Messages in this thread | | | From | Arnd Bergmann <> | Date | Mon, 22 Mar 2021 22:18:17 +0100 | Subject | Re: [PATCH] static_call: fix function type mismatch |
| |
On Mon, Mar 22, 2021 at 9:47 PM Peter Zijlstra <peterz@infradead.org> wrote: > On Mon, Mar 22, 2021 at 03:32:14PM -0400, Steven Rostedt wrote: > > On Mon, 22 Mar 2021 18:06:37 +0100 > > Arnd Bergmann <arnd@kernel.org> wrote: > > > > > From: Arnd Bergmann <arnd@arndb.de> > > > > > > The __static_call_return0() function is declared to return a 'long', > > > while it aliases a couple of functions that all return 'int'. When > > > building with 'make W=1', gcc warns about this: > > > > > > kernel/sched/core.c:5420:37: error: cast between incompatible function types from 'long int (*)(void)' to 'int (*)(void)' [-Werror=cast-function-type] > > > 5420 | static_call_update(might_resched, (typeof(&__cond_resched)) __static_call_return0); > > > > > > Change the function to return 'int' as well, but remove the cast to > > > ensure we get a warning if any of the types ever change. > > > > I think the answer is the other way around. That is, to make the functions > > it references return long instead. __static_call_return0 is part of the > > dynamic call infrastructure. Perhaps it is currently only used by functions > > that return int, but what happens when it is used for a function that > > returns a pointer?
I've done a little testing on the replacement patch now, will send in a bit.
> Steve is correct. Also, why is that warning correct? On x86 we return in > RAX, and using int will simply not inspect the upper 32 bits there.
I think the code works correctly on all architectures we support because both 'int' and 'long' are returned in a register with any unused bits cleared. It is however undefined behavior in C because 'int' and 'long' are not compatible types, and the calling conventions don't have to allow this.
> And I'm fairly sure I had a pointer user somewhere recently.
I've only tested my series with 5.12-rc so far, but don't get any other such warnings. Maybe it's in linux-next?
Arnd
| |