>Number:         10653
>Category:       kern
>Synopsis:       sigreturn can return nonsensical values with x86 Linux emulation
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Jul 21 14:29:00 PDT 2000
>Closed-Date:
>Last-Modified:
>Originator:     tls@cs.stevens-tech.edu
>Release:        NetBSD-1.5_ALPHA as of 20000711
>Organization:
	Computer Science Department, Stevens Institute of Technology
>Environment:
System: NetBSD breakup-bock.cs.stevens-tech.edu 1.5_ALPHA NetBSD 1.5_ALPHA (BOCK) #0: Tue Jul 11 18:29:15 EDT 2000 root@amstel.cs.stevens-tech.edu:/usr/src/sys/arch/i386/compile/BOCK i386


>Description:
	While attempting to run the PVS theorem prover, which includes a
	modified Allegro Common Lisp binary for Linux, the interpreter
	failed with the evident belief that it'd received a SIGEMT.

	Investigation with ktrace/kdump revealed that in fact the process
	had received a SIGBUS, which *might* be a normal consequence of
	its garbage collection algorithm; we are investigating this with
	SRI and Allegro currently.  In any event, the kdump output reveals
	an abnormal return from sigreturn(2) with a nonsensical error
	value which does not appear to be present in the code (-2).

	Conclusion: *something* is hosed with Linux emulation signal
	handling, whether the registration of handlers or the actual
	delivery of signals, in 1.5_ALPHA on the x86.  Here is the kdump
	output:

   562 pvs-allegro5.0 EMUL  "linux"
   562 pvs-allegro5.0 RET   select 1
   562 pvs-allegro5.0 CALL  read(0,0x20118855,0x200)
   562 pvs-allegro5.0 GIO   fd 0 read 8 bytes
       "(grind)
       "
   562 pvs-allegro5.0 RET   read 8
   562 pvs-allegro5.0 CALL  times(0xbfbf61e8)
   562 pvs-allegro5.0 RET   times 33063190/0x1f88116
   562 pvs-allegro5.0 CALL  times(0xbfbf5c38)
   562 pvs-allegro5.0 RET   times 33063193/0x1f88119
   562 pvs-allegro5.0 CALL  times(0xbfbf5c38)
   562 pvs-allegro5.0 RET   times 33063193/0x1f88119
   562 pvs-allegro5.0 CALL  times(0xbfbf5d38)
   562 pvs-allegro5.0 RET   times 33063194/0x1f8811a
   562 pvs-allegro5.0 CALL  times(0xbfbf5d38)
   562 pvs-allegro5.0 RET   times 33063194/0x1f8811a
   562 pvs-allegro5.0 CALL  brk(0x806b000)
   562 pvs-allegro5.0 RET   brk 134656000/0x806b000
   562 pvs-allegro5.0 CALL  old_mmap(0xbfbf5cb4)
   562 pvs-allegro5.0 RET   old_mmap 1210507264/0x4826e000
   562 pvs-allegro5.0 CALL  rt_sigprocmask(0,0,0x4824c1d0,0x8)
   562 pvs-allegro5.0 RET   rt_sigprocmask 0
   562 pvs-allegro5.0 CALL  rt_sigaction(0x2,0xbfbf5b28,0xbfbf5a9c,0x8)
   562 pvs-allegro5.0 RET   rt_sigaction 0
   562 pvs-allegro5.0 CALL  rt_sigaction(0x2,0xbfbf5c3c,0,0x8)
   562 pvs-allegro5.0 RET   rt_sigaction 0
   562 pvs-allegro5.0 PSIG  SIGBUS caught handler=0x481c4318 mask=() code=0x4
   562 pvs-allegro5.0 CALL  sigreturn(0xbfbf5e7c)
   562 pvs-allegro5.0 RET   sigreturn -1 errno -2 No such file or directory
   562 pvs-allegro5.0 CALL  write(0x1,0x20118a5c,0x6a)
   562 pvs-allegro5.0 GIO   fd 1 wrote 106 bytes
       "Error: Received signal number 7 (EMT instruction)
	  [condition type: SYNCHRONOUS-OPERATING-SYSTEM-SIGNAL]
       "
   562 pvs-allegro5.0 RET   write 106/0x6a
   562 pvs-allegro5.0 CALL  write(0x1,0x20118a5c,0x2b)
   562 pvs-allegro5.0 GIO   fd 1 wrote 43 bytes
       "
	Restart actions (select using :continue):
       "
   562 pvs-allegro5.0 RET   write 43/0x2b
   562 pvs-allegro5.0 CALL  write(0x1,0x20118a5c,0x4)
   562 pvs-allegro5.0 GIO   fd 1 wrote 4 bytes
       " 0: "
   562 pvs-allegro5.0 RET   write 4
   562 pvs-allegro5.0 CALL  write(0x1,0x20118a5c,0x28)
   562 pvs-allegro5.0 GIO   fd 1 wrote 40 bytes
       "Return to Top Level (an "abort" restart)"
   562 pvs-allegro5.0 RET   write 40/0x28
   562 pvs-allegro5.0 CALL  select(0x1,0xbfbf5770,0,0,0xbfbf57f0)
   562 pvs-allegro5.0 RET   select 0
   562 pvs-allegro5.0 CALL  write(0x1,0x20118a5c,0x1)
   562 pvs-allegro5.0 GIO   fd 1 wrote 1 bytes
       "
       "
   562 pvs-allegro5.0 RET   write 1
   562 pvs-allegro5.0 CALL  write(0x1,0x20118a5c,0xd)
   562 pvs-allegro5.0 GIO   fd 1 wrote 13 bytes
       "[1] PVS(20): "
   562 pvs-allegro5.0 RET   write 13/0xd
   562 pvs-allegro5.0 CALL  select(0x1,0xbfbf5548,0,0,0)

>How-To-Repeat:
	To repeat this problem, you'll probably need a copy of the PVS
	distribution or another source of the Allegro interpreter binary.

	To get PVS, follow the links from http://pvs.csl.sri.com/announce.html
	to the download directory on the SRI FTP server.  You'll want the
	"Redhat Linux 5" version.  You'll then need to frob the "pvs" shell
	script so it thinks NetBSD is a version of Linux, and make symlinks
	named "libdl.so" and "libm.so" in your /emul/linux/lib because for
	some reason the Allegro interpreter has no shared library version
	numbers in the names it tries to load.  Then all you need to do is
	try to prove a conjecture; one of the ones from the PVS tutorial
	will do nicely.  When the (grind) call to actually do the work is
	executed, the Allegro interpreter splodes as described above.  I'm
	sure there are simpler Lisp procedures that will splode the 
	interpreter but I haven't found one yet.

>Fix:
	Unknown.  This needs attention from someone who understands the
	intricacies of signal delivery on x86 and the Linux compat code.
>Release-Note:
>Audit-Trail:
>Unformatted: