Re: [RFC PATCH 09/11] kprobes: core logic of eraly kprobes

[Wang Nan <wangnan0@huawei.com>]

On 2015/1/7 17:21, Hillf Danton wrote:
>>
>> +struct early_kprobe_slot {
>> +	struct optimized_kprobe op;
>> +};
>> +
> [...]
>>
>>  /* Free optimized instructions and optimized_kprobe */
>> +static int ek_free_early_kprobe(struct early_kprobe_slot *slot);
> 
> [2] How is it implemented? In subsequent patches?
> 

It is implemented using macro. Please see patch 7/11 and

DEFINE_EKPROBE_ALLOC_OPS(struct early_kprobe_slot, early_kprobe, static);

following.

>>  static void free_aggr_kprobe(struct kprobe *p)
>>  {
>>  	struct optimized_kprobe *op;
>> +	struct early_kprobe_slot *ep;
>>
>>  	op = container_of(p, struct optimized_kprobe, kp);
>>  	arch_remove_optimized_kprobe(op);
>>  	arch_remove_kprobe(p);
>> -	kfree(op);
>> +	ep = container_of(op, struct early_kprobe_slot, op);
>> +	if (likely(!ek_free_early_kprobe(ep)))
>> +		kfree(op);
> 
> [1] s/op/ep/   yes?

Which one? Do you mean kfree(op) --> kfree(ep)?

If ek_free_early_kprobe(ep) fail (not in early_kprobe area defined by
DEFINE_EKPROBE_ALLOC_OPS), then this is a normal aggr probe, allocated
using kzalloc() as struct optimized_kprobe, see alloc_aggr_kprobe().
So kfree corresponding op structure.

If ek_free_early_kprobe(ep) success, then this is an struct early_kprobe_slot
and allocated statically.

>>  }
>>
> [...]
>> +#else
>> +static int register_early_kprobe(struct kprobe *p) { return -ENOSYS; }
>> +static int ek_free_early_kprobe(struct early_kprobe_slot *slot) { return 0; }
> 
> [3] Compile-able with CONFIG_EARLY_KPROBES enabled?

These empty functions are for CONFIG_EARLY_KPROBES disabled. They won't be compiled
when CONFIG_EARLY_KPROBES=y. I have tested both cases on x86 and ARM.

> 
>> +static void convert_early_kprobes(void) {};
>> +#endif
>> --
>> 1.8.4
> 

Thanks.