2014-01-15 19:31:15

by David Long

[permalink] [raw]
Subject: Re: [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" <[email protected]>
>>
>> Add an emulate flag into the instruction interpreter, primarily for uprobes
>> support.
>>
>> Signed-off-by: David A. Long <[email protected]>
>> ---
>> 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


2014-01-16 09:19:28

by Jon Medhurst (Tixy)

[permalink] [raw]
Subject: Re: [PATCH v4 13/16] ARM: Add an emulate flag to the kprobes/uprobes instruction decode functions

On Wed, 2014-01-15 at 14:31 -0500, David Long wrote:
> On 12/20/13 09:58, Jon Medhurst (Tixy) wrote:
> > On Sun, 2013-12-15 at 23:08 -0500, David Long wrote:
[...]
> >> {
> >> #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.

'insn' is the local variable containing the instruction value we're
processing. It doesn't matter if we change that, we just need to avoid
updating the instruction in memory, which the code in the next chunk
already correctly checks for...

> >> }
> >>
> >> - *pinsn = insn;
> >> + if (modify)
> >> + *pinsn = insn;
> >> +
> >> return true;
> >>

So only one of these 'if (modify)' checks is required for code
correctness, and I suggest keeping the second one as it's more explicit
and defensive.


--
Tixy

2014-01-16 18:12:42

by David Long

[permalink] [raw]
Subject: Re: [PATCH v4 13/16] ARM: Add an emulate flag to the kprobes/uprobes instruction decode functions

On 01/16/14 04:18, Jon Medhurst (Tixy) wrote:
> On Wed, 2014-01-15 at 14:31 -0500, David Long wrote:
>> On 12/20/13 09:58, Jon Medhurst (Tixy) wrote:
>>> On Sun, 2013-12-15 at 23:08 -0500, David Long wrote:
> [...]
>>>> {
>>>> #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.
>
> 'insn' is the local variable containing the instruction value we're
> processing. It doesn't matter if we change that, we just need to avoid
> updating the instruction in memory, which the code in the next chunk
> already correctly checks for...
>
>>>> }
>>>>
>>>> - *pinsn = insn;
>>>> + if (modify)
>>>> + *pinsn = insn;
>>>> +
>>>> return true;
>>>>
>
> So only one of these 'if (modify)' checks is required for code
> correctness, and I suggest keeping the second one as it's more explicit
> and defensive.
>
>

OK, I see your point. I shall simplify the code as you have suggested.

-dl