| Age | Commit message (Collapse) | Author |
|---|
|
The current calculation takes dep->trb_dequeue - dep->trb_enqueue to
find the TRB space left.
If you enqueue 1, that results in:
(u8) 0 - (u8) 1 = 0xff = 255 TRBs left.
This is correct if DWC3_TRB_NUM == 256.
If DWC3_TRB_NUM is less than 256 (but still a power of 2) you need to
mod the result by DWC3_TRB_NUM.
For example the same calculation with DWC3_TRB_NUM = 8, results in:
255 % 6 = 7 TRBs left.
Signed-off-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
If trbs_left == 0, we don't have any space left in the TRB ring so don't
prepare anything.
Signed-off-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Clears out all the TRBs in the ring to clean up any stale data that
might be in them from the previous time the endpoint was enabled.
Also removed the existing clear of the LINK trb since the entire ring is
cleard just before.
Signed-off-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Make the skipping of the link TRBS built-in to the increment operation.
This simplifies the code wherever we increment the trb index and ensures
that we never end up pointing to a link trb.
Signed-off-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Sparse complains even though it looks ok. Probably it cannot detect that
the wValue, wIndex, and wLength are declared __le16 due to the macro
magic.
Redeclare them as CPU endianness and make the conversion on assignment.
Signed-off-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Cleans up the sparse warning:
warning: dubious: x | !y
Since we do want a bitwise OR here, don't use a logical (true/false)
value. Probably is not a real issue but it cleans up the warning.
Signed-off-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
u2sel and u2pel should be __le16. Doesn't fix any issue.
Found with sparse.
Signed-off-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
The wIndex passed in here is CPU endianness, but the function expects
little endian.
Found with sparse.
Signed-off-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Just like we did for endpoint commands, let's have a
single trace output for the command and its
status. This will improve trace readability
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
We want commands to finish ASAP, so let's remove
that udelay() call.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Just like we did for endpoint commands, let's use a
single return point for generic commands as
well. This aids readability.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Instead of printing command's status with a separate
trace printout, let's print it within a single call.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
instead of having infinite loop and always checking
timeout value as a break condition, we can just
decrement timeout inside while's condition.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
We don't need the extra %s when command fails.
Let's remove it
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
This will make it more human-friendly to read trace
output from dwc3.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
I really thought this would be useful, but as it
turns out, it creates more problems than fixes. The
amount of times we had to fix this because some
other commit shuffled things around and ended up
regressing this tiny little string manupulation...
Might as well remove it, since it has a negligible
added benefit.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Improve trb tracing by showing trb flags, interrupts
trb type.
trb flags:
- h - hardware owner of descriptor
- l - last TRB
- c - chain buffers
- s - continue on short packet
interrupt flags:
- s - interrupt on short packet
- c - interrupt on complete
Capital letter means that bit is set, while
lowercase letter means bit is cleared.
Signed-off-by: Janusz Dziedzic <januszx.dziedzic@linux.intel.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
This will allow us to process several endpoints at a
time by making sure that we lock only shared
resources.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Allow for dwc3-pci to reach D3 and enable pm_runtime
by providing dummy PM hooks. Without them, PCI
subsystem won't put device to D3.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
this patch implements the most basic pm_runtime
support for dwc3. Whenever USB cable is dettached,
then we will allow core to runtime_suspend.
Runtime suspending will involve completely tearing
down event buffers and require a full soft-reset of
the IP.
Note that a further optimization could be
implemented once we decide to support hibernation,
which is to allow runtime_suspend with cable
connected when bus is in U3. That's subject to a
separate patch, however.
Tested-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
when we call dwc3_gadget_giveback(), we end up
releasing our controller's lock. Another thread
could get scheduled and disable the endpoint,
subsequently setting dep->endpoint.desc to NULL.
In that case, we would end up dereferencing a NULL
pointer which would result in a Kernel Oops. Let's
avoid the problem by simply returning early if we
have a NULL descriptor.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
commit f3af36511e60 ("usb: dwc3: gadget: always
enable IOC on bulk/interrupt transfers") ended up
regressing Isochronous endpoints by clearing
DWC3_EP_BUSY flag too early, which resulted in
choppy audio playback over USB.
Fix that by partially reverting original commit and
making sure that we check for isochronous endpoints.
Fixes: f3af36511e60 ("usb: dwc3: gadget: always enable IOC
on bulk/interrupt transfers")
Cc: <stable@vger.kernel.org>
Signed-off-by: Konrad Leszczynski <konrad.leszczynski@intel.com>
Signed-off-by: Rafal Redzimski <rafal.f.redzimski@intel.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
As a micro-power optimization, let's only resume the
USB2 PHY if we're working on <=HIGHSPEED. If we're
gonna work on SUPERSPEED or SUPERSPEED+, there's no
point in resuming the USB2 PHY.
Fixes: 2b0f11df84bb ("usb: dwc3: gadget: clear SUSPHY bit before ep cmds")
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
by holding gadget's IRQ number in dwc->irq_gadget,
it'll be simpler to free_irq() and disable the IRQ
in case an IRQ fires while we are runtime suspended.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
now that we have re-factored dwc3_core_init() and
dwc3_core_exit() we can use them for suspend/resume
operations.
This will help us avoid some common mistakes when
patching code when we have duplicated pieces of code
doing the same thing.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
The idea of this patch is for dwc3_core_init() to
abstract all the details about how to initialize
dwc3 and dwc3_core_exit() to do the same for
teardown.
With this, we can simplify suspend/resume operations
by a large margin and always know that we're going
to start dwc3 from a known starting point.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
this patch is in preparation for some further
re-factoring in dwc3 initialization. No functional
changes.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
By adding a pointer to endpoint registers' base
address, we can avoid using our controller-wide
struct dwc3 pointer for everything. At some point
this will allow us to have per-endpoint locks which
will, in turn, let us queue requests to separate
endpoints in parallel.
Because of this change our debugfs interface and io
accessors need to be changed accordingly.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
In all call sites of dwc3_send_gadget_ep_cmd() we
already had a valid dep pointer, so instead of
passing dwc and dep->number, which would be used to
fetch the same pointer we already had, just pass dep
directly.
In other words, we're changing:
struct dwc3_ep *dep = dwc[dep->number];
to just passing struct dwc3_ep *dep as argument.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Instead of using burst size to configure NUMP, we
should be using RxFIFO Size instead. DWC3 is smart
enough to know that it shouldn't burst in case burst
size is 0.
Reported-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
To aid code readability, we're gonna split
__dwc3_gadget_kick_transfer() into its constituent
parts: scatter gather and linear buffers.
That way, it's easier to follow the code and focus
debug effort when one or the other fails.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Instead of returning -EINVAL when someone calls
__dwc3_gadget_wakeup() in speeds > highspeed, let's
return 0. There are no problems for the driver for
calling it in superspeed as we cleanly just return.
This avoids an annoying WARN_ONCE() always
triggering during superspeed enumeration with LPM
enabled.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
When we send an endpoint command, we want that to
complete as soon as possible, so let's remove the
unnecessary udelay(1) call.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
sg_is_last() and list_is_last() will encode the
required information for the driver to make
decisions WRT CHN and LST bits.
While at that, also replace '1' with 'true' for
consistency.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
as it turns out, we don't need the extra 'start_new'
argument as that can be inferred from DWC3_EP_BUSY
flag.
Because of that, we can simplify
__dwc3_gadget_kick_transfer() by quite a bit, even
allowing us to prepare more TRBs unconditionally.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
If we're updating transfers, we can also prepare as
many TRBs as we can fit in the ring. Let's start
doing that.
This patch 'solves' a limitation of how many TRBs we
can prepare when we're getting close the end of the
ring. Instead driver to prepare only up to end of
the ring, we check if we have space to wrap around
the ring properly.
Note that this only happens when our enqueue and
dequeue pointers are equal (which is the case for
bulk endpoints after an XferComplete event).
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
that macro is unnecessary and just adds pointless
obfuscation. Let's remove it.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
Instead of trying hard to stay connected to the
host, it's best (and far easier) to disconnect from
the host already.
Anything relying on KEEP_CONNECT will just have that
ignored, but we don't have proper hibernation
implementation yet, so there are no regressions.
In any case, hibernation is only useful for runtime
PM, not system sleep.
While at that, also remove dwc3.dcfg which has been
rendered unnecessary.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
we will be re-using it for suspend/resume, so
instead of duplicating code, let's just re-factor
the functions so they can be re-used.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
The PLX USB2380 is a PCIe version of the NET2280 and behaves more like the
USB338x but without the USB3.0 superspeed support.
This was tested with g_ether, g_serial, g_mass_storage on a Gateworks
Ventana GW2383.
Cc: Justin DeFields <justindefields@gmail.com>
Signed-off-by: Tim Harvey <tharvey@gateworks.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
With a default size of 16kiB and with maximum of 32
buffers, we can transfer up to 512kiB, however Linux
can transfer up to 1MiB in a single mass storage
block transfer to USB3 storage devices.
Because of this, 1MiB block transfers end up being
slower than 512kiB block transfers. Let's increase
maximum number of storage buffers to a ridiculous
amount (256) so that anybody wanting to test maximum
achievable throughput can do so.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
valid range for storage buffers is encoded in
Kconfig already. Instead of checking again, let's
drop fsg_num_buffers_validate() altogether.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>
|
|
As per commit:
b7fa30c9cc48 ("sched/fair: Fix post_init_entity_util_avg() serialization")
> the code generated from update_cfs_rq_load_avg():
>
> if (atomic_long_read(&cfs_rq->removed_load_avg)) {
> s64 r = atomic_long_xchg(&cfs_rq->removed_load_avg, 0);
> sa->load_avg = max_t(long, sa->load_avg - r, 0);
> sa->load_sum = max_t(s64, sa->load_sum - r * LOAD_AVG_MAX, 0);
> removed_load = 1;
> }
>
> turns into:
>
> ffffffff81087064: 49 8b 85 98 00 00 00 mov 0x98(%r13),%rax
> ffffffff8108706b: 48 85 c0 test %rax,%rax
> ffffffff8108706e: 74 40 je ffffffff810870b0 <update_blocked_averages+0xc0>
> ffffffff81087070: 4c 89 f8 mov %r15,%rax
> ffffffff81087073: 49 87 85 98 00 00 00 xchg %rax,0x98(%r13)
> ffffffff8108707a: 49 29 45 70 sub %rax,0x70(%r13)
> ffffffff8108707e: 4c 89 f9 mov %r15,%rcx
> ffffffff81087081: bb 01 00 00 00 mov $0x1,%ebx
> ffffffff81087086: 49 83 7d 70 00 cmpq $0x0,0x70(%r13)
> ffffffff8108708b: 49 0f 49 4d 70 cmovns 0x70(%r13),%rcx
>
> Which you'll note ends up with sa->load_avg -= r in memory at
> ffffffff8108707a.
So I _should_ have looked at other unserialized users of ->load_avg,
but alas. Luckily nikbor reported a similar /0 from task_h_load() which
instantly triggered recollection of this here problem.
Aside from the intermediate value hitting memory and causing problems,
there's another problem: the underflow detection relies on the signed
bit. This reduces the effective width of the variables, IOW its
effectively the same as having these variables be of signed type.
This patch changes to a different means of unsigned underflow
detection to not rely on the signed bit. This allows the variables to
use the 'full' unsigned range. And it does so with explicit LOAD -
STORE to ensure any intermediate value will never be visible in
memory, allowing these unserialized loads.
Note: GCC generates crap code for this, might warrant a look later.
Note2: I say 'full' above, if we end up at U*_MAX we'll still explode;
maybe we should do clamping on add too.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yuyang Du <yuyang.du@intel.com>
Cc: bsegall@google.com
Cc: kernel@kyup.com
Cc: morten.rasmussen@arm.com
Cc: pjt@google.com
Cc: steve.muckle@linaro.org
Fixes: 9d89c257dfb9 ("sched/fair: Rewrite runnable load and utilization average tracking")
Link: http://lkml.kernel.org/r/20160617091948.GJ30927@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
I misread the inline asm. It uses a rare construct to provide an input
to a previously declared output to do the atomic_read().
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andreas Schwab <schwab@linux-m68k.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Cc: linux-m68k@lists.linux-m68k.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
Resolve conflict between commits:
fbffe892e525 ("locking/atomic, arch/arc: Implement atomic_fetch_{add,sub,and,andnot,or,xor}()")
and:
ed6aefed726a ("Revert "ARCv2: spinlock/rwlock/atomics: Delayed retry of failed SCOND with exponential backoff"")
Reported-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Nigel Topham <ntopham@synopsys.com>
Cc: Noam Camus <noamc@ezchip.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vineet Gupta <vgupta@synopsys.com>
Cc: linux-kernel@vger.kernel.org
Cc: linux-snps-arc@lists.infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
This patchs adds basic support for the bytewerk.org candleLight interface,
a open hardware (CERN OHL) USB CAN adapter.
Signed-off-by: Hubert Denkmair <hubert@denkmair.de>
Signed-off-by: Maximilian Schneider <max@schneidersoft.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
At high bus load it could happen that "at91_poll()" enters with all RX
message boxes filled up. If then at the end the "quota" is exceeded as
well, "rx_next" will not be reset to the first RX mailbox and hence the
interrupts remain disabled.
Signed-off-by: Wolfgang Grandegger <wg@grandegger.com>
Tested-by: Amr Bekhit <amrbekhit@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
access
When testing CAN write floods on Altera's CycloneV, the first 2 bytes
are sometimes 0x00, 0x00 or corrupted instead of the values sent. Also
observed bytes 4 & 5 were corrupted in some cases.
The D_CAN Data registers are 32 bits and changing from 16 bit writes to
32 bit writes fixes the problem.
Testing performed on Altera CycloneV (D_CAN). Requesting tests on other
C_CAN & D_CAN platforms.
Reported-by: Richard Andrysek <richard.andrysek@gomtec.de>
Signed-off-by: Thor Thayer <tthayer@opensource.altera.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
|
|
|
|
Yuval Mintz says:
====================
qed*: Fixes series
This series contains several small fixes to driver behavior
[4th patch is the only one containing a 'fatal' fix, but the error
is only theoretical for qede; if would require another protocol
driver yet unsubmitted to reach it].
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
