[andrewbird]

After loading country.sys the Upcase function address was being zeroed out rather than updated correctly. Traced to a NEAR/FAR pointer mismatch. Discovered whilst looking for the cause of #269

Using the Dosemu2 test suite command test/test_dosemu.py FRDOSGITTestCase.test_get_country

Before:

INFO: sym_currency = '0x9c,0x00,0x00,0x00,0x00'
INFO: sep_thousand = ','
INFO: sep_decimal = '.'
INFO: sep_date = '/'
INFO: sep_time = ':'
INFO: fmt_currency = '0'
INFO: num_curdigits = '2'
INFO: fmt_time = '1'
INFO: func_upcase = '0000:0000'
INFO: sep_datalist = ','
<crash>

After:

INFO: sym_currency = '0x9c,0x00,0x00,0x00,0x00'
INFO: sep_thousand = ','
INFO: sep_decimal = '.'
INFO: sep_date = '/'
INFO: sep_time = ':'
INFO: fmt_currency = '0'
INFO: num_curdigits = '2'
INFO: fmt_time = '1'
INFO: func_upcase = '00D9:1338'
INFO: sep_datalist = ','
INFO: upcased_char = 'B'

Tested with Ia16-gcc and Watcom compilers only.

[ecm-pushbx]

LGTM, assuming you tested this. Although, your description of the problem, "a NEAR/FAR pointer mismatch", doesn't account for the +2 that you had to add.

[ecm-pushbx]

If it was previously handled as a wrong + near pointer, did the write corrupt memory in some other segment?

[andrewbird]

did the write corrupt memory in some other segment?

Well it was writing somewhere so quite possibly, but I didn't find where only that it didn't change what it should have.

Tested fine with the dosemu2 test_get_country https://github.com/dosemu2/dosemu2/blob/devel/test/func_country.py. It was a little late last night for me, so I'll run the full test suite now and check nothing new breaks, but I doubt it as that code path is only run if country.sys is loaded, and I don't do that anywhere else in the tests.

[ecm-pushbx]

The patch breaks gcc-ia16 build exactly in the way the bug was originally reported. 21.6501.bx=FFFF.dx=FFFF returns the function pointer as 0:0 and VC v4.05 crashes.

[ecm-pushbx]

Prior to patch, gcc-ia16 build works for me.

[andrewbird]

How strange, both gcc-ia16 and watcom fail without the patch for me and pass with it (but I only tested with int21/38h). So it looks like more work to do then, I'll convert to draft. Thanks!

[andrewbird]

@ecm-pushbx I thought I'd better write some more tests to see if I can replicate your experience, however I don't see where the upcase function is supposed to be returned in int21/6501. Here's the RBIL https://github.com/LoopZ/TheList/blob/main/source/Interrupt%20List/INT%2021%20DOS%20Function%20Calls/INT%202165%20DOS%2033%20GET%20EXTENDED%20COUNTRY%20INFORMATION.txt . What tool did you use to see the missing value?

[ecm-pushbx]

As to how to query the data using 21.65, see https://github.com/LoopZ/TheList/blob/main/source/Interrupt%20List/INT%2021%20DOS%20Function%20Calls/INT%202165%20DOS%2033%20GET%20EXTENDED%20COUNTRY%20INFORMATION.txt#L44 - the 21.38 data is included at offset 7 for info ID = 1.

I placed everything in a directory at https://pushbx.org/ecm/test/20261030/

This includes the kernel file with your patch, built using the gcc-ia16 toolchain on our server.

Here's the log showing how to query the internationalisation data using both 21.6501 and 21.38:

-t=100
AX=6501 BX=0031 CX=0200 DX=0400 SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=0103 NV UP EI PL ZR NA PE NC
2A2F:0103 BAFFFF            mov     dx, FFFF
-
AX=6501 BX=0031 CX=0200 DX=FFFF SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=0106 NV UP EI PL ZR NA PE NC
2A2F:0106 BBFFFF            mov     bx, FFFF
-
AX=6501 BX=FFFF CX=0200 DX=FFFF SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=0109 NV UP EI PL ZR NA PE NC
2A2F:0109 B90002            mov     cx, 0200
-
AX=6501 BX=FFFF CX=0200 DX=FFFF SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=010C NV UP EI PL ZR NA PE NC
2A2F:010C BF0002            mov     di, 0200
-
AX=6501 BX=FFFF CX=0200 DX=FFFF SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=010F NV UP EI PL ZR NA PE NC
2A2F:010F F9                stc
-
AX=6501 BX=FFFF CX=0200 DX=FFFF SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=0110 NV UP EI PL ZR NA PE CY
2A2F:0110 CD21              int     21
-
AX=6501 BX=FFFF CX=0200 DX=FFFF SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=0112 NV UP EI PL ZR NA PE NC
2A2F:0112 CC                int3
-
Unexpected breakpoint interrupt
AX=6501 BX=FFFF CX=0200 DX=FFFF SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=0113 NV UP EI PL ZR NA PE NC
2A2F:0113 B80038            mov     ax, 3800
-d es:di l 40
2A2F:0200  01 1C 00 31 00 5A 03 01-00 D5 00 00 00 00 2E 00 ...1.Z..........
2A2F:0210  2C 00 2E 00 3A 00 03 02-01 00 00 00 00 2C 00 20 ,...:........,.
2A2F:0220  20 20 20 20 20 20 20 20-20 20 20 20 20 20 20 20
2A2F:0230  20 20 20 20 20 20 20 20-20 20 20 20 20 20 20 20
-

-t
AX=3800 BX=FFFF CX=0200 DX=FFFF SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=0116 NV UP EI PL ZR NA PE NC
2A2F:0116 BA0004            mov     dx, 0400
-
AX=3800 BX=FFFF CX=0200 DX=0400 SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=0119 NV UP EI PL ZR NA PE NC
2A2F:0119 F9                stc
-
AX=3800 BX=FFFF CX=0200 DX=0400 SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=011A NV UP EI PL ZR NA PE CY
2A2F:011A CD21              int     21
-
AX=0031 BX=0031 CX=0200 DX=0400 SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=011C NV UP EI PL ZR NA PE NC
2A2F:011C 90                nop
-
AX=0031 BX=0031 CX=0200 DX=0400 SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=011D NV UP EI PL ZR NA PE NC
2A2F:011D CC                int3
-
Unexpected breakpoint interrupt
AX=0031 BX=0031 CX=0200 DX=0400 SP=FFFE BP=0000 SI=0000 DI=0200
DS=2A2F ES=2A2F SS=2A2F CS=2A2F IP=011E NV UP EI PL ZR NA PE NC
2A2F:011E 90                nop
-d ds:dx l 40
2A2F:0400  01 00 D5 00 00 00 00 2E-00 2C 00 2E 00 3A 00 03 .........,...:..
2A2F:0410  02 01 00 00 00 00 2C 00-20 20 20 20 20 20 20 20 ......,.
2A2F:0420  20 20 20 20 20 20 20 20-20 20 20 20 20 20 20 20
2A2F:0430  20 20 20 20 20 20 20 20-20 20 20 20 20 20 20 20
-

And here's the kernel patch I applied (failing), which should match yours:

ulukai2:~/proj/kernel$ ls -lgG bin/kernel.sys kernel/config.c
-rw-r--r-- 1 81652 Oct 30 10:24 bin/kernel.sys
-rw-r--r-- 1 76740 Oct 30 10:24 kernel/config.c
ulukai2:~/proj/kernel$ git diff
diff --git a/kernel/config.c b/kernel/config.c
index e3470f6..63febef 100644
--- a/kernel/config.c
+++ b/kernel/config.c
@@ -1664,7 +1664,7 @@ err:printf("%s has invalid format\n", filename);
                   /* skip length ^*/  subf_data.length);
           if (hdr[i].id == 1) {
               /* fixup user callable address in case we overwrote it */
-              ((struct CountrySpecificInfo *)ptable)->CharMapFn = CharMapFn;
+              ((struct CountrySpecificInfo FAR *)ptable + 2)->CharMapFn = CharMapFn;
           }
       }
     }
ulukai2:~/proj/kernel$

[ecm-pushbx]

Argh! I forgot parens around the addition. Retesting now.

[ecm-pushbx]

Okay, it ran fine with the proper patch. That still leaves the mystery of how it could have worked prior to this patch.

[ecm-pushbx]

Aaaaand I cannot reproduce the lack of failure on an earlier revision - 12f4f7d No idea how I got the wrong result (no failure) earlier.

[ecm-pushbx]

So yeah, LGTM and tested every which way now. Proper patch applied rather than a fool trying to manually copy it into her repo makes this work as expected.

[andrewbird]

@ecm-pushbx Thanks for looking at it again, and for your info about int21/6501 too. I'm going to extend the dosemu2 test for 6501 anyway now, so it'll be good to get confirmation that way as well.

[andrewbird]

So I'm happy that the int21/6501 returns the same data. I did notice a problem with the 6501 specific fields and CX register regarding sizes (they are different from MS and OpenDOS), but that's a subject for another issue/PR I think. Thanks again for your help!