On Mon, Nov 25, 2019 at 09:32:23PM +1100, Michael Ellerman wrote:

Segher Boessenkool [off-list ref] writes:

quoted

quoted

quoted

+static inline void call_do_irq(struct pt_regs *regs, void *sp)
+{
+	register unsigned long r3 asm("r3") = (unsigned long)regs;
+
+	/* Temporarily switch r1 to sp, call __do_irq() then restore r1 */
+	asm volatile(
+		"	"PPC_STLU"	1, %2(%1);\n"
+		"	mr		1, %1;\n"
+		"	bl		%3;\n"
+		"	"PPC_LL"	1, 0(1);\n" :
+		"+r"(r3) :
+		"b"(sp), "i"(THREAD_SIZE - STACK_FRAME_OVERHEAD), "i"(__do_irq) :
+		"lr", "xer", "ctr", "memory", "cr0", "cr1", "cr5", "cr6", "cr7",
+		"r0", "r2", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11", "r12");
+}

If we add a nop after the bl, so the linker could insert a TOC restore,
then I don't think there's any circumstance under which we expect this
to actually clobber r2, is there?

That is mostly correct.

That's the standard I aspire to :P

quoted

If call_do_irq was a no-inline function, there would not be problems.

What TOC does __do_irq require in r2 on entry, and what will be there
when it returns?

The kernel TOC, and also the kernel TOC, unless something's gone wrong
or I'm missing something.

If that is the case, we can just do the bl, no nop at all?  And that works
for all of our ABIs.

If we can be certain that we have the kernel TOC in r2 on entry to
call_do_irq, that is!  (Or it establishes it itself).


Segher