lkml.org 
[lkml]   [2014]   [Jan]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v4 13/16] ARM: Add an emulate flag to the kprobes/uprobes instruction decode functions
On 12/20/13 09:58, Jon Medhurst (Tixy) wrote:
> On Sun, 2013-12-15 at 23:08 -0500, David Long wrote:
>> From: "David A. Long" <dave.long@linaro.org>
>>
>> Add an emulate flag into the instruction interpreter, primarily for uprobes
>> support.
>>
>> Signed-off-by: David A. Long <dave.long@linaro.org>
>> ---
>> arch/arm/kernel/kprobes.c | 3 ++-
>> arch/arm/kernel/kprobes.h | 1 +
>> arch/arm/kernel/probes-arm.c | 4 ++--
>> arch/arm/kernel/probes-arm.h | 2 +-
>> arch/arm/kernel/probes-thumb.c | 8 ++++----
>> arch/arm/kernel/probes-thumb.h | 4 ++--
>> arch/arm/kernel/probes.c | 32 +++++++++++++++++++++++---------
>> arch/arm/kernel/probes.h | 2 +-
>> 8 files changed, 36 insertions(+), 20 deletions(-)
>>
>> diff --git a/arch/arm/kernel/kprobes.c b/arch/arm/kernel/kprobes.c
>> index 0d9d49b..04690f9 100644
>> --- a/arch/arm/kernel/kprobes.c
>> +++ b/arch/arm/kernel/kprobes.c
>> @@ -87,7 +87,8 @@ int __kprobes arch_prepare_kprobe(struct kprobe *p)
>> p->opcode = insn;
>> p->ainsn.insn = tmp_insn;
>>
>> - switch ((*decode_insn)(insn, &p->ainsn, actions)) {
>> + switch ((*decode_insn)(insn, &p->ainsn,
>> + true, actions)) {
>
> Any reason why the function args need splitting over two lines?

I undid the that change.

>> case INSN_REJECTED: /* not supported */
>> return -EINVAL;
>>
>
> [...]
>
>> --- a/arch/arm/kernel/probes.c
>> +++ b/arch/arm/kernel/probes.c
>> @@ -193,7 +193,7 @@ void __kprobes probes_emulate_none(probes_opcode_t opcode,
>> */
>> static probes_opcode_t __kprobes
>> prepare_emulated_insn(probes_opcode_t insn, struct arch_specific_insn *asi,
>> - bool thumb)
>> + bool thumb)
>
> Seems like a spurious indentation change.

Fixed.

>> {
>> #ifdef CONFIG_THUMB2_KERNEL
>> if (thumb) {
>> @@ -218,7 +218,7 @@ prepare_emulated_insn(probes_opcode_t insn, struct arch_specific_insn *asi,
>> */
>> static void __kprobes
>> set_emulated_insn(probes_opcode_t insn, struct arch_specific_insn *asi,
>> - bool thumb)
>> + bool thumb)
>
> Another spurious whitespace change.

Fixed.

>> {
>> #ifdef CONFIG_THUMB2_KERNEL
>> if (thumb) {
>> @@ -253,7 +253,7 @@ set_emulated_insn(probes_opcode_t insn, struct arch_specific_insn *asi,
>> * non-zero value, the corresponding nibble in pinsn is validated and modified
>> * according to the type.
>> */
>> -static bool __kprobes decode_regs(probes_opcode_t *pinsn, u32 regs)
>> +static bool __kprobes decode_regs(probes_opcode_t *pinsn, u32 regs, bool modify)
>> {
>> probes_opcode_t insn = *pinsn;
>> probes_opcode_t mask = 0xf; /* Start at least significant nibble */
>> @@ -317,9 +317,16 @@ static bool __kprobes decode_regs(probes_opcode_t *pinsn, u32 regs)
>> /* Replace value of nibble with new register number... */
>> insn &= ~mask;
>> insn |= new_bits & mask;
>> + if (modify) {
>> + /* Replace value of nibble with new register number */
>> + insn &= ~mask;
>> + insn |= new_bits & mask;
>> + }
>
> Huh? As is, the above addition doesn't do anything because insn has
> already been modified. I guess you played with the idea that you needed
> to avoid changing insn (you don't) and then didn't undo the experiment
> quite right. :-)
>

The conditional modification of the instruction was part of Rabin's
original work for uprobes, but I messed up the merge from an earlier
working version of my patches. My intention was/is to delete the old
unconditional code. Sounds like maybe you disagree though. The intent
is to only modify the instruction in the kprobes case.

>> }
>>
>> - *pinsn = insn;
>> + if (modify)
>> + *pinsn = insn;
>> +
>> return true;
>>
>> reject:
>> @@ -380,14 +387,15 @@ static const int decode_struct_sizes[NUM_DECODE_TYPES] = {
>> */
>> int __kprobes
>> probes_decode_insn(probes_opcode_t insn, struct arch_specific_insn *asi,
>> - const union decode_item *table, bool thumb,
>> - const union decode_item *actions)
>> + const union decode_item *table, bool thumb,
>> + bool emulate, const union decode_item *actions)
>> {
>> struct decode_header *h = (struct decode_header *)table;
>> struct decode_header *next;
>> bool matched = false;
>>
>> - insn = prepare_emulated_insn(insn, asi, thumb);
>> + if (emulate)
>> + insn = prepare_emulated_insn(insn, asi, thumb);
>>
>> for (;; h = next) {
>> enum decode_type type = h->type_regs.bits & DECODE_TYPE_MASK;
>> @@ -402,7 +410,7 @@ probes_decode_insn(probes_opcode_t insn, struct arch_specific_insn *asi,
>> if (!matched && (insn & h->mask.bits) != h->value.bits)
>> continue;
>>
>> - if (!decode_regs(&insn, regs))
>> + if (!decode_regs(&insn, regs, emulate))
>> return INSN_REJECTED;
>>
>> switch (type) {
>> @@ -415,7 +423,8 @@ probes_decode_insn(probes_opcode_t insn, struct arch_specific_insn *asi,
>>
>> case DECODE_TYPE_CUSTOM: {
>> struct decode_custom *d = (struct decode_custom *)h;
>> - return actions[d->decoder.bits].decoder(insn, asi, h);
>> + return actions[d->decoder.bits].decoder(insn,
>> + asi, h);
>
> No need to split the above line, you haven't changed it and it doesn't
> exceed 80 characters anyway.

Fixed.

> [Rest of patch cut]
>

-dl



\
 
 \ /
  Last update: 2014-01-15 21:01    [W:0.115 / U:0.176 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site