During DFU download the destination start address is checked to not exceed the
RAM or flash end address, but it is also necessary to check if the end of the
data to be downloaded is also within the allowed range.
When downloading to RAM it is also necessary to check if the data to be
downloaded does not overwrite (i.e. corrupt) the stack.
Each board can define its own conditions on which the controller should
boot into DFU mode rather than normal application mode. Let's move the
"UART loopback jumper" to QMOD specific part. For SIMtrace we have an
actual button and can use that in a future patch.
We now generalize the USB communiction and abandon the 'req_ctx'
structure inherited from openpcd. Instead we use the libosmocore 'msgb'
structure to handle incoming and outgoing USB tranfers. We also use
linuxlist-based msgb-queues for each endpoint.
At power-up we need to initialize g_dfu once, to ensure a consistent
state. Afterwards we want to keep it across (software) reset, but on
power-up the memory would otherwise be filled with random data, causing
issues with detection of DFU/Runtime switching.
Using the USBDFU_OverrideEnterDFU() function, a board/application can
define extra conditions when the system should boot in DFU mode, even if
it was not explicitly switched to DFU mode from the application.
The app/dfu/main.c uses this mechanism to boot into DFU mode if the
stack + reset vector addresses are not plausible (i.e. some random junk
appears to be flashed in the application partition) or if the user
places a jumper accross the RxD+TxD lines of the debug UART. The idea
is that the system can be recovered by placing this jumper and then
re-installing the application from DFU.
This makes sure that we'll re-enumerate on the USB, as a CPU reset
apparently doesn't automatically release the pull-up and notify the hub
that we were gone?
For some strange reason my output is garbled in both the 'screen' and
'cu' teerminal programs and 'raw' terminal (stty) mode. I fail to
understand why, but let's simply adjust the code as needed for now.
We need to refer to g_dfu as a pointer from all code. In DFU mode, it
gets assigned to the address of _g_dfu, which is placed by the linker
script at the start of RAM. In runtime mode, the pointer is statically
defined to point at the start of RAM. The linker script for the runtime
(dfu environment) is adjusted to save the first 32 bytes for the _g_dfu
structure.
* we now actually route the EP0 control requests in DFU mode to the
correct handler (weak linker symbols are tricky)
* we now actually call code to read/write data from/to RAM/FLASH
