| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Mon, 26 Apr 2021 19:13:33 +0200
From: Peter Zijlstra <peterz@...radead.org>
To: Segher Boessenkool <segher@...nel.crashing.org>
Cc: David Malcolm <dmalcolm@...hat.com>,
Ard Biesheuvel <ardb@...nel.org>,
linux-toolchains@...r.kernel.org,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Josh Poimboeuf <jpoimboe@...hat.com>,
Jason Baron <jbaron@...mai.com>,
"Steven Rostedt (VMware)" <rostedt@...dmis.org>,
yuanzhaoxiong@...du.com
Subject: Re: static_branch/jump_label vs branch merging
On Thu, Apr 22, 2021 at 01:31:51PM -0500, Segher Boessenkool wrote:
> It sounds like you want it to behave like attribute((pure)) already is
> documented as doing. Please open a PR? https://gcc.gnu.org/bugs.html
> (We need buildable stand-alone example code, with what flags to use, and
> something like what should happen and what did happen).
Something like so?
---
gcc (Debian 10.2.1-6) 10.2.1 20210110
Clear fail on both counts, the first emits _3_ nops, where 1 is
expected, and the second emits the nop _inside_ the loop.
0000000000001180 <elide_branches>:
1180: 48 83 ec 08 sub $0x8,%rsp
1184: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1)
1189: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1)
118e: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1)
1193: 48 83 c4 08 add $0x8,%rsp
1197: c3 retq
1198: 0f 1f 84 00 00 00 00 00 nopl 0x0(%rax,%rax,1)
11a0: 48 8d 3d 5d 0e 00 00 lea 0xe5d(%rip),%rdi # 2004 <_IO_stdin_used+0x4>
11a7: 31 c0 xor %eax,%eax
11a9: e8 a2 fe ff ff callq 1050 <printf@plt>
11ae: eb d9 jmp 1189 <elide_branches+0x9>
11b0: 48 8d 3d 5a 0e 00 00 lea 0xe5a(%rip),%rdi # 2011 <_IO_stdin_used+0x11>
11b7: 48 83 c4 08 add $0x8,%rsp
11bb: e9 80 fe ff ff jmpq 1040 <puts@plt>
11c0: 48 8d 3d 45 0e 00 00 lea 0xe45(%rip),%rdi # 200c <_IO_stdin_used+0xc>
11c7: 31 c0 xor %eax,%eax
11c9: e8 82 fe ff ff callq 1050 <printf@plt>
11ce: eb be jmp 118e <elide_branches+0xe>
00000000000011d0 <hoist>:
11d0: 53 push %rbx
11d1: 31 db xor %ebx,%ebx
11d3: eb 16 jmp 11eb <hoist+0x1b>
11d5: 0f 1f 00 nopl (%rax)
11d8: 89 de mov %ebx,%esi
11da: 48 8d 3d 36 0e 00 00 lea 0xe36(%rip),%rdi # 2017 <_IO_stdin_used+0x17>
11e1: 31 c0 xor %eax,%eax
11e3: 83 c3 01 add $0x1,%ebx
11e6: e8 65 fe ff ff callq 1050 <printf@plt>
11eb: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1)
11f0: b8 0a 00 00 00 mov $0xa,%eax
11f5: 39 c3 cmp %eax,%ebx
11f7: 7c df jl 11d8 <hoist+0x8>
11f9: bf 0a 00 00 00 mov $0xa,%edi
11fe: 5b pop %rbx
11ff: e9 2c fe ff ff jmpq 1030 <putchar@plt>
1204: 0f 1f 40 00 nopl 0x0(%rax)
1208: b8 05 00 00 00 mov $0x5,%eax
120d: eb e6 jmp 11f5 <hoist+0x25>
120f: 90 nop
---
/* gcc -O2 -o pure-fail pure-fail.c */
#include <stdio.h>
#include <stdbool.h>
#define __pure __attribute__((__pure__))
#define __stringify_1(x...) #x
#define __stringify(x...) __stringify_1(x)
#ifndef __x86_64__
#define BYTES_NOP4 0x8d,0x74,0x26,0x00
#define BYTES_NOP5 0x3e,BYTES_NOP4
#else
#define BYTES_NOP5 0x0f,0x1f,0x44,0x00,0x00
#endif
# define __ASM_FORM(x) " " __stringify(x) " "
# define __ASM_FORM_RAW(x) __stringify(x)
# define __ASM_FORM_COMMA(x) " " __stringify(x) ","
#ifndef __x86_64__
/* 32 bit */
# define __ASM_SEL(a,b) __ASM_FORM(a)
# define __ASM_SEL_RAW(a,b) __ASM_FORM_RAW(a)
#else
/* 64 bit */
# define __ASM_SEL(a,b) __ASM_FORM(b)
# define __ASM_SEL_RAW(a,b) __ASM_FORM_RAW(b)
#endif
#define __ASM_SIZE(inst, ...) __ASM_SEL(inst##l##__VA_ARGS__, \
inst##q##__VA_ARGS__)
#define __ASM_REG(reg) __ASM_SEL_RAW(e##reg, r##reg)
#define _ASM_PTR __ASM_SEL(.long, .quad)
#define _ASM_ALIGN __ASM_SEL(.balign 4, .balign 8)
#define asm_volatile_goto(x...) do { asm goto(x); asm (""); } while (0)
/* --- */
struct static_key {
int enabled;
};
#define STATIC_KEY_INIT_TRUE { .enabled = 1, }
#define STATIC_KEY_INIT_FALSE { .enabled = 0, }
static __always_inline __pure bool arch_static_branch(struct static_key * const key, const bool branch)
{
asm_volatile_goto("1:"
".byte " __stringify(BYTES_NOP5) "\n\t"
".pushsection __jump_table, \"aw\" \n\t"
_ASM_ALIGN "\n\t"
".long 1b - ., %l[l_yes] - . \n\t"
_ASM_PTR "%c0 + %c1 - .\n\t"
".popsection \n\t"
: : "i" (key), "i" (branch) : : l_yes);
return false;
l_yes:
return true;
}
static __always_inline __pure bool static_key_false(struct static_key * const key)
{
return arch_static_branch(key, false);
}
static __always_inline __pure bool static_key_true(struct static_key * const key)
{
return !arch_static_branch(key, true);
}
/* --- */
static struct static_key key_A = STATIC_KEY_INIT_FALSE;
/*
* Expect:
*
* if (static_key_false(&key_A)) {
* printf("ponies ");
* printf("are ");
* printf("small\n");
* }
*/
void elide_branches(void)
{
if (static_key_false(&key_A))
printf("ponies ");
if (static_key_false(&key_A))
printf("are ");
if (static_key_false(&key_A))
printf("small\n");
}
/* --- */
static struct static_key key_B = STATIC_KEY_INIT_TRUE;
static __pure int count(void)
{
if (static_key_true(&key_B))
return 10;
return 5;
}
/*
* Expect:
*
* tmp = count();
* for (i = 0; i < tmp; i++)
* printf("%d, ", i);
* printf("\n");
*/
void hoist(void)
{
int i;
for (i = 0; i < count(); i++)
printf("%d, ", i);
printf("\n");
}
/* --- */
int main(int argc, char **argv)
{
elide_branches();
hoist();
return 0;
}
Powered by blists - more mailing lists