Date:	Fri, 8 Jan 93 10:38:36 JST
From:	Makoto Kobayashi <na70103 AT cc DOT mie-u DOT ac DOT jp>
To:	djgpp AT sun DOT soe DOT clarkson DOT edu
Subject:	paging mechanism trouble in go32

I find a trouble in go32. 
I show you a sample program for the trouble. Please compile 
it with djgpp, and run with go32 (not debug32).

/* this is file go32chk.c */
#include <stdio.h>
char a[24576]={1};
void main()
{	
    printf ("1: a[12288]= %d\n",a[12288]);
    a[12288]=-1;
    printf ("2: a[12288]= %d\n",a[12288]);

    system("echo child process");

    printf ("   a[8192]= %d\n",a[8192]);
    printf ("3: a[12288]= %d\n",a[12288]);
}
/* end of file */

     > gcc go32chk.c
     > go32 a.out

Of cource the expected result is:

1: a[12288]= 0
2: a[12288]= -1
child process
   a[8192]= 0
3: a[12288]= -1

but some results may be:

1: a[12288]= 0
2: a[12288]= -1
child process
   a[8192]= 0
3: a[12288]= 0

I think that this is due to a bug in the paging mechanism.
This bug is rare but I feel it is very dangerous...

Patch for go32 1.09 to fix this trouble follows.
Please try and test it.

*** paging.c0	Sun Nov  1 07:54:38 1992
--- paging.c	Fri Jan  8 09:27:46 1993
***************
*** 110,116 ****
  word32 far *vcpi_pt = 0;
  extern word32 graphics_pt_lin;
  char paging_buffer[4096*MAX_PAGING_NUM];
- static int paged_out_something = 0;
  
  word32 screen_primary, screen_secondary;
  
--- 110,115 ----
***************
*** 567,573 ****
      return 0;
    }
    count = MAX_PAGING_NUM;
!   if ((count > mem_avail/4) || (paged_out_something))
      count = 1;
    if (pti + count > 1024)
      count = 1024 - pti;
--- 566,572 ----
      return 0;
    }
    count = MAX_PAGING_NUM;
!   if (count > mem_avail/4)
      count = 1;
    if (pti + count > 1024)
      count = 1024 - pti;
***************
*** 582,587 ****
--- 581,588 ----
    {
      if (pt[pti] & PT_P)
        break;
+     if (eaddr != -1 && (pt[pti] & PT_I) != 0) 
+       break;  /* EastWind 1993 */
      dblock = pt[pti] >> 12;
      pn = valloc(VA_1M);
      pt[pti] &= 0xfffL & ~(word32)(PT_A | PT_D);
***************
*** 629,635 ****
        }
        pt[pti] |= PT_I;
      }
-     if (paged_out_something)
        break;
    }
    if (eaddr != -1)
--- 630,635 ----
***************
*** 680,687 ****
  #if TOPLINEINFO
    old_status = update_status('>' | 0x0a00, 79);
  #endif
-   if (!fInPageOutEverything)
-     paged_out_something = 1;
    start_pdi = last_po_pdi;
    start_pti = last_po_pti;
    if (where == VA_640)
--- 680,685 ----

The variable "paged_out_something" seemed to be expected to 
fix the paging problem, but it was imperfect.  I took away 
it.

At last, I want to hear DJ's view.
Do you think multi-page-loading mechanism is effective?
(Should MAX_PAGING_NUM be 4 ?)

Thank you very much.

                     Faculty of Eng., Mi'e Univ., Tsu, Japan.
                     na70103 AT mips DOT cc DOT mie-u DOT ac DOT jp    Makoto Kobayashi

- Raw text -