Le 19/08/2019 à 16:37, Segher Boessenkool a écrit :

On Mon, Aug 19, 2019 at 04:08:43PM +0200, Christophe Leroy wrote:

quoted

Le 19/08/2019 à 15:23, Segher Boessenkool a écrit :

quoted

On Mon, Aug 19, 2019 at 01:06:31PM +0000, Christophe Leroy wrote:

quoted

Note that we keep using an assembly text using "twi 31, 0, 0" for
inconditional traps because GCC drops all code after
__builtin_trap() when the condition is always true at build time.

As I said, it can also do this for conditional traps, if it can prove
the condition is always true.

But we have another branch for 'always true' and 'always false' using
__builtin_constant_p(), which don't use __builtin_trap(). Is there
anything wrong with that ?:

The compiler might not realise it is constant when it evaluates the
__builtin_constant_p, but only realises it later.  As the documentation
for the builtin says:
   A return of 0 does not indicate that the
   value is _not_ a constant, but merely that GCC cannot prove it is a
   constant with the specified value of the '-O' option.

So you mean GCC would not be able to prove that 
__builtin_constant_p(cond) is always true but it would be able to prove 
that if (cond)  is always true ?

And isn't there a away to tell GCC that '__builtin_trap()' is 
recoverable in our case ?

(and there should be many more and more serious warnings here).

quoted

#define BUG_ON(x) do {						\
	if (__builtin_constant_p(x)) {				\
		if (x)						\
			BUG();					\
	} else {						\
		if (x)						\
			__builtin_trap();			\
		BUG_ENTRY("", 0);				\
	}							\
} while (0)

I think it may work if you do

#define BUG_ON(x) do {						\
	if (__builtin_constant_p(x)) {				\
		if (x)						\
			BUG();					\
	} else {						\
		BUG_ENTRY("", 0);				\
		if (x)						\
			__builtin_trap();			\
	}							\
} while (0)

It doesn't work:

void test_bug1(unsigned long long a)
{
	BUG_ON(a);
}

00000090 <test_bug1>:
   90:	7c 63 23 78 	or      r3,r3,r4
   94:	0f 03 00 00 	twnei   r3,0
   98:	4e 80 00 20 	blr

RELOCATION RECORDS FOR [__bug_table]:
OFFSET   TYPE              VALUE
00000084 R_PPC_ADDR32      .text+0x00000090

As you see, the relocation in __bug_table points to the 'or' and not to 
the 'twnei'.

or even just

#define BUG_ON(x) do {						\
	BUG_ENTRY("", 0);					\
	if (x)							\
		__builtin_trap();				\
	}							\
} while (0)

if BUG_ENTRY can work for the trap insn *after* it.

quoted

quoted

Can you put the bug table asm *before* the __builtin_trap maybe?  That
should make it all work fine...  If you somehow can tell what machine
instruction is that trap, anyway.

And how can I tell that ?

I don't know how BUG_ENTRY works exactly.

It's basic, maybe too basic: it adds an inline asm with a label, and 
adds a .long in the __bug_table section with the address of that label.

When putting it after the __builtin_trap(), I changed it to using the 
address before the one of the label which is always the twxx instruction 
as far as I can see.

#define BUG_ENTRY(insn, flags, ...)			\
	__asm__ __volatile__(				\
		"1:	" insn "\n"			\
		".section __bug_table,\"aw\"\n"		\
		"2:\t" PPC_LONG "1b, %0\n"		\
		"\t.short %1, %2\n"			\
		".org 2b+%3\n"				\
		".previous\n"				\
		: : "i" (__FILE__), "i" (__LINE__),	\
		  "i" (flags),				\
		  "i" (sizeof(struct bug_entry)),	\
		  ##__VA_ARGS__)

Christophe