On at 2023-09-03 20:31 +0200, C. Masloch via Freedos-user wrote:
On at 2023-08-26 16:30 +0200, C. Masloch via Freedos-user wrote:
Hello list,

I finished release 6 of lDebug (with a small L) today. This is my advanced 86-DOS debugger project based on FreeDOS Debug/X (in turn based on MS-DOS Debug), with some ideas from DR-DOS Debug. The duration since the prior release 5 is back to less than 6 months as opposed to the year between releases 4 and 5. Apart from the usual amount of bugfixes, there are some new features.

I just found a bug in the release 6 revision of lDebug. It makes the DIL command (Dump Interrupt chains with discovery of hidden chains through AMIS Interrupt Lists) not work properly. Depending on the prior use of (and left-over contents in) the auxiliary buffer, it could either seem to appear to work, it could display an error immediately on any hidden chain being detected (very common), or it could overflow the auxiliary buffer (only possible if hundreds and hundreds of interrupt entrypoints are found for one particular interrupt number).

It is unfortunate that this bug slipped by me into the release. I am considering whether to make a "minor" release about it. As a halfway remedy for now, I created a Script for lDebug that can patch an uncompressed executable (ldebugu.com, ldebugxu.com, lcdebugu.com, etc) to work around the bug. I have tested this patch on several different files. It is described some on my blog [1]. A copy of the patch itself is also hosted on our server [2]. The files that can be patched in the FreeDOS package are found in the package's directory tree at these positions:

Today I figured I could make an online patch in which the debugger that needs to be patched can, itself, run a Script for lDebug file to patch its own code. The patch result is the same as for the offline patch, however the online method works on any executable, including the compressed executables.

With the insight into the exact location of the code section (read from the read-only variables DCODE1SEG and DCODE1LEN), the online patch is even more reliable than the offline one. However, the detection of the correct offset used by the instructions to replace is a bit different as, at script run time, the init section is no longer accessible. Because the online debugger is already loaded, the patch can also patch in the exact allowable size of the auxiliary buffer (calculated as DAUXBUFLEN - #24 if this variable is available), as at this point this size cannot change after the control flow has left init behind. (The /A switch to init can be used to enlarge the buffer, which is also the cause of this bug to begin with.)

As an addition, this script will end in one of five different paths: Success, patch not needed (source patch already applied), patch not needed (binary patch is already applied or auxiliary buffer size is hardcoded as before introduction of the /A switch), attempt to run the script in Protected Mode (which is not supported), and other error.

Here is the full script:

=== snip patchl6.sld
if (dif & 800) then goto :nopm
; 205 <1> protectedmode equ 800h ; in (DPMI) protected mode
s dcode1seg:0 l dcode1len 88 65 04 80 65 05 FD 89 4D 06
; 2491 00001C74 886504 <1> mov byte [di + 4], ah ; store the multiplex number ; 2492 00001C77 806505FD <1> clropt [di + 4], 200h ; indicate it is claimed ; 2493 00001C7B 894D06 <1> mov word [di + 6], cx ; = how many list entries before ours,
;  2494                              <1>                                  ;  or 
= -1 if not from a list
;  2495 00001C7E E9B900              <1>  jmp .done
;  2496                              <1>
;  2497                              <1> @@:
; 2498 <1> ; ds:di -> second terminator (will be overwritten)
;  2499                              <1> %if _AUXBUFFSIZE == _AUXBUFFMAXSIZE
; 2500 <1> cmp di, _AUXBUFFSIZE - 8 * 3 ; enough for 1 entry + 2 terminators ?
;  2501                              <1> %else
; 2502 00001C81 3B3E[540A] <1> cmp di, word [auxbuff_current_size_minus_24]
;  2503                              <1> %endif
if (src != 1) then goto :error
r v0 := 2
if (byte [srs:sro + count] == EB) then goto :short
r v0 += 1
if (byte [srs:sro + count] != E9) then goto :error
:short
if (byte [srs:sro + count + v0] == 36) then goto :notneedednew
if (word [srs:sro + count + v0] == FF81) then goto :notneededold
if (word [srs:sro + count + v0] != 3E3B) then goto :error
r word [srs:sro + count + v0 + 2] .
s dcode1seg:0 l dcode1len range srs:sro + count + v0 l 4
if (src != 3) then goto :error
r v1 := 8 * #1024 + #16 - #24
if exists r dauxbuflen then r v1 := dauxbuflen - #24
a srs:sro0
 cmp di, (v1)
 .
a srs:sro1
 cmp di, (v1)
 .
a srs:sro2
 cmp di, (v1)
 .
@goto :eof

:nopm
; Cannot operate in PM
@goto :eof

:notneededold
; Patch not needed, auxiliary buffer has fixed size (or already patched)
@goto :eof

:notneedednew
; Patch not needed, comparison uses ss: segment override
@goto :eof

:error
; Error detected
=== snip

It can also be downloaded from our server [1].

The patch is applied by simply running a command like "y patchl6.sld" in the affected debugger. It will write the SRC (Search Result Count), SRS (Search Result Segment), SRO (Search Result Offsets), COUNT (search pattern length), AAS (Address Assembly Segment), AAO (Address Assembly Offset), and V0 and V1 variables.

Regards,
ecm


[1]: https://pushbx.org/ecm/test/20230904/patchl6.sld


_______________________________________________
Freedos-user mailing list
Freedos-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freedos-user

Reply via email to