Trigger the gdb_bootloader programmatically about fabooh HOT 8 CLOSED

I think you would be hard pressed to make many modifications to the current code as it is pretty close to the 1k limit. If you are you willing to let it get larger than 1K you could write some data into a specific ram location and have the gdb_bootloader main code look at that location to decide if the gdb code should start.

from fabooh.

Ahh, is RAM preserved across RESETs?
Or, exposing a call address should not add any extra code.
Also, hooking the bootloader to NMI (as per some compile-time config options) should be free, size-wise. (Useful if the fw doesn't use NMI otherwise.)
Or do I misunderstand something?

from fabooh.

The gdb_bootloader code uses a custom c runtime startup routine. It doesn't touch ram, nor does it use a data section. It does however initialize the stack to the top of ram. You can see the code here:
https://github.com/RickKimball/fabooh/blob/master/include/msp430/core/main.h#L126

So if you use memory at the lower end ( 0x200 ... ) it will be there when you reset. Here is the memory map used by the linker script for the gdb_bootloader:
https://github.com/RickKimball/fabooh/blob/master/board/msp430g2553in20/apps/gdb_bootloader/msp430g2553in20/_memory.x#L5

I would probably use a couple bytes with a magic word and the inverse of your magic word to trigger.

from fabooh.

It is pretty tight .. maybe ~40 bytes left for code

$ msp430-size gdb_bootloader.elf 
   text    data     bss     dec     hex filename
   1048       0      66    1114     45a gdb_bootloader.elf

from fabooh.

Yeah, I know, thanks! I've also seen the memory map and the code, I'm just not familiar with the bootstrap stuff in general (and I didn't know the chip itself has a promise to keep the RAM intact across resets).
I just hoped that checking for being called via NMI might be cheap, considering code-size.
Or, it'd still also be configurable compile-time, with no GPIO checking code in NMI mode.

from fabooh.

(Also, I could replace the LED stuff with memory reads in a copy of the bootloader code if I wanted (I can still check BSL mode "visually" from UART, sending e.g. $#12 and receiving a '-'), but still would prefer a "free" (less intrusive) alternative without code changes, which an exposed call address could probably do. I guess it's easier with assembly than C, I'm not sure if it's even possible with GCC without contorted tricks.)

from fabooh.

This fits:

diff --git a/board/msp430g2553in20/apps/gdb_bootloader/gdb_bootloader.cpp b/board/msp430g2553in20/apps/gdb_bootloader/g
index a7daa55..af8483a 100644
--- a/board/msp430g2553in20/apps/gdb_bootloader/gdb_bootloader.cpp
+++ b/board/msp430g2553in20/apps/gdb_bootloader/gdb_bootloader.cpp
@@ -51,7 +51,7 @@
 #if 1 /* might need to comment this out to gain space for the software only UART */
 __attribute__((section(".copyr"),used))
 const char _copyright[] = {
-    "Copyright \xC2\xA9 2013 Kimball Software"
+    "Copyright \xC2\xA9 KimballSoftware"
     };
 #endif

@@ -124,7 +124,7 @@ void gdb_bootloader() {
   PUSH2::setmode_inputpullup();

   // if user didn't press button P1.3, just run from the user reset vector
-  if ( !PUSH2::is_low() ) {
+  if ( !(PUSH2::is_low() || (*(uint16_t *)0x0200 == 0xb002)) ) {
     register unsigned work_reg = USER_RESET_VECTOR; // check for valid user vector in flash hiding spot

     // TODO: probably should reset clock and P1.3 button
@@ -140,6 +140,7 @@ void gdb_bootloader() {
         );
     }
   }
+  *(unsigned *)0x200 = 0x0;

   RED_LED::setmode_output();
   RED_LED::high();
diff --git a/examples/basic/blink/blink.cpp b/examples/basic/blink/blink.cpp
index 99bffc1..7f587e2 100644
--- a/examples/basic/blink/blink.cpp
+++ b/examples/basic/blink/blink.cpp
@@ -6,6 +6,7 @@

 void setup(void) {
   RED_LED::setmode_output();
+  PUSH2::setmode_inputpullup();
 }

 void loop(void) {
@@ -13,6 +14,11 @@ void loop(void) {
   const uint32_t msec_delay = 1000 / freq / 2;

   while (1) {
+    if ( PUSH2::is_low() ) {
+       *(unsigned *)0x0200 = 0xb002;
+        WDTCTL = 0;
+        while(1);
+    }
     RED_LED::toggle();
     delay(msec_delay);
   }

from fabooh.

OK, I like it, thanks! :) I'll close the issue.
(In the meantime, I finally realized that the bootloader must be started via RESET, since it itself also relies on a clean PUC state for proper init, so direct jumping or NMI wouldn't cut it for a generic code branch.)

from fabooh.