system/freebsd-src
system/freebsd-src
1
0
Fork 0
mirror of https://github.com/freebsd/freebsd-src synced 2024-07-22 02:37:15 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

x86: remove gcov kernel support

Browse source 
Reviewed by:	jhb
Sponsored by:	The FreeBSD Foundation
Differential revision:	https://reviews.freebsd.org/D29529
This commit is contained in:
Konstantin Belousov 2021-04-01 00:04:55 +03:00
parent 76b1b5ce6d
commit aa3ea612be
42 changed files with 79 additions and 1914 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
sys/amd64/amd64/apic_vector.S
View file

@ -81,7 +81,6 @@ as_lapic_eoi:
*/
.macro ISR_VEC index, vec_name
INTR_HANDLER \vec_name
FAKE_MCOUNT(TF_RIP(%rsp))
cmpl $0,x2apic_mode
je 1f
movl $(MSR_APIC_ISR0 + \index),%ecx
@ -98,7 +97,6 @@ as_lapic_eoi:
movl %eax, %edi /* pass the IRQ */
call lapic_handle_intr
3:
MEXITCOUNT
jmp doreti
.endm
@ -127,28 +125,22 @@ IDTVEC(spuriousint)
* Local APIC periodic timer handler.
*/
INTR_HANDLER timerint
FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp, %rdi
call lapic_handle_timer
MEXITCOUNT
jmp doreti
/*
* Local APIC CMCI handler.
*/
INTR_HANDLER cmcint
FAKE_MCOUNT(TF_RIP(%rsp))
call lapic_handle_cmc
MEXITCOUNT
jmp doreti
/*
* Local APIC error interrupt handler.
*/
INTR_HANDLER errorint
FAKE_MCOUNT(TF_RIP(%rsp))
call lapic_handle_error
MEXITCOUNT
jmp doreti
#ifdef XENHVM
@ -157,10 +149,8 @@ IDTVEC(spuriousint)
* Only used when the hypervisor supports direct vector callbacks.
*/
INTR_HANDLER xen_intr_upcall
FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp, %rdi
call xen_intr_handle_upcall
MEXITCOUNT
jmp doreti
#endif
@ -184,9 +174,7 @@ IDTVEC(spuriousint)
*/
INTR_HANDLER ipi_intr_bitmap_handler
call as_lapic_eoi
FAKE_MCOUNT(TF_RIP(%rsp))
call ipi_bitmap_handler
MEXITCOUNT
jmp doreti
/*
@ -210,9 +198,7 @@ IDTVEC(spuriousint)
*/
INTR_HANDLER ipi_swi
call as_lapic_eoi
FAKE_MCOUNT(TF_RIP(%rsp))
call ipi_swi_handler
MEXITCOUNT
jmp doreti
/*

2
sys/amd64/amd64/atpic_vector.S
View file

@ -44,11 +44,9 @@
*/
.macro INTR irq_num, vec_name
INTR_HANDLER \vec_name
FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp, %rsi
movl $\irq_num, %edi /* pass the IRQ */
call atpic_handle_intr
MEXITCOUNT
jmp doreti
.endm

25
sys/amd64/amd64/exception.S
View file

@ -101,9 +101,6 @@ dtrace_invop_calltrap_addr:
* registers are reloaded on return to the usermode.
*/
MCOUNT_LABEL(user)
MCOUNT_LABEL(btrap)
/* Traps that we leave interrupts disabled for. */
.macro TRAP_NOEN l, trapno
PTI_ENTRY \l,\l\()_pti_k,\l\()_pti_u
@ -257,7 +254,6 @@ alltraps_pushregs_no_rax:
pushfq
andq $~(PSL_D | PSL_AC),(%rsp)
popfq
FAKE_MCOUNT(TF_RIP(%rsp))
#ifdef KDTRACE_HOOKS
/*
* DTrace Function Boundary Trace (fbt) probes are triggered
@ -288,7 +284,6 @@ alltraps_pushregs_no_rax:
calltrap:
movq %rsp,%rdi
call trap_check
MEXITCOUNT
jmp doreti /* Handle any pending ASTs */
/*
@ -576,7 +571,6 @@ fast_syscall_common:
movq %r14,TF_R14(%rsp) /* C preserved */
movq %r15,TF_R15(%rsp) /* C preserved */
movl $TF_HASSEGS,TF_FLAGS(%rsp)
FAKE_MCOUNT(TF_RIP(%rsp))
movq PCPU(CURTHREAD),%rdi
movq %rsp,TD_FRAME(%rdi)
movl TF_RFLAGS(%rsp),%esi
@ -594,7 +588,6 @@ fast_syscall_common:
call handle_ibrs_exit
callq *mds_handler
/* Restore preserved registers. */
MEXITCOUNT
movq TF_RDI(%rsp),%rdi /* bonus; preserve arg 1 */
movq TF_RSI(%rsp),%rsi /* bonus: preserve arg 2 */
movq TF_RDX(%rsp),%rdx /* return value 2 */
@ -621,7 +614,6 @@ fast_syscall_common:
jmp 1b
4: /* Requested full context restore, use doreti for that. */
MEXITCOUNT
jmp doreti
/*
@ -692,10 +684,8 @@ IDTVEC(dbg)
rdmsr
movl %eax,%r14d
call handle_ibrs_entry
2: FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp,%rdi
2: movq %rsp,%rdi
call trap
MEXITCOUNT
testl $CPUID_STDEXT3_IBPB,cpu_stdext_feature3(%rip)
je 3f
movl %r14d,%eax
@ -866,10 +856,8 @@ nmi_fromuserspace:
3:
/* Note: this label is also used by ddb and gdb: */
nmi_calltrap:
FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp,%rdi
call trap
MEXITCOUNT
#ifdef HWPMC_HOOKS
/*
* Capture a userspace callchain if needed.
@ -1055,10 +1043,8 @@ mchk_fromuserspace:
1: call handle_ibrs_entry
/* Note: this label is also used by ddb and gdb: */
mchk_calltrap:
FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp,%rdi
call mca_intr
MEXITCOUNT
testl %ebx,%ebx /* %ebx != 0 => return to userland */
jnz doreti_exit
/*
@ -1089,7 +1075,6 @@ ENTRY(fork_trampoline)
movq %rbx,%rsi /* arg1 */
movq %rsp,%rdx /* trapframe pointer */
call fork_exit
MEXITCOUNT
jmp doreti /* Handle any ASTs */
/*
@ -1115,8 +1100,6 @@ ENTRY(fork_trampoline)
.p2align 4
.text
SUPERALIGN_TEXT
MCOUNT_LABEL(bintr)
#include <amd64/amd64/apic_vector.S>
#ifdef DEV_ATPIC
@ -1128,9 +1111,6 @@ MCOUNT_LABEL(bintr)
#include <amd64/amd64/atpic_vector.S>
#endif
.text
MCOUNT_LABEL(eintr)
/*
* void doreti(struct trapframe)
*
@ -1141,7 +1121,6 @@ MCOUNT_LABEL(eintr)
.type doreti,@function
.globl doreti
doreti:
FAKE_MCOUNT($bintr) /* init "from" bintr -> doreti */
/*
* Check if ASTs can be handled now.
*/
@ -1171,7 +1150,6 @@ doreti_ast:
* registers. The fault is handled in trap.c.
*/
doreti_exit:
MEXITCOUNT
movq PCPU(CURPCB),%r8
/*
@ -1332,7 +1310,6 @@ doreti_iret_fault:
movl $T_PROTFLT,TF_TRAPNO(%rsp)
movq $0,TF_ERR(%rsp) /* XXX should be the error code */
movq $0,TF_ADDR(%rsp)
FAKE_MCOUNT(TF_RIP(%rsp))
jmp calltrap
ALIGN_TEXT

10
sys/amd64/amd64/locore.S
View file

@ -54,7 +54,7 @@
*
* We are already in long mode, on a 64 bit %cs and running at KERNBASE.
*/
NON_GPROF_ENTRY(btext)
ENTRY(btext)
/* Tell the bios to warmboot next time */
movw $0x1234,0x472
@ -79,7 +79,7 @@ NON_GPROF_ENTRY(btext)
jmp 0b
/* la57_trampoline(%rdi pml5) */
NON_GPROF_ENTRY(la57_trampoline)
ENTRY(la57_trampoline)
movq %rsp,%r11
movq %rbx,%r10
leaq la57_trampoline_end(%rip),%rsp
@ -118,11 +118,11 @@ l2: movq %r11,%rsp
movq %r10,%rbx
retq
.p2align 4,0
NON_GPROF_ENTRY(la57_trampoline_gdt_desc)
ENTRY(la57_trampoline_gdt_desc)
.word la57_trampoline_end - la57_trampoline_gdt
.long 0 /* filled by pmap_bootstrap_la57 */
.p2align 4,0
NON_GPROF_ENTRY(la57_trampoline_gdt)
ENTRY(la57_trampoline_gdt)
.long 0x00000000 /* null desc */
.long 0x00000000
.long 0x00000000 /* 64bit code */
@ -132,7 +132,7 @@ NON_GPROF_ENTRY(la57_trampoline_gdt)
.long 0x0000ffff /* universal data */
.long 0x00cf9300
.dcb.l 16,0
NON_GPROF_ENTRY(la57_trampoline_end)
ENTRY(la57_trampoline_end)
.bss
ALIGN_DATA /* just to be sure */

323
sys/amd64/amd64/prof_machdep.c
View file

@ -1,323 +0,0 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 1996 Bruce D. Evans.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#ifdef GUPROF
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/cpu.h>
#include <sys/eventhandler.h>
#include <sys/gmon.h>
#include <sys/kernel.h>
#include <sys/smp.h>
#include <sys/sysctl.h>
#include <machine/clock.h>
#include <machine/timerreg.h>
#define CPUTIME_CLOCK_UNINITIALIZED 0
#define CPUTIME_CLOCK_I8254 1
#define CPUTIME_CLOCK_TSC 2
#define CPUTIME_CLOCK_I8254_SHIFT 7
int cputime_bias = 1; /* initialize for locality of reference */
static int cputime_clock = CPUTIME_CLOCK_UNINITIALIZED;
static int cputime_prof_active;
#endif /* GUPROF */
#ifdef __GNUCLIKE_ASM
#if defined(SMP) && defined(GUPROF)
#define MPLOCK " \n\
movl $1,%edx \n\
9: \n\
xorl %eax,%eax \n\
lock \n\
cmpxchgl %edx,mcount_lock \n\
jne 9b \n"
#define MPUNLOCK "movl $0,mcount_lock \n"
#else /* !(SMP && GUPROF) */
#define MPLOCK
#define MPUNLOCK
#endif /* SMP && GUPROF */
__asm(" \n\
GM_STATE = 0 \n\
GMON_PROF_OFF = 3 \n\
\n\
.text \n\
.p2align 4,0x90 \n\
.globl __mcount \n\
.type __mcount,@function \n\
__mcount: \n\
# \n\
# Check that we are profiling. Do it early for speed. \n\
# \n\
cmpl $GMON_PROF_OFF,_gmonparam+GM_STATE \n\
je .mcount_exit \n\
# \n\
# __mcount is the same as [.]mcount except the caller \n\
# hasn't changed the stack except to call here, so the \n\
# caller's raddr is above our raddr. \n\
# \n\
pushq %rax \n\
pushq %rdx \n\
pushq %rcx \n\
pushq %rsi \n\
pushq %rdi \n\
pushq %r8 \n\
pushq %r9 \n\
movq 7*8+8(%rsp),%rdi \n\
jmp .got_frompc \n\
\n\
.p2align 4,0x90 \n\
.globl .mcount \n\
.mcount: \n\
cmpl $GMON_PROF_OFF,_gmonparam+GM_STATE \n\
je .mcount_exit \n\
# \n\
# The caller's stack frame has already been built, so \n\
# %rbp is the caller's frame pointer. The caller's \n\
# raddr is in the caller's frame following the caller's \n\
# caller's frame pointer. \n\
# \n\
pushq %rax \n\
pushq %rdx \n\
pushq %rcx \n\
pushq %rsi \n\
pushq %rdi \n\
pushq %r8 \n\
pushq %r9 \n\
movq 8(%rbp),%rdi \n\
.got_frompc: \n\
# \n\
# Our raddr is the caller's pc. \n\
# \n\
movq 7*8(%rsp),%rsi \n\
\n\
pushfq \n\
cli \n"
MPLOCK " \n\
call mcount \n"
MPUNLOCK " \n\
popfq \n\
popq %r9 \n\
popq %r8 \n\
popq %rdi \n\
popq %rsi \n\
popq %rcx \n\
popq %rdx \n\
popq %rax \n\
.mcount_exit: \n\
ret $0 \n\
");
#else /* !__GNUCLIKE_ASM */
#error "this file needs to be ported to your compiler"
#endif /* __GNUCLIKE_ASM */
#ifdef GUPROF
/*
* [.]mexitcount saves the return register(s), loads selfpc and calls
* mexitcount(selfpc) to do the work. Someday it should be in a machine
* dependent file together with cputime(), __mcount and [.]mcount. cputime()
* can't just be put in machdep.c because it has to be compiled without -pg.
*/
#ifdef __GNUCLIKE_ASM
__asm(" \n\
.text \n\
# \n\
# Dummy label to be seen when gprof -u hides [.]mexitcount. \n\
# \n\
.p2align 4,0x90 \n\
.globl __mexitcount \n\
.type __mexitcount,@function \n\
__mexitcount: \n\
nop \n\
\n\
GMON_PROF_HIRES = 4 \n\
\n\
.p2align 4,0x90 \n\
.globl .mexitcount \n\
.mexitcount: \n\
cmpl $GMON_PROF_HIRES,_gmonparam+GM_STATE \n\
jne .mexitcount_exit \n\
pushq %rax \n\
pushq %rdx \n\
pushq %rcx \n\
pushq %rsi \n\
pushq %rdi \n\
pushq %r8 \n\
pushq %r9 \n\
movq 7*8(%rsp),%rdi \n\
pushfq \n\
cli \n"
MPLOCK " \n\
call mexitcount \n"
MPUNLOCK " \n\
popfq \n\
popq %r9 \n\
popq %r8 \n\
popq %rdi \n\
popq %rsi \n\
popq %rcx \n\
popq %rdx \n\
popq %rax \n\
.mexitcount_exit: \n\
ret $0 \n\
");
#endif /* __GNUCLIKE_ASM */
/*
* Return the time elapsed since the last call. The units are machine-
* dependent.
*/
int
cputime()
{
u_int count;
int delta;
u_char high, low;
static u_int prev_count;
if (cputime_clock == CPUTIME_CLOCK_TSC) {
/*
* Scale the TSC a little to make cputime()'s frequency
* fit in an int, assuming that the TSC frequency fits
* in a u_int. Use a fixed scale since dynamic scaling
* would be slower and we can't really use the low bit
* of precision.
*/
count = (u_int)rdtsc() & ~1u;
delta = (int)(count - prev_count) >> 1;
prev_count = count;
return (delta);
}
/*
* Read the current value of the 8254 timer counter 0.
*/
outb(TIMER_MODE, TIMER_SEL0 | TIMER_LATCH);
low = inb(TIMER_CNTR0);
high = inb(TIMER_CNTR0);
count = ((high << 8) | low) << CPUTIME_CLOCK_I8254_SHIFT;
/*
* The timer counts down from TIMER_CNTR0_MAX to 0 and then resets.
* While profiling is enabled, this routine is called at least twice
* per timer reset (for mcounting and mexitcounting hardclock()),
* so at most one reset has occurred since the last call, and one
* has occurred iff the current count is larger than the previous
* count. This allows counter underflow to be detected faster
* than in microtime().
*/
delta = prev_count - count;
prev_count = count;
if ((int) delta <= 0)
return (delta + (i8254_max_count << CPUTIME_CLOCK_I8254_SHIFT));
return (delta);
}
static int
sysctl_machdep_cputime_clock(SYSCTL_HANDLER_ARGS)
{
int clock;
int error;
clock = cputime_clock;
error = sysctl_handle_opaque(oidp, &clock, sizeof clock, req);
if (error == 0 && req->newptr != NULL) {
if (clock < 0 || clock > CPUTIME_CLOCK_TSC)
return (EINVAL);
cputime_clock = clock;
}
return (error);
}
SYSCTL_PROC(_machdep, OID_AUTO, cputime_clock,
CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(u_int),
sysctl_machdep_cputime_clock, "I",
"");
/*
* The start and stop routines need not be here since we turn off profiling
* before calling them. They are here for convenience.
*/
void
startguprof(gp)
struct gmonparam *gp;
{
uint64_t freq;
freq = atomic_load_acq_64(&tsc_freq);
if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED) {
if (freq != 0 && mp_ncpus == 1)
cputime_clock = CPUTIME_CLOCK_TSC;
else
cputime_clock = CPUTIME_CLOCK_I8254;
}
if (cputime_clock == CPUTIME_CLOCK_TSC) {
gp->profrate = freq >> 1;
cputime_prof_active = 1;
} else
gp->profrate = i8254_freq << CPUTIME_CLOCK_I8254_SHIFT;
cputime_bias = 0;
cputime();
}
void
stopguprof(gp)
struct gmonparam *gp;
{
if (cputime_clock == CPUTIME_CLOCK_TSC)
cputime_prof_active = 0;
}
/* If the cpu frequency changed while profiling, report a warning. */
static void
tsc_freq_changed(void *arg, const struct cf_level *level, int status)
{
/*
* If there was an error during the transition or
* TSC is P-state invariant, don't do anything.
*/
if (status != 0 || tsc_is_invariant)
return;
if (cputime_prof_active && cputime_clock == CPUTIME_CLOCK_TSC)
printf("warning: cpu freq changed while profiling active\n");
}
EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL,
EVENTHANDLER_PRI_ANY);
#endif /* GUPROF */

2
sys/amd64/amd64/sigtramp.S
View file

@ -38,7 +38,7 @@
* Signal trampoline, copied to top of user stack
*
*/
NON_GPROF_ENTRY(sigcode)
ENTRY(sigcode)
call *SIGF_HANDLER(%rsp) /* call signal handler */
lea SIGF_UC(%rsp),%rdi /* get ucontext_t */
pushq $0 /* junk to fake return addr. */

1
sys/amd64/amd64/support.S
View file

@ -1490,7 +1490,6 @@ ENTRY(lgdt)
popq %rax
pushq $KCSEL
pushq %rax
MEXITCOUNT
lretq
END(lgdt)

6
sys/amd64/amd64/xen-locore.S
View file

@ -69,11 +69,11 @@
.text
.p2align PAGE_SHIFT, 0x90 /* Hypercall_page needs to be PAGE aligned */
NON_GPROF_ENTRY(hypercall_page)
ENTRY(hypercall_page)
.skip 0x1000, 0x90 /* Fill with "nop"s */
/* Legacy PVH entry point, to be removed. */
NON_GPROF_ENTRY(xen_start)
ENTRY(xen_start)
/* Don't trust what the loader gives for rflags. */
pushq $PSL_KERNEL
popfq
@ -97,7 +97,7 @@ NON_GPROF_ENTRY(xen_start)
/* PVH entry point. */
.code32
NON_GPROF_ENTRY(xen_start32)
ENTRY(xen_start32)
/* Load flat GDT */
movl $VTOP(gdtdesc32), %eax

4
sys/amd64/conf/NOTES
View file

@ -7,10 +7,6 @@
# $FreeBSD$
#
#
# We want LINT to cover profiling as well.
profile 2
#
# Enable the kernel DTrace hooks which are required to load the DTrace
# kernel modules.

2
sys/amd64/ia32/ia32_exception.S
View file

@ -73,8 +73,6 @@ int0x80_syscall_common:
pushfq
andq $~(PSL_D | PSL_AC),(%rsp)
popfq
FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp, %rdi
call ia32_syscall
MEXITCOUNT
jmp doreti

82
sys/amd64/include/asmacros.h
View file

@ -55,92 +55,14 @@
#define CNAME(csym) csym
#define ALIGN_DATA .p2align 3 /* 8 byte alignment, zero filled */
#ifdef GPROF
#define ALIGN_TEXT .p2align 4,0x90 /* 16-byte alignment, nop filled */
#else
#define ALIGN_TEXT .p2align 4,0x90 /* 16-byte alignment, nop filled */
#endif
#define SUPERALIGN_TEXT .p2align 4,0x90 /* 16-byte alignment, nop filled */
#define GEN_ENTRY(name) ALIGN_TEXT; .globl CNAME(name); \
.type CNAME(name),@function; CNAME(name):
#define NON_GPROF_ENTRY(name) GEN_ENTRY(name)
#define NON_GPROF_RET .byte 0xc3 /* opcode for `ret' */
#define END(name) .size name, . - name
#ifdef GPROF
/*
* __mcount is like [.]mcount except that doesn't require its caller to set
* up a frame pointer. It must be called before pushing anything onto the
* stack. gcc should eventually generate code to call __mcount in most
* cases. This would make -pg in combination with -fomit-frame-pointer
* useful. gcc has a configuration variable PROFILE_BEFORE_PROLOGUE to
* allow profiling before setting up the frame pointer, but this is
* inadequate for good handling of special cases, e.g., -fpic works best
* with profiling after the prologue.
*
* [.]mexitcount is a new function to support non-statistical profiling if an
* accurate clock is available. For C sources, calls to it are generated
* by the FreeBSD extension `-mprofiler-epilogue' to gcc. It is best to
* call [.]mexitcount at the end of a function like the MEXITCOUNT macro does,
* but gcc currently generates calls to it at the start of the epilogue to
* avoid problems with -fpic.
*
* [.]mcount and __mcount may clobber the call-used registers and %ef.
* [.]mexitcount may clobber %ecx and %ef.
*
* Cross-jumping makes non-statistical profiling timing more complicated.
* It is handled in many cases by calling [.]mexitcount before jumping. It
* is handled for conditional jumps using CROSSJUMP() and CROSSJUMP_LABEL().
* It is handled for some fault-handling jumps by not sharing the exit
* routine.
*
* ALTENTRY() must be before a corresponding ENTRY() so that it can jump to
* the main entry point. Note that alt entries are counted twice. They
* have to be counted as ordinary entries for gprof to get the call times
* right for the ordinary entries.
*
* High local labels are used in macros to avoid clashes with local labels
* in functions.
*
* Ordinary `ret' is used instead of a macro `RET' because there are a lot
* of `ret's. 0xc3 is the opcode for `ret' (`#define ret ... ret' can't
* be used because this file is sometimes preprocessed in traditional mode).
* `ret' clobbers eflags but this doesn't matter.
*/
#define ALTENTRY(name) GEN_ENTRY(name) ; MCOUNT ; MEXITCOUNT ; jmp 9f
#define CROSSJUMP(jtrue, label, jfalse) \
jfalse 8f; MEXITCOUNT; jmp __CONCAT(to,label); 8:
#define CROSSJUMPTARGET(label) \
ALIGN_TEXT; __CONCAT(to,label): ; MCOUNT; jmp label
#define ENTRY(name) GEN_ENTRY(name) ; 9: ; MCOUNT
#define FAKE_MCOUNT(caller) pushq caller ; call __mcount ; popq %rcx
#define MCOUNT call __mcount
#define MCOUNT_LABEL(name) GEN_ENTRY(name) ; nop ; ALIGN_TEXT
#ifdef GUPROF
#define MEXITCOUNT call .mexitcount
#define ret MEXITCOUNT ; NON_GPROF_RET
#else
#define MEXITCOUNT
#endif
#else /* !GPROF */
/*
* ALTENTRY() has to align because it is before a corresponding ENTRY().
* ENTRY() has to align to because there may be no ALTENTRY() before it.
* If there is a previous ALTENTRY() then the alignment code for ENTRY()
* is empty.
*/
#define ALTENTRY(name) GEN_ENTRY(name)
#define CROSSJUMP(jtrue, label, jfalse) jtrue label
#define CROSSJUMPTARGET(label)
#define ENTRY(name) GEN_ENTRY(name)
#define FAKE_MCOUNT(caller)
#define MCOUNT
#define MCOUNT_LABEL(name)
#define MEXITCOUNT
#endif /* GPROF */
#define ALTENTRY(name) GEN_ENTRY(name)
#define END(name) .size name, . - name
/*
* Convenience for adding frame pointers to hand-coded ASM. Useful for

81
sys/amd64/include/profile.h
View file

@ -35,74 +35,9 @@
#ifndef _MACHINE_PROFILE_H_
#define _MACHINE_PROFILE_H_
#ifndef _SYS_CDEFS_H_
#error this file needs sys/cdefs.h as a prerequisite
#endif
#ifndef _KERNEL
#ifdef _KERNEL
/*
* Config generates something to tell the compiler to align functions on 16
* byte boundaries. A strict alignment is good for keeping the tables small.
*/
#define FUNCTION_ALIGNMENT 16
/*
* The kernel uses assembler stubs instead of unportable inlines.
* This is mainly to save a little time when profiling is not enabled,
* which is the usual case for the kernel.
*/
#define _MCOUNT_DECL void mcount
#define MCOUNT
#ifdef GUPROF
#define MCOUNT_DECL(s)
#define MCOUNT_ENTER(s)
#define MCOUNT_EXIT(s)
#ifdef __GNUCLIKE_ASM
#define MCOUNT_OVERHEAD(label) \
__asm __volatile("pushq %0; call __mcount; popq %%rcx" \
: \
: "i" (label) \
: "cx", "r10", "r11", "memory")
#define MEXITCOUNT_OVERHEAD() \
__asm __volatile("call .mexitcount; 1:" \
: : \
: "r10", "r11", "memory")
#define MEXITCOUNT_OVERHEAD_GETLABEL(labelp) \
__asm __volatile("movq $1b,%0" : "=rm" (labelp))
#else
#error this file needs to be ported to your compiler
#endif /* !__GNUCLIKE_ASM */
#else /* !GUPROF */
#define MCOUNT_DECL(s) register_t s;
#ifdef SMP
extern int mcount_lock;
#define MCOUNT_ENTER(s) { s = intr_disable(); \
while (!atomic_cmpset_acq_int(&mcount_lock, 0, 1)) \
/* nothing */ ; }
#define MCOUNT_EXIT(s) { atomic_store_rel_int(&mcount_lock, 0); \
intr_restore(s); }
#else
#define MCOUNT_ENTER(s) { s = intr_disable(); }
#define MCOUNT_EXIT(s) (intr_restore(s))
#endif
#endif /* GUPROF */
void bintr(void);
void btrap(void);
void eintr(void);
void user(void);
#define MCOUNT_FROMPC_USER(pc) \
((pc < (uintfptr_t)VM_MAXUSER_ADDRESS) ? (uintfptr_t)user : pc)
#define MCOUNT_FROMPC_INTR(pc) \
((pc >= (uintfptr_t)btrap && pc < (uintfptr_t)eintr) ? \
((pc >= (uintfptr_t)bintr) ? (uintfptr_t)bintr : \
(uintfptr_t)btrap) : ~0UL)
#else /* !_KERNEL */
#include <sys/cdefs.h>
#define FUNCTION_ALIGNMENT 4
@ -172,28 +107,18 @@ mcount() \
typedef u_long uintfptr_t;
#endif /* _KERNEL */
/*
* An unsigned integral type that can hold non-negative difference between
* function pointers.
*/
typedef u_long fptrdiff_t;
#ifdef _KERNEL
void mcount(uintfptr_t frompc, uintfptr_t selfpc);
#else /* !_KERNEL */
#include <sys/cdefs.h>
__BEGIN_DECLS
#ifdef __GNUCLIKE_ASM
void mcount(void) __asm(".mcount");
#endif
__END_DECLS
#endif /* _KERNEL */
#endif /* !_KERNEL */
#endif /* !_MACHINE_PROFILE_H_ */

10
sys/amd64/linux/linux_locore.asm
View file

@ -17,7 +17,7 @@ linux_platform:
* To avoid excess stack frame the signal trampoline code emulates
* the 'call' instruction.
*/
NON_GPROF_ENTRY(linux_rt_sigcode)
ENTRY(linux_rt_sigcode)
movq %rsp, %rbx /* preserve sigframe */
call .getip
.getip:
@ -32,28 +32,28 @@ NON_GPROF_ENTRY(linux_rt_sigcode)
.endrtsigcode:
0: jmp 0b
NON_GPROF_ENTRY(__vdso_clock_gettime)
ENTRY(__vdso_clock_gettime)
movq $LINUX_SYS_linux_clock_gettime,%rax
syscall
ret
.weak clock_gettime
.set clock_gettime, __vdso_clock_gettime
NON_GPROF_ENTRY(__vdso_time)
ENTRY(__vdso_time)
movq $LINUX_SYS_linux_time,%rax
syscall
ret
.weak time
.set time, __vdso_time
NON_GPROF_ENTRY(__vdso_gettimeofday)
ENTRY(__vdso_gettimeofday)
movq $LINUX_SYS_gettimeofday,%rax
syscall
ret
.weak gettimeofday
.set gettimeofday, __vdso_gettimeofday
NON_GPROF_ENTRY(__vdso_getcpu)
ENTRY(__vdso_getcpu)
movq $-38,%rax /* not implemented */
ret
.weak getcpu

6
sys/amd64/linux32/linux32_locore.asm
View file

@ -18,7 +18,7 @@ linux_platform:
* To avoid excess stack frame the signal trampoline code emulates
* the 'call' instruction.
*/
NON_GPROF_ENTRY(linux32_sigcode)
ENTRY(linux32_sigcode)
movl %esp, %ebx /* preserve sigframe */
call .getip0
.getip0:
@ -33,7 +33,7 @@ NON_GPROF_ENTRY(linux32_sigcode)
.endsigcode:
0: jmp 0b
NON_GPROF_ENTRY(linux32_rt_sigcode)
ENTRY(linux32_rt_sigcode)
leal LINUX_RT_SIGF_UC(%esp),%ebx /* linux ucp */
leal LINUX_RT_SIGF_SC(%ebx),%ecx /* linux sigcontext */
movl %esp, %edi
@ -49,7 +49,7 @@ NON_GPROF_ENTRY(linux32_rt_sigcode)
.endrtsigcode:
0: jmp 0b
NON_GPROF_ENTRY(linux32_vsyscall)
ENTRY(linux32_vsyscall)
.startvsyscall:
int $0x80
ret

3
sys/cddl/dev/dtrace/amd64/dtrace_asm.S
View file

@ -37,7 +37,6 @@
#include "assym.inc"
#define INTR_POP \
MEXITCOUNT; \
movq TF_RDI(%rsp),%rdi; \
movq TF_RSI(%rsp),%rsi; \
movq TF_RDX(%rsp),%rdx; \
@ -72,7 +71,7 @@
movq %rsp, %rsi
movq TF_RAX(%rsp), %rdx
call dtrace_invop
ALTENTRY(dtrace_invop_callsite)
ENTRY(dtrace_invop_callsite)
cmpl $DTRACE_INVOP_PUSHL_EBP, %eax
je bp_push
cmpl $DTRACE_INVOP_LEAVE, %eax

2
sys/conf/Makefile.arm
View file

@ -35,7 +35,7 @@ INCLUDES+= -I$S/contrib/libfdt -I$S/contrib/device-tree/include
LINUX_DTS_VERSION!= awk '/freebsd,dts-version/ { sub(/;$$/,"", $$NF); print $$NF }' $S/dts/freebsd-compatible.dts
CFLAGS += -DLINUX_DTS_VERSION=\"${LINUX_DTS_VERSION}\"
.if !defined(DEBUG) && !defined(PROFLEVEL)
.if !defined(DEBUG)
STRIP_FLAGS = -S
.endif

27
sys/conf/files
View file

@ -855,43 +855,43 @@ dev/aic7xxx/aic7xxx_pci.c optional ahc pci
dev/aic7xxx/aic7xxx_reg_print.c optional ahc ahc_reg_pretty_print
dev/al_eth/al_eth.c optional al_eth fdt \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
dev/al_eth/al_init_eth_lm.c optional al_eth fdt \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
dev/al_eth/al_init_eth_kr.c optional al_eth fdt \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/al_hal_iofic.c optional al_iofic \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/al_hal_serdes_25g.c optional al_serdes \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/al_hal_serdes_hssp.c optional al_serdes \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/al_hal_udma_config.c optional al_udma \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/al_hal_udma_debug.c optional al_udma \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/al_hal_udma_iofic.c optional al_udma \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/al_hal_udma_main.c optional al_udma \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/al_serdes.c optional al_serdes \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/eth/al_hal_eth_kr.c optional al_eth \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
contrib/alpine-hal/eth/al_hal_eth_main.c optional al_eth \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
dev/alc/if_alc.c optional alc pci
dev/ale/if_ale.c optional ale pci
dev/alpm/alpm.c optional alpm pci
@ -4069,7 +4069,6 @@ libkern/inet_ntop.c standard
libkern/inet_pton.c standard
libkern/jenkins_hash.c standard
libkern/murmur3_32.c standard
libkern/mcount.c optional profiling-routine
libkern/memcchr.c standard
libkern/memchr.c standard
libkern/memmem.c optional gdb

1
sys/conf/files.amd64
View file

@ -121,7 +121,6 @@ amd64/amd64/minidump_machdep.c standard
amd64/amd64/mp_machdep.c optional smp
amd64/amd64/mpboot.S optional smp
amd64/amd64/pmap.c standard
amd64/amd64/prof_machdep.c optional profiling-routine
amd64/amd64/ptrace_machdep.c standard
amd64/amd64/sigtramp.S standard
amd64/amd64/support.S standard

2
sys/conf/files.arm
View file

@ -152,4 +152,4 @@ arm/annapurna/alpine/alpine_pci.c optional al_pci fdt
arm/annapurna/alpine/alpine_pci_msix.c optional al_pci fdt
arm/annapurna/alpine/alpine_serdes.c optional al_serdes fdt \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"

18
sys/conf/files.arm64
View file

@ -109,32 +109,32 @@ dev/iommu/iommu_gas.c optional iommu
crypto/armv8/armv8_crypto.c optional armv8crypto
armv8_crypto_wrap.o optional armv8crypto \
dependency "$S/crypto/armv8/armv8_crypto_wrap.c" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc:N-mgeneral-regs-only} -I$S/crypto/armv8/ ${WERROR} ${NO_WCAST_QUAL} ${PROF} -march=armv8-a+crypto ${.IMPSRC}" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc:N-mgeneral-regs-only} -I$S/crypto/armv8/ ${WERROR} ${NO_WCAST_QUAL} -march=armv8-a+crypto ${.IMPSRC}" \
no-implicit-rule \
clean "armv8_crypto_wrap.o"
aesv8-armx.o optional armv8crypto \
dependency "$S/crypto/openssl/aarch64/aesv8-armx.S" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc:N-mgeneral-regs-only} -I$S/crypto/armv8/ -I$S/crypto/openssl/crypto ${WERROR} ${NO_WCAST_QUAL} ${PROF} -march=armv8-a+crypto ${.IMPSRC}" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc:N-mgeneral-regs-only} -I$S/crypto/armv8/ -I$S/crypto/openssl/crypto ${WERROR} ${NO_WCAST_QUAL} -march=armv8-a+crypto ${.IMPSRC}" \
no-implicit-rule \
clean "aesv8-armx.o"
ghashv8-armx.o optional armv8crypto \
dependency "$S/crypto/openssl/aarch64/ghashv8-armx.S" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc:N-mgeneral-regs-only} -I$S/crypto/armv8/ -I$S/crypto/openssl/crypto ${WERROR} ${NO_WCAST_QUAL} ${PROF} -march=armv8-a+crypto ${.IMPSRC}" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc:N-mgeneral-regs-only} -I$S/crypto/armv8/ -I$S/crypto/openssl/crypto ${WERROR} ${NO_WCAST_QUAL} -march=armv8-a+crypto ${.IMPSRC}" \
no-implicit-rule \
clean "ghashv8-armx.o"
crypto/des/des_enc.c optional netsmb
crypto/openssl/ossl_aarch64.c optional ossl
crypto/openssl/aarch64/chacha-armv8.S optional ossl \
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${PROF} ${.IMPSRC}"
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${.IMPSRC}"
crypto/openssl/aarch64/poly1305-armv8.S optional ossl \
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${PROF} ${.IMPSRC}"
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${.IMPSRC}"
crypto/openssl/aarch64/sha1-armv8.S optional ossl \
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${PROF} ${.IMPSRC}"
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${.IMPSRC}"
crypto/openssl/aarch64/sha256-armv8.S optional ossl \
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${PROF} ${.IMPSRC}"
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${.IMPSRC}"
crypto/openssl/aarch64/sha512-armv8.S optional ossl \
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${PROF} ${.IMPSRC}"
compile-with "${CC} -c ${CFLAGS:N-mgeneral-regs-only} ${WERROR} ${.IMPSRC}"
dev/acpica/acpi_bus_if.m optional acpi
dev/acpica/acpi_if.m optional acpi
@ -355,7 +355,7 @@ arm/annapurna/alpine/alpine_pci.c optional al_pci fdt
arm/annapurna/alpine/alpine_pci_msix.c optional al_pci fdt
arm/annapurna/alpine/alpine_serdes.c optional al_serdes fdt \
no-depend \
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}"
compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${.IMPSRC}"
# Broadcom
arm64/broadcom/brcmmdio/mdio_mux_iproc.c optional soc_brcm_ns2 fdt

1
sys/conf/files.i386
View file

@ -190,7 +190,6 @@ i386/i386/perfmon.c optional perfmon
i386/i386/pmap_base.c standard
i386/i386/pmap_nopae.c standard
i386/i386/pmap_pae.c standard
i386/i386/prof_machdep.c optional profiling-routine
i386/i386/ptrace_machdep.c standard
i386/i386/sigtramp.s standard
i386/i386/support.s standard

10
sys/conf/files.x86
View file

@ -23,27 +23,27 @@ cddl/dev/dtrace/x86/instr_size.c optional dtrace_fbt | dtraceall compile-with
crypto/aesni/aesni.c optional aesni
aesni_ghash.o optional aesni \
dependency "$S/crypto/aesni/aesni_ghash.c" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} ${NO_WCAST_QUAL} ${PROF} -mmmx -msse -msse4 -maes -mpclmul ${.IMPSRC}" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} ${NO_WCAST_QUAL} -mmmx -msse -msse4 -maes -mpclmul ${.IMPSRC}" \
no-implicit-rule \
clean "aesni_ghash.o"
aesni_ccm.o optional aesni \
dependency "$S/crypto/aesni/aesni_ccm.c" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} ${NO_WCAST_QUAL} ${PROF} -mmmx -msse -msse4 -maes -mpclmul ${.IMPSRC}" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} ${NO_WCAST_QUAL} -mmmx -msse -msse4 -maes -mpclmul ${.IMPSRC}" \
no-implicit-rule \
clean "aesni_ccm.o"
aesni_wrap.o optional aesni \
dependency "$S/crypto/aesni/aesni_wrap.c" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} ${NO_WCAST_QUAL} ${PROF} -mmmx -msse -msse4 -maes ${.IMPSRC}" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} ${NO_WCAST_QUAL} -mmmx -msse -msse4 -maes ${.IMPSRC}" \
no-implicit-rule \
clean "aesni_wrap.o"
intel_sha1.o optional aesni \
dependency "$S/crypto/aesni/intel_sha1.c" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} ${PROF} -mmmx -msse -msse4 -msha ${.IMPSRC}" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} -mmmx -msse -msse4 -msha ${.IMPSRC}" \
no-implicit-rule \
clean "intel_sha1.o"
intel_sha256.o optional aesni \
dependency "$S/crypto/aesni/intel_sha256.c" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} ${PROF} -mmmx -msse -msse4 -msha ${.IMPSRC}" \
compile-with "${CC} -c ${CFLAGS:C/^-O2$/-O3/:N-nostdinc} ${WERROR} -mmmx -msse -msse4 -msha ${.IMPSRC}" \
no-implicit-rule \
clean "intel_sha256.o"
crypto/openssl/ossl_x86.c optional ossl

3
sys/conf/kern.post.mk
View file

@ -463,7 +463,4 @@ embedfs_${MFS_IMAGE:T:R}.o: ${MFS_IMAGE} $S/dev/md/embedfs.S
.endif
.endif
# XXX strictly, everything depends on Makefile because changes to ${PROF}
# only appear there, but we don't handle that.
.include "kern.mk"

42
sys/conf/kern.pre.mk
View file

@ -92,21 +92,6 @@ CFLAGS+= -fno-common
# XXX LOCORE means "don't declare C stuff" not "for locore.s".
ASM_CFLAGS= -x assembler-with-cpp -DLOCORE ${CFLAGS} ${ASM_CFLAGS.${.IMPSRC:T}}
.if defined(PROFLEVEL) && ${PROFLEVEL} >= 1
CFLAGS+= -DGPROF
CFLAGS.gcc+= -falign-functions=16
.if ${PROFLEVEL} >= 2
CFLAGS+= -DGPROF4 -DGUPROF
PROF= -pg
.if ${COMPILER_TYPE} == "gcc"
PROF+= -mprofiler-epilogue
.endif
.else
PROF= -pg
.endif
.endif
DEFINED_PROF= ${PROF}
KCSAN_ENABLED!= grep KCSAN opt_global.h || true ; echo
.if !empty(KCSAN_ENABLED)
SAN_CFLAGS+= -fsanitize=thread
@ -138,8 +123,8 @@ GCOV_CFLAGS+= -fprofile-arcs -ftest-coverage
CFLAGS+= ${GCOV_CFLAGS}
# Put configuration-specific C flags last (except for ${PROF}) so that they
# can override the others.
# Put configuration-specific C flags last so that they can override
# the others.
CFLAGS+= ${CONF_CFLAGS}
.if defined(LINKER_FEATURES) && ${LINKER_FEATURES:Mbuild-id}
@ -176,13 +161,12 @@ CFLAGS+= -fPIE
.endif
.endif
NORMAL_C= ${CC} -c ${CFLAGS} ${WERROR} ${PROF} ${.IMPSRC}
NORMAL_C= ${CC} -c ${CFLAGS} ${WERROR} ${.IMPSRC}
NORMAL_S= ${CC:N${CCACHE_BIN}} -c ${ASM_CFLAGS} ${WERROR} ${.IMPSRC}
PROFILE_C= ${CC} -c ${CFLAGS} ${WERROR} ${.IMPSRC}
NORMAL_C_NOWERROR= ${CC} -c ${CFLAGS} ${PROF} ${.IMPSRC}
NORMAL_C_NOWERROR= ${CC} -c ${CFLAGS} ${.IMPSRC}
NORMAL_M= ${AWK} -f $S/tools/makeobjops.awk ${.IMPSRC} -c ; \
${CC} -c ${CFLAGS} ${WERROR} ${PROF} ${.PREFIX}.c
${CC} -c ${CFLAGS} ${WERROR} ${.PREFIX}.c
NORMAL_FW= uudecode -o ${.TARGET} ${.ALLSRC}
NORMAL_FWO= ${CC:N${CCACHE_BIN}} -c ${ASM_CFLAGS} ${WERROR} -o ${.TARGET} \
@ -192,9 +176,7 @@ NORMAL_FWO= ${CC:N${CCACHE_BIN}} -c ${ASM_CFLAGS} ${WERROR} -o ${.TARGET} \
# for ZSTD in the kernel (include zstd/lib/freebsd before other CFLAGS)
ZSTD_C= ${CC} -c -DZSTD_HEAPMODE=1 -I$S/contrib/zstd/lib/freebsd ${CFLAGS} \
-I$S/contrib/zstd/lib -I$S/contrib/zstd/lib/common ${WERROR} \
-Wno-missing-prototypes ${PROF} -U__BMI__ \
-DZSTD_NO_INTRINSICS \
${.IMPSRC}
-Wno-missing-prototypes -U__BMI__ -DZSTD_NO_INTRINSICS ${.IMPSRC}
# https://github.com/facebook/zstd/commit/812e8f2a [zstd 1.4.1]
# "Note that [GCC] autovectorization still does not do a good job on the
# optimized version, so it's turned off via attribute and flag. I found
@ -237,7 +219,7 @@ CDDL_CFLAGS= \
-include ${ZINCDIR}/os/freebsd/spl/sys/ccompile.h \
-I$S/cddl/contrib/opensolaris/uts/common \
-I$S -I$S/cddl/compat/opensolaris
CDDL_C= ${CC} -c ${CDDL_CFLAGS} ${WERROR} ${PROF} ${.IMPSRC}
CDDL_C= ${CC} -c ${CDDL_CFLAGS} ${WERROR} ${.IMPSRC}
# Special flags for managing the compat compiles for ZFS
ZFS_CFLAGS+= ${CDDL_CFLAGS} -DBUILDING_ZFS -DHAVE_UIO_ZEROCOPY \
@ -258,8 +240,8 @@ ZFS_CFLAGS+= -DBITS_PER_LONG=64
ZFS_ASM_CFLAGS= -x assembler-with-cpp -DLOCORE ${ZFS_CFLAGS}
ZFS_C= ${CC} -c ${ZFS_CFLAGS} ${WERROR} ${PROF} ${.IMPSRC}
ZFS_RPC_C= ${CC} -c ${ZFS_CFLAGS} -DHAVE_RPC_TYPES ${WERROR} ${PROF} ${.IMPSRC}
ZFS_C= ${CC} -c ${ZFS_CFLAGS} ${WERROR} ${.IMPSRC}
ZFS_RPC_C= ${CC} -c ${ZFS_CFLAGS} -DHAVE_RPC_TYPES ${WERROR} ${.IMPSRC}
ZFS_S= ${CC} -c ${ZFS_ASM_CFLAGS} ${WERROR} ${.IMPSRC}
@ -272,7 +254,7 @@ DTRACE_CFLAGS+= -I$S/cddl/contrib/opensolaris/uts/intel -I$S/cddl/dev/dtrace/x86
DTRACE_CFLAGS+= -I$S/cddl/contrib/opensolaris/common/util -I$S -DDIS_MEM -DSMP -I$S/cddl/compat/opensolaris
DTRACE_CFLAGS+= -I$S/cddl/contrib/opensolaris/uts/common
DTRACE_ASM_CFLAGS= -x assembler-with-cpp -DLOCORE ${DTRACE_CFLAGS}
DTRACE_C= ${CC} -c ${DTRACE_CFLAGS} ${WERROR} ${PROF} ${.IMPSRC}
DTRACE_C= ${CC} -c ${DTRACE_CFLAGS} ${WERROR} ${.IMPSRC}
DTRACE_S= ${CC} -c ${DTRACE_ASM_CFLAGS} ${WERROR} ${.IMPSRC}
# Special flags for managing the compat compiles for DTrace/FBT
@ -280,7 +262,7 @@ FBT_CFLAGS= -DBUILDING_DTRACE -nostdinc -I$S/cddl/dev/fbt/${MACHINE_CPUARCH} -I$
.if ${MACHINE_CPUARCH} == "amd64" || ${MACHINE_CPUARCH} == "i386"
FBT_CFLAGS+= -I$S/cddl/dev/fbt/x86
.endif
FBT_C= ${CC} -c ${FBT_CFLAGS} ${WERROR} ${PROF} ${.IMPSRC}
FBT_C= ${CC} -c ${FBT_CFLAGS} ${WERROR} ${.IMPSRC}
.if ${MK_CTF} != "no"
NORMAL_CTFCONVERT= ${CTFCONVERT} ${CTFFLAGS} ${.TARGET}
@ -300,7 +282,7 @@ OFEDINCLUDES= -I$S/ofed/include -I$S/ofed/include/uapi ${LINUXKPI_INCLUDES}
OFEDNOERR= -Wno-cast-qual -Wno-pointer-arith
OFEDCFLAGS= ${CFLAGS:N-I*} -DCONFIG_INFINIBAND_USER_MEM \
${OFEDINCLUDES} ${CFLAGS:M-I*} ${OFEDNOERR}
OFED_C_NOIMP= ${CC} -c -o ${.TARGET} ${OFEDCFLAGS} ${WERROR} ${PROF}
OFED_C_NOIMP= ${CC} -c -o ${.TARGET} ${OFEDCFLAGS} ${WERROR}
OFED_C= ${OFED_C_NOIMP} ${.IMPSRC}
# mlxfw C flags.

1
sys/conf/options.i386
View file

@ -8,7 +8,6 @@ COUNT_XINVLTLB_HITS opt_smp.h
COUNT_IPIS opt_smp.h
DISABLE_PG_G opt_pmap.h
DISABLE_PSE opt_pmap.h
I586_PMC_GUPROF opt_i586_guprof.h
MAXMEM
MPTABLE_FORCE_HTT
MP_WATCHDOG

2
sys/dev/hyperv/vmbus/amd64/vmbus_vector.S
View file

@ -39,8 +39,6 @@
.text
SUPERALIGN_TEXT
INTR_HANDLER vmbus_isr
FAKE_MCOUNT(TF_RIP(%rsp))
movq %rsp, %rdi
call vmbus_handle_intr
MEXITCOUNT
jmp doreti

2
sys/dev/hyperv/vmbus/i386/vmbus_vector.S
View file

@ -44,9 +44,7 @@ IDTVEC(vmbus_isr)
SET_KERNEL_SREGS
cld
KENTER
FAKE_MCOUNT(TF_EIP(%esp))
pushl %esp
call vmbus_handle_intr
add $4, %esp
MEXITCOUNT
jmp doreti

5
sys/i386/conf/NOTES
View file

@ -7,10 +7,6 @@
# $FreeBSD$
#
#
# We want LINT to cover profiling as well.
profile 2
#
# Enable the kernel DTrace hooks which are required to load the DTrace
# kernel modules.
@ -907,7 +903,6 @@ options PV_STATS
options FB_INSTALL_CDEV # install a CDEV entry in /dev
options I586_PMC_GUPROF=0x70000
options KBDIO_DEBUG=2
options KBD_MAXRETRY=4
options KBD_MAXWAIT=6

14
sys/i386/i386/apic_vector.s
View file

@ -79,7 +79,6 @@ X\()\vec_name:
SET_KERNEL_SREGS
cld
KENTER
FAKE_MCOUNT(TF_EIP(%esp))
cmpl $0,x2apic_mode
je 2f
movl $(MSR_APIC_ISR0 + \index),%ecx
@ -98,7 +97,6 @@ X\()\vec_name:
call *%eax
addl $8, %esp /* discard parameter */
4:
MEXITCOUNT
jmp doreti
.endm
@ -136,12 +134,10 @@ IDTVEC(timerint)
SET_KERNEL_SREGS
cld
KENTER
FAKE_MCOUNT(TF_EIP(%esp))
pushl %esp
movl $lapic_handle_timer, %eax
call *%eax
add $4, %esp
MEXITCOUNT
jmp doreti
/*
@ -155,10 +151,8 @@ IDTVEC(cmcint)
SET_KERNEL_SREGS
cld
KENTER
FAKE_MCOUNT(TF_EIP(%esp))
movl $lapic_handle_cmc, %eax
call *%eax
MEXITCOUNT
jmp doreti
/*
@ -172,10 +166,8 @@ IDTVEC(errorint)
SET_KERNEL_SREGS
cld
KENTER
FAKE_MCOUNT(TF_EIP(%esp))
movl $lapic_handle_error, %eax
call *%eax
MEXITCOUNT
jmp doreti
#ifdef XENHVM
@ -190,12 +182,10 @@ IDTVEC(xen_intr_upcall)
SET_KERNEL_SREGS
cld
KENTER
FAKE_MCOUNT(TF_EIP(%esp))
pushl %esp
movl $xen_intr_handle_upcall, %eax
call *%eax
add $4, %esp
MEXITCOUNT
jmp doreti
#endif
@ -272,10 +262,8 @@ IDTVEC(ipi_intr_bitmap_handler)
cld
KENTER
call as_lapic_eoi
FAKE_MCOUNT(TF_EIP(%esp))
movl $ipi_bitmap_handler, %eax
call *%eax
MEXITCOUNT
jmp doreti
/*
@ -319,10 +307,8 @@ IDTVEC(ipi_swi)
cld
KENTER
call as_lapic_eoi
FAKE_MCOUNT(TF_EIP(%esp))
movl $ipi_swi_handler, %eax
call *%eax
MEXITCOUNT
jmp doreti
/*

2
sys/i386/i386/atpic_vector.s
View file

@ -55,14 +55,12 @@ X\()\vec_name:
SET_KERNEL_SREGS
cld
KENTER
FAKE_MCOUNT(TF_EIP(%esp))
pushl %esp
pushl $\irq_num /* pass the IRQ */
movl $atpic_handle_intr, %eax
call *%eax
addl $8, %esp /* discard the parameters */
MEXITCOUNT
jmp doreti
.endm

30
sys/i386/i386/exception.s
View file

@ -96,9 +96,6 @@ tramp_idleptd: .long 0
* must be PIC.
*/
MCOUNT_LABEL(user)
MCOUNT_LABEL(btrap)
#define TRAP(a) pushl $(a) ; jmp alltraps
IDTVEC(div)
@ -172,7 +169,6 @@ alltraps_with_regs_pushed:
SET_KERNEL_SREGS
cld
KENTER
FAKE_MCOUNT(TF_EIP(%esp))
calltrap:
pushl %esp
movl $trap,%eax
@ -182,7 +178,6 @@ calltrap:
/*
* Return via doreti to handle ASTs.
*/
MEXITCOUNT
jmp doreti
.globl irettraps
@ -227,8 +222,7 @@ irettraps:
movl $(2 * TF_SZ - TF_FS), %ecx
jmp 6f
/* kernel mode, normal */
5: FAKE_MCOUNT(TF_EIP(%esp))
jmp calltrap
5: jmp calltrap
6: cmpl $PMAP_TRM_MIN_ADDRESS, %esp /* trampoline stack ? */
jb 5b /* if not, no need to change stacks */
movl (tramp_idleptd - 1b)(%ebx), %eax
@ -239,7 +233,6 @@ irettraps:
movl %esp, %esi
rep; movsb
movl %edx, %esp
FAKE_MCOUNT(TF_EIP(%esp))
jmp calltrap
/*
@ -307,7 +300,6 @@ IDTVEC(dbg)
1: popl %eax
movl (tramp_idleptd - 1b)(%eax), %eax
movl %eax, %cr3
FAKE_MCOUNT(TF_EIP(%esp))
testl $PSL_VM, TF_EFLAGS(%esp)
jnz dbg_user
testb $SEL_RPL_MASK,TF_CS(%esp)
@ -321,7 +313,6 @@ dbg_user:
call *%eax
add $4, %esp
movl $T_RESERVED, TF_TRAPNO(%esp)
MEXITCOUNT
jmp doreti
IDTVEC(mchk)
@ -352,7 +343,6 @@ nmi_mchk_common:
1: popl %eax
movl (tramp_idleptd - 1b)(%eax), %eax
movl %eax, %cr3
FAKE_MCOUNT(TF_EIP(%esp))
jmp calltrap
/*
@ -375,12 +365,10 @@ IDTVEC(int0x80_syscall)
movl $handle_ibrs_entry,%eax
call *%eax
sti
FAKE_MCOUNT(TF_EIP(%esp))
pushl %esp
movl $syscall, %eax
call *%eax
add $4, %esp
MEXITCOUNT
jmp doreti
ENTRY(fork_trampoline)
@ -395,25 +383,13 @@ ENTRY(fork_trampoline)
/*
* Return via doreti to handle ASTs.
*/
MEXITCOUNT
jmp doreti
/*
* To efficiently implement classification of trap and interrupt handlers
* for profiling, there must be only trap handlers between the labels btrap
* and bintr, and only interrupt handlers between the labels bintr and
* eintr. This is implemented (partly) by including files that contain
* some of the handlers. Before including the files, set up a normal asm
* environment so that the included files doen't need to know that they are
* included.
*/
.data
.p2align 4
.text
SUPERALIGN_TEXT
MCOUNT_LABEL(bintr)
#ifdef DEV_ATPIC
#include <i386/i386/atpic_vector.s>
@ -437,7 +413,6 @@ MCOUNT_LABEL(bintr)
#include <i386/i386/vm86bios.s>
.text
MCOUNT_LABEL(eintr)
#include <i386/i386/copyout_fast.s>
@ -451,7 +426,6 @@ MCOUNT_LABEL(eintr)
.type doreti,@function
.globl doreti
doreti:
FAKE_MCOUNT($bintr) /* init "from" bintr -> doreti */
doreti_next:
/*
* Check if ASTs can be handled now. ASTs cannot be safely
@ -504,8 +478,6 @@ doreti_ast:
* registers. The fault is handled in trap.c.
*/
doreti_exit:
MEXITCOUNT
cmpl $T_NMI, TF_TRAPNO(%esp)
je doreti_iret_nmi
cmpl $T_MCHK, TF_TRAPNO(%esp)

4
sys/i386/i386/locore.s
View file

@ -79,7 +79,7 @@ bootinfo: .space BOOTINFO_SIZE /* bootinfo that we can handle */
* This is where the bootblocks start us, set the ball rolling...
*
*/
NON_GPROF_ENTRY(btext)
ENTRY(btext)
/* Tell the bios to warmboot next time */
movw $0x1234,0x472
@ -451,6 +451,6 @@ END(identify_cpu)
.text
.p2align PAGE_SHIFT, 0x90 /* Hypercall_page needs to be PAGE aligned */
NON_GPROF_ENTRY(hypercall_page)
ENTRY(hypercall_page)
.skip 0x1000, 0x90 /* Fill with "nop"s */
#endif

4
sys/i386/i386/mpboot.s
View file

@ -68,7 +68,7 @@
*/
.p2align 4
NON_GPROF_ENTRY(MPentry)
ENTRY(MPentry)
CHECKPOINT(0x36, 3)
/*
* Enable features on this processor. We don't support SMP on
@ -142,7 +142,7 @@ mp_begin: /* now running relocated at KERNBASE */
BOOTMP1:
NON_GPROF_ENTRY(bootMP)
ENTRY(bootMP)
.code16
cli
CHECKPOINT(0x34, 1)

404
sys/i386/i386/prof_machdep.c
View file

@ -1,404 +0,0 @@
/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 1996 Bruce D. Evans.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#ifdef GUPROF
#include "opt_i586_guprof.h"
#include "opt_perfmon.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/cpu.h>
#include <sys/eventhandler.h>
#include <sys/gmon.h>
#include <sys/kernel.h>
#include <sys/smp.h>
#include <sys/sysctl.h>
#include <machine/clock.h>
#include <machine/perfmon.h>
#include <machine/timerreg.h>
#define CPUTIME_CLOCK_UNINITIALIZED 0
#define CPUTIME_CLOCK_I8254 1
#define CPUTIME_CLOCK_TSC 2
#define CPUTIME_CLOCK_I586_PMC 3
#define CPUTIME_CLOCK_I8254_SHIFT 7
int cputime_bias = 1; /* initialize for locality of reference */
static int cputime_clock = CPUTIME_CLOCK_UNINITIALIZED;
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
static u_int cputime_clock_pmc_conf = I586_PMC_GUPROF;
static int cputime_clock_pmc_init;
static struct gmonparam saved_gmp;
#endif
#if defined(I586_CPU) || defined(I686_CPU)
static int cputime_prof_active;
#endif
#endif /* GUPROF */
#ifdef __GNUCLIKE_ASM
#if defined(SMP) && defined(GUPROF)
#define MPLOCK " \n\
movl $1,%edx \n\
9: \n\
xorl %eax,%eax \n\
lock \n\
cmpxchgl %edx,mcount_lock \n\
jne 9b \n"
#define MPUNLOCK "movl $0,mcount_lock \n"
#else /* !(SMP && GUPROF) */
#define MPLOCK
#define MPUNLOCK
#endif /* SMP && GUPROF */
__asm(" \n\
GM_STATE = 0 \n\
GMON_PROF_OFF = 3 \n\
\n\
.text \n\
.p2align 4,0x90 \n\
.globl __mcount \n\
.type __mcount,@function \n\
__mcount: \n\
# \n\
# Check that we are profiling. Do it early for speed. \n\
# \n\
cmpl $GMON_PROF_OFF,_gmonparam+GM_STATE \n\
je .mcount_exit \n\
# \n\
# __mcount is the same as [.]mcount except the caller \n\
# hasn't changed the stack except to call here, so the \n\
# caller's raddr is above our raddr. \n\
# \n\
movl 4(%esp),%edx \n\
jmp .got_frompc \n\
\n\
.p2align 4,0x90 \n\
.globl .mcount \n\
.mcount: \n\
cmpl $GMON_PROF_OFF,_gmonparam+GM_STATE \n\
je .mcount_exit \n\
# \n\
# The caller's stack frame has already been built, so \n\
# %ebp is the caller's frame pointer. The caller's \n\
# raddr is in the caller's frame following the caller's \n\
# caller's frame pointer. \n\
# \n\
movl 4(%ebp),%edx \n\
.got_frompc: \n\
# \n\
# Our raddr is the caller's pc. \n\
# \n\
movl (%esp),%eax \n\
\n\
pushfl \n\
pushl %eax \n\
pushl %edx \n\
cli \n"
MPLOCK " \n\
call mcount \n"
MPUNLOCK " \n\
addl $8,%esp \n\
popfl \n\
.mcount_exit: \n\
ret $0 \n\
");
void __mcount(void);
void (*__mcountp)(void) = __mcount;
#else /* !__GNUCLIKE_ASM */
#error "this file needs to be ported to your compiler"
#endif /* __GNUCLIKE_ASM */
#ifdef GUPROF
/*
* [.]mexitcount saves the return register(s), loads selfpc and calls
* mexitcount(selfpc) to do the work. Someday it should be in a machine
* dependent file together with cputime(), __mcount and [.]mcount. cputime()
* can't just be put in machdep.c because it has to be compiled without -pg.
*/
#ifdef __GNUCLIKE_ASM
__asm(" \n\
.text \n\
# \n\
# Dummy label to be seen when gprof -u hides [.]mexitcount. \n\
# \n\
.p2align 4,0x90 \n\
.globl __mexitcount \n\
.type __mexitcount,@function \n\
__mexitcount: \n\
nop \n\
\n\
GMON_PROF_HIRES = 4 \n\
\n\
.p2align 4,0x90 \n\
.globl .mexitcount \n\
.mexitcount: \n\
cmpl $GMON_PROF_HIRES,_gmonparam+GM_STATE \n\
jne .mexitcount_exit \n\
pushl %edx \n\
pushl %eax \n\
movl 8(%esp),%eax \n\
pushfl \n\
pushl %eax \n\
cli \n"
MPLOCK " \n\
call mexitcount \n"
MPUNLOCK " \n\
addl $4,%esp \n\
popfl \n\
popl %eax \n\
popl %edx \n\
.mexitcount_exit: \n\
ret $0 \n\
");
#endif /* __GNUCLIKE_ASM */
void __mexitcount(void);
void (*__mexitcountp)(void) = __mexitcount;
/*
* Return the time elapsed since the last call. The units are machine-
* dependent.
*/
int
cputime()
{
u_int count;
int delta;
#if (defined(I586_CPU) || defined(I686_CPU)) && \
defined(PERFMON) && defined(I586_PMC_GUPROF) && !defined(SMP)
u_quad_t event_count;
#endif
u_char high, low;
static u_int prev_count;
#if defined(I586_CPU) || defined(I686_CPU)
if (cputime_clock == CPUTIME_CLOCK_TSC) {
/*
* Scale the TSC a little to make cputime()'s frequency
* fit in an int, assuming that the TSC frequency fits
* in a u_int. Use a fixed scale since dynamic scaling
* would be slower and we can't really use the low bit
* of precision.
*/
count = (u_int)rdtsc() & ~1u;
delta = (int)(count - prev_count) >> 1;
prev_count = count;
return (delta);
}
#if defined(PERFMON) && defined(I586_PMC_GUPROF) && !defined(SMP)
if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
/*
* XXX permon_read() should be inlined so that the
* perfmon module doesn't need to be compiled with
* profiling disabled and so that it is fast.
*/
perfmon_read(0, &event_count);
count = (u_int)event_count;
delta = (int)(count - prev_count);
prev_count = count;
return (delta);
}
#endif /* PERFMON && I586_PMC_GUPROF && !SMP */
#endif /* I586_CPU || I686_CPU */
/*
* Read the current value of the 8254 timer counter 0.
*/
outb(TIMER_MODE, TIMER_SEL0 | TIMER_LATCH);
low = inb(TIMER_CNTR0);
high = inb(TIMER_CNTR0);
count = ((high << 8) | low) << CPUTIME_CLOCK_I8254_SHIFT;
/*
* The timer counts down from TIMER_CNTR0_MAX to 0 and then resets.
* While profiling is enabled, this routine is called at least twice
* per timer reset (for mcounting and mexitcounting hardclock()),
* so at most one reset has occurred since the last call, and one
* has occurred iff the current count is larger than the previous
* count. This allows counter underflow to be detected faster
* than in microtime().
*/
delta = prev_count - count;
prev_count = count;
if ((int) delta <= 0)
return (delta + (i8254_max_count << CPUTIME_CLOCK_I8254_SHIFT));
return (delta);
}
static int
sysctl_machdep_cputime_clock(SYSCTL_HANDLER_ARGS)
{
int clock;
int error;
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
int event;
struct pmc pmc;
#endif
clock = cputime_clock;
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
if (clock == CPUTIME_CLOCK_I586_PMC) {
pmc.pmc_val = cputime_clock_pmc_conf;
clock += pmc.pmc_event;
}
#endif
error = sysctl_handle_opaque(oidp, &clock, sizeof clock, req);
if (error == 0 && req->newptr != NULL) {
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
if (clock >= CPUTIME_CLOCK_I586_PMC) {
event = clock - CPUTIME_CLOCK_I586_PMC;
if (event >= 256)
return (EINVAL);
pmc.pmc_num = 0;
pmc.pmc_event = event;
pmc.pmc_unit = 0;
pmc.pmc_flags = PMCF_E | PMCF_OS | PMCF_USR;
pmc.pmc_mask = 0;
cputime_clock_pmc_conf = pmc.pmc_val;
cputime_clock = CPUTIME_CLOCK_I586_PMC;
} else
#endif
{
if (clock < 0 || clock >= CPUTIME_CLOCK_I586_PMC)
return (EINVAL);
cputime_clock = clock;
}
}
return (error);
}
SYSCTL_PROC(_machdep, OID_AUTO, cputime_clock,
CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(u_int),
sysctl_machdep_cputime_clock, "I",
"");
/*
* The start and stop routines need not be here since we turn off profiling
* before calling them. They are here for convenience.
*/
void
startguprof(gp)
struct gmonparam *gp;
{
#if defined(I586_CPU) || defined(I686_CPU)
uint64_t freq;
freq = atomic_load_acq_64(&tsc_freq);
if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED) {
if (freq != 0 && mp_ncpus == 1)
cputime_clock = CPUTIME_CLOCK_TSC;
else
cputime_clock = CPUTIME_CLOCK_I8254;
}
if (cputime_clock == CPUTIME_CLOCK_TSC) {
gp->profrate = freq >> 1;
cputime_prof_active = 1;
} else
gp->profrate = i8254_freq << CPUTIME_CLOCK_I8254_SHIFT;
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
if (cputime_clock == CPUTIME_CLOCK_I586_PMC) {
if (perfmon_avail() &&
perfmon_setup(0, cputime_clock_pmc_conf) == 0) {
if (perfmon_start(0) != 0)
perfmon_fini(0);
else {
/* XXX 1 event == 1 us. */
gp->profrate = 1000000;
saved_gmp = *gp;
/* Zap overheads. They are invalid. */
gp->cputime_overhead = 0;
gp->mcount_overhead = 0;
gp->mcount_post_overhead = 0;
gp->mcount_pre_overhead = 0;
gp->mexitcount_overhead = 0;
gp->mexitcount_post_overhead = 0;
gp->mexitcount_pre_overhead = 0;
cputime_clock_pmc_init = TRUE;
}
}
}
#endif /* PERFMON && I586_PMC_GUPROF */
#else /* !(I586_CPU || I686_CPU) */
if (cputime_clock == CPUTIME_CLOCK_UNINITIALIZED)
cputime_clock = CPUTIME_CLOCK_I8254;
gp->profrate = i8254_freq << CPUTIME_CLOCK_I8254_SHIFT;
#endif /* I586_CPU || I686_CPU */
cputime_bias = 0;
cputime();
}
void
stopguprof(gp)
struct gmonparam *gp;
{
#if defined(PERFMON) && defined(I586_PMC_GUPROF)
if (cputime_clock_pmc_init) {
*gp = saved_gmp;
perfmon_fini(0);
cputime_clock_pmc_init = FALSE;
}
#endif
#if defined(I586_CPU) || defined(I686_CPU)
if (cputime_clock == CPUTIME_CLOCK_TSC)
cputime_prof_active = 0;
#endif
}
#if defined(I586_CPU) || defined(I686_CPU)
/* If the cpu frequency changed while profiling, report a warning. */
static void
tsc_freq_changed(void *arg, const struct cf_level *level, int status)
{
/*
* If there was an error during the transition or
* TSC is P-state invariant, don't do anything.
*/
if (status != 0 || tsc_is_invariant)
return;
if (cputime_prof_active && cputime_clock == CPUTIME_CLOCK_TSC)
printf("warning: cpu freq changed while profiling active\n");
}
EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL,
EVENTHANDLER_PRI_ANY);
#endif /* I586_CPU || I686_CPU */
#endif /* GUPROF */

2
sys/i386/i386/sigtramp.s
View file

@ -48,7 +48,7 @@
/*
* Signal trampoline, copied to top of user stack
*/
NON_GPROF_ENTRY(sigcode)
ENTRY(sigcode)
calll *SIGF_HANDLER(%esp)
leal SIGF_UC(%esp),%eax /* get ucontext */
pushl %eax

2
sys/i386/i386/support.s
View file

@ -155,7 +155,6 @@ ENTRY(bcopy)
movl 8(%esp),%edx
movl %eax,8(%esp)
movl %edx,4(%esp)
MEXITCOUNT
jmp memmove
END(bcopy)
@ -284,7 +283,6 @@ ENTRY(lgdt)
movl (%esp),%eax
pushl %eax
movl $KCSEL,4(%esp)
MEXITCOUNT
lret
END(lgdt)

1
sys/i386/i386/vm86bios.s
View file

@ -130,7 +130,6 @@ ENTRY(vm86_bioscall)
/*
* Return via doreti
*/
MEXITCOUNT
jmp doreti

82
sys/i386/include/asmacros.h
View file

@ -48,92 +48,14 @@
#define CNAME(csym) csym
#define ALIGN_DATA .p2align 2 /* 4 byte alignment, zero filled */
#ifdef GPROF
#define ALIGN_TEXT .p2align 4,0x90 /* 16-byte alignment, nop filled */
#else
#define ALIGN_TEXT .p2align 2,0x90 /* 4-byte alignment, nop filled */
#endif
#define SUPERALIGN_TEXT .p2align 4,0x90 /* 16-byte alignment, nop filled */
#define GEN_ENTRY(name) ALIGN_TEXT; .globl CNAME(name); \
.type CNAME(name),@function; CNAME(name):
#define NON_GPROF_ENTRY(name) GEN_ENTRY(name)
#define NON_GPROF_RET .byte 0xc3 /* opcode for `ret' */
#define END(name) .size name, . - name
#ifdef GPROF
/*
* __mcount is like [.]mcount except that doesn't require its caller to set
* up a frame pointer. It must be called before pushing anything onto the
* stack. gcc should eventually generate code to call __mcount in most
* cases. This would make -pg in combination with -fomit-frame-pointer
* useful. gcc has a configuration variable PROFILE_BEFORE_PROLOGUE to
* allow profiling before setting up the frame pointer, but this is
* inadequate for good handling of special cases, e.g., -fpic works best
* with profiling after the prologue.
*
* [.]mexitcount is a new function to support non-statistical profiling if an
* accurate clock is available. For C sources, calls to it are generated
* by the FreeBSD extension `-mprofiler-epilogue' to gcc. It is best to
* call [.]mexitcount at the end of a function like the MEXITCOUNT macro does,
* but gcc currently generates calls to it at the start of the epilogue to
* avoid problems with -fpic.
*
* [.]mcount and __mcount may clobber the call-used registers and %ef.
* [.]mexitcount may clobber %ecx and %ef.
*
* Cross-jumping makes non-statistical profiling timing more complicated.
* It is handled in many cases by calling [.]mexitcount before jumping. It
* is handled for conditional jumps using CROSSJUMP() and CROSSJUMP_LABEL().
* It is handled for some fault-handling jumps by not sharing the exit
* routine.
*
* ALTENTRY() must be before a corresponding ENTRY() so that it can jump to
* the main entry point. Note that alt entries are counted twice. They
* have to be counted as ordinary entries for gprof to get the call times
* right for the ordinary entries.
*
* High local labels are used in macros to avoid clashes with local labels
* in functions.
*
* Ordinary `ret' is used instead of a macro `RET' because there are a lot
* of `ret's. 0xc3 is the opcode for `ret' (`#define ret ... ret' can't
* be used because this file is sometimes preprocessed in traditional mode).
* `ret' clobbers eflags but this doesn't matter.
*/
#define ALTENTRY(name) GEN_ENTRY(name) ; MCOUNT ; MEXITCOUNT ; jmp 9f
#define CROSSJUMP(jtrue, label, jfalse) \
jfalse 8f; MEXITCOUNT; jmp __CONCAT(to,label); 8:
#define CROSSJUMPTARGET(label) \
ALIGN_TEXT; __CONCAT(to,label): ; MCOUNT; jmp label
#define ENTRY(name) GEN_ENTRY(name) ; 9: ; MCOUNT
#define FAKE_MCOUNT(caller) pushl caller ; call *__mcountp ; popl %ecx
#define MCOUNT call *__mcountp
#define MCOUNT_LABEL(name) GEN_ENTRY(name) ; nop ; ALIGN_TEXT
#ifdef GUPROF
#define MEXITCOUNT call *__mexitcountp
#define ret MEXITCOUNT ; NON_GPROF_RET
#else
#define MEXITCOUNT
#endif
#else /* !GPROF */
/*
* ALTENTRY() has to align because it is before a corresponding ENTRY().
* ENTRY() has to align to because there may be no ALTENTRY() before it.
* If there is a previous ALTENTRY() then the alignment code for ENTRY()
* is empty.
*/
#define ALTENTRY(name) GEN_ENTRY(name)
#define CROSSJUMP(jtrue, label, jfalse) jtrue label
#define CROSSJUMPTARGET(label)
#define ENTRY(name) GEN_ENTRY(name)
#define FAKE_MCOUNT(caller)
#define MCOUNT
#define MCOUNT_LABEL(name)
#define MEXITCOUNT
#endif /* GPROF */
#define ALTENTRY(name) GEN_ENTRY(name)
#define END(name) .size name, . - name
#ifdef LOCORE

94
sys/i386/include/profile.h
View file

@ -35,87 +35,9 @@
#ifndef _MACHINE_PROFILE_H_
#define _MACHINE_PROFILE_H_
#ifndef _SYS_CDEFS_H_
#error this file needs sys/cdefs.h as a prerequisite
#endif
#ifndef _KERNEL
#ifdef _KERNEL
/*
* Config generates something to tell the compiler to align functions on 16
* byte boundaries. A strict alignment is good for keeping the tables small.
*/
#define FUNCTION_ALIGNMENT 16
/*
* The kernel uses assembler stubs instead of unportable inlines.
* This is mainly to save a little time when profiling is not enabled,
* which is the usual case for the kernel.
*/
#define _MCOUNT_DECL void mcount
#define MCOUNT
#ifdef GUPROF
#define MCOUNT_DECL(s)
#define MCOUNT_ENTER(s)
#define MCOUNT_EXIT(s)
#ifdef __GNUCLIKE_ASM
#define MCOUNT_OVERHEAD(label) \
__asm __volatile("pushl %0; call __mcount; popl %%ecx" \
: \
: "i" (label) \
: "ax", "dx", "cx", "memory")
#define MEXITCOUNT_OVERHEAD() \
__asm __volatile("call .mexitcount; 1:" \
: : \
: "cx", "memory")
#define MEXITCOUNT_OVERHEAD_GETLABEL(labelp) \
__asm __volatile("movl $1b,%0" : "=rm" (labelp))
#else
#error
#endif /* !__GNUCLIKE_ASM */
#else /* !GUPROF */
#define MCOUNT_DECL(s) register_t s;
#ifdef SMP
extern int mcount_lock;
#define MCOUNT_ENTER(s) { s = intr_disable(); \
while (!atomic_cmpset_acq_int(&mcount_lock, 0, 1)) \
/* nothing */ ; }
#define MCOUNT_EXIT(s) { atomic_store_rel_int(&mcount_lock, 0); \
intr_restore(s); }
#else
#define MCOUNT_ENTER(s) { s = intr_disable(); }
#define MCOUNT_EXIT(s) (intr_restore(s))
#endif
#endif /* GUPROF */
void bintr(void);
void btrap(void);
void eintr(void);
#if 0
void end_exceptions(void);
void start_exceptions(void);
#else
#include <machine/pmc_mdep.h> /* XXX */
#endif
void user(void);
#include <machine/md_var.h> /* XXX for setidt_disp */
#define MCOUNT_DETRAMP(pc) do { \
if ((pc) >= (uintfptr_t)start_exceptions + setidt_disp && \
(pc) < (uintfptr_t)end_exceptions + setidt_disp) \
(pc) -= setidt_disp; \
} while (0)
#define MCOUNT_FROMPC_INTR(pc) \
((pc >= (uintfptr_t)btrap && pc < (uintfptr_t)eintr) ? \
((pc >= (uintfptr_t)bintr) ? (uintfptr_t)bintr : \
(uintfptr_t)btrap) : ~0U)
#define MCOUNT_USERPC ((uintfptr_t)user)
#else /* !_KERNEL */
#include <sys/cdefs.h>
#define FUNCTION_ALIGNMENT 4
@ -159,28 +81,18 @@ mcount() \
typedef u_int uintfptr_t;
#endif /* _KERNEL */
/*
* An unsigned integral type that can hold non-negative difference between
* function pointers.
*/
typedef u_int fptrdiff_t;
#ifdef _KERNEL
void mcount(uintfptr_t frompc, uintfptr_t selfpc);
#else /* !_KERNEL */
#include <sys/cdefs.h>
__BEGIN_DECLS
#ifdef __GNUCLIKE_ASM
void mcount(void) __asm(".mcount");
#endif
__END_DECLS
#endif /* _KERNEL */
#endif /* !_KERNEL */
#endif /* !_MACHINE_PROFILE_H_ */

6
sys/i386/linux/linux_locore.asm
View file

@ -11,7 +11,7 @@
* To avoid excess stack frame the signal trampoline code emulates
* the 'call' instruction.
*/
NON_GPROF_ENTRY(linux_sigcode)
ENTRY(linux_sigcode)
movl %esp, %ebx /* preserve sigframe */
call .getip0
.getip0:
@ -26,7 +26,7 @@ NON_GPROF_ENTRY(linux_sigcode)
.endsigcode:
0: jmp 0b
NON_GPROF_ENTRY(linux_rt_sigcode)
ENTRY(linux_rt_sigcode)
leal LINUX_RT_SIGF_UC(%esp),%ebx /* linux ucp */
leal LINUX_RT_SIGF_SC(%ebx),%ecx /* linux sigcontext */
movl %esp, %edi
@ -42,7 +42,7 @@ NON_GPROF_ENTRY(linux_rt_sigcode)
.endrtsigcode:
0: jmp 0b
NON_GPROF_ENTRY(linux_vsyscall)
ENTRY(linux_vsyscall)
.startvsyscall:
int $0x80
ret

345
sys/kern/subr_prof.c
View file

@ -46,351 +46,6 @@ __FBSDID("$FreeBSD$");
#include <machine/cpu.h>
#ifdef GPROF
#include <sys/malloc.h>
#include <sys/gmon.h>
#undef MCOUNT
static MALLOC_DEFINE(M_GPROF, "gprof", "kernel profiling buffer");
static void kmstartup(void *);
SYSINIT(kmem, SI_SUB_KPROF, SI_ORDER_FIRST, kmstartup, NULL);
struct gmonparam _gmonparam = { GMON_PROF_OFF };
#ifdef GUPROF
void
nullfunc_loop_profiled()
{
int i;
for (i = 0; i < CALIB_SCALE; i++)
nullfunc_profiled();
}
#define nullfunc_loop_profiled_end nullfunc_profiled /* XXX */
void
nullfunc_profiled()
{
}
#endif /* GUPROF */
/*
* Update the histograms to support extending the text region arbitrarily.
* This is done slightly naively (no sparse regions), so will waste slight
* amounts of memory, but will overall work nicely enough to allow profiling
* of KLDs.
*/
void
kmupetext(uintfptr_t nhighpc)
{
struct gmonparam np; /* slightly large */
struct gmonparam *p = &_gmonparam;
char *cp;
GIANT_REQUIRED;
bcopy(p, &np, sizeof(*p));
np.highpc = ROUNDUP(nhighpc, HISTFRACTION * sizeof(HISTCOUNTER));
if (np.highpc <= p->highpc)
return;
np.textsize = np.highpc - p->lowpc;
np.kcountsize = np.textsize / HISTFRACTION;
np.hashfraction = HASHFRACTION;
np.fromssize = np.textsize / HASHFRACTION;
np.tolimit = np.textsize * ARCDENSITY / 100;
if (np.tolimit < MINARCS)
np.tolimit = MINARCS;
else if (np.tolimit > MAXARCS)
np.tolimit = MAXARCS;
np.tossize = np.tolimit * sizeof(struct tostruct);
cp = malloc(np.kcountsize + np.fromssize + np.tossize,
M_GPROF, M_WAITOK);
/*
* Check for something else extending highpc while we slept.
*/
if (np.highpc <= p->highpc) {
free(cp, M_GPROF);
return;
}
np.tos = (struct tostruct *)cp;
cp += np.tossize;
np.kcount = (HISTCOUNTER *)cp;
cp += np.kcountsize;
np.froms = (u_short *)cp;
#ifdef GUPROF
/* Reinitialize pointers to overhead counters. */
np.cputime_count = &KCOUNT(&np, PC_TO_I(&np, cputime));
np.mcount_count = &KCOUNT(&np, PC_TO_I(&np, mcount));
np.mexitcount_count = &KCOUNT(&np, PC_TO_I(&np, mexitcount));
#endif
critical_enter();
bcopy(p->tos, np.tos, p->tossize);
bzero((char *)np.tos + p->tossize, np.tossize - p->tossize);
bcopy(p->kcount, np.kcount, p->kcountsize);
bzero((char *)np.kcount + p->kcountsize, np.kcountsize -
p->kcountsize);
bcopy(p->froms, np.froms, p->fromssize);
bzero((char *)np.froms + p->fromssize, np.fromssize - p->fromssize);
cp = (char *)p->tos;
bcopy(&np, p, sizeof(*p));
critical_exit();
free(cp, M_GPROF);
}
static void
kmstartup(void *dummy)
{
char *cp;
struct gmonparam *p = &_gmonparam;
#ifdef GUPROF
int cputime_overhead;
int empty_loop_time;
int i;
int mcount_overhead;
int mexitcount_overhead;
int nullfunc_loop_overhead;
int nullfunc_loop_profiled_time;
uintfptr_t tmp_addr;
#endif
/*
* Round lowpc and highpc to multiples of the density we're using
* so the rest of the scaling (here and in gprof) stays in ints.
*/
p->lowpc = ROUNDDOWN((u_long)btext, HISTFRACTION * sizeof(HISTCOUNTER));
p->highpc = ROUNDUP((u_long)etext, HISTFRACTION * sizeof(HISTCOUNTER));
p->textsize = p->highpc - p->lowpc;
printf("Profiling kernel, textsize=%lu [%jx..%jx]\n",
p->textsize, (uintmax_t)p->lowpc, (uintmax_t)p->highpc);
p->kcountsize = p->textsize / HISTFRACTION;
p->hashfraction = HASHFRACTION;
p->fromssize = p->textsize / HASHFRACTION;
p->tolimit = p->textsize * ARCDENSITY / 100;
if (p->tolimit < MINARCS)
p->tolimit = MINARCS;
else if (p->tolimit > MAXARCS)
p->tolimit = MAXARCS;
p->tossize = p->tolimit * sizeof(struct tostruct);
cp = (char *)malloc(p->kcountsize + p->fromssize + p->tossize,
M_GPROF, M_WAITOK | M_ZERO);
p->tos = (struct tostruct *)cp;
cp += p->tossize;
p->kcount = (HISTCOUNTER *)cp;
cp += p->kcountsize;
p->froms = (u_short *)cp;
p->histcounter_type = FUNCTION_ALIGNMENT / HISTFRACTION * NBBY;
#ifdef GUPROF
/* Signed counters. */
p->histcounter_type = -p->histcounter_type;
/* Initialize pointers to overhead counters. */
p->cputime_count = &KCOUNT(p, PC_TO_I(p, cputime));
p->mcount_count = &KCOUNT(p, PC_TO_I(p, mcount));
p->mexitcount_count = &KCOUNT(p, PC_TO_I(p, mexitcount));
/*
* Disable interrupts to avoid interference while we calibrate
* things.
*/
critical_enter();
/*
* Determine overheads.
* XXX this needs to be repeated for each useful timer/counter.
*/
cputime_overhead = 0;
startguprof(p);
for (i = 0; i < CALIB_SCALE; i++)
cputime_overhead += cputime();
empty_loop();
startguprof(p);
empty_loop();
empty_loop_time = cputime();
nullfunc_loop_profiled();
/*
* Start profiling. There won't be any normal function calls since
* interrupts are disabled, but we will call the profiling routines
* directly to determine their overheads.
*/
p->state = GMON_PROF_HIRES;
startguprof(p);
nullfunc_loop_profiled();
startguprof(p);
for (i = 0; i < CALIB_SCALE; i++)
MCOUNT_OVERHEAD(sys_profil);
mcount_overhead = KCOUNT(p, PC_TO_I(p, sys_profil));
startguprof(p);
for (i = 0; i < CALIB_SCALE; i++)
MEXITCOUNT_OVERHEAD();
MEXITCOUNT_OVERHEAD_GETLABEL(tmp_addr);
mexitcount_overhead = KCOUNT(p, PC_TO_I(p, tmp_addr));
p->state = GMON_PROF_OFF;
stopguprof(p);
critical_exit();
nullfunc_loop_profiled_time = 0;
for (tmp_addr = (uintfptr_t)nullfunc_loop_profiled;
tmp_addr < (uintfptr_t)nullfunc_loop_profiled_end;
tmp_addr += HISTFRACTION * sizeof(HISTCOUNTER))
nullfunc_loop_profiled_time += KCOUNT(p, PC_TO_I(p, tmp_addr));
#define CALIB_DOSCALE(count) (((count) + CALIB_SCALE / 3) / CALIB_SCALE)
#define c2n(count, freq) ((int)((count) * 1000000000LL / freq))
printf("cputime %d, empty_loop %d, nullfunc_loop_profiled %d, mcount %d, mexitcount %d\n",
CALIB_DOSCALE(c2n(cputime_overhead, p->profrate)),
CALIB_DOSCALE(c2n(empty_loop_time, p->profrate)),
CALIB_DOSCALE(c2n(nullfunc_loop_profiled_time, p->profrate)),
CALIB_DOSCALE(c2n(mcount_overhead, p->profrate)),
CALIB_DOSCALE(c2n(mexitcount_overhead, p->profrate)));
cputime_overhead -= empty_loop_time;
mcount_overhead -= empty_loop_time;
mexitcount_overhead -= empty_loop_time;
/*-
* Profiling overheads are determined by the times between the
* following events:
* MC1: mcount() is called
* MC2: cputime() (called from mcount()) latches the timer
* MC3: mcount() completes
* ME1: mexitcount() is called
* ME2: cputime() (called from mexitcount()) latches the timer
* ME3: mexitcount() completes.
* The times between the events vary slightly depending on instruction
* combination and cache misses, etc. Attempt to determine the
* minimum times. These can be subtracted from the profiling times
* without much risk of reducing the profiling times below what they
* would be when profiling is not configured. Abbreviate:
* ab = minimum time between MC1 and MC3
* a = minimum time between MC1 and MC2
* b = minimum time between MC2 and MC3
* cd = minimum time between ME1 and ME3
* c = minimum time between ME1 and ME2
* d = minimum time between ME2 and ME3.
* These satisfy the relations:
* ab <= mcount_overhead (just measured)
* a + b <= ab
* cd <= mexitcount_overhead (just measured)
* c + d <= cd
* a + d <= nullfunc_loop_profiled_time (just measured)
* a >= 0, b >= 0, c >= 0, d >= 0.
* Assume that ab and cd are equal to the minimums.
*/
p->cputime_overhead = CALIB_DOSCALE(cputime_overhead);
p->mcount_overhead = CALIB_DOSCALE(mcount_overhead - cputime_overhead);
p->mexitcount_overhead = CALIB_DOSCALE(mexitcount_overhead
- cputime_overhead);
nullfunc_loop_overhead = nullfunc_loop_profiled_time - empty_loop_time;
p->mexitcount_post_overhead = CALIB_DOSCALE((mcount_overhead
- nullfunc_loop_overhead)
/ 4);
p->mexitcount_pre_overhead = p->mexitcount_overhead
+ p->cputime_overhead
- p->mexitcount_post_overhead;
p->mcount_pre_overhead = CALIB_DOSCALE(nullfunc_loop_overhead)
- p->mexitcount_post_overhead;
p->mcount_post_overhead = p->mcount_overhead
+ p->cputime_overhead
- p->mcount_pre_overhead;
printf(
"Profiling overheads: mcount: %d+%d, %d+%d; mexitcount: %d+%d, %d+%d nsec\n",
c2n(p->cputime_overhead, p->profrate),
c2n(p->mcount_overhead, p->profrate),
c2n(p->mcount_pre_overhead, p->profrate),
c2n(p->mcount_post_overhead, p->profrate),
c2n(p->cputime_overhead, p->profrate),
c2n(p->mexitcount_overhead, p->profrate),
c2n(p->mexitcount_pre_overhead, p->profrate),
c2n(p->mexitcount_post_overhead, p->profrate));
printf(
"Profiling overheads: mcount: %d+%d, %d+%d; mexitcount: %d+%d, %d+%d cycles\n",
p->cputime_overhead, p->mcount_overhead,
p->mcount_pre_overhead, p->mcount_post_overhead,
p->cputime_overhead, p->mexitcount_overhead,
p->mexitcount_pre_overhead, p->mexitcount_post_overhead);
#endif /* GUPROF */
}
/*
* Return kernel profiling information.
*/
static int
sysctl_kern_prof(SYSCTL_HANDLER_ARGS)
{
int *name = (int *) arg1;
u_int namelen = arg2;
struct gmonparam *gp = &_gmonparam;
int error;
int state;
/* all sysctl names at this level are terminal */
if (namelen != 1)
return (ENOTDIR); /* overloaded */
switch (name[0]) {
case GPROF_STATE:
state = gp->state;
error = sysctl_handle_int(oidp, &state, 0, req);
if (error)
return (error);
if (!req->newptr)
return (0);
if (state == GMON_PROF_OFF) {
gp->state = state;
PROC_LOCK(&proc0);
stopprofclock(&proc0);
PROC_UNLOCK(&proc0);
stopguprof(gp);
} else if (state == GMON_PROF_ON) {
gp->state = GMON_PROF_OFF;
stopguprof(gp);
gp->profrate = profhz;
PROC_LOCK(&proc0);
startprofclock(&proc0);
PROC_UNLOCK(&proc0);
gp->state = state;
#ifdef GUPROF
} else if (state == GMON_PROF_HIRES) {
gp->state = GMON_PROF_OFF;
PROC_LOCK(&proc0);
stopprofclock(&proc0);
PROC_UNLOCK(&proc0);
startguprof(gp);
gp->state = state;
#endif
} else if (state != gp->state)
return (EINVAL);
return (0);
case GPROF_COUNT:
return (sysctl_handle_opaque(oidp,
gp->kcount, gp->kcountsize, req));
case GPROF_FROMS:
return (sysctl_handle_opaque(oidp,
gp->froms, gp->fromssize, req));
case GPROF_TOS:
return (sysctl_handle_opaque(oidp,
gp->tos, gp->tossize, req));
case GPROF_GMONPARAM:
return (sysctl_handle_opaque(oidp, gp, sizeof *gp, req));
default:
return (EOPNOTSUPP);
}
/* NOTREACHED */
}
static SYSCTL_NODE(_kern, KERN_PROF, prof,
CTLFLAG_RW | CTLFLAG_MPSAFE, sysctl_kern_prof,
"");
#endif /* GPROF */
/*
* Profiling system call.
*

316
sys/libkern/mcount.c
View file

@ -1,316 +0,0 @@
/*-
* SPDX-License-Identifier: BSD-3-Clause
*
* Copyright (c) 1983, 1992, 1993
* The Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/gmon.h>
#ifdef _KERNEL
#ifndef GUPROF
#include <sys/systm.h>
#endif
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#endif
/*
* mcount is called on entry to each function compiled with the profiling
* switch set. _mcount(), which is declared in a machine-dependent way
* with _MCOUNT_DECL, does the actual work and is either inlined into a
* C routine or called by an assembly stub. In any case, this magic is
* taken care of by the MCOUNT definition in <machine/profile.h>.
*
* _mcount updates data structures that represent traversals of the
* program's call graph edges. frompc and selfpc are the return
* address and function address that represents the given call graph edge.
*
* Note: the original BSD code used the same variable (frompcindex) for
* both frompcindex and frompc. Any reasonable, modern compiler will
* perform this optimization.
*/
/* _mcount; may be static, inline, etc */
_MCOUNT_DECL(uintfptr_t frompc, uintfptr_t selfpc)
{
#ifdef GUPROF
int delta;
#endif
fptrdiff_t frompci;
u_short *frompcindex;
struct tostruct *top, *prevtop;
struct gmonparam *p;
long toindex;
#ifdef _KERNEL
MCOUNT_DECL(s)
#endif
p = &_gmonparam;
#ifndef GUPROF /* XXX */
/*
* check that we are profiling
* and that we aren't recursively invoked.
*/
if (p->state != GMON_PROF_ON)
return;
#endif
#ifdef _KERNEL
MCOUNT_ENTER(s);
#else
p->state = GMON_PROF_BUSY;
#endif
#ifdef _KERNEL
/* De-relocate any addresses in a (single) trampoline. */
#ifdef MCOUNT_DETRAMP
MCOUNT_DETRAMP(frompc);
MCOUNT_DETRAMP(selfpc);
#endif
/*
* When we are called from an exception handler, frompc may be
* a user address. Convert such frompc's to some representation
* in kernel address space.
*/
#ifdef MCOUNT_FROMPC_USER
frompc = MCOUNT_FROMPC_USER(frompc);
#elif defined(MCOUNT_USERPC)
/*
* For separate address spaces, we can only guess that addresses
* in the range known to us are actually kernel addresses. Outside
* of this range, conerting to the user address is fail-safe.
*/
if (frompc < p->lowpc || frompc - p->lowpc >= p->textsize)
frompc = MCOUNT_USERPC;
#endif
#endif /* _KERNEL */
frompci = frompc - p->lowpc;
if (frompci >= p->textsize)
goto done;
#ifdef GUPROF
if (p->state == GMON_PROF_HIRES) {
/*
* Count the time since cputime() was previously called
* against `frompc'. Compensate for overheads.
*
* cputime() sets its prev_count variable to the count when
* it is called. This in effect starts a counter for
* the next period of execution (normally from now until
* the next call to mcount() or mexitcount()). We set
* cputime_bias to compensate for our own overhead.
*
* We use the usual sampling counters since they can be
* located efficiently. 4-byte counters are usually
* necessary. gprof will add up the scattered counts
* just like it does for statistical profiling. All
* counts are signed so that underflow in the subtractions
* doesn't matter much (negative counts are normally
* compensated for by larger counts elsewhere). Underflow
* shouldn't occur, but may be caused by slightly wrong
* calibrations or from not clearing cputime_bias.
*/
delta = cputime() - cputime_bias - p->mcount_pre_overhead;
cputime_bias = p->mcount_post_overhead;
KCOUNT(p, frompci) += delta;
*p->cputime_count += p->cputime_overhead;
*p->mcount_count += p->mcount_overhead;
}
#endif /* GUPROF */
#ifdef _KERNEL
/*
* When we are called from an exception handler, frompc is faked
* to be for where the exception occurred. We've just solidified
* the count for there. Now convert frompci to an index that
* represents the kind of exception so that interruptions appear
* in the call graph as calls from those index instead of calls
* from all over.
*/
frompc = MCOUNT_FROMPC_INTR(selfpc);
if ((frompc - p->lowpc) < p->textsize)
frompci = frompc - p->lowpc;
#endif
/*
* check that frompc is a reasonable pc value.
* for example: signal catchers get called from the stack,
* not from text space. too bad.
*/
if (frompci >= p->textsize)
goto done;
frompcindex = &p->froms[frompci / (p->hashfraction * sizeof(*p->froms))];
toindex = *frompcindex;
if (toindex == 0) {
/*
* first time traversing this arc
*/
toindex = ++p->tos[0].link;
if (toindex >= p->tolimit)
/* halt further profiling */
goto overflow;
*frompcindex = toindex;
top = &p->tos[toindex];
top->selfpc = selfpc;
top->count = 1;
top->link = 0;
goto done;
}
top = &p->tos[toindex];
if (top->selfpc == selfpc) {
/*
* arc at front of chain; usual case.
*/
top->count++;
goto done;
}
/*
* have to go looking down chain for it.
* top points to what we are looking at,
* prevtop points to previous top.
* we know it is not at the head of the chain.
*/
for (; /* goto done */; ) {
if (top->link == 0) {
/*
* top is end of the chain and none of the chain
* had top->selfpc == selfpc.
* so we allocate a new tostruct
* and link it to the head of the chain.
*/
toindex = ++p->tos[0].link;
if (toindex >= p->tolimit)
goto overflow;
top = &p->tos[toindex];
top->selfpc = selfpc;
top->count = 1;
top->link = *frompcindex;
*frompcindex = toindex;
goto done;
}
/*
* otherwise, check the next arc on the chain.
*/
prevtop = top;
top = &p->tos[top->link];
if (top->selfpc == selfpc) {
/*
* there it is.
* increment its count
* move it to the head of the chain.
*/
top->count++;
toindex = prevtop->link;
prevtop->link = top->link;
top->link = *frompcindex;
*frompcindex = toindex;
goto done;
}
}
done:
#ifdef _KERNEL
MCOUNT_EXIT(s);
#else
p->state = GMON_PROF_ON;
#endif
return;
overflow:
p->state = GMON_PROF_ERROR;
#ifdef _KERNEL
MCOUNT_EXIT(s);
#endif
return;
}
/*
* Actual definition of mcount function. Defined in <machine/profile.h>,
* which is included by <sys/gmon.h>.
*/
MCOUNT
#ifdef GUPROF
void
mexitcount(uintfptr_t selfpc)
{
struct gmonparam *p;
uintfptr_t selfpcdiff;
p = &_gmonparam;
#ifdef MCOUNT_DETRAMP
MCOUNT_DETRAMP(selfpc);
#endif
selfpcdiff = selfpc - (uintfptr_t)p->lowpc;
if (selfpcdiff < p->textsize) {
int delta;
/*
* Count the time since cputime() was previously called
* against `selfpc'. Compensate for overheads.
*/
delta = cputime() - cputime_bias - p->mexitcount_pre_overhead;
cputime_bias = p->mexitcount_post_overhead;
KCOUNT(p, selfpcdiff) += delta;
*p->cputime_count += p->cputime_overhead;
*p->mexitcount_count += p->mexitcount_overhead;
}
}
#ifndef __GNUCLIKE_ASM
#error "This file uses null asms to prevent timing loops being optimized away."
#endif
void
empty_loop(void)
{
int i;
for (i = 0; i < CALIB_SCALE; i++)
__asm __volatile("");
}
void
nullfunc(void)
{
__asm __volatile("");
}
void
nullfunc_loop(void)
{
int i;
for (i = 0; i < CALIB_SCALE; i++)
nullfunc();
}
#endif /* GUPROF */