This patch fixes a bug in the XHCI stack that occurs when a multi-TRB TD
times out before the last TRB is processed. The driver will correctly
issue a Stop Endpoint command in that case, but the xHC will still
preserve the transfer state and just pick up right after that on the
next doorbell ring. It will then process the leftover TRBs from the old
TD the next time a transfer is issued. (cf. XHCI 4.6.9)
We fix this by changing the existing xhci_reset_endpoint() calls in
transfer functions to not only trigger on Halted (2) and Error (4), but
also on Stopped (3). That function will not actually issue a Reset
Endpoint command in this case, but it will nuke the whole transfer ring
and issue a Set TR Dequeue Pointer command, which is sufficient (though
slightly overkill) to solve our problem.
Change-Id: I3abbe30ff9d4911a8af1f792324e018d427019e8
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/170833
Reviewed-by: Ronald Minnich <rminnich@chromium.org>
Reviewed-by: Kees Cook <keescook@chromium.org>
(cherry picked from commit f12424af0e29ac12963e8e5a7970fadcc0bb6cee)
Signed-off-by: Isaac Christensen <isaac.christensen@se-eng.com>
Reviewed-on: http://review.coreboot.org/6787
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Tested-by: build bot (Jenkins)
This patch updates the libpayload XHCI stack to run on ARM CPUs (tested
with the DWC3 controller on an Exynos5420). Firstly, it adds support for
64-byte Slot/Endpoint Context sizes. Since the existing context handling
code represented the whole device context as a C struct (whose size has
to be known at compile time), it was necessary to refactor the input and
device context structures to consist of pointers to the actual contexts
instead.
Secondly, it moves all data structures that the xHC accesses through DMA
to cache-coherent memory. With a similar rationale as in the ARM patches
for EHCI, using explicit cache maintenance functions to correctly handle
the actual transfer buffers in all cases is presumably impossible.
Instead this patch also chooses to create a DMA bounce buffer in the
XHCI stack where transfer buffers which are not already cache-coherent
will be copied to/from.
Change-Id: I14e82fffb43b4d52d687b65415f2e33920e088de
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/169453
Reviewed-by: Stefan Reinauer <reinauer@google.com>
(cherry picked from commit 1fa9964063cce6cbd87ba68334806dde8aa2354c)
Signed-off-by: Isaac Christensen <isaac.christensen@se-eng.com>
Reviewed-on: http://review.coreboot.org/6643
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
The USB bulk and control transfer functions in libpayload currently
always return 0 for success and 1 for all errors. This is sufficient for
current use cases (essentially just mass storage), but other classes
(like certain Ethernet adapters) need to be able to tell if a transfer
reached the intended amount of bytes, or if it fell short.
This patch slightly changes that USB API to return -1 on errors, and the
amount of transferred bytes on successes. All drivers in the current
libpayload mainline are modified to conform to the new error detection
model. Any third party users of this API will need to adapt their
if (...<controller>->bulk/control(...)) checks to
if (...<controller>->bulk/control(...) < 0) as well.
The host controller drivers for OHCI and EHCI correctly implement the
new behavior. UHCI and the XHCI stub just comply with the new API by
returning 0 or -1, but do not actually count the returned bytes.
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://gerrit.chromium.org/gerrit/48308
Reviewed-by: Gabe Black <gabeblack@chromium.org>
Reviewed-by: Stefan Reinauer <reinauer@google.com>
Tested-by: Gabe Black <gabeblack@chromium.org>
Commit-Queue: Gabe Black <gabeblack@chromium.org>
Updated the patch to support XHCI as well.
Change-Id: Ic2ea2810c5edb992cbe185bc9711d2f8f557cae6
(cherry picked from commit e39e2d84762a3804653d950a228ed2269c651458)
Signed-off-by: Isaac Christensen <isaac.christensen@se-eng.com>
Reviewed-on: http://review.coreboot.org/6390
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-by: Patrick Georgi <patrick@georgi-clan.de>
The current XHCI code only sets IOC on the last TRB of a TD, and
doesn't set ISP anywhere. On my Synopsys DesignWare3 controller, this
won't generate an event at all when we have a short transfer that is not
on the last TRB of a TD, resulting in event ring desync and everyone
having a bad time. However, just setting ISP on other TRBs doesn't
really make for a nice solution: we then need to do ugly special casing
to fish out the spurious second transfer event you get for short
packets, and we still need a way to figure out how many bytes were
transferred. Since the Short Packet transfer event only reports
untransferred bytes for the current TRB, we would have to manually walk
the rest of the unprocessed TRB chain and add up the bytes. Check out
U-Boot and the Linux kernel to see how complicated this looks in
practice.
Now what if we had a way to just tell the HC "I want an event at exactly
*this* point in the TD, I want it to have the right completion code for
the whole TD, and to contain the exact number of bytes written"? Enter
the Event Data TRB: this little gizmo really does pretty much exactly
what any sane XHCI driver would want, and I have no idea why it isn't
used more often. It solves both the short packet event generation and
counting the transferred bytes without requiring any special magic in
software.
Change-Id: Idab412d61edf30655ec69c80066bfffd80290403
Signed-off-by: Julius Werner <jwerner@chromium.org>
Reviewed-on: https://chromium-review.googlesource.com/170980
Reviewed-by: Stefan Reinauer <reinauer@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
(cherry picked from commit e512c8bcaa5b8e05cae3b9d04cd4947298de999d)
Signed-off-by: Isaac Christensen <isaac.christensen@se-eng.com>
Reviewed-on: http://review.coreboot.org/6516
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
When libpayload header files are included in the payload itself, it's possible
that the payloads config settings will conflict with the ones in libpayload.
It's also possible for the libpayload config settings to conflict with the
payloads. To avoid that, the libpayload config settings have _LP_ (for
libpayload) added to them. The symbols themselves as defined in the Config.in files
are still the same, but the prefix added to them is now CONFIG_LP_ instead of just
CONFIG_.
Change-Id: Ib8a46d202e7880afdeac7924d69a949bfbcc5f97
Signed-off-by: Gabe Black <gabeblack@google.com>
Reviewed-on: https://gerrit.chromium.org/gerrit/65303
Reviewed-by: Stefan Reinauer <reinauer@google.com>
Tested-by: Gabe Black <gabeblack@chromium.org>
Commit-Queue: Gabe Black <gabeblack@chromium.org>
(cherry picked from commit 23e866da20862cace0ed2a67d6fb74056bc9ea9a)
Signed-off-by: Isaac Christensen <isaac.christensen@se-eng.com>
Reviewed-on: http://review.coreboot.org/6427
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Marc Jones <marc.jones@se-eng.com>
Setting of `controller->reg_base` is of no use here, as it is never read
(in another function) later. Looks like this pattern originated from uhci.c
where it makes sense.
By removing the indirection through `reg_base` we also fix a possible
truncation to u32.
Change-Id: I5c99c5bf1f5b1d6c04bd84d87fd3e275fd7d0411
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: http://review.coreboot.org/6251
Tested-by: build bot (Jenkins)
Reviewed-by: Patrick Georgi <patrick@georgi-clan.de>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Fix a possible null-pointer dereference (hopefully) before anyone runs
into this. Also don't switch ports to xHCI if initialization failed.
Change-Id: I5dbaeb435a98ead0b50d27fde13c9f1433ea3e81
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: http://review.coreboot.org/6245
Tested-by: build bot (Jenkins)
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
As the controller structure is never fully cleared, this one wasn't
initialized for non-pci controllers (but checked for non-null later).
Change-Id: I852671c5f55650bdb6cd97f4ec74b1f95ee894c7
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: http://review.coreboot.org/6246
Tested-by: build bot (Jenkins)
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Using void* for physical addresses leads to much casting and confuses
developers when to convert from physical to virtual addresses or
the other way around. When using plain integers for physical addresses
and pointers for virtual addresses things become much cleaner and we
won't ever end up dereferencing a physical address.
Change-Id: I24cd53b81c7863b6d14f0cbb4ce8937728b37c1c
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: http://review.coreboot.org/6244
Tested-by: build bot (Jenkins)
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Remove a redundant phys_to_virt() that sneaked in the initialization of
PCI xHCI controllers. The use of casts from void* to u32 (and vice versa)
prompts for things going wrong here. That will be addressed in a later
commit.
Change-Id: Ibc71ed6ee7016529c0e3a51559aaec07aaaba315
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: http://review.coreboot.org/6243
Tested-by: build bot (Jenkins)
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
being a good citizen on the box, libpayload tries to return to EHCI
mode on shutdown, so a non-XHCI capable USB driver after it (eg. in
the OS) finds something to work with.
Change-Id: Id227d646e08a258b841c644263112f0815dd486c
Signed-off-by: Patrick Georgi <patrick@georgi-clan.de>
Reviewed-on: http://review.coreboot.org/4547
Tested-by: build bot (Jenkins)
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Restructure USB stack to not depend on PCI, and
make PCI stub available on x86, but provide fixed
BARs for ARM (Exynos 5)
Change-Id: Iee7c8b134c22b661a9a515e24943470c9dbadd1f
Signed-off-by: Stefan Reinauer <reinauer@google.com>
Reviewed-on: https://gerrit.chromium.org/gerrit/49970
Reviewed-on: http://review.coreboot.org/4175
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
On Intel's Panther Point the xHCI ports are shared with an EHCI
controller. Our xHCI driver switches them to xHCI, naturally. But
we forgot to switch them back on shutdown, which left them
unusable by a non-xHCI aware operating system.
Change-Id: I70ef08655a603b42ee939935d50cf77ea97878a3
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: http://review.coreboot.org/3791
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Patrick Georgi <patrick@georgi-clan.de>
Well, it turned out to be more as some gaps ;)
but we finally have xHCI running. It's well tested against a QM77 Ivy
Bridge board.
We have no SuperSpeed support (yet). On Ivy Bridge, SuperSpeed is not
advertised and USB 3 devices will just work at HighSpeed.
There are still some bit fields in xhci_private.h, so this might need
little more work to run on ARM.
Change-Id: I7a2cb3f226d24573659142565db38b13acdc218c
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Signed-off-by: Patrick Georgi <patrick.georgi@secunet.com>
Reviewed-on: http://review.coreboot.org/3452
Tested-by: build bot (Jenkins)
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Previously printf()'s were used to show USB messages
which results in lots of USB information being shown
when it isn't needed. This will now use the usb_debug()
printing funtion that already exists in usb.h.
Change-Id: I2199814de3327417417eb2e26a660f4a5557cb9f
Signed-off-by: Dave Frodin <dave.frodin@se-eng.com>
Reviewed-on: http://review.coreboot.org/2044
Tested-by: build bot (Jenkins)
Reviewed-by: Anton Kochkov <anton.kochkov@gmail.com>
uhci_reset() differs in semantics compared to the other HCI's reset()
implementations. uhci_reset() does some initialization work after a
controller reset. So move the initialization part to a new function,
uhci_reinit(), which get's exported through a new entry in hci_t:
hci_t.init().
Warning: This breaks code that relies on the current, special,
counterintuitive behaviour of uhci_reset(). If one wants a working host
controller after calling hci_t.reset(), he should call hci_t.init()
afterwards.
Change-Id: Ia7ce80865d12d11157645ce251f77f349f8e3c34
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: http://review.coreboot.org/1851
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
When ohci_reset() was implemented, OHCI controllers stopped working
since the stub ohci_reset() is called at the end of ohci_init().
This is fixed by removing the call. To prevent further problems the call
to the xhci_reset() stub is removed, too.
Change-Id: If89825c8e6caf40f7f4fe078e8b2e90054a54ba2
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: http://review.coreboot.org/1850
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
The "debug" macro used internally in the libpayload USB subsystem was very
generically named and would leak into consumers of the library that included
usb.h directly or indirectly. This change turns that #define from a macro into
a static inline function to move away from the preprocessor, and also renames
it to usb_debug so it's less likely to collide with something unrelated.
Change-Id: I18717df111aa9671495f8a2a5bdb2c6311fa7acf
Signed-off-by: Gabe Black <gabeblack@google.com>
Reviewed-on: http://review.coreboot.org/1738
Reviewed-by: Patrick Georgi <patrick@georgi-clan.de>
Tested-by: build bot (Jenkins)
Add controller type (UHCI, OHCI, EHCI or XHCI)
into usbdev_hc (hci_t) struct, so now we know
which type selected controller have. It needed
to access controller specific data, if access
usb tree outside of libpayload (e.g. in payload
intself)
Change-Id: I7df947bbb56a50d0d792ccd4d3a6b021ee95e2ea
Signed-off-by: Anton Kochkov <anton.kochkov@gmail.com>
Reviewed-on: http://review.coreboot.org/1145
Tested-by: build bot (Jenkins)
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
We have fatal(), which is just as good.
Coccinelle script:
@@
expression E;
@@
-usb_fatal(E)
+fatal(E)
Change-Id: Iabecbcc7d068cc0f82687bf51d89c2626642cd86
Signed-off-by: Patrick Georgi <patrick.georgi@secunet.com>
Reviewed-on: http://review.coreboot.org/395
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Tested-by: build bot (Jenkins)
Improve scanning for USB controllers.
Limitations:
- OHCI doesn't support interrupt transfers yet (ie. no keyboards)
- xHCI just does initialization and device attach/detach so far
Signed-off-by: Patrick Georgi <patrick@georgi-clan.de>
Acked-by: Peter Stuge <peter@stuge.se>
git-svn-id: svn://svn.coreboot.org/coreboot/trunk@5691 2b7e53f0-3cfb-0310-b3e9-8179ed1497e1
Julius Werner