kernel_optimize_test/arch/arm/kernel/entry-ftrace.S
Ard Biesheuvel c241cfd0a5 ARM: ftrace: avoid redundant loads or clobbering IP
[ Upstream commit d11967870815b5ab89843980e35aab616c97c463 ]

Tweak the ftrace return paths to avoid redundant loads of SP, as well as
unnecessary clobbering of IP.

This also fixes the inconsistency of using MOV to perform a function
return, which is sub-optimal on recent micro-architectures but more
importantly, does not perform an interworking return, unlike compiler
generated function returns in Thumb2 builds.

Let's fix this by popping PC from the stack like most ordinary code
does.

Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2022-04-08 14:40:34 +02:00

266 lines
5.8 KiB
ArmAsm

/* SPDX-License-Identifier: GPL-2.0-only */
#include <asm/assembler.h>
#include <asm/ftrace.h>
#include <asm/unwind.h>
#include "entry-header.S"
/*
* When compiling with -pg, gcc inserts a call to the mcount routine at the
* start of every function. In mcount, apart from the function's address (in
* lr), we need to get hold of the function's caller's address.
*
* Newer GCCs (4.4+) solve this problem by using a version of mcount with call
* sites like:
*
* push {lr}
* bl __gnu_mcount_nc
*
* With these compilers, frame pointers are not necessary.
*
* mcount can be thought of as a function called in the middle of a subroutine
* call. As such, it needs to be transparent for both the caller and the
* callee: the original lr needs to be restored when leaving mcount, and no
* registers should be clobbered.
*
* When using dynamic ftrace, we patch out the mcount call by a "pop {lr}"
* instead of the __gnu_mcount_nc call (see arch/arm/kernel/ftrace.c).
*/
.macro mcount_adjust_addr rd, rn
bic \rd, \rn, #1 @ clear the Thumb bit if present
sub \rd, \rd, #MCOUNT_INSN_SIZE
.endm
.macro __mcount suffix
mcount_enter
ldr r0, =ftrace_trace_function
ldr r2, [r0]
adr r0, .Lftrace_stub
cmp r0, r2
bne 1f
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
ldr r1, =ftrace_graph_return
ldr r2, [r1]
cmp r0, r2
bne ftrace_graph_caller\suffix
ldr r1, =ftrace_graph_entry
ldr r2, [r1]
ldr r0, =ftrace_graph_entry_stub
cmp r0, r2
bne ftrace_graph_caller\suffix
#endif
mcount_exit
1: mcount_get_lr r1 @ lr of instrumented func
mcount_adjust_addr r0, lr @ instrumented function
badr lr, 2f
mov pc, r2
2: mcount_exit
.endm
#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
.macro __ftrace_regs_caller
str lr, [sp, #-8]! @ store LR as PC and make space for CPSR/OLD_R0,
@ OLD_R0 will overwrite previous LR
ldr lr, [sp, #8] @ get previous LR
str r0, [sp, #8] @ write r0 as OLD_R0 over previous LR
str lr, [sp, #-4]! @ store previous LR as LR
add lr, sp, #16 @ move in LR the value of SP as it was
@ before the push {lr} of the mcount mechanism
push {r0-r11, ip, lr}
@ stack content at this point:
@ 0 4 48 52 56 60 64 68 72
@ R0 | R1 | ... | IP | SP + 4 | previous LR | LR | PSR | OLD_R0 |
mov r3, sp @ struct pt_regs*
ldr r2, =function_trace_op
ldr r2, [r2] @ pointer to the current
@ function tracing op
ldr r1, [sp, #S_LR] @ lr of instrumented func
ldr lr, [sp, #S_PC] @ get LR
mcount_adjust_addr r0, lr @ instrumented function
.globl ftrace_regs_call
ftrace_regs_call:
bl ftrace_stub
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
.globl ftrace_graph_regs_call
ftrace_graph_regs_call:
mov r0, r0
#endif
@ pop saved regs
pop {r0-r11, ip, lr} @ restore r0 through r12
ldr lr, [sp], #4 @ restore LR
ldr pc, [sp], #12
.endm
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
.macro __ftrace_graph_regs_caller
sub r0, fp, #4 @ lr of instrumented routine (parent)
@ called from __ftrace_regs_caller
ldr r1, [sp, #S_PC] @ instrumented routine (func)
mcount_adjust_addr r1, r1
mov r2, fp @ frame pointer
bl prepare_ftrace_return
@ pop registers saved in ftrace_regs_caller
pop {r0-r11, ip, lr} @ restore r0 through r12
ldr lr, [sp], #4 @ restore LR
ldr pc, [sp], #12
.endm
#endif
#endif
.macro __ftrace_caller suffix
mcount_enter
mcount_get_lr r1 @ lr of instrumented func
mcount_adjust_addr r0, lr @ instrumented function
#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
ldr r2, =function_trace_op
ldr r2, [r2] @ pointer to the current
@ function tracing op
mov r3, #0 @ regs is NULL
#endif
.globl ftrace_call\suffix
ftrace_call\suffix:
bl ftrace_stub
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
.globl ftrace_graph_call\suffix
ftrace_graph_call\suffix:
mov r0, r0
#endif
mcount_exit
.endm
.macro __ftrace_graph_caller
sub r0, fp, #4 @ &lr of instrumented routine (&parent)
#ifdef CONFIG_DYNAMIC_FTRACE
@ called from __ftrace_caller, saved in mcount_enter
ldr r1, [sp, #16] @ instrumented routine (func)
mcount_adjust_addr r1, r1
#else
@ called from __mcount, untouched in lr
mcount_adjust_addr r1, lr @ instrumented routine (func)
#endif
mov r2, fp @ frame pointer
bl prepare_ftrace_return
mcount_exit
.endm
/*
* __gnu_mcount_nc
*/
.macro mcount_enter
/*
* This pad compensates for the push {lr} at the call site. Note that we are
* unable to unwind through a function which does not otherwise save its lr.
*/
UNWIND(.pad #4)
stmdb sp!, {r0-r3, lr}
UNWIND(.save {r0-r3, lr})
.endm
.macro mcount_get_lr reg
ldr \reg, [sp, #20]
.endm
.macro mcount_exit
ldmia sp!, {r0-r3}
ldr lr, [sp, #4]
ldr pc, [sp], #8
.endm
ENTRY(__gnu_mcount_nc)
UNWIND(.fnstart)
#ifdef CONFIG_DYNAMIC_FTRACE
push {lr}
ldr lr, [sp, #4]
ldr pc, [sp], #8
#else
__mcount
#endif
UNWIND(.fnend)
ENDPROC(__gnu_mcount_nc)
#ifdef CONFIG_DYNAMIC_FTRACE
ENTRY(ftrace_caller)
UNWIND(.fnstart)
__ftrace_caller
UNWIND(.fnend)
ENDPROC(ftrace_caller)
#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
ENTRY(ftrace_regs_caller)
UNWIND(.fnstart)
__ftrace_regs_caller
UNWIND(.fnend)
ENDPROC(ftrace_regs_caller)
#endif
#endif
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
ENTRY(ftrace_graph_caller)
UNWIND(.fnstart)
__ftrace_graph_caller
UNWIND(.fnend)
ENDPROC(ftrace_graph_caller)
#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
ENTRY(ftrace_graph_regs_caller)
UNWIND(.fnstart)
__ftrace_graph_regs_caller
UNWIND(.fnend)
ENDPROC(ftrace_graph_regs_caller)
#endif
#endif
.purgem mcount_enter
.purgem mcount_get_lr
.purgem mcount_exit
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
.globl return_to_handler
return_to_handler:
stmdb sp!, {r0-r3}
mov r0, fp @ frame pointer
bl ftrace_return_to_handler
mov lr, r0 @ r0 has real ret addr
ldmia sp!, {r0-r3}
ret lr
#endif
ENTRY(ftrace_stub)
.Lftrace_stub:
ret lr
ENDPROC(ftrace_stub)