readme

用于TM1300/PNX1300系列DSP（主要用于视频处理）的下载程序

This example shows how to boot a standalone multiprocessor
system from a single object file in EEPROM, using a staged 
downloader which runs on one of the processors, and which 
loads the object file on each of the processors.

The basic idea is simple: assumed that the EEPROM is accessible
via PCI, downloading can be performed using the multiprocessor
loading function described in file MP_downloader.c. This function
takes the EEPROM memory address, together with arrays of basic processor
information for all processors used. After this function has been used 
to load the object from eeprom to each of the processors, all 
processors can be started.

MP_downloader.c is intended for demonstration on tmsim, and therefore
contains the following adaptations:

     - the EEPROM is simulated by a piece of malloced SDRAM
       into which the object file "program.out" is read.

     - the value of the tmsim specific download symbol __syscall 
       is copied to the loaded program. The names of the corresponding
       download symbols for MacOS and Win95, respectively, are:
        _TMMANSharedPatch and __HostCall_commvar_init. After the
       appropriate changes, the demo will run also on MacOS or Win95.
       Standalone executables (host= nohost) do not have such 
       specific download symbols (unless user specified).

     - The top of SRDAM is avoided for downloading (see TOP_MEM_HOLE),
       because tmsim has mapped some undocumented I/O locations there.

The only problem is that the MP_downloader runs on a processor that
is also itself a download target. While the downloader runs, a program
must be loaded onto the same processor whithout corrupting the  downloader.

This can be handled as follows: Modify the L1 boot code for the boot
processor so that it loads the executable somewhere in the stack/heap gap
of the final object to be loaded (here the stack heap gap of program.out).
This stack heap gap is not needed anyway until the final program 
starts executing, and therefore is free for use during downloading.
For this tmsim demo, the L1 boot code is simulated by L1.c, by means 
of a call to the downloader to download MP_downloader in an area 5 megabytes
below the end of SDRAM, with a length of 4 Megabytes. Note that 
the stack heap gap is dependent on the final executable; however, 
the sizes used here are usually ok.

The MP_downloader as described can now be used. It will download
to all processors, while neatly avoiding the memory in which it itself
runs on the current processor. All executables on the other processors 
can be started using the familiar fiddling with their BUI_CTL's; 
the executable on the MP_downloader's node itself can be started by 
a jump to its SDRAM base, after the proper cache invalidations. 

The demo loads on one processor only, but this already 
demonstrates how to handle the major problems, i.e. 
downloading on the current node.

Note that each loaded object (L1.out, MP_downloader.out, program.out)
is a complete executable, an will start setting its required endian, 
its stack and heap area, and that the code of the previous boot stage 
is not used, and can safely be overwritten.