0 when (re)entering peripheral mode */ musb_ep_select(musb->mregs, 0); musb_writeb(musb->mregs, MUSB_FADDR, 0);

From sunxi_musb_disable?

So, I was affected by this issue on a SinA33, even when not switching roles (since the SinA33 is peripheral only).

Missed this the first reply, are you sure, that is somewhat weird. I mean it is possible if they did not hook up the id pin, or do not allow you to send out +5v, but it is weird. I guess they did route the real usb-host to an USB-A receptacle on that board, so it is not a big deal, right ?

Regards,

Hans