Lpc2000

Hello Bill. Thank you for your comments.

At 11:02 18-12-2003, you wrote:
>   I downloaded the new release this morning then recompiled my
>operational thumb-interwork(ing) program.  A quick look at the return
>code in the ISRs showed they were not returning via the 'bx'
>instruction as required to restore the appropriate ARM/THUMB mode.  I
>loaded the code into my eval board via a JTAG interface & gdb
>anyway and it crashed out almost immediately.  I reloaded it into the
>simulator in gdb and got the same results.  Think I'll hold off until
>the final version of 3.4 is available.

  Please, can you send me your compilation options? I *assume* you are 
using the "thumb-interwork" switch, aren't you?

  BTW, ISRs do not return with the bx instruction. ISRs restore the CPSR 
saved in the link register (r14). This will preserve the mode the processor 
was in before the interrupt.

  For example, here is the listing for a simple program with IRQ and FIQ 
calls. Disassembly seems right to me.

---8<---
00000000 <do_it>:
int
do_it() {
    0:   e1a0c00d        mov     ip, sp
    4:   e92dd800        stmdb   sp!, {fp, ip, lr, pc}
    8:   e24cb004        sub     fp, ip, #4      ; 0x4
    c:   e1a00003        mov     r0, r3
   10:   e89d6800        ldmia   sp, {fp, sp, lr}
   14:   e12fff1e        bx      lr

00000018 <do_irq>:
}

int __attribute__ ((interrupt ("IRQ")))
do_irq() {
   18:   e52dc004        str     ip, [sp, -#4]!
   1c:   e1a0c00d        mov     ip, sp
   20:   e92dd809        stmdb   sp!, {r0, r3, fp, ip, lr, pc}
   24:   e24cb004        sub     fp, ip, #4      ; 0x4
   28:   e1a00003        mov     r0, r3
   2c:   e89d6809        ldmia   sp, {r0, r3, fp, sp, lr}
   30:   e8bd1000        ldmia   sp!, {ip}
   34:   e25ef004        subs    pc, lr, #4      ; 0x4

00000038 <do_fiq>:
}

void __attribute__ ((interrupt ("FIQ")))
do_fiq() {
   38:   e52dc004        str     ip, [sp, -#4]!
   3c:   e1a0c00d        mov     ip, sp
   40:   e92dd800        stmdb   sp!, {fp, ip, lr, pc}
   44:   e24cb004        sub     fp, ip, #4      ; 0x4
   48:   e89d6800        ldmia   sp, {fp, sp, lr}
   4c:   e8bd1000        ldmia   sp!, {ip}
   50:   e25ef004        subs    pc, lr, #4      ; 0x4

00000054 <main>:
}

int
main(int argc, char **argv) {
   54:   e1a0c00d        mov     ip, sp
   58:   e92dd800        stmdb   sp!, {fp, ip, lr, pc}
   5c:   e24cb004        sub     fp, ip, #4      ; 0x4
   60:   e24dd008        sub     sp, sp, #8      ; 0x8
   64:   e50b0010        str     r0, [fp, -#16]
   68:   e50b1014        str     r1, [fp, -#20]
         do_it();
   6c:   ebfffffe        bl      0 <do_it>
                         6c: R_ARM_PC24  do_it
         do_irq();
   70:   ebfffffe        bl      0 <do_it>
                         70: R_ARM_PC24  do_irq
         do_fiq();
   74:   ebfffffe        bl      0 <do_it>
                         74: R_ARM_PC24  do_fiq
         return 0;
   78:   e3a03000        mov     r3, #0  ; 0x0
}
   7c:   e1a00003        mov     r0, r3
   80:   e24bd00c        sub     sp, fp, #12     ; 0xc
   84:   e89d6800        ldmia   sp, {fp, sp, lr}
   88:   e12fff1e        bx      lr

--->8---


>   This probably isn't the place to ask seeing as I just complained
>about the compiler but here goes anyway.  Would it be possible in
>future releases to include the Insight version of gdb?  I realize it
>makes the distribution bigger but I do find the graphical interface
>indespensible.  If it is too big to include, could it be made a
>seperate distribution by itself?  Thanks.

  Insight is planned to go into the distribution. I am having problems 
compiling the TCL libraries (it seems the TCL/TK guys made some API changes 
to the latest distro that Insight doesn't like). As soon as I resolve these 
issues I will include Insight in the toolchain. Please bear with me.

>-Bill Knight
>R O SoftWare
>
>PS - The distributions I use with thumb-interwork always seem to have
>seperate directories for those libraries.  I do not see them with this
>distribution.  Am I missing something or is it possible the
>thumb-interwork code & libraries are missing?  Though I expected the
>compiler to complain if I requested that mode and it wasn't supported.
>It did not.

  That is part of the problem ;^)

  As I understand it, old GCC ARM interworking code was a mess of a hack. 
All libraries were compiled against all possibilities of the 
thumb/interwork switches. This added a lot of redundancy and was getting 
difficult to maintain. The compilation process was rewritten in the "right 
way" in the GCC-3 series and now GCC is "aware" of ARM interworking. Now 
there are only two set of libraries, plain ARM and thumb (look under the 
lib/gcc/arm-elf/[gcc-version] toolchain subdirectory -- the thumb directory 
is there). GCC should be smart enough to generate the correct mode swapping 
code (it "should" -- but it is not there yet, some small issues are 
remaining but are getting continuously detected and fixed).

  Cheers.