- 03 Oct, 2011 17 commits
-
-
Jonathan Cameron authored
Part never existed under that name. Ooops Signed-off-by:
Jonathan Cameron <jic23@cam.ac.uk> Signed-off-by:
Greg Kroah-Hartman <gregkh@suse.de>
-
Jonathan Cameron authored
It's going away. Signed-off-by:
Michael Hennerich <michael.hennerich@analog.com> Signed-off-by:
-
Jonathan Cameron authored
These are now handled through the chan_spec arrays and no one is using them anymore. Signed-off-by:
-
Jonathan Cameron authored
id wasn't used anywhere and st->irq can be removed by simply passing it into the core remove function (trivially available in the two bus implementations). Signed-off-by:
-
Jonathan Cameron authored
Other than a few slight refactorings the local version was pretty standard so replace it and rework to get rid of st->d_size which it setup. Signed-off-by:
-
Jonathan Cameron authored
To my mind, if a gpio is specified in the board file, yet fails to be successfully requested, that is an error condidtion and the driver should not muddle on regardless. This does mean unwinding the gpios on error. Also the free_gpios function is reordered so that it is consistent with the request one (reverse order obviously). This patch is the category of not technically fixing anything, just making the driver be more in line with what a reviewer will expect. Signed-off-by:
-
Jonathan Cameron authored
Signed-off-by:
-
Jonathan Cameron authored
IIO_CHAN is being phased out and in this case things are so simple it makes sense to have a local one parameter equivalent. Signed-off-by:
-
Jonathan Cameron authored
Saves on setting the value of address for the simple situation seen in this device. They are already used interchangably to get data from the buffer. Signed-off-by:
-
Jonathan Cameron authored
Signed-off-by:
-
Jonathan Cameron authored
Signed-off-by:
-
Jonathan Cameron authored
Signed-off-by:
-
Jonathan Cameron authored
Signed-off-by:
-
Jonathan Cameron authored
This is much cleaner than bouncing through the various structures to get to the same thing. Signed-off-by:
-
Jonathan Cameron authored
Preparation for moving driver out of staging. That macro is a nightmare to maintain so it is going away. Signed-off-by:
-
Jonathan Cameron authored
A few of these had the wrong shifts, which would lead to userspace hacking off the top couple of bits. Also, one part had the wrong accuracy. Signed-off-by:
-
Greg Kroah-Hartman authored
Now that we are in the 3.x days, "2.6" doesn't make sense. Signed-off-by:
-
- 30 Sep, 2011 23 commits
-
-
Marc Dietrich authored
The suspend commands need to be sent using the synchronous method, otherwise the power gets disabled before the messages are transferred. Signed-off-by:
Marc Dietrich <marvin24@gmx.de> [jak@jak-linux.org: Rewrote commit message] Signed-off-by:
Julian Andres Klode <jak@jak-linux.org> Signed-off-by:
-
Julian Andres Klode authored
As the comment indicates, adding that udelay seems to improve the stability of the communication, although it is not known why this is the case. Signed-off-by:
-
Julian Andres Klode authored
Add the includes that are currently missing in nvec.h and nvec.c and reorder them alphabetically. Signed-off-by:
-
Julian Andres Klode authored
I intent to support this code, especially the parts I wrote; and will thus enter as co-maintainer. Signed-off-by:
-
Julian Andres Klode authored
The nvec_power system polls nvec for battery information. In some cases, that part seems to be overloaded and unable to respond fast in which case it sends an incomplete response. We need to mark the transfer as completed, though, in order to prevent endless retries which can kill nvec. Signed-off-by:
-
Julian Andres Klode authored
Get 4 bytes of data from nvec at once instead of just a single byte. This makes the driver more similar to nvidias and might improve reliability. Signed-off-by:
-
Julian Andres Klode authored
Return the return value of nvec_write_async() in the methods returning an int. Signed-off-by:
-
Julian Andres Klode authored
Reject incomplete messages, causing the request to be transmitted again. This should fix various problems out there. Signed-off-by:
-
Julian Andres Klode authored
If no RX buffer is available in state 1, jump to state 0 again. This will produce an incredible amount of warnings, but it is not supposed to happen anyway. Signed-off-by:
-
Julian Andres Klode authored
Allow TX buffers to be allocated only in the upper 75% of the pool to avoid a completely filled buffer preventing the driver from processing responses. This also improves performance, as RX allocations do not require checking buffers allocated for TX unless there are more than 16 incoming messages -- which is highly unlikely. An earlier version used the lower 75% for TX messages, but that was considered to be not that effective due to the overlaps of RX and TX buffers mentioned above. Signed-off-by:
-
Julian Andres Klode authored
Client code wishing to make use of nvec_write_sync() must have a way to free the returned pointer, otherwise we run out of poool memory fairly soon. Signed-off-by:
-
Julian Andres Klode authored
Those macros are needed only for implementation purposes and do not have any use for other code wishing to use nvec. Signed-off-by:
-
Julian Andres Klode authored
Add kernel-doc comments describing the functions and structs we currently have. Signed-off-by:
-
Julian Andres Klode authored
Change nvec_write_async() to return an integer, 0 by default, a negative error on failure. Change nvec_write_sync() to check the return value and abort if it is negative. Signed-off-by:
-
Julian Andres Klode authored
When the caps lock key is pressed, toggle the associated LED. According to Nvidia code, we should send 0x01 where we sent 0x07, but this does not appear to work correctly on the AC100. Signed-off-by:
-
Julian Andres Klode authored
Forgotten in the last commit(s) which contained the copyrightable material, so let's add it now. I believe that only my nvec.c contributions are copyrightable, nvec.h is just interface naming, so does not deserve that yet. Signed-off-by:
-
Julian Andres Klode authored
The scratch area is not part of the pool and thus gets no allocation message. Printing a free message would be confusing, and the pointer subtraction would be undefined behavior. Signed-off-by:
-
Julian Andres Klode authored
Rewrite the interrupt handler to use a state machine similar to that found in the various kernels for the Advent Vega. This also changes the code to use the new functions introduced in the previous commits. This also merges the rewrite sent in August 2011 by Marc Dietrich, and thus also includes code by him. His original patch can be found on the mailing list. Signed-off-by:
-
Julian Andres Klode authored
Add functions nvec_msg_is_event() and nvec_msg_size() which do just what the say: tell whether the message is an event, and getting the size of the message. Signed-off-by:
-
Julian Andres Klode authored
Introduce nvec_gpio_set_value(), which works like gpio_set_value(), but also creates a debugging message, if that's enabled. Signed-off-by:
-
Julian Andres Klode authored
Introduce two new functions nvec_msg_alloc() and nvec_msg_free() that allocate and free message buffers from the internal pool of messages. Signed-off-by:
-
Marc Dietrich authored
This patch adds config dependencies for the NVEC keyboard, mouse, and power drivers. Signed-off-by:
-
Marc Dietrich authored
This adds the relevant info to the MAINTAINERS file so people can find the right person to blame. Signed-off-by:
-
