tyler.durden/kernel
1
0
Fork 0
mirror of https://github.com/FDOS/kernel.git synced 2025-07-24 14:24:46 +02:00
Code Issues Packages Projects Releases Wiki Activity

Add Tom Ehlert's freeing of near f_node's work-around for problems with

Browse Source 
INT24 handling.


git-svn-id: https://svn.code.sf.net/p/freedos/svn/kernel/trunk@1317 6ac86273-5f31-0410-b378-82cca8765d1b
This commit is contained in:
Bart Oldeman 2007-05-15 17:50:22 +00:00
parent 1e98a86033
commit d99cfc38c1
1 changed files with 48 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
kernel/inthndlr.c
View File

@ -425,7 +425,29 @@ dispatch:
}
/* Clear carry by default for these functions */
/* We force clear the near fnodes,
On a normal int21h entry these are unused, so should already
be clear, thus this code is effectively redundant.
However when re-entering int21h, such as a device driver
(e.g. shsufdrv), the fnodes may contain the values of the
in progress call; as long as the driver/tsr causing the
reentrancey saves and restores the SDA across the call then
any changes we do (ie clearing these) should go unnoticed
and if the cause of the reentrancy does not save/restore the
SDA then the caller should expected the unexpected.
Failure to do this will at a minimal result in the extra
output indicating fnodes not 0 (and possibly clearing at end
of call) when the re-entered int21h call completes, but could
result in the re-entered call to fail with no free near fnodes
if both are already in use by the original int21h call (?bug 1879?).
See also PATCH TE 5 jul 04 explanation at end
*/
if (/*ErrorMode && */lr.AH > 0x0c && lr.AH != 0x30 && lr.AH != 0x59)
{
/*if (ErrorMode)*/ ErrorMode = 0;
fnode[0].f_count = 0; /* don't panic - THEY ARE unused !! */
fnode[1].f_count = 0;
}
/* Check for Ctrl-Break */
if (break_ena || (lr.AH >= 1 && lr.AH <= 5) || (lr.AH >= 8 && lr.AH <= 0x0b))
check_handle_break(&syscon);
@ -1545,6 +1567,31 @@ exit_dispatch:
r->ES = lr.ES;
real_exit:;
/* PATCH !! TE 5 JUL 04
what happened:
Application does FindFirst("I:\*.*");
this fails, and causes Int24
this sets ErrorMode, and calls Int24
Application decides NOT to return to DOS,
but instead pop the stack and return to itself
(this is legal; see RBIL/INT 24 description
a) now the alloc()'ed fnode[0] never gets free()'ed
b) errormode NEVER gets set back to 0 unyil exit()
I have NO idea how real DOS handles this;
the appended patch cures the worst symptoms
*/
if (fnode[0].f_count != 0 ||
fnode[1].f_count != 0 )
{
if (ErrorMode == 0)
put_string("near_fnodes not 0"); /* panic ?? */
fnode[0].f_count = 0; /* don't panic - THEY ARE unused !! */
fnode[1].f_count = 0;
}
/* PATCH !! END TE 5 JUL 04 */
#ifdef DEBUG
if (bDumpRegs)
{