Messages in this thread Patch in this message | | | Subject | Re: [PATCH 3/5] perf_counter: rework ioctl()s | From | Peter Zijlstra <> | Date | Tue, 12 May 2009 09:10:25 +0200 |
| |
On Tue, 2009-05-12 at 08:27 +0200, Peter Zijlstra wrote: > On Tue, 2009-05-12 at 16:22 +1000, Paul Mackerras wrote: > > Peter Zijlstra writes: > > > > > Hmm, are you saying that the 3rd argument to unlocked_ioctl is actually > > > (void __user *) instead of unsigned long? > > > > He's saying (correctly) that using _IOR or _IOW implies that the ioctl > > is going to read or write the memory location pointed to by the 3rd > > argument to unlocked_ioctl. If the 3rd argument is just a number, not > > an address, I believe you should use _IO. > > Oh, somewhat confusing all this. Would be good to spell out these things > somewhere. Documentation/ioctl/ seems less than helpful.
Ah, so _IO() gets an unsigned long 3rd argument. _IOW() treats the 3rd arg as a (type __user *) and copies the bits over _IOR() copies the bits back out to userspace _IORW() does both
In which case the below should fix things up, no?
--- diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h index 614f921..6a9cebc 100644 --- a/include/linux/perf_counter.h +++ b/include/linux/perf_counter.h @@ -159,10 +159,10 @@ struct perf_counter_hw_event { /* * Ioctls that can be done on a perf counter fd: */ -#define PERF_COUNTER_IOC_ENABLE _IOW('$', 0, u32) -#define PERF_COUNTER_IOC_DISABLE _IOW('$', 1, u32) -#define PERF_COUNTER_IOC_REFRESH _IOW('$', 2, u32) -#define PERF_COUNTER_IOC_RESET _IOW('$', 3, u32) +#define PERF_COUNTER_IOC_ENABLE _IO('$', 0) +#define PERF_COUNTER_IOC_DISABLE _IO('$', 1) +#define PERF_COUNTER_IOC_REFRESH _IO('$', 2) +#define PERF_COUNTER_IOC_RESET _IO('$', 3) enum perf_counter_ioc_flags { PERF_IOC_FLAG_GROUP = 1U << 0,
| |