Re: [GIT pull] locking/urgent for v5.12-rc3

From: Peter Zijlstra <peterz@infradead.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	the arch/x86 maintainers <x86@kernel.org>,
	Steven Rostedt <rostedt@goodmis.org>,
	Josh Poimboeuf <jpoimboe@redhat.com>
Subject: Re: [GIT pull] locking/urgent for v5.12-rc3
Date: Mon, 15 Mar 2021 12:26:12 +0100	[thread overview]
Message-ID: <YE9EVNG4/5UMJvFp@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <YE9AkgbqL+eVO6p1@hirez.programming.kicks-ass.net>

On Mon, Mar 15, 2021 at 12:10:10PM +0100, Peter Zijlstra wrote:
> On Mon, Mar 15, 2021 at 09:33:45AM +0100, Peter Zijlstra wrote:
> > On Sun, Mar 14, 2021 at 01:15:25PM -0700, Linus Torvalds wrote:
> > > On Sun, Mar 14, 2021 at 8:40 AM Thomas Gleixner <tglx@linutronix.de> wrote:
> > > >
> > > >  - A fix for the static_call mechanism so it handles unaligned
> > > >    addresses correctly.
> > > 
> > > I'm not disputing the fix in any way, but why weren't the relocation
> > > info and function start addresses mutually aligned?
> > > 
> > > Are we perhaps missing some .align directive somewhere?
> > > 
> > > Or am I missing something?
> > 
> > So I considered looking into that, but since carrying the flag on the
> > absolute address is always correct I figured it was the more robust fix.
> > 
> > I suppose I can try and figure out where alignment went wobbly.
> 
> include/asm-generic/vmlinux.lds.h:
> 
> #define STATIC_CALL_DATA						\
> 	. = ALIGN(8);							\
> 	__start_static_call_sites = .;					\
> 	KEEP(*(.static_call_sites))					\
> 	__stop_static_call_sites = .;					\
> 	__start_static_call_tramp_key = .;				\
> 	KEEP(*(.static_call_tramp_key))					\
> 	__stop_static_call_tramp_key = .;
> 
> #ifndef RO_AFTER_INIT_DATA
> #define RO_AFTER_INIT_DATA						\
> 	. = ALIGN(8);							\
> 	__start_ro_after_init = .;					\
> 	*(.data..ro_after_init)						\
> 	JUMP_TABLE_DATA							\
> 	STATIC_CALL_DATA						\
> 	__end_ro_after_init = .;
> #endif
> 
> 	.rodata           : AT(ADDR(.rodata) - LOAD_OFFSET) {		\
> 		__start_rodata = .;					\
> 		*(.rodata) *(.rodata.*)					\
> 		SCHED_DATA						\
> 		RO_AFTER_INIT_DATA	/* Read only after init */	\
> 		. = ALIGN(8);						\
> 		__start___tracepoints_ptrs = .;				\
> 		KEEP(*(__tracepoints_ptrs)) /* Tracepoints: pointer array */ \
> 		__stop___tracepoints_ptrs = .;				\
> 		*(__tracepoints_strings)/* Tracepoints: strings */	\
> 	}								\
> 
> 
> $ nm defconfig-build/vmlinux | grep static_call_sites
> ffffffff82916dc0 D __start_static_call_sites
> ffffffff8291aaf0 D __stop_static_call_sites
> 
> 
> Which all reads to me like it *SHOULD* work. Howver when I was debugging
> the crash as reported by Steve (using his .config), I definitely saw
> non-aligned base offsets causing mayhem.

Ooooh, modules don't have this. They still have regular
.static_call_sites sections, and *those* are unaligned.

Section Headers:
[Nr] Name              Type            Address          Off    Size   ES Flg Lk Inf Al

[16] .static_call_sites PROGBITS        0000000000000000 008aa1 0006f0 00  WA  0   0  1

And that goes *BOOM*.. Let me ses if I can figure out how to make
objtool align those sections.