Mandatory on machines with 4 G memory, IOMMU enabled or DMA debugging
active. And it also simplifies the driver code.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Introduce the URB flag URB_ALLOC_TRANSFER_BUF. When set, a transfer
buffer of max_packet_size is allocated on URB registration and also
premapped for DMA usage. The buffer is freed again on unregistration.
This avoids having to perform the potentially RT-incompatible mapping
operations from a real-time task or interrupt context.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Add the proper GFP flag for this driver memory.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Avoid duplicating cleanup code via goto.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

We already resolved the generic URB pointer in those functions. Use it.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

The DMA API changed since the original code was written. Nowadays you
need to call an error checking functions to test if a DMA mapping was
successful.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

The setup packet is only transferred from the host to the device. So
also unmap it as PCI_DMA_TODEVICE. The kernel will complain otherwise.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

The kernel dislikes us for using stack memory as DMA buffer. Use a
kmalloc'ed buffer instead. We can also avoid the string buffer by using
the device's config copies while retrieving the strings.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Too much code duplications which makes it hard to add resources that
required cleanup.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Some gcc versions complain about uninitialized p_td, though we don't use
it in that case. Silence the warning nevertheless.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Use kzalloc instead of kmalloc + zero-initialization.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Already declared in rt_usb.h.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Make internal functions static, move prototypes to headers, use kernel
coding style. A first step. No functional changes.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This tests the rt_cdc_acm driver against the USB orientation sensors
3DM-GX3-25 and 3DM-GX3-45 by MicroStrain. It configures the sensors to
deliver a continuous data stream at 1 kHz (GX3-25) and 100 Hz (GX3-45),
then receives the stream and displays the data as well as the timing.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Replace the hard-coded Microstrain endpoint numbers with values obtained
from the device configuration.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Add support for dropping all queued data from the input queue.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This allows to set/retrieve the read and write timeout values.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

We only use a single write URB with up to 512 bytes so far. If more data
has to be written or a second requests is issued before the first
completed, the writer has to wait. This can easily be extended later on
by using a pool of write URBs.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This is a basic version of a CDC ACM driver, only supporting read access
so far and hard-coding the input endpoint.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Nothing good seems to come from it, so leave it off by default.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Mixed implicit and normal rule - we only need the latter.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Just wait for the frame number in the corresponding UHCI register to
change once. As we are polling anyway, there is no need for fiddling
with IOC of the termination TD.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Implement the necessary bits to support bulk ring buffer URBs. The
section size equals the maximum transfer packet size, thus the URB's
buffer is split into transfer_buffer_length / max_packet_size pieces.
Only callback-based handling is supported so far.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Ring buffer URBs are input URBs that are continuously queued, reading
input in a ring-like fashion into the associated buffer. The buffer may
be filled non-contiguously (depending on the transaction size).
Therefore, the concept of URB section is introduced. After a ring buffer
URB (partially) completed, rt_usb_next_urb_section will return the next
filled section. Furthermore, rt_usb_next_urb_section releases the
section it returned in the previous invocation, if any.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Wait at least two frame cycles after removing an active URB from the
schedule. This ensures we do not release resources before the controller
stopped using them.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

usb_device::lock is a regular Linux spin lock, thus we need to use Linux
lock/unlock services. Moreover, there is no need for irq protection
here.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

A TD is either registered with the active or the free list, so there is
no need for two list links. At this change, use advance list
manipulation functions to move an entry between two lists or fold on
list into the other. Makes things simpler.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

To properly handle short transactions, we have to walk the full TD list,
stop on errors or on the last inactive TD and then update the toggle
state.

Also, the SPD bit was set incorrectly. We want to detect short
transactions unconditionally but only report them as error if
URB_SHORT_NOT_OK. This aligns us with the Linux UHCI driver.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

URBs that don't have the state PURB_IN_PROGRESS are not registered with
the host controller, thus cannot cause any errors. Skip them while
iterating the list of registered URBs on interrupts.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

There is no point in letting the final TD in a frame point to itself,
creating a strange loop. For that we have bandwidth reclamation mode.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

This avoids potentially corrupting the list due to concurrent schedule/
unschedule operations.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

We can lose events if we clear interrupt reasons after walking the list
of URBs: If we missed a completed URB during the walk, the late clearing
will prevent that the controller reasserts the interrupt.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>

rt_schedule/unschedule_ctrl_bulk_urb may race with each other changing
the queue heads. Protect this critical region with a per-controller spin
lock.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>