RE: [ECOS] Network driver problem only with larger programs (cache issues?)

["Joe Porthouse" <jporthouse at toptech dot com>]

Ok, I'm positive I'm going nuts now....  Code seems good, if I can only get
the processor to correctly execute it.

Started seeing the code lock up after 10-30 minutes without a JTAG attached
so I started to look further into the possible cache issues.

Attaching the JTAG and again the code can not make it through the
cyg_do_net_init().  Same problem, a jump to address 0x0100330 from a place
in the code that it should not occur.

Next I modified the platform setup code to never enable the ICache and
DCache.  Code runs slower, but same problem.

Next modified the MMU tables to make the FLASH and SDRAM _UNCACHED and
_NOBUF.  Still the same problem.

In the trace I noticed DABORT, UNDEF, and PABORT states.  I did not know if
this was a byproduct of setting break points (also saw FIRQs when I don't
use them).  To help rule out memory timing issues, I modified the platform
setup to including the max timing parameters on all memory and I/O selects
(0x7ff0, 0x7ff4, 0x7ffc, etc).  Still the same problem.

Upon a closer look at the trace I see the jump is now going to 0x010032c
(data not code) from a slightly different place in the code, but further
into the cyg_do_net_init() routine.

Below in the trace with a breakpoint set at 0x0010032c.  Notice that just
before -00018 the "BL cyg_callout_init" instruction is not executed and
instead execution simply falls through to the next address.  I also see
UNDEF, PABORT, DABORT and IBRA states.  Are these my problem?  What else
should I do to try to eliminate them?

Any help would be greatly appreciated.

-00020 IBRA   07  001003F0 000FFAAC 
              000FFAAC e91ba870  LDMDB     R11,{R4-R6,R11,SP,PC}
              timeout.c Line 274 (cyg_timeout)
              274               for (e = _timeouts, i = 0;  i < NTIMEOUTS;
i++, e++) {
              001003F0 e59f4060  LDR       R4,00100458
              timeout.c Line 273 (cyg_timeout)
              273               stamp = 0;  // Assume no slots available
              001003F4 e3a05000  MOV       R5,#00000000
              001003F8 e1a06000  MOV       R6,R0
              001003FC e1a02005  MOV       R2,R5
              timeout.c Line 275 (cyg_timeout)
              275               if ((e->flags & CALLOUT_PENDING) == 0) {
              00100400 e5943014  LDR       R3,[R4,#014]
              00100404 e2822001  ADD       R2,R2,#00000001
              00100408 e3130004  TST       R3,#00000004
-00019 DBRA   02  00000000 0010040C 
              0010040C 0a000006  BEQ       cyg_timeout+58
              timeout.c Line 277 (cyg_timeout)
              277               callout_init(e);
              0010042C e1a00004  MOV       R0,R4
              00100430 eb000009  BL        cyg_callout_init
-00018 DBRA   03  00000000 00000000 
-00017        00  00000000 00000000 
-00016 UNDEF  00  00000000 00000000 
-00015        00  00000000 00000000 
-00014 PABORT 00  00000000 00000000 
-00013 IBRA   00  00100438 00100434 
              timeout.c Line 278 (cyg_timeout)
              278               e->flags = CALLOUT_LOCAL;
              00100434 e3a0c001  MOV       R12,#00000001
-00012 DBRA   03  00000000 00000000 
-00011        00  00000000 00000000 
-00010 UNDEF  00  00000000 00000000 
-00009        00  00000000 00000000 
-00008 PABORT 00  00000000 00000000 
-00007 IBRA   00  0010043C 00100438 
              timeout.c Line 279 (cyg_timeout)
              279               callout_reset(e, delta, fun, arg);
              00100438 e1a01007  MOV       R1,R7
-00006 DBRA   03  00000000 00000000 
-00005        00  00000000 00000000 
-00004 UNDEF  00  00000000 00000000 
-00003        00  00000000 00000000 
-00002 DABORT 00  00000000 00000000 
-00001 IBRA   00  00100440 0010043C 
              0010043C e1a02008  MOV       R2,R8
  TRIG DBRA   00  00000000 00000000
PC is now 0x0010032c!!!

Thanks,
Joe Porthouse
Toptech Systems, Inc.
Longwood, FL 32750




-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss