5 Replies Latest reply on Jun 10, 2016 10:39 AM by jenoc_293011

    Code execution error (ICE Cube)

      Why on Earth would it do this? (Debug log)

      PC / SYMBOL           INSTRUCTION               DATA    A    X    SP   FLAGS   SRAM PAGE  006A                  HALT                       11     01   02   07    00         00  0069                  HALT                       11     01   02   07    00         00  0068                  HALT                       11     01   02   07    00         00  27E3                  JZ 27FDh                   00     01   02   04    81         00  2087                  RET                        00     01   02   04    81         03  2085                  MOV A, 01h                 00     01   02   06    81         00

      How did a short jump go from 27E5 to 0068, clear the flags and add 3 to SP in one fell swoop? (And it shouldn't have taken the jump anyway, it wanted to fall through.)


      Full disclosure: hand-built pod and cable. Code was free-running until it hit HALT, I hit Run a few more times so the debug buffer would show the HALTs. Here's a snippit from the .lst file just to be irritating:

      27E0: 7C 20 66 LCALL HIDReport2Ready    (0463) lcall  HIDReport2Ready  27E3: A0 19    JZ    0x27FD             (0464) jz            .return                                                             ; If not, just skip the rest and wait some more.                                          (0465)                                          (0466) ; First we see if we just finished a full report. If so, send a "full report finished" message this time.  27E5: 47 A6 FF TST   [0xA6],0xFF        (0467) tst           [switches_full_report_done],0xFF         ; See if ANY bits are set.  

      Designer 5.4, built for CY8C24794, running on CY8C24094 of course.

        • 1. Re: Code execution error (ICE Cube)

          Flash address 0x27fd does not exist. Probably stack return address destroyed.





          • 2. Re: Code execution error (ICE Cube)

            That's an odd assertion, Bob. How does flash address 0x27FD not exist?

            • 3. Re: Code execution error (ICE Cube)

              Sorry, mis-calculated. Still assuming clobbered stack.





              • 4. Re: Code execution error (ICE Cube)

                I checked that, but turns out it doesn't track. First, 27E3 is a valid return address (shown by the listing snippet) and is where the last RET in the trace returns to. (I didn't paste enough for you to see this, but I also verified that the code being returned from was legitimately executed from the call.) So as far as I can tell, it reached 27E3 exactly when it should have, and with its marbles intact. It's also in the stack exactly where you'd expect it, given the stack pointer values seen in the trace.


                Second, the stack doesn't show anything that would explain a return to 0068. I didn't post that before, it didn't actually seem relevant since the stack wasn't in operation, but I've pasted it below to satisfy any curiosity.


                But none of this involved the stack, or at least, it shouldn't have. JZ doesn't use the stack for anything. The fact is that a local jump to 27FD actually landed at 0068... the return had already been executed, so even a borked stack wouldn't explain the wayward jump. Also note that the stack pointer jumped from 4 to 7 and the flags all cleared. All during a simple JZ which, according to documentation, does not affect the flags at all, and can't jump more than 2048 bytes from the current PC+1.


                This was recorded via Run, not Step, so any interrupts during that execution would have been recorded in the debug trace.


                I did save a copy of the stack, so here it is.


                0300  1B 03 26 DB 27 E3 07 00  ..&Û'ã..


                0308  20 6F 85 C4 85 14 90 81   o.Ä....


                0310  91 01 73 81 69 81 81 81  ..s.i...


                0318  83 81 19 25 00 00 00 00  ...%....


                The rest of page 3 is all zeros.




                ...On a separate note, I find it both odd and annoying that this forum allows you to apply all kinds of nice formatting options to your post, and then throws it away when you click Submit. The code in my OP should have been all nicely lined up with a fixed-width font. I don't know what you're seeing, but I'm not seeing that. For me, the entire post shows in a sans-serif proportional font.




                ... Oh, fer cryin' out loud. I also find it egregious that this forum allowed me to complete the above post without warning me that I wasn't logged in. When I clicked Submit it brought me to a login page and threw away all I'd written. Fortunately I was able to hit Back and copy my post into the clipboard before completing the action. I hereby invoke the Principle of Least Astonishment, violations of which Cypress has a lot to answer for. *sigh* I'll just add this to the list. (At least the post was still in a buffer and I was able to retrieve it. Google Plus, for one, will throw away everything I wrote when a glitch happens.)

                • 5. Re: Code execution error (ICE Cube)

                  Oh now, see, there. The reply kept the fixed-width font (stack trace) and shows it to me. Why not on OP? Jeez, Cypress, you're not even consistent. I'll see if I can edit the post and... uh... fix it in post. Hah hah.

                  • 6. Re: Code execution error (ICE Cube)

                    Nope. Formatting shows in the edit screen but isn't applied. *sigh* I'll add it to the list.