Compiling with CONFIG_USB_XHCI_PCI and CONFIG_PCI=n results in

    usb/host/xhci-pci.c:48:(.text.xhci_pci_probe+0x44):
    undefined reference to `dm_pci_write_config32

Add the missing Kconfig dependency.

Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Mark Kettenis <kettenis@openbsd.org>
Reviewed-by: Marek Vasut <marex@denx.de>

Like Rockchip RK3328 and RK3568, the RK3588 also have a single node to
represent the glue and ctrl for USB 3.0.

Use rk_ops as driver data to select correct ctrl node for RK3588 DWC3.

Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Reviewed-by: Kever Yang <kever.yang@rock-chips.com>
Reviewed-by: Marek Vasut <marex@denx.de>

We need to get the DMA address before incrementing the pointer, as that
might move us onto another segment.

Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>

Now that we always check the return value, just return NULL on timeouts.
We can still log the error since this is a problem, but it's not reason
to panic.

Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>

This isn't going to work, don't pretend it will and then end up timing
out.

Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>

There is currently no codepath to recover from this case. In principle
we could require that the upper layer do this explicitly, but let's just
do it in xHCI when the next bulk transfer is started, since that
reasonably implies whatever caused the problem has been dealt with.

Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>

There is a race where an endpoint may halt by itself while we are trying
to halt it, which results in a context state error. See xHCI 4.6.9 which
mentions this case.

This also avoids BUGging when we attempt to stop an endpoint which was
already stopped to begin with, which is probably a bug elsewhere but
not a good reason to crash.

Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>

If the xHC has a problem with our STOP ENDPOINT command, it is likely to
return a completion directly instead of first a transfer event for the
in-progress transfer. Handle that more gracefully.

We still BUG() on the error code, but at least we don't end up timing
out on the event and ending up with unexpected event errors.

Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>

xhci_wait_for_event returns NULL on timeout, so the caller always has to
check for that. This addresses immediate explosions in this part
of the code when timeouts happen, but not the root cause for the
timeout.

Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Marek Vasut <marex@denx.de>

Add missing newline in pr_err.

Signed-off-by: Simon Holesch <simon@holesch.de>
Reviewed-by: Marek Vasut <marex@denx.de>
Reviewed-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>
Link: https://lore.kernel.org/r/20231120002024.32865-2-simon@holesch.de
Signed-off-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>