RE: [PATCH] compiler.h: Fix barrier_data() on clang
From: David Laight
Date: Thu Oct 15 2020 - 04:50:13 EST
From: Arvind Sankar
> Sent: 14 October 2020 22:27
...
> +/*
> + * This version is i.e. to prevent dead stores elimination on @ptr
> + * where gcc and llvm may behave differently when otherwise using
> + * normal barrier(): while gcc behavior gets along with a normal
> + * barrier(), llvm needs an explicit input variable to be assumed
> + * clobbered. The issue is as follows: while the inline asm might
> + * access any memory it wants, the compiler could have fit all of
> + * @ptr into memory registers instead, and since @ptr never escaped
> + * from that, it proved that the inline asm wasn't touching any of
> + * it. This version works well with both compilers, i.e. we're telling
> + * the compiler that the inline asm absolutely may see the contents
> + * of @ptr. See also: https://llvm.org/bugs/show_bug.cgi?id=15495
> + */
> +# define barrier_data(ptr) __asm__ __volatile__("": :"r"(ptr) :"memory")
That comment doesn't actually match the asm statement.
Although the asm statement probably has the desired effect.
The "r"(ptr) constraint only passes the address of the buffer
into the asm - it doesn't say anything at all about the associated
memory.
What the "r"(ptr) actually does is to force the address of the
associated data to be taken.
This means that on-stack space must actually be allocated.
The "memory" clobber will then force the registers caching
the variable be written out to stack.
If you only want to force stores on a single data structure
you actually want:
#define barrier_data(ptr) asm volatile("" :: "m"(*ptr))
although it would be best then to add an explicit size
and associated cast.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)