diff options
| author | Vincent Whitchurch <vincent.whitchurch@axis.com> | 2018-07-13 11:12:22 +0100 |
|---|---|---|
| committer | Greg Kroah-Hartman <gregkh@linuxfoundation.org> | 2019-03-13 14:05:01 -0700 |
| commit | 88bc59cff5226a25dc20017eee2cb470b364f97a (patch) | |
| tree | be7e6863254fb3f4baa7c47bfc3c02e5e6698a79 /arch/arm/kernel | |
| parent | 660e1bf847546a36b09c210a1f40f988800a2894 (diff) | |
| download | linux-stable-88bc59cff5226a25dc20017eee2cb470b364f97a.tar.gz linux-stable-88bc59cff5226a25dc20017eee2cb470b364f97a.tar.bz2 linux-stable-88bc59cff5226a25dc20017eee2cb470b364f97a.zip | |
ARM: 8781/1: Fix Thumb-2 syscall return for binutils 2.29+
[ Upstream commit afc9f65e01cd114cb2cedf544d22239116ce0cc6 ]
When building the kernel as Thumb-2 with binutils 2.29 or newer, if the
assembler has seen the .type directive (via ENDPROC()) for a symbol, it
automatically handles the setting of the lowest bit when the symbol is
used with ADR. The badr macro on the other hand handles this lowest bit
manually. This leads to a jump to a wrong address in the wrong state
in the syscall return path:
Internal error: Oops - undefined instruction: 0 [#2] SMP THUMB2
Modules linked in:
CPU: 0 PID: 652 Comm: modprobe Tainted: G D 4.18.0-rc3+ #8
PC is at ret_fast_syscall+0x4/0x62
LR is at sys_brk+0x109/0x128
pc : [<80101004>] lr : [<801c8a35>] psr: 60000013
Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment none
Control: 50c5387d Table: 9e82006a DAC: 00000051
Process modprobe (pid: 652, stack limit = 0x(ptrval))
80101000 <ret_fast_syscall>:
80101000: b672 cpsid i
80101002: f8d9 2008 ldr.w r2, [r9, #8]
80101006: f1b2 4ffe cmp.w r2, #2130706432 ; 0x7f000000
80101184 <local_restart>:
80101184: f8d9 a000 ldr.w sl, [r9]
80101188: e92d 0030 stmdb sp!, {r4, r5}
8010118c: f01a 0ff0 tst.w sl, #240 ; 0xf0
80101190: d117 bne.n 801011c2 <__sys_trace>
80101192: 46ba mov sl, r7
80101194: f5ba 7fc8 cmp.w sl, #400 ; 0x190
80101198: bf28 it cs
8010119a: f04f 0a00 movcs.w sl, #0
8010119e: f3af 8014 nop.w {20}
801011a2: f2af 1ea2 subw lr, pc, #418 ; 0x1a2
To fix this, add a new symbol name which doesn't have ENDPROC used on it
and use that with badr. We can't remove the badr usage since that would
would cause breakage with older binutils.
Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Diffstat (limited to 'arch/arm/kernel')
| -rw-r--r-- | arch/arm/kernel/entry-common.S | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S index 56be67ecf0fa..d69adfb3d79e 100644 --- a/arch/arm/kernel/entry-common.S +++ b/arch/arm/kernel/entry-common.S @@ -32,6 +32,7 @@ * features make this path too inefficient. */ ret_fast_syscall: +__ret_fast_syscall: UNWIND(.fnstart ) UNWIND(.cantunwind ) disable_irq_notrace @ disable interrupts @@ -57,6 +58,7 @@ fast_work_pending: * r0 first to avoid needing to save registers around each C function call. */ ret_fast_syscall: +__ret_fast_syscall: UNWIND(.fnstart ) UNWIND(.cantunwind ) str r0, [sp, #S_R0 + S_OFF]! @ save returned r0 @@ -223,7 +225,7 @@ local_restart: tst r10, #_TIF_SYSCALL_WORK @ are we tracing syscalls? bne __sys_trace - invoke_syscall tbl, scno, r10, ret_fast_syscall + invoke_syscall tbl, scno, r10, __ret_fast_syscall add r1, sp, #S_OFF 2: cmp scno, #(__ARM_NR_BASE - __NR_SYSCALL_BASE) |