On Thu 2016-02-25 23:11:45, Balbir Singh wrote:
> This applies on top of the patches posted by Michael today
> Enable livepatching. This takes patch 6/8 and 7/8 of v8 as the base.
> Removes the extra strict check in gcc-profile-kernel-notrace.sh
> and adds logic for checking offsets in livepatch. The patch
> for HAVE_C_RECORDMCOUNT is not required and not used here.
>
> Depending on whether or not a TOC is generated, the offset
> for _mcount can be +16 or +8. The changes are such that the
> offset checks are specific to powerpc.
>
> Comments? Testing? I tested the sample in the livepatch
> directory
>
> --- /dev/null
> +++ b/arch/powerpc/include/asm/livepatch.h
> +#define ARCH_HAVE_KLP_MATCHADDR
> +static inline int klp_matchaddr(struct ftrace_ops *ops, unsigned long ip,
> + int remove, int reset)
> +{
> + int offsets[] = {8, 16};
> + int i;
> + int ret = 1;
> +
> + for (i = 0; i < ARRAY_SIZE(offsets); i++) {
> + ret = ftrace_set_filter_ip(ops, ip+offsets[i], remove, reset);
The serach for the right address might get replaced by
ftrace_location call.
But instead of blindly trying to get the right address. I would
suggest to implement a function that will search the ftrace
location for the given function address. I means something like:
/**
* function_to_ftrace_location - get ftrace location for the given
* function address.
* @addr: function address
*
* Returns the address of the ftrace location for the given function.
* Returns 0 if the address does not correspond to any function
* or if the function can not be traced.
*/
unsigned long function_to_ftrace_location(unsigned long addr)
{
const struct dyn_ftrace *rec;
const struct ftrace_page *pg;
unsigned long symbol_size, offset, post_addr;
unsigned long ret = 0UL;
if (!kallsyms_lookup_size_offset(addr, &symbol_size, &offset))
return 0UL;
addr -= offset;
post_addr = addr += symbol_size;
mutex_lock(&ftrace_lock);
do_for_each_ftrace_rec(pg, rec) {
if (rec->ip >= addr && rec->ip < post_addr ) {
ret = rec->ip;
goto end;
}
} while_for_each_ftrace_rec()
end:
mutex_unlock(&ftrace_lock);
return ret;
}
The function is just compile tested on x86_64. We had similar function
in the SUSE-specific LivePatch implementation before we realized
that only fentry (zero offset) could be reasonable supported
on x86_64.
I will prepare a proper patch/patches for PPC early next week.
I have attended a training this week and am snowed under mails
a bit.
Best Regards,
Petr
--
To unsubscribe from this list: send the line "unsubscribe live-patching" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
