Skip to contentSkip to conversation

bug in handle_sys?

DISCUSSION Fuxin Zhang 1 participant 2 messages Mar 28, 2004 – Mar 29, 2004 MIPS

lore
Share & tools
lore
#0 lore raw

Hi,

My colleague finds that there is probably a bug in handle_sys:

.align 5
NESTED(handle_sys, PT_SIZE, sp)
.set noat
SAVE_SOME
STI
.set at

lw t1, PT_EPC(sp) # skip syscall on return

sltiu t0, v0, MAX_SYSCALL_NO + 1 # check syscall number
addiu t1, 4 # skip to next instruction
beqz t0, illegal_syscall
sw t1, PT_EPC(sp)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This code is not guarded by .set no reorder,so it won't be the delay slot instruction,thus illegal_syscall with num > MAX_SYSCALL_NO will return with EPC unchanged. The reason it works is that the syscall number register v0 will be changed to ENOSYS. ENOSYS is fortunately another illegal syscall number that will take another illegal_syscall return path.

Newer glibc of debian(2.3.2+?) will generate sys_4246,and that lead to real problem for mips64. Put the line ahead of the beqz solve it.

#1 lore raw

Hi,all

It seems the fix is a little more complex than I had thought because we've to make sure every configuration works,so a patch is attached below.
And during the reading of related code,we can't find code in mips64 to ensure correct handle of too small a syscall number? For mips we have reserved space for nr<4000,but now it seems random. Do we need to fix that too?

===================================================================
RCS file: /home/cvs/linux-godson2/arch/mips64/kernel/scall_o32.S,v retrieving revision 1.3
retrieving revision 1.4

scall_o32.S 27 Mar 2004 14:58:11 -0000 1.4 renamed from scall_o32.S 16 Oct 2003 16:18:15 -0000 1.3 +0 -0
diff -u -r1.3 -r1.4
@@ -33,8 +33,8 @@

subu t0, v0, __NR_O32_Linux # check syscall number sltiu t0, t0, __NR_O32_Linux_syscalls + 1 daddiu t1, 4 # skip to next instruction
- beqz t0, not_o32_scall
sd t1, PT_EPC(sp)
+ beqz t0, not_o32_scall
#if 0
SAVE_ALL
move a1, v0

Index: scall_n32.S
===================================================================
RCS file: /home/cvs/linux-godson2/arch/mips64/kernel/scall_n32.S,v retrieving revision 1.4
retrieving revision 1.6

scall_n32.S 29 Mar 2004 13:41:45 -0000 1.6 renamed from scall_n32.S 16 Oct 2003 16:18:15 -0000 1.4 +0 -0
diff -u -r1.4 -r1.6
@@ -35,13 +35,16 @@

STI
.set at
#endif
- ld t1, PT_EPC(sp) # skip syscall on return

subu t0, v0, __NR_N32_Linux # check syscall number sltiu t0, t0, __NR_N32_Linux_syscalls + 1 +
+#ifndef CONFIG_MIPS32_O32
+ ld t1, PT_EPC(sp) # skip syscall on return daddiu t1, 4 # skip to next instruction
- beqz t0, not_n32_scall
sd t1, PT_EPC(sp)
+#endif
+ beqz t0, not_n32_scall

dsll t0, v0, 3 # offset into table
ld t2, (sysn32_call_table - (__NR_N32_Linux * 8))(t0)

Index: scall_64.S
===================================================================
RCS file: /home/cvs/linux-godson2/arch/mips64/kernel/scall_64.S,v retrieving revision 1.2
retrieving revision 1.4

scall_64.S 29 Mar 2004 13:41:44 -0000 1.4 renamed from scall_64.S 16 Oct 2003 16:18:15 -0000 1.2 +0 -0
diff -u -r1.2 -r1.4
@@ -31,13 +31,15 @@

STI
.set at
#endif
- ld t1, PT_EPC(sp) # skip syscall on return
-
subu t0, v0, __NR_Linux # check syscall number
sltiu t0, t0, __NR_Linux_syscalls + 1
+
+#if !defined(CONFIG_MIPS32_O32) && !defined(CONFIG_MIPS32_N32)
+ ld t1, PT_EPC(sp) # skip syscall on return
daddiu t1, 4 # skip to next instruction
- beqz t0, illegal_syscall
sd t1, PT_EPC(sp)
+#endif
+ beqz t0, illegal_syscall

dsll t0, v0, 3 # offset into table
ld t2, (sys_call_table - (__NR_Linux * 8))(t0) # syscall routine

Quoted 33 lines · Fuxin Zhang

Hi,

My colleague finds that there is probably a bug in handle_sys:

.align 5
NESTED(handle_sys, PT_SIZE, sp)
.set noat
SAVE_SOME
STI
.set at

lw t1, PT_EPC(sp) # skip syscall on return

sltiu t0, v0, MAX_SYSCALL_NO + 1 # check syscall number
addiu t1, 4 # skip to next instruction
beqz t0, illegal_syscall
sw t1, PT_EPC(sp)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This code is not guarded by .set no reorder,so it won't be the delay slot instruction,thus illegal_syscall with num > MAX_SYSCALL_NO will return with EPC unchanged. The reason it works is that the syscall number register v0 will be changed to ENOSYS. ENOSYS is fortunately another illegal syscall number that will take another illegal_syscall return path.

Newer glibc of debian(2.3.2+?) will generate sys_4246,and that lead to real problem for mips64. Put the line ahead of the beqz solve it.