Re: [myrinet] active_mm

=> With 2.4.0-test10 and 1.4pre43 I found I was getting an oops every time
=> a process which had /dev/gm* opened would exit.  Here's the fix:
=>         {
=> -               pgd = pgd_offset(gm_current->mm, addr);
=> +               pgd = pgd_offset(gm_current->active_mm, addr);
=>         }
=> The problem is that an exiting process has already entered lazy mm state
=> before exit_files is called to close /dev/gm*, so access through current->mm
=> is invalid.  This is x86 smp no highmem, hopefully similar for other arches
=> and configs.


Thanks! We'll code it as the following to allow the code to
still compile under linux-2.2 and linux-2.0.



diff -u -r1.182 gm_arch.c
--- drivers/linux/gm/gm_arch.c  2000/12/20 23:42:08     1.182
+++ drivers/linux/gm/gm_arch.c  2000/12/21 22:38:15
@@ -623,7 +623,11 @@
 #warning **************************************************************
 #endif
        {
-               pgd = pgd_offset(gm_current->mm, addr);
+#if LINUX_XX >= 24
+               pgd = pgd_offset(gm_current->active_mm, addr);
+#else
+               pgd = pgd_offset(gm_current->mm, addr); 
+#endif
        }
        GM_PRINT (GM_DEBUG_KVIRT_TO_PHYS,
                          ("kvirt_to_phys pgd = 0x%lx\n", *pgd));