Re: [hatari-devel] Most wanted debugger/profiler feature or convenience? (DSP)

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On torstai 11 huhtikuu 2013, Laurent Sallafranque wrote:
> I did first implement these options for DSP debugging itself (when I
> used to code it) and then to help me finding bugs in the DSP (by looking
> at the exchanges between the 68030 and the DSP).
> Of course, if someone has another needs, just ask and I'll see what I
> can do.
> 
> Actually, I use the dsp_host_command and the dsp_host_interface commands
> (nearly never the SSI actually).

Attached is a patch to do the ring buffer thing, but I don't know
what information would be relevant to show.

Currently it looks like this:
---
> info dsp 
DSP core information:
- SSH stack: 0000 0000 007b 00b3 00b9 00f2 0162 014e 01e1 01cf 01cb 0000 
0000 0000 0000 0000
- SSL stack: 0000 0000 0010 8058 00ad 8055 0008 8051 8050 0001 8050 0000 
0000 0000 0000 0000
- Interrupt IPL: 0003 0003 0003 0003 0003 0002 0002 0002 ffff ffff ffff ffff
- Pending ints:  0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
- Hostport: 80 12 06 0f 00 00 00 00 00 00 07 05
- Hostport IO history, from newest item to oldest (R=read,W=write,m=multi):
  05 (W/m) 07 (W/m) 00 (W/m) 00 (W) 06 (R) 02 (W/m) 07 (W/m) 00 (W/m) 00 (W) 
06 (R) ff (W/m) 06 (W/m)
---

As you can see, DSP core's hostport array seems to contain some of
same data bytes as what were stored in the hostport IO history ring-buffer.
What this (already shown) dsp_core.hostport array content is?


	- Eero

> 
> Best regards
> Laurent
> 
> Le 11/04/2013 21:56, Eero Tamminen a écrit :
> > Hi,
> > 
> > On torstai 11 huhtikuu 2013, Douglas Little wrote:
> >> Making it deeper can't solve problems where a block is several kwords
> >> in size and the fault happened near the start, and surfaced near the
> >> end. Nobody would search through that much history anyway, unless the
> >> debugger did sanity checks on the history itself (which is another
> >> possibility - some common faults are detectable)
> > 
> > Note that Hatari has already quite a few options for tracing DSP
> > activity: $ hatari --trace help 2>&1 | grep dsp
> > 
> >    dsp_host_interface
> >    dsp_host_command
> >    dsp_host_ssi
> >    dsp_interrupt
> >    dsp_disasm
> >    dsp_disasm_reg
> >    dsp_disasm_mem
> >    dsp_state
> >    dsp_all
> >    dsp_symbols
> > 
> > For example for the freezing Stocasto demo:
> > 	--trace dsp_host_interface,dsp_host_command
> > 
> > Will end at this:
> > -----------
> > Dsp: (Host->DSP): Direct Transfer 0x0006ff
> > Dsp: (Host->DSP): Dsp HRDF set
> > Dsp: (Host->DSP): Dsp HRDF cleared
> > Dsp: (Host->DSP): Direct Transfer 0x000702
> > Dsp: (Host->DSP): Dsp HRDF set
> > Dsp: (Host->DSP): Dsp HRDF cleared
> > Dsp: (Host->DSP): Direct Transfer 0x000705
> > Dsp: (Host->DSP): Dsp HRDF set
> > Dsp: (Host->DSP): Dsp HRDF cleared
> > -----------
> > 
> > (Whereas SSI tracing output didn't end and was *much* more verbose.)
> > 
> > Is that enough for what you were looking for, or what kind
> > of output you were wishing for?
> > 
> >>> How short?  Could it be fixed size (e.g. 20), or does it need to
> >>> be configurable?
> >> 
> >> Good question. I think something like 20 would be enough to be useful.
> > 
> > In that case I guess it could be added to the "info dsp" command
> > (implemented in falcon/dsp.c) I think, with a small ring buffer.
> > 
> > Laurent, any comments on this?
> > 
> > 	- Eero

diff -r 0227a8a7dd35 src/falcon/dsp.c
--- a/src/falcon/dsp.c	Wed Apr 10 18:32:56 2013 +0300
+++ b/src/falcon/dsp.c	Thu Apr 11 23:53:46 2013 +0300
@@ -64,6 +64,16 @@
 	"", "", "", "", "", "", "BCR", "IPR"	
 };
 
+typedef struct {
+	Uint8 value;
+	bool write;
+	bool multi;
+} host_io_t;
+
+/* ring buffer for extra debugger info */
+static int host_io_idx;
+static host_io_t host_io_buffer[ARRAYSIZE(dsp_core.hostport)];
+
 
 /**
  * Trigger HREQ interrupt at the host CPU.
@@ -405,6 +415,14 @@
 	return dsp_memdump_upper+1;
 }
 
+static void add_host_io(Uint8 value, bool multi, bool write)
+{
+	host_io_buffer[host_io_idx].value = value;
+	host_io_buffer[host_io_idx].multi = multi;
+	host_io_buffer[host_io_idx].write = write;
+	host_io_idx = (host_io_idx + 1) % ARRAYSIZE(host_io_buffer);
+}
+
 /**
  * Show information on DSP core state which isn't
  * shown by any of the other commands (dd, dm, dr).
@@ -442,6 +460,18 @@
 		fprintf(stderr, " %02x", dsp_core.hostport[i]);
 	}
 	fputs("\n", stderr);
+
+	fprintf(stderr, "- Hostport IO history, from newest item to oldest (R=read,W=write,m=multi):\n ");
+	for (i = 0; i < ARRAYSIZE(host_io_buffer); i++) {
+		if (--host_io_idx < 0) {
+			host_io_idx += ARRAYSIZE(host_io_buffer);
+		}
+		fprintf(stderr, " %02x (%c%s)",
+			host_io_buffer[host_io_idx].value,
+			host_io_buffer[host_io_idx].write ? 'W' : 'R',
+			host_io_buffer[host_io_idx].multi ? "/m": "");
+	}
+	fputs("\n", stderr);
 #endif
 }
 
@@ -759,6 +789,7 @@
 	{
 #if ENABLE_DSP_EMU
 		value = dsp_core_read_host(addr-DSP_HW_OFFSET);
+		add_host_io(value, multi_access, false);
 		if (multi_access == true)
 			M68000_AddCycles(4);
 		multi_access = true;
@@ -788,6 +819,7 @@
 		value = IoMem_ReadByte(addr);
 		Dprintf(("HWput_b(0x%08x,0x%02x) at 0x%08x\n", addr, value, m68k_getpc()));
 #if ENABLE_DSP_EMU
+		add_host_io(value, multi_access, true);
 		dsp_core_write_host(addr-DSP_HW_OFFSET, value);
 		if (multi_access == true)
 			M68000_AddCycles(4);