usb: ci_udc: remove controller.items array

There's no need to store an array of QTD pointers in the controller.
Since the calculation is so simple, just have ci_get_qtd() perform it
at run-time, rat

usb: ci_udc: remove controller.items array

There's no need to store an array of QTD pointers in the controller.
Since the calculation is so simple, just have ci_get_qtd() perform it
at run-time, rather than pre-calculating everything.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: fix items array size/stride calculation

2 QTDs are allocated for each EP. The current allocation scheme aligns
the first QTD in each pair, but simply adds the struct size to calculate
t

usb: ci_udc: fix items array size/stride calculation

2 QTDs are allocated for each EP. The current allocation scheme aligns
the first QTD in each pair, but simply adds the struct size to calculate
the second QTD's address. This will result in a non-cache-aligned
addresss IF the system's ARCH_DMA_MINALIGN is not 32 bytes (i.e. the
size of struct ept_queue_item).

Similarly, the original ilist_ent_sz calculation aligned the value to
ARCH_DMA_MINALIGN but didn't take the USB HW's 32-byte alignment
requirement into account. This doesn't cause a practical issue unless
ARCH_DMA_MINALIGN < 32 (which I suspect is quite unlikely), but we may
as well fix the code to be explicit, so it's obviously completely
correct.

The new value of ILIST_ENT_SZ takes all alignment requirements into
account, so we can simplify ci_{flush,invalidate}_qtd() by simply using
that macro rather than calling roundup().

Similarly, the calculation of controller.items[i] can be simplified,
since each QTD is evenly spaced at its individual alignment requirement,
rather than each pair being aligned, and entries within the pair being
spaced apart only by structure size.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: lift ilist size calculations to global scope

This will allow functions other than ci_udc_probe() to make use of the
constants in a future change.

This in turn requires converting the c

usb: ci_udc: lift ilist size calculations to global scope

This will allow functions other than ci_udc_probe() to make use of the
constants in a future change.

This in turn requires converting the const int variables to #defines,
since the initialization of one global const int can't depend on the
value of another const int; the compiler thinks it's non-constant if
that dependency exists.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: don't assume QTDs are adjacent when transmitting ZLPs

Fix ci_ep_submit_next_request()'s ZLP transmission code to explicitly
call ci_get_qtd() to find the address of the other QTD to use

usb: ci_udc: don't assume QTDs are adjacent when transmitting ZLPs

Fix ci_ep_submit_next_request()'s ZLP transmission code to explicitly
call ci_get_qtd() to find the address of the other QTD to use. This
will allow us to correctly align each QTD individually in the future,
which may involve leaving a gap between the QTDs.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: fix ci_flush_{qh,qtd} calls in ci_udc_probe()

ci_udc_probe() initializes a pair of QHs and QTDs for each EP. After
each pair has been initialized, the pair is cache-flushed. The
convers

usb: ci_udc: fix ci_flush_{qh,qtd} calls in ci_udc_probe()

ci_udc_probe() initializes a pair of QHs and QTDs for each EP. After
each pair has been initialized, the pair is cache-flushed. The
conversion from QH/QTD index [0..2*NUM_END_POINTS) to EP index
[0..NUM_ENDPOINTS] is incorrect; it simply subtracts 1 (which yields
the QH/QTD index of the first entry in the pair) rather than dividing
by two (which scales the range). Fix this.

On my system, this avoids cache debug prints due to requests to flush
unaligned ranges. This is caused because the flush calls happen before
the items[] array entries are initialized for all but EP0.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: fix typo in debug message

s/ot/to/

Signed-off-by: Stephen Warren <swarren@nvidia.com>

usb: ci_udc: fix interaction with CONFIG_USB_ETH_CDC

ci_udc.c's usb_gadget_unregister_driver() doesn't call driver->unbind()
unlike other USB gadget drivers. Fix it to do this.

Without this, when e

usb: ci_udc: fix interaction with CONFIG_USB_ETH_CDC

ci_udc.c's usb_gadget_unregister_driver() doesn't call driver->unbind()
unlike other USB gadget drivers. Fix it to do this.

Without this, when ether.c's CDC Ethernet device is torn down,
eth_unbind() is never called, so dev->gadget is never set to NULL.
For some reason, usb_eth_halt() is called both at the end of the first
use of the Ethernet device, and prior to any subsequent use. Since
dev->gadget is never cleared, all calls to usb_eth_halt() attempt to
stop, disconnect, and clean up the device, resulting in double cleanup,
which hangs U-Boot on my Tegra device at least.

ci_udc allocates its own singleton EP0 request object, and cleans it up
during usb_gadget_unregister_driver(). This appears necessary when using
the USB gadget framework in U-Boot, since that does not allocate/free
the EP0 request. However, the CDC Ethernet driver *does* allocate and
free its own EP0 requests. Consequently, we must protect
ci_ep_free_request() against double-freeing the request.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

Merge branch 'u-boot/master' into 'u-boot-arm/master'

Merge branch 'master' of git://git.denx.de/u-boot-usb

usb: ci_udc: terminate ep0 INs with a zlp when required

Sometimes, a zero-length packet is required at the end of an IN
transaction so that the host knows the device is done sending data.
Enhance ci

usb: ci_udc: terminate ep0 INs with a zlp when required

Sometimes, a zero-length packet is required at the end of an IN
transaction so that the host knows the device is done sending data.
Enhance ci_udc to send a zlp when necessary. See the comments for
more details.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: clean up all allocations in unregister

usb_gadget_unregister_driver() is called to tear down the USB device mode
stack. Fix the driver to stop the USB HW (which causes any attached host

usb: ci_udc: clean up all allocations in unregister

usb_gadget_unregister_driver() is called to tear down the USB device mode
stack. Fix the driver to stop the USB HW (which causes any attached host
to notice the disappearance of the device), and free all allocations
(which obviously prevents memory leaks).

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: fix probe error cleanup

If allocation of the ep0 req fails, clean up all the allocations that
were made in ci_udc_probe().

Signed-off-by: Stephen Warren <swarren@nvidia.com>

usb: ci_udc: fix freeing of ep0 req

ci_ep_alloc_request() avoids allocating multiple request objects for ep0
by keeping a record of the first req allocated for ep0, and always
returning that instead

usb: ci_udc: fix freeing of ep0 req

ci_ep_alloc_request() avoids allocating multiple request objects for ep0
by keeping a record of the first req allocated for ep0, and always
returning that instead of allocating a new req. However, if this req is
ever freed, the record of the previous allocation is not cleared, so
ci_ep_alloc_request() will keep returning this stale pointer. Fix
ci_ep_free_request() to clear the record of the previous allocation.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: call udc_disconnect() from ci_pullup()

ci_pullup()'s !is_on path contains a cut/paste copy of udc_disconnect().
Remove the duplication by simply calling udc_disconnect() instead.

Signe

usb: ci_udc: call udc_disconnect() from ci_pullup()

ci_pullup()'s !is_on path contains a cut/paste copy of udc_disconnect().
Remove the duplication by simply calling udc_disconnect() instead.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: complete ep0 direction handling

handle_setup() currently assumes that the response to a Setup transaction
will be an OUT transaction, and any subsequent packet (if any) will be an
IN tr

usb: ci_udc: complete ep0 direction handling

handle_setup() currently assumes that the response to a Setup transaction
will be an OUT transaction, and any subsequent packet (if any) will be an
IN transaction. This appears to be valid in many cases; both USB
enumeration and Mass Storage work OK with this restriction. However, DFU
uses ep0 to transfer data in both directions. This renders the assumption
invalid; when sending data from device to host, the Data Stage is an IN
transaction, and the Status Stage is an OUT transaction. Enhance
handle_setup() to deduce the correct direction for the USB transactions
based on Setup transaction data.

ep0's request object only needs to be automatically re-queued when the
Data Stage completes, in order to implement the Status Stage. Once the
Status Stage transaction is complete, there is no need to re-queue the
USB request, so don't do that.

Don't sent USB request completion callbacks for Status Stage transactions.
These were queued by ci_udc itself, and only serve to confuse the USB
function code. For example, f_dfu attempts to interpret the 0-length data
buffers for Status Stage transactions as DFU packets. These buffers
contain stale data from the previous transaction. This causes f_dfu to
complain about a sequence number mismatch.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: pre-allocate ep0 req

Allocate ep0's USB request object when the UDC driver is probed. This
solves a couple of issues in the current code:

a) A request object always exists for ep0. Pri

usb: ci_udc: pre-allocate ep0 req

Allocate ep0's USB request object when the UDC driver is probed. This
solves a couple of issues in the current code:

a) A request object always exists for ep0. Prior to this patch, if setup
transactions arrived in an unexpected order, handle_setup() would need
to reply to a setup transaction before any ep0 usb_req was created.

This issue was introduced in commit 2813006fecda "usb: ci_udc: allow
multiple buffer allocs per ep."

b) handle_ep_complete no longer /has/ to queue the ep0 request again
after every single request completion. This is currently required, since
handle_setup() assumes it can find some request object in ep0's request
queue. This patch doesn't actually stop handle_ep_complete() from always
requeueing the request, but the next patch will.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: use a single descriptor for ep0

ci_udc currently points ep->desc at separate descriptors for IN and OUT.
These descriptors only differ in the ep address IN/OUT field. Modify the
code to

usb: ci_udc: use a single descriptor for ep0

ci_udc currently points ep->desc at separate descriptors for IN and OUT.
These descriptors only differ in the ep address IN/OUT field. Modify the
code to use a single descriptor, and change that descriptor's ep address
to indicate IN/OUT as required. This removes some data duplication.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

usb: ci_udc: detect queued requests on ep0

The flipping of ep0 between IN and OUT relies on ci_ep_queue() consuming
the current IN/OUT setting immediately. If this is deferred to a later
point when

usb: ci_udc: detect queued requests on ep0

The flipping of ep0 between IN and OUT relies on ci_ep_queue() consuming
the current IN/OUT setting immediately. If this is deferred to a later
point when the req is pulled out of ci_req->queue, then the IN/OUT
setting may have been changed since the req was queued, and state will
get out of sync. This condition doesn't occur today, but could if bugs
were introduced later, and this error-check will save a lot of debugging
time.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

Merge branch 'pr-15052014' of git://git.denx.de/u-boot-usb

Merge branch 'master' of git://git.denx.de/u-boot-usb

Merge branch 'master' of git://git.denx.de/u-boot-arm

usb: ci_udc: parse QTD before over-writing it

ci_udc only allocates a single QTD structure per EP. All data needs to be
extracted from the DTD prior to calling ci_ep_submit_next_request(), since
tha

usb: ci_udc: parse QTD before over-writing it

ci_udc only allocates a single QTD structure per EP. All data needs to be
extracted from the DTD prior to calling ci_ep_submit_next_request(), since
that fills the QTD with next transaction's parameters. Fix
handle_ep_complete() to extract the transaction (remaining) length before
kicking off the next transaction.

In practice, this only causes writes to UMS devices to fail for me. I may
have tested the final versions of my previous ci_udc patch only with
reads. More recently, I had patches applied locally that allocated a QTD
per USB request rather than per USB EP, although since that doesn't give
any performance benefit, I'm dropping those.

Fixes: 2813006fecda ("usb: ci_udc: allow multiple buffer allocs per ep")
Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

Merge branch 'u-boot/master'

Conflicts:
drivers/net/Makefile

(trivial merge)

usb: ci_udc: allow multiple buffer allocs per ep

Modify ci_ep_alloc_request() to return a dynamically allocated request
object, rather than a singleton that's part of the endpoint. This
requires mov

usb: ci_udc: allow multiple buffer allocs per ep

Modify ci_ep_alloc_request() to return a dynamically allocated request
object, rather than a singleton that's part of the endpoint. This
requires moving various state from the endpoint structure to the request
structure, since we need one copy per request.

The "fast bounce buffer" b_fast is removed by this change rather than
moved to the request object. Instead, we enhance the bounce buffer logic
in ci_bounce()/ci_debounce() to keep the bounce buffer around between
request submissions. This avoids the need to allocate an arbitrarily-
sized bounce buffer up-front, yet avoids incurring the allocation
overhead each time a request is submitted.

A future enhancement would be to actually submit multiple requests to HW
at once. The Linux driver shows that this is possible. That might improve
throughput (depending on the USB protocol in use), since USB could be
performing a transfer to one HW buffer in parallel with whatever SW
actions U-Boot performs on another buffer. However, I have not made this
change as part of this patch, in order to keep SW changes related to
buffer management separate from any change in the way the HW is
programmed.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

show more ...

Merge branch 'master' of git://git.denx.de/u-boot-usb