Sometimes check_function_exists does not work on Windows, it is then worth
testing to see if the LIBUSBX_API_VERSION is equal or greater than 0x01000102
(1.0.16). This version contains the neccessary libusb hotplug features needed to
make bladeRF-flash work.
- Split RF, FPGA load, and firmware methods into seperate .c files
- Split several vendor requests to eliminate magic glEp0Idx and to allow
for better error handling
- Updated libbladeRF to use new vendor requests and properly check return
code.
- Changed glSpiPageSize from global to #define
- Move OTP handling into firmware.c
- Version bump
Conflicts:
fx3_firmware/bladeRF.c
The bitflipping look takes about 180 instruction cycles to flip 1 byte. The new
method takes less than 5 clock cycles, making both images load times seem
almost instantenous. Further optimizations can be made by adding a Duff's
machine.
- Move and use interface numbers
- Renumber USB_IF_CONFIG from 0 to 3
- Introduced 0 alt-interface as a NULL interface
- Fixed FS descriptor length
- Implemented GET_STATUS, SET_FEATURE, CLEAR_FEATURE w.r.t. halting bulk streams
- Added include guards for firmware_common/bladeRF.h
- Changed guard on suspend/resume requests to key on glUsbConfiguration instead
of glUsbAltInterface. This is because suspend/resume requests are on
interfaces, not endpoints.
- Addresses github tickets #97, #98, #99
Fixed defect introduced in 9015bdbb5c, where a device handle's device identifier information was being populated from a (most likely) uninitialized bladerf_devinfo, rather than the one fully populated by a call to lusb_get_devinfo().
Initializing, extracting, and comparing device identifier information
(struct bladerf_devinfo) is now exposed through the API. Among other
things, this allow users to check if a device specifier string matches a
device they already have a handle for.
To support this, a copy of bladerf_devinfo information is now kept around in
the device handle; backends are responsible for populating this in their
open() implementation. A couple of the fields were already present, so
this only adds a couple small variables worth of overhead.
Some re-arranging of the linux and libusb probe and open calls has been done
to achieve this. A few warnings and bugs were addressed in this process:
- The Linux backend was attempting to fill in a serial number via
a cached entry.
- There was some potential for a read of an uninitialized portion
of an otp buffer.
- Missing serial # has been made non-fatal. (Some developers'
device do not have serial numbers.)
Files associated with this functionality was moved from host/common
back into the library, which should now be used instead. bladeRF-flash
has been updated to use these new API functions.
Setting the SPI flash interface is not required for reading. Only for writing. Turns out, if you try to set that interface twice then things do not go so well.
I am not 100% sure if this is just a host side fix for a firmware issue, but it alleviates the problem for now.
Also fixed some tabs that snuck in there.
All executables and libraries are now placed in a single "output" directory.
For Windows builds, third-party DLLs are copied into this directory.
This commit also moves some CFLAGS changes in the top-level CMakeLists.txt
down to where some other compiler flags are set up. (Non-functional change)