We will reject probing of unprivileged load and store instructions.
These rarely occur and writing test cases for them is difficult.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
We'll treat the preload instructions as nops as they are just
performance hints.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
The kernel doesn't currently support VFP or Neon code, and probing of
code with CP15 operations is fraught with bad consequences. So we will
just reject probing these instructions.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
We reject probing of load/store exclusive instructions because any
emulation routine could never succeed in gaining exclusive access as the
exception framework clears the exclusivity monitor when a probes
breakpoint is hit.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This patch improves the performance of LDM and STM instruction
emulation. This is desirable because.
- jprobes and kretprobes probe the first instruction in a function and,
when the frame pointer is omitted, this instruction is often a STM
used to push registers onto the stack.
- The STM and LDM instructions are common in the body and tail of
functions.
- At the same time as being a common instruction form, they also have
one of the slowest and most complicated simulation routines.
The approach taken to optimisation is to use emulation rather than
simulation, that is, a modified form of the instruction is run with
an appropriate register context.
Benchmarking on an OMAP3530 shows the optimised emulation is between 2
and 3 times faster than the simulation routines. On a Kirkwood based
device the relative performance was very significantly better than this.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
The encoding of these instructions is substantially the same for both
ARM and Thumb, so we can have common decoding and simulation functions.
This patch moves the simulation functions from kprobes-arm.c to
kprobes-common.c. It also adds a new simulation function
(simulate_ldm1_pc) for the case where we load into PC because this may
need to interwork.
The instruction decoding is done by a custom function
(kprobe_decode_ldmstm) rather than just relying on decoding table
entries because we will later be adding optimisation code.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This writes a value to PC which was obtained as the result of a
LDR or LDM instruction. For ARMv5T and later this must perform
interworking.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
For hints which may have observable effects, like SEV (send event), we
use kprobe_emulate_none which emulates the hint by executing the
original instruction.
For NOP we simulate the instruction using kprobe_simulate_nop, which
does nothing. As probes execute with interrupts disabled this is also
used for hints which may block for an indefinite time, like WFE (wait
for event).
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
These are very rare and/or problematic to emulate so we will take the
easy option and disallow probing them (as does the existing ARM
implementation).
Rejecting these instructions doesn't actually require any entries in the
decoding table as it is the default case for instructions which aren't
found.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
We previously changed the behaviour of probes so that conditional
instructions don't fire when the condition isn't met. For ARM branches,
and Thumb branches in IT blocks, this means they don't fire if the
branch isn't taken.
For consistency, we implement the same for Thumb conditional branch
instructions. This involves setting up insn_check_cc to point to the
relevant condition checking function. As the emulation routine is only
called when this condition passes, it doesn't need to check again and
can unconditionally update PC.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
SVC (SWI) instructions shouldn't occur in kernel code so we don't
need to be able to probe them.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
The normal Thumb singlestepping routine updates the IT state after
calling the instruction handler. We don't what this to happen after the
IT instruction simulation sets the IT state, therefore we need to
provide a custom singlestep routine.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
These instructions are equivalent to
stmdb sp!,{r0-r7,lr}
ldmdb sp!,{r0-r7,pc}
and we emulate them by transforming them into the 32-bit Thumb
instructions
stmdb r9!,{r0-r7,r8}
ldmdb r9!,{r0-r7,r8}
This is simpler, and almost certainly executes faster, than writing
simulation functions.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Most of these instructions only operate on the low registers R0-R7
so they can make use of t16_emulate_loregs_rwflags.
The instructions which use SP or PC for addressing have their own
simulation functions.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
These data-processing instructions operate on the full range of CPU
registers, so to simulate them we have to modify the registers used
by the instruction. We can't make use of the decoding table framework to
do this because the registers aren't encoded cleanly in separate
nibbles, therefore we need a custom decode function.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This writes a value to PC, with interworking. I.e. switches to Thumb or
ARM mode depending on the state of the least significant bit.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
These instructions only operate on the low registers R0-R7, therefore
it is possible to emulate them by executing the original instruction
unaltered if we restore and save these registers. This is what
t16_emulate_loregs does.
Some of these instructions don't update the PSR when they execute in an
IT block, so there are two flavours of emulation functions:
t16_emulate_loregs_{noit}rwflags
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
For hints which may have observable effects, like SEV (send event), we
use kprobe_emulate_none which emulates the hint by executing the
original instruction.
For NOP we simulate the instruction using kprobe_simulate_nop, which
does nothing. As probes execute with interrupts disabled this is also
used for hints which may block for an indefinite time, like WFE (wait
for event).
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
The existing ARM instruction decoding functions are a mass of if/else
code. Rather than follow this pattern for Thumb instruction decoding
this patch implements an infrastructure for a new table driven scheme.
This has several advantages:
- Reduces the kernel size by approx 2kB. (The ARM instruction decoding
will eventually have -3.1kB code, +1.3kB data; with similar or better
estimated savings for Thumb decoding.)
- Allows programmatic checking of decoding consistency and test case
coverage.
- Provides more uniform source code and is therefore, arguably, clearer.
For a detailed explanation of how decoding tables work see the in-source
documentation in kprobes.h, and also for kprobe_decode_insn().
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
When we come to emulating Thumb instructions then, to interwork
correctly, the code on in the instruction slot must be invoked with a
function pointer which has the least significant bit set. Rather that
set this by hand in every Thumb emulation function we will add a new
field for this purpose to arch_specific_insn, called insn_fn.
This also enables us to seamlessly share emulation functions between ARM
and Thumb code.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
When a probe fires we must single-step the instruction which was
replaced by a breakpoint. As the steps to do this vary between ARM and
Thumb instructions we need a way to customise single-stepping.
This is done by adding a new hook called insn_singlestep to
arch_specific_insn which is initialised by the instruction decoding
functions.
These single-step hooks must update PC and call the instruction handler.
For Thumb instructions an additional step of updating ITSTATE is needed.
We do this after calling the handler because some handlers will need to
test if they are running in an IT block.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Now we no longer trigger probes on conditional instructions when the
condition is false, we can make use of conditional instructions as
breakpoints in ARM code to avoid taking unnecessary exceptions.
Note, we can't rely on not getting an exception when the condition check
fails, as that is Implementation Defined on newer ARM architectures. We
therefore still need to perform manual condition checks as well.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This patch changes the behavior of kprobes on ARM so that:
Kprobes on conditional instructions don't trigger when the
condition is false. For conditional branches, this means that
they don't trigger in the branch not taken case.
Rationale:
When probes are placed onto conditionally executed instructions in a
Thumb IT block, they may not fire if the condition is not met. This
is because we use invalid instructions for breakpoints and "it is
IMPLEMENTATION DEFINED whether the instruction executes as a NOP or
causes an Undefined Instruction exception". Therefore, for consistency,
we will ignore all probes on any conditional instructions when the
condition is false. Alternative solutions seem to be too complex to
implement or inconsistent.
This issue was discussed on linux.arm.kernel in the thread titled
"[RFC] kprobes with thumb2 conditional code" See
http://comments.gmane.org/gmane.linux.linaro.devel/2985
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This advances the ITSTATE bits in CPSR to their values for the next
instruction.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Extend the breakpoint insertion and catching functions to support Thumb
code.
As breakpoints are no longer of a fixed size, the flush_insns macro
is modified to take a size argument instead of an instruction count.
Note, we need both 16- and 32-bit Thumb breakpoints, because if we
were to use a 16-bit breakpoint to replace a 32-bit instruction which
was in an IT block, and the condition check failed, then the breakpoint
may not fire (it's unpredictable behaviour) and the CPU could then try
and execute the second half of the 32-bit Thumb instruction.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Extend arch_prepare_kprobe to support probing of Thumb code. For
the actual decoding of Thumb instructions, stub functions are
added which currently just reject the probe.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Fix up kprobes framework so that it builds and correctly interworks on
Thumb-2 kernels.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
The str_pc_offset value is architecturally defined on ARMv7 onwards so
we can make it a compile time constant. This means on Thumb kernels the
runtime checking code isn't needed, which saves us from having to fix it
to work for Thumb.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Move str_pc_offset into kprobes-common.c as it will be needed by common
code later.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This file will contain the instruction decoding and emulation code
which is common to both ARM and Thumb instruction sets.
For now, we will just move over condition_checks from kprobes-arm.c
This table is also renamed to kprobe_condition_checks to avoid polluting
the public namespace with a too generic name.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Later, we will be adding a considerable amount of internal
implementation definitions to kprobe header files and it would be good
to have these in local header file along side the source code, rather
than pollute the existing header which is include by all users of
kprobes.
To this end, we add arch/arm/kernel/kprobes.h and move into this the
existing internal defintions from arch/arm/include/asm/kprobes.h
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This file contains decoding and emulation functions for the ARM
instruction set. As we will later be adding a file for Thumb and a
file with common decoding functions, this renaming makes things clearer.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
This patch allows undef_hook's to be specified for 32-bit Thumb
instructions and also to be used for thumb kernel-side code.
32-bit Thumb instructions are specified in the form:
((first_half << 16 ) | second_half)
which matches the layout used by the ARM ARM.
ptrace was handling 32-bit Thumb instructions by hooking the first
halfword and manually checking the second half. This method would be
broken by this patch so it is migrated to make use of the new Thumb-2
support.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
The implementation of svc_exit didn't take into account any stack hole
created by svc_entry; as happens with the undef handler when kprobes are
configured. The fix is to read the saved value of SP rather than trying
to calculate it.
Signed-off-by: Jon Medhurst <tixy@yxit.co.uk>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
On Mon, Jul 11, 2011 at 3:52 PM, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:
...
> The __exception annotation on a function causes this to happen:
>
> [<c002406c>] (asm_do_IRQ+0x6c/0x8c) from [<c0024b84>]
> (__irq_svc+0x44/0xcc)
> Exception stack(0xc3897c78 to 0xc3897cc0)
> 7c60: 4022d320 4022e000
> 7c80: 08000075 00001000 c32273c0 c03ce1c0 c2b49b78 4022d000 c2b420b4 00000001
> 7ca0: 00000000 c3897cfc 00000000 c3897cc0 c00afc54 c002edd8 00000013 ffffffff
>
> Where that stack dump represents the pt_regs for the exception which
> happened. Any function found in while unwinding will cause this to
> be printed.
>
> If you insert a C function between the IRQ assembly and asm_do_IRQ,
> the
> dump you get from asm_do_IRQ will be the stack for your function,
> not
> the pt_regs. That makes the feature useless.
>
When __irq_svc - or any of the other exception handling assembly code -
calls the C code, the stack pointer will be pointing at the pt_regs
structure.
All the entry points into C code from the exception handling code are
marked with __exception or __exception_irq_enter to indicate that they
are one of the functions which has pt_regs above them.
Normally, when you've entered asm_do_IRQ() you will have this stack
layout (higher address towards top):
pt_regs
asm_do_IRQ frame
If you insert a C function between the exception assembly code and
asm_do_IRQ, you end up with this stack layout instead:
pt_regs
your function frame
asm_do_IRQ frame
This means when we unwind, we'll get to asm_do_IRQ, and rather than
dumping out the pt_regs, we'll dump out your functions stack frame
instead, because that's what is above the asm_do_IRQ stack frame
rather than the expected pt_regs structure.
The fix is to introduce handle_IRQ() for no exception stack dump, so
it can be called with MULTI_IRQ_HANDLER is selected and a C function
is between the assembly code and the actual IRQ handling code.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Eric Miao <eric.y.miao@gmail.com>
Fix a hole in the VFP thread migration. Lets define two threads.
Thread 1, we'll call 'interesting_thread' which is a thread which is
running on CPU0, using VFP (so vfp_current_hw_state[0] =
&interesting_thread->vfpstate) and gets migrated off to CPU1, where
it continues execution of VFP instructions.
Thread 2, we'll call 'new_cpu0_thread' which is the thread which takes
over on CPU0. This has also been using VFP, and last used VFP on CPU0,
but doesn't use it again.
The following code will be executed twice:
cpu = thread->cpu;
/*
* On SMP, if VFP is enabled, save the old state in
* case the thread migrates to a different CPU. The
* restoring is done lazily.
*/
if ((fpexc & FPEXC_EN) && vfp_current_hw_state[cpu]) {
vfp_save_state(vfp_current_hw_state[cpu], fpexc);
vfp_current_hw_state[cpu]->hard.cpu = cpu;
}
/*
* Thread migration, just force the reloading of the
* state on the new CPU in case the VFP registers
* contain stale data.
*/
if (thread->vfpstate.hard.cpu != cpu)
vfp_current_hw_state[cpu] = NULL;
The first execution will be on CPU0 to switch away from 'interesting_thread'.
interesting_thread->cpu will be 0.
So, vfp_current_hw_state[0] points at interesting_thread->vfpstate.
The hardware state will be saved, along with the CPU number (0) that
it was executing on.
'thread' will be 'new_cpu0_thread' with new_cpu0_thread->cpu = 0.
Also, because it was executing on CPU0, new_cpu0_thread->vfpstate.hard.cpu = 0,
and so the thread migration check is not triggered.
This means that vfp_current_hw_state[0] remains pointing at interesting_thread.
The second execution will be on CPU1 to switch _to_ 'interesting_thread'.
So, 'thread' will be 'interesting_thread' and interesting_thread->cpu now
will be 1. The previous thread executing on CPU1 is not relevant to this
so we shall ignore that.
We get to the thread migration check. Here, we discover that
interesting_thread->vfpstate.hard.cpu = 0, yet interesting_thread->cpu is
now 1, indicating thread migration. We set vfp_current_hw_state[1] to
NULL.
So, at this point vfp_current_hw_state[] contains the following:
[0] = &interesting_thread->vfpstate
[1] = NULL
Our interesting thread now executes a VFP instruction, takes a fault
which loads the state into the VFP hardware. Now, through the assembly
we now have:
[0] = &interesting_thread->vfpstate
[1] = &interesting_thread->vfpstate
CPU1 stops due to ptrace (and so saves its VFP state) using the thread
switch code above), and CPU0 calls vfp_sync_hwstate().
if (vfp_current_hw_state[cpu] == &thread->vfpstate) {
vfp_save_state(&thread->vfpstate, fpexc | FPEXC_EN);
BANG, we corrupt interesting_thread's VFP state by overwriting the
more up-to-date state saved by CPU1 with the old VFP state from CPU0.
Fix this by ensuring that we have sane semantics for the various state
describing variables:
1. vfp_current_hw_state[] points to the current owner of the context
information stored in each CPUs hardware, or NULL if that state
information is invalid.
2. thread->vfpstate.hard.cpu always contains the most recent CPU number
which the state was loaded into or NR_CPUS if no CPU owns the state.
So, for a particular CPU to be a valid owner of the VFP state for a
particular thread t, two things must be true:
vfp_current_hw_state[cpu] == &t->vfpstate && t->vfpstate.hard.cpu == cpu.
and that is valid from the moment a CPU loads the saved VFP context
into the hardware. This gives clear and consistent semantics to
interpreting these variables.
This patch also fixes thread copying, ensuring that t->vfpstate.hard.cpu
is invalidated, otherwise CPU0 may believe it was the last owner. The
hole can happen thus:
- thread1 runs on CPU2 using VFP, migrates to CPU3, exits and thread_info
freed.
- New thread allocated from a previously running thread on CPU2, reusing
memory for thread1 and copying vfp.hard.cpu.
At this point, the following are true:
new_thread1->vfpstate.hard.cpu == 2
&new_thread1->vfpstate == vfp_current_hw_state[2]
Lastly, this also addresses thread flushing in a similar way to thread
copying. Hole is:
- thread runs on CPU0, using VFP, migrates to CPU1 but does not use VFP.
- thread calls execve(), so thread flush happens, leaving
vfp_current_hw_state[0] intact. This vfpstate is memset to 0 causing
thread->vfpstate.hard.cpu = 0.
- thread migrates back to CPU0 before using VFP.
At this point, the following are true:
thread->vfpstate.hard.cpu == 0
&thread->vfpstate == vfp_current_hw_state[0]
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Jon Medhurst