On 01/31/2011 04:41 PM, Earl Chew wrote:
I notice that a 32 bit process running on a 64 bit kernel is expected to
know that it should fill in mips64_watch_regs --- even though it is running
against a 32 bit ABI.
Is this an oversight, or am I missing something ?
It is intentional.
[ That same 32 bit process must fill in mips32_watch_regs when running on
a 32 bit kernel. ]
In arch/mips/include/asm/ptrace.h:
struct mips32_watch_regs {
unsigned int watchlo[8];
...
};
and
struct mips64_watch_regs {
unsigned long long watchlo[8];
...
};
These are used in a union, but sizeof(mips64_watch_regs.watchlo) will not
match sizeof(mips32_watch_regs.watchlo).
The sizes are different and thus are ... different. struct
pt_watch_regs however, has a well defined size.
The important thing is that the style element of struct pt_watch_regs is
always in the same place.
For a 64 bit kernel, the code in arch/mips/kernel/ptrace.c reads:
/* Check the values. */
for (i = 0; i< current_cpu_data.watch_reg_use_cnt; i++) {
__get_user(lt[i],&addr->WATCH_STYLE.watchlo[i]);
#ifdef CONFIG_32BIT
if (lt[i]& __UA_LIMIT)
return -EINVAL;
#else
if (test_tsk_thread_flag(child, TIF_32BIT_ADDR)) {
if (lt[i]& 0xffffffff80000000UL)
return -EINVAL;
} else {
if (lt[i]& __UA_LIMIT)
return -EINVAL;
}
#endif
Thus for a 64 bit kernel, WATCH_STYLE is defined to be mips64, and the code
goes on to obtain:
addr->mips64.watchlo[i]
and to verify it based on TIF_32BIT_ADDR.
In other words, the 32 bit process is expected to fill in mips64_watch_regs
when it is running on a 64 bit kernel, and mips32_watch_regs when it is running
on a 32 bit kernel.
Yes. The debugging agent must check struct pt_watch_regs style to
determine which type of watch registers the hardware is using.
Take a look at how GDB handles this in mips-linux-nat.c.
David Daney
