Linux usb/ch9.h seems to have all the same information (and more)
as usbdescriptors.h so use the former instead of the later one.
As a consequense of this change USB_SPEED_* values don't correspond
directly to EHCI speed encoding anymore, I've added necessary
recoding in EHCI driver. Also there is no point to put speed into
pipe anymore so it's removed and a bunch of host drivers fixed to
look at usb_device->speed instead.
Old usbdescriptors.h included is not removed as it seems to be
used by old USB device code.
This makes usb.h and usbdevice.h incompatible. Fortunately the
only place that tries to include both are the old MUSB code and
it needs usb.h only for USB_DMA_MINALIGN used in aligned attribute
on musb_regs structure but this attribute seems to be unneeded
(old MUSB code doesn't support any DMA at all).
Signed-off-by: Ilya Yanok <ilya.yanok@cogentembedded.com>
Kill off ehci-core.h
It was used to specify some static controller data. To support more than
one controller being active at any time we have to carry the controller
data ourselfes. Change the ehci interface accordingly.
NOTE: OMAP implemented the ehci stuff a bit backwards and should be fixed
to do the same thing as other platforms. But the change for now is at least
compile clean.
Signed-off-by: Lucas Stach <dev@lynxeye.de>
Reviewed-by: Marek Vasut <marex@denx.de>
Carry an index in the lowlevel usb functions to make specify the
respective usb controller.
Also pass through an controller struct from lowlevel_init to the
creation of the root usb device of this controller.
Signed-off-by: Lucas Stach <dev@lynxeye.de>
Reviewed-by: Marek Vasut <marex@denx.de>
usb_hub_descriptor has to be packed as it's used for
communication with the device. Member wHubCharacteristics
violates the natural alignment rules.
Use explicit unaligned access functions for this member.
Fixes ARMv7 traping while using USB.
v2: fix typo found by Thomas Langer
v3: rebased on top of u-boot-usb/master
Signed-off-by: Lucas Stach <dev@lynxeye.de>
Relax the qTD transfer alignment constraints in order to need less qTDs for
buffers that are aligned to 512 bytes but not to pages.
Signed-off-by: Benoît Thébaudeau <benoit.thebaudeau@advansee.com>
Cc: Marek Vasut <marex@denx.de>
Cc: Ilya Yanok <ilya.yanok@cogentembedded.com>
Cc: Stefan Herbrechtsmeier <stefan@herbrechtsmeier.net>
This patch takes advantage of the hardware EHCI qTD queuing mechanism to avoid
software and transfer splitting overhead so as to make transfers as fast as
possible.
The only drawback is a call to memalign. However, this is fast compared to the
transfer timings, and the heap size to allocate is small, e.g. 128 kiB in the
worst case for a transfer length of 65535 packets of 512 bytes.
Tested on i.MX25, i.MX35 and i.MX51. In my test conditions, the speed gain was
very significant (several times faster), which is really appreciable when
accessing large files.
Signed-off-by: Benoît Thébaudeau <benoit.thebaudeau@advansee.com>
Cc: Marek Vasut <marex@denx.de>
Cc: Ilya Yanok <ilya.yanok@cogentembedded.com>
Cc: Stefan Herbrechtsmeier <stefan@herbrechtsmeier.net>
Make some light cosmetic code cleanup by the way.
Signed-off-by: Benoît Thébaudeau <benoit.thebaudeau@advansee.com>
Cc: Marek Vasut <marex@denx.de>
Cc: Ilya Yanok <ilya.yanok@cogentembedded.com>
Cc: Stefan Herbrechtsmeier <stefan@herbrechtsmeier.net>
Interrupt transfers requiring several transactions are not supported by
submit_int_msg() because bInterval is ignored. This patch returns a failure code
and prints an error message in this case.
Signed-off-by: Benoît Thébaudeau <benoit.thebaudeau@advansee.com>
Cc: Marek Vasut <marex@denx.de>
Cc: Ilya Yanok <ilya.yanok@cogentembedded.com>
Cc: Stefan Herbrechtsmeier <stefan@herbrechtsmeier.net>
Buffer coming from upper layers should be cacheline aligned/padded
to perform safe cache operations. For now we don't do bounce
buffering so getting unaligned buffer is an upper layer error.
We can't check if the buffer is properly padded with current
interface so just assume it is (consider changing with in the
future). The following changes are done:
1. Remove useless length alignment check. We get actual transfer
length not the size of the underlying buffer so it's perfectly
valid for it to be unaligned.
2. Move flush_dcache_range() out of while loop or it will
flush too much.
3. Don't try to fix buffer address before calling invalidate:
if it's unaligned it's an error anyway so let cache subsystem
cry about that.
4. Fix end buffer address to be cacheline aligned assuming upper
layer reserved enough space. This is potentially dangerous
operation so upper layers should be careful about that.
Signed-off-by: Ilya Yanok <ilya.yanok@cogentembedded.com>
The USB spec says that 32 bytes is the minimum required alignment.
However on some platforms we have a larger minimum requirement for cache
coherency. In those cases, use that value rather than the USB spec
minimum. We add a cpp check to <usb.h> to define USB_DMA_MINALIGN and
make use of it in ehci-hcd.c and musb_core.h. We cannot use MAX() here
as we are not allowed to have tests inside of align(...).
Signed-off-by: Tom Rini <trini@ti.com>
[marek.vasut]: introduce some crazy macro voodoo
Signed-off-by: Marek Vasut <marex@denx.de>
[ilya.yanok]: moved external buffer fixes to separate patch,
we use {ALLOC,DEFINE}_ALIGN_BUFFER macros with alignment of USB_DMA_MINALIGN
for qh_list, qh and qtd structures to make sure they are proper aligned
for both controller and cache operations.
Signed-off-by: Ilya Yanok <ilya.yanok@cogentembedded.com>
Move or_asynclistaddr programming to ehci_submit_async()
function to make sure queue head is properly programmed
before every transfer. This solves the problem with changing
qh address.
Also remove unneeded qh_list->qh_link reprogramming at the
end of transfer.
Signed-off-by: Ilya Yanok <ilya.yanok@cogentembedded.com>
The construction of QH and qTD lists in ehci_submit_async() call is cryptic
business, add at least a bit of comments so if someone is reading it, he can at
least reference the intel ehci manual (ehci-r10.pdf).
Signed-off-by: Marek Vasut <marek.vasut@gmail.com>
Discard the creepy cache flushing mechanisms in ehci-hcd.c and replace them with
more straightforward flushing. In the new approach, the flushing takes place
directly in ehci_submit_async() call instead of going through the QH list and
flushing all members and buffers. This discards a lot of weird bit operations
on the members of QH and qTD structures.
NOTE: Certainly, this flushes even qTDs which are possibly unused in some
transactions, though the overhead of the previous code was much higher than is
the overhead of flushing two more cache lines (which most probably aren't even
cached).
Signed-off-by: Marek Vasut <marek.vasut@gmail.com>
These two functions were called only from ehci_submit_async(), therefore
dissolve them as part of ehci_submit_async() to get rid of all those static
variables.
Signed-off-by: Marek Vasut <marex@denx.de>
CONFIG_USB_EHCI_TXFIFO_THRESH enables setting of the txfilltuning
field in the EHCI controller on reset.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Tom Warren <twarren@nvidia.com>
Common code has a mdelay() func, so use that instead of the usb-specific
wait_ms() func. This also fixes the build errors:
ohci-hcd.c: In function 'submit_common_msg':
/usr/local/src/u-boot/blackfin/include/usb.h:202:44: sorry, unimplemented: inlining failed in call to 'wait_ms': function body not available
ohci-hcd.c:1519:9: sorry, unimplemented: called from here
/usr/local/src/u-boot/blackfin/include/usb.h:202:44: sorry, unimplemented: inlining failed in call to 'wait_ms': function body not available
ohci-hcd.c:1816:10: sorry, unimplemented: called from here
/usr/local/src/u-boot/blackfin/include/usb.h:202:44: sorry, unimplemented: inlining failed in call to 'wait_ms': function body not available
ohci-hcd.c:1827:10: sorry, unimplemented: called from here
/usr/local/src/u-boot/blackfin/include/usb.h:202:44: sorry, unimplemented: inlining failed in call to 'wait_ms': function body not available
ohci-hcd.c:1844:10: sorry, unimplemented: called from here
/usr/local/src/u-boot/blackfin/include/usb.h:202:44: sorry, unimplemented: inlining failed in call to 'wait_ms': function body not available
ohci-hcd.c:1563:11: sorry, unimplemented: called from here
/usr/local/src/u-boot/blackfin/include/usb.h:202:44: sorry, unimplemented: inlining failed in call to 'wait_ms': function body not available
ohci-hcd.c:1583:9: sorry, unimplemented: called from here
make[1]: *** [ohci-hcd.o] Error 1
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Marek Vasut <marex@denx.de>
This moves keyboard polling logic from USB HCD drivers into USB
keyboard driver. Remove usb_event_poll() as keyboard polling was
the only user of this API. With this patch USB keyboard works with
EHCI controllers again. Tested on a tegra2 seaboard.
Signed-off-by: Allen Martin <amartin@nvidia.com>
This patch allows USB to work on some hosts, which need additional frobing after
the host was powered up via regular USB powerup sequence.
Signed-off-by: Marek Vasut <marek.vasut@gmail.com>
Cc: Stefano Babic <sbabic@denx.de>
Cc: Remy Bohmer <linux@bohmer.net>
Cc: Wolfgang Grandegger <wg@denx.de>
Cc: Jason Liu <r64343@freescale.com>
According to EHCI specification v1.0, the controller should stabilize
the power on a port at most 20 ms after the port power bit transition.
So, we put this setting in the virtual descriptor corresponding field,
(bPwrOn2PwrGood = 10 => 10 x 2ms = 20ms), this saves about 500ms at each
controller initialization/enumeration.
Signed-off-by: Vincent Palatin <vpalatin@chromium.org>
With this, the EHCI seems to "recover" from a timeout. This is particularly
observable if you were to ping the wrong IP Address and then ping the correct
one or if there was a temporary failure during tftp sessions.
All it takes is one timeout to disable it. If you have a noisy network (lot
of traffic), even if the traffic is not for the board, the timeouts don't occur.
Signed-off-by: Joel A Fernandes <agnel.joel@gmail.com>
Signed-off-by: Sandeep Paulraj <s-paulraj@ti.com>
Changed both to use a common timeout for URB submission, since they were using
different values and EHCI's was too short.
Also fixed EHCI to actually check if urb submission succeeded, rather than
silently continuing into the weeds.
Change-Id: I7f71499ffaa05187d8e5618db2419e1606007b82
Signed-off-by: Simon Glass <sjg@chromium.org>
This patch fixes a problem noticed on lwmon5 (PPC440EPx) using the
common EHCI driver, when "usb reset" is issued multiple times.
Upon the 2nd (and further) "usb reset" command, the command fails
with the following messages:
=> usb reset
(Re)start USB...
USB: Register 1111 NbrPorts 1
USB EHCI 1.00
scanning bus for devices... 5 USB Device(s) found
scanning bus for storage devices... 2 Storage Device(s) found
=> usb reset
(Re)start USB...
USB: EHCI fail to reset
Error, couldn't init Lowlevel part
This patch fixes this problem. Now "usb reset" can be called multiple
times.
Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Remy Bohmer <linux@bohmer.net>
This watchdog reset call is needed here, otherwise the lwmon5 board
(PPC440EPx based) will reset upon the "usb reset" command.
Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Remy Bohmer <linux@bohmer.net>
Checking the status field of the qTD token in the current code
do not take into acount cases where endpoint stall (halted) bit
is set together with XactErr status bit. As a result clearing
stall on an endpoint won't be done if this status bit was also
set. Check for halted bit and report USB_ST_STALLED status
if the host controller also indicates endpoit stall condition.
Signed-off-by: Anatolij Gustschin <agust@denx.de>
This patch solves a problem with USB hanging under higher load on a
i.MX31 board. It falls into class of typical USB problems and fixes:
if you don't understand the real cause, add a delay somewhere.
The problem appeared after introduction of ELF relocation, which
results in smaller code, which appears to run faster (probably because
it fits better in the cache); turning off the instruction cache,
adding debug printf()s and increasing the delay have all been found to
make the problem go away.
Moving the original "udelay(1)" up in the code to it's new place made
the problem appear much less frequently. Increasing the delay to 2
microseconds then made the code run reliably in all (hour-long) tests.
To be on the safe side, we set it to 5 microseconds here.
Signed-off-by: Heiko schocher <hs@denx.de>
Signed-off-by: Wolfgang Denk <wd@denx.de>
Cc: Remy Bohmer <linux@bohmer.net>
Cc: Stefano Babic <sbabic@denx.de>
Appendix B "EHCI 64-Bit Data Structures" of the "Enhanced Host
Controller Interface Specification for Universal Serial Bus" (Rev.
1.0, March 12, 2002) defines additional fields which were missing in
U-Boot's struct qTD; as these are also present in recent versions of
struct ehci_qtd in the Linux kernel, we add them here, too.
This fixes some nasty memory corruption problems.
Reported-by: Dan Lykowski <lykowdk@gmail.com>
See http://permalink.gmane.org/gmane.comp.boot-loaders.u-boot/76942
Signed-off-by: Wolfgang Denk <wd@denx.de>
Cc: Remy Bohmer <linux@bohmer.net>
Cc: Dan Lykowski <lykowdk@gmail.com>
Cc: Stefano Babic <sbabic@denx.de>
Tested-by: Stefano Babic <sbabic@denx.de>
ehci_submit_async() doesn't really zero out the QH transfer overlay (as the EHCI
specification suggests) which leads to the controller seeing the "token" field
as the previous call has left it, i.e.:
- if a timeout occured on the previous call (Active bit left as 1), controller
incorrectly tries to complete a previous transaction on a newly programmed
endpoint;
- if a halt occured on the previous call (Halted bit set to 1), controller just
ignores the newly programmed TD(s) and the function then keeps returning error
ad infinitum.
This turned out to be caused by the wrong orger of the arguments to the memset()
call in ehci_alloc(), so the allocated TDs weren't cleared either.
While at it, stop needlessly initializing the alternate next TD pointer in the
QH transfer overlay...
Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Acked-by: Remy Bohmer <linux@bohmer.net>
Commit b416191a14 (Fix EHCI port reset.) didn't
move the code that checked for successful clearing of the port reset bit from
ehci_submit_root(), relying on wait_ms() call instead. The mentioned code also
erroneously reported port reset state when the reset was already completed.
Signed-off-by: Sergei Shtylyov <sshtylyov@mvista.com>
USB devices on the 2nd port are not detected and I get the following message:
The request port(1) is not configured
That's with default CONFIG_SYS_USB_EHCI_MAX_ROOT_PORTS value of 2. 'req->index'
is 1-based, so the comparison in ehci_submit_root() can't be correct.
Signed-off-by: Sergei Shtylyov <sshtylyov@mvista.com>
On little endian machines, EHCI root hub's USB revision is reported as 0.2 --
cpu_to_le16() was missed in the initializer for the 'bcdUSB' descriptor field.
The same should be done for the 'bcdDevice' field.
Signed-off-by: Sergei Shtylyov <sshtylyov@mvista.com>
In USB ehci driver, the port reset is not terminated. EHCI spec says "A host
controller must terminate the reset and stabilize the state of the port within
2 milliseconds". Without termination, a port stays at reset state. This is
observed on ppc4xx(sequoia) boards.
Signed-off-by: Chris Zhang <chris@seamicro.com>
The header files usb.h and usbdescriptors.h have the same nameed
structure definitions for
usb_config_descriptor
usb_interface_descriptor
usb_endpoint_descriptor
usb_device_descriptor
usb_string_descriptor
These are out right duplicates in usb.h
usb_device_descriptor
usb_string_descriptor
This one has extra unused elements
usb_endpoint_descriptor
unsigned char bRefresh
unsigned char bSynchAddress;
These in usb.h have extra elements at the end of the usb 2.0
specified descriptor and are used.
usb_config_descriptor
usb_interface_descriptor
The change is to consolidate the definition of the descriptors
to usbdescriptors.h. The dublicates in usb.h are removed.
The extra element structure will have their name shorted by
removing the '_descriptor' suffix.
So
usb_config_descriptor -> usb_config
usb_interface_descriptor -> usb_interface
For these, the common descriptor elements are accessed now
by an element 'desc'.
As an example
- if (iface->bInterfaceClass != USB_CLASS_HUB)
+ if (iface->desc.bInterfaceClass != USB_CLASS_HUB)
This has been compile tested on MAKEALL arm, ppc and mips.
Signed-off-by: Tom Rix <Tom.Rix@windriver.com>
This change is cheked in Linux source and fix found to be in sync.
This patch is tested for USB host interface on Kirkwood based
Sheevaplug platform (ARM little endian board)
Risk: the impact of this patch is not validated on big endian board.
This need to be checked...
Signed-off-by: Prafulla Wadaskar <prafulla@marvell.com>
Signed-off-by: Remy Bohmer <linux@bohmer.net>
ehci-hcd.c: In function 'ehci_submit_root':
ehci-hcd.c:719: warning: value computed is not used
ehci-hcd.c:748: warning: value computed is not used
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Remy Bohmer <linux@bohmer.net>
move to linux usb driver organisation
as following
drivers/usb/gadget
drivers/usb/host
drivers/usb/musb
Signed-off-by: Jean-Christophe PLAGNIOL-VILLARD <plagnioj@jcrosoft.com>
Signed-off-by: Remy Bohmer <linux@bohmer.net>
