On Tuesday 21 June 2005 04:00 pm, Andrew Gallatin wrote: > John Baldwin writes: > > On Sunday 29 May 2005 01:50 pm, Kris Kennaway wrote: > > > On Tue, May 10, 2005 at 03:36:36PM -0700, Kris Kennaway wrote: > > > > Got this on a dual amd64 with 8GB RAM running 6.0 from last week: > > > > > > > > Fatal trap 12: page fault while in kernel mode > > > > cpuid = 1; apic id = 01 > > > > fault virtual address = 0xffffffffa9cdc000 > > > > fault code = supervisor read, page not present > > > > instruction pointer = 0x8:0xffffffff8037759f > > > > stack pointer = 0x10:0xffffffffba1637d0 > > > > frame pointer = 0x10:0xffffffffba163820 > > > > code segment = base 0x0, limit 0xfffff, type 0x1b > > > > = DPL 0, pres 1, long 1, def32 0, gran 1 > > > > processor eflags = interrupt enabled, resume, IOPL = 0 > > > > current process = 52247 (sh) > > > > [thread pid 52247 tid 100149 ] > > > > Stopped at exec_copyout_strings+0x12f: > > > > db> wh > > > > Tracing pid 52247 tid 100149 td 0xffffff016e5724c0 > > > > exec_copyout_strings() at exec_copyout_strings+0x12f > > > > do_execve() at do_execve+0x39a > > > > kern_execve() at kern_execve+0xab > > > > execve() at execve+0x49 > > > > syscall() at syscall+0x382 > > > > Xfast_syscall() at Xfast_syscall+0xa8 > > > > --- syscall (59, FreeBSD ELF64, execve), rip = 0x80090622c, rsp = > > > > 0x7fffffffe058, rbp = 0xffffffff --- db> > > > > > > I've got this panic twice more since. > > > > Do you have a kernel.debug? Can you do 'list > > *exec_copyout_strings+0x12f'? I think I've seen reports of the > > linux32_exec_copyout_strings() having a similar fault as well on amd64. > > I just got this on my freshly installed UP, 512MB athlon64. For me, > its 100% reproducable when running a cross-compiler built on > FreeBSD-4. > > (kgdb) p *imgp->args > $33 = { > buf = 0xffffffff90ba3000 <Address 0xffffffff90ba3000 out of bounds>, > begin_argv = 0xffffffff90ba3000 <Address 0xffffffff90ba3000 out of > bounds>, begin_envv = 0xffffffff90ba313d <Address 0xffffffff90ba313d out of > bounds>, endp = 0xffffffff90ba389f <Address 0xffffffff90ba389f out of > bounds>, fname = 0xffffffff90be3000 > "/home/gallatin/lanaitools/intel_FreeBSD/lib/gcc-lib/lanai/2.95.2..1.6/cc1" >, stringspace = 259937, > argc = 23, > envc = 46 > } > > I'm puzzled. fname seems to be buf+ARGV_MAX, so its not > like something randomly scribbled on this memory. > > In the debugger, the memory just below buf+ARGV_MAX seems to be > unmapped. But we've done copyins in freebsd32_exec_copyin_args(), > otherwise endp would not have been advanced. So we've written to this > memory. > > It is almost like somebody freed buf through buf + 262144. I think I figured it out. sobomax_at_ changed how much memory exec_copyin_args() and exec_free_args() allocated and freed without updating freebsd32_exec_copyin_args() and linux_exec_copyin_args(), so more memory was freed than was allocated which would free memory out from other execs. Patch is below. Let me know if it fixes the problem. Index: amd64/linux32/linux32_machdep.c =================================================================== RCS file: /usr/cvs/src/sys/amd64/linux32/linux32_machdep.c,v retrieving revision 1.9 diff -u -r1.9 linux32_machdep.c --- amd64/linux32/linux32_machdep.c 5 Apr 2005 15:28:06 -0000 1.9 +++ amd64/linux32/linux32_machdep.c 22 Jun 2005 14:26:03 -0000 _at__at_ -113,7 +113,8 _at__at_ * Allocate temporary demand zeroed space for argument and * environment strings */ - args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX); + args->buf = (char *) kmem_alloc_wait(exec_map, + PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); if (args->buf == NULL) return (ENOMEM); args->begin_argv = args->buf; Index: compat/freebsd32/freebsd32_misc.c =================================================================== RCS file: /usr/cvs/src/sys/compat/freebsd32/freebsd32_misc.c,v retrieving revision 1.35 diff -u -r1.35 freebsd32_misc.c --- compat/freebsd32/freebsd32_misc.c 11 Jun 2005 14:58:20 -0000 1.35 +++ compat/freebsd32/freebsd32_misc.c 22 Jun 2005 14:26:11 -0000 _at__at_ -237,7 +237,8 _at__at_ * Allocate temporary demand zeroed space for argument and * environment strings */ - args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX); + args->buf = (char *) kmem_alloc_wait(exec_map, + PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); if (args->buf == NULL) return (ENOMEM); args->begin_argv = args->buf; -- John Baldwin <jhb_at_FreeBSD.org> <>< http://www.FreeBSD.org/~jhb/ "Power Users Use the Power to Serve" = http://www.FreeBSD.orgReceived on Wed Jun 22 2005 - 12:31:37 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:37 UTC