This adds a bladerf_get_fpga_bytes() function, which returns the
expected length of the FPGA image.
This is a dynamic replacement for BLADERF_FLASH_BYTE_LEN_FPGA,
which has been removed.
In particular, this is used in spi_flash_write_fpga_bitstream() and
spi_flash_erase_fpga() to determine the number of blocks to erase
during FPGA flashing operations.
This fixes an issue where xA9 boards were not being adequately
erased before writing.
Replaced these macros with appropriate calculations using
BLADERF_FLASH_ADDR_FIRMWARE and the page size held in the device's flash
architecture structure. Page size may change depending on the flash
used.
Replaced this macro with an appropriate calculation using the
BLADERF_FLASH_ADDR_FPGA macro and the page size held in the device's
flash architecture structure. Page size may change depending on the
flash used.
API change:
- bladerf_image_write(): new parameter: bladeRF device handle
Change is inspired by the pending removal of BLADERF_FLASH_EB_SIZE
macro. This macro really shouldn't be exposed to the world, nor
should it be constant as it may change depending on the flash
used on the board.
API new function:
- bladerf_erase_flash_bytes(): new function
Similar to the existing bladerf_erase_flash() function, but
handles the conversion from bytes to flash erase block pages
internally. This function is recommended for new designs because
the erase block size may change depending on the board. This
function also checks for proper alignment along erase block
boundaries.
General changes:
Although the macro still exists (for now), all uses of
BLADERF_FLASH_EB_SIZE have been replaced with equivalent
dev->flash_arch->ebsize_bytes because the value may change depending
on the flash used on the board.
The BLADERF_FLASH_PAGE_SIZE macro was replaced with a similar value from
the from the device's flash architecture
structure (e.g. dev->flash_arch->psize_bytes).
The BLADERF_FLASH_TO_PAGES() macro was replaced by using the new(er) API
functions bladerf_{read,write}_flash_bytes(). The conversion from bytes
to pages occurs in those functions, removing the need for the
macro. Going forward, the *_bytes() functions should be preferred
because the flash architecture may change.
New API functions:
- bladerf_read_flash_bytes()
This function is nearly identical to the existing
bladerf_read_flash() function, but takes the address and length in
units of bytes instead of flash pages. The function performs the
necessary conversions from bytes to pages internally based on the
specific flash architecture in use.
- bladerf_write_flash_bytes()
Same as above, for writes.
New code is advised to use these byte-unit functions instead of the
older page-unit functions because the page size may not be guaranteed to
be 256 bytes in future bladeRF platforms.
The BLADERF_FLASH_NUM_EBS macro was replaced with data from the device's
flash architecture structure (e.g. dev->flash_arch->num_ebs).
Functions with new parameters include:
- spi_flash.c: check_eb_access() - bladeRF device handle
The BLADERF_FLASH_NUM_PAGES macro was replaced with data from the
device's flash architecture structure (e.g. dev->flash_arch->num_pages).
Functions with new parameters include:
- spi_flash.c: check_page_access() - bladeRF device handle
API changes:
- bladerf_alloc_image(): new parameter - bladeRF device handle
- bladerf_alloc_cal_image(): new parameter - bladeRF device handle
General changes:
Although the macro still exists (for now), all uses of the
BLADERF_FLASH_TOTAL_SIZE macro have been replaced with equivalent
dev->flash_arch->tsize_bytes because the value may change depending on
teh flash used on the board.
During this process, some function prototypes were modified so the
device handle could be passed in, thus allowing flash architecture
data to be extracted.
- calibration.c: dc_cal_tbl_image_load()
- image.c: is_valid_addr_len()
_rfic_fpga_get_status_wqlen:
C4267: 'return': conversion from 'size_t' to 'int'
_rfic_fpga_spinwait:
C2065: 'useconds_t': undeclared identifier
C2120: 'void' illegal with all types
_rfic_fpga_set_gain:
C4244: conversion from 'float' to 'int'
After the RFIC control split, we now directly call the rfic
functions to get/set filters when changing the sample rate.
So, we can go back to MUTEXing the API functions.
This undos the changes in 9e8bd237bc
host: bladerf2: remove unneeded pre-declarations
In most cases, BLADERF_ERR_TIMEOUT should be reported to the user.
So, for the NIOS_PKT_16x64_TARGET_RFIC target, nios_access_quiet
can be used to suppress all the log_errors.
For expedience, quick retunes directly control the RFIC via SPI.
This works fine as it is, but it puts the RFIC into an unexpected
state, and a reset is needed.
This commit fully de-initializes the RFIC during bladerf_close, if
a bladerf_get_quick_tune is performed sometime during the session.
Moves much of the low-level control etc into rfic_{fpga,host}.c
Implements the necessary plumbing to make FPGA RFIC control work,
while still supporting host-based RFIC control.
host: bladerf2: verify that FPGA supports RFIC control before using it
FPGA-based RFIC control is rather memory-heavy, so omitting it
from builds is possible, but the usual capabilities mechanism
won't detect this.
This will throw a couple annoying errors from nios_16x64_read
when checked, but they are ignorable.
host: bladerf2: split bladerf2_get_quick_tune into host/fpga control
also:
common.h: remove vestigial comments
host: bladerf2: support standby state for FPGA RFIC mode
this lets us ensure the RFIC is in a known state when we
call bladerf_close()
host: bladerf2: move rfic_fpga init stuff to FPGA
During the recent refactor, the RFIC FIR filter functions got
wrapped in a MUTEX_LOCK. This is incompatible with their use
by bladerf2_set_sample_rate, which is called within a MUTEX_LOCK.
🤦♀️
General documentation cleanup, function name adjustments, merging of
redundant structures, etc.
fpga_common/ad936x_helpers:
Contains wrappers for AD936x RFIC functionality.
fpga_common/bladerf2_common:
Contains bladeRF2-specific helper functions and constants.
fpga_common/ad936x_params:
Reformat (no more tabs, phew)
Move ina219 shunt const to bladerf2_common
host/common/range:
Move functions and macros for handling bladerf_range structs
../libbladeRF/include/bladeRF2.h:
Explicitly index RX/TX FIR enums
Update references to ad936x_params.c
../libbladeRF/src/board/bladerf2/common:
New macros for boilerplate:
- WITH_MUTEX: wrap around locked blocks
- CHECK_STATUS: calls function, checks return code
- CHECK_AD936X: calls function, checks return code, translates
from AD936x to libbladeRF error codes
- CHECK_*_LOCKED: above, but unlocks mutex before erroring
- NULL_CHECK: checks a var for null
Move helper functions into here
../libbladeRF/src/board/bladerf2/bladerf2.c:
Use new macros
Code style fixes
Remove redundant code
Clear out any scheduled retunes that may exist in the Nios during
bladerf2_open() and bladerf2_close(). This helps prevent unexpected
behavior if the device is (re)opened later.
Let libbladeRF handle the quick tune profile numbering for bladeRF2
devices. This makes it so that existing bladeRF1 applications using the
quick tune feature should "just work" with bladeRF2.
A few changes to the bladerf_quick_tune structure:
- rename 'profile' to 'rffe_profile' for clarity
- delete 'data[16]' not needed anymore and preserves ABI compatibility
- add 'nios_profile' for indexing the profile stored in the BBP
- add 'port' for storing the RFFE port settings
- add 'spdt' for storing the external RF switch settings
Improvements to bladerf2_get_quick_tune():
- Send rffe_fastlock_save packet to instruct Nios to save a copy of
the RFFE fast lock profile to itself. This means the fast lock
profile data (16 bytes) no longer needs to be transmitted to/from
the Nios and the host machine (retune speedup).
- Save the RFFE port setting to the quick tune structure so it may be
sent down to the Nios when scheduling a retune.
Saving the 16-bytes of fast lock profile data to the host is no longer
necessary because it is stored on the Nios itself. This has two
additional benefits: quicker retunes (the profile data does not need to
be shuffled across USB), and it preserves ABI compatibility because the
size of the bladerf_quick_tune struct does not increase compared to
previous versions of libbladeRF.
Update bladerf_quick_tune structure for bladeRF2 data. It now contains:
- profile (profile number, integer)
- data (profile data, 16 bytes)
Fill in the bladerf2_get_quick_tune() function. The bladeRF2's RFIC uses
the concept of fast lock profiles, of which there are 8 for TX and 8 for
RX. These profiles store the VCO information needed to quickly change
frequencies. To use more than 8 profiles, the profiles may be stored
off-chip, which is what we're doing here.
1. It is assumed the RFIC has been tuned to a particular frequency.
2. The profile is created using ad9361_fastlock_store().
3. The profile is read back and stored in host memory using
ad9361_fastlock_save().
_apportion_gain did not pay heed to the step size of the gain
range, causing it to end up with invalid values for LNA and
RXVGA2.
We now ensure that the allocation mod <stepsize> is 0, so that
other, more flexible gain stages can be used to pick up the
remainder.