ref: 28465682cf7707532583d3a386991b39c683ee1c
parent: 4de9d8a511c1403ef849147e82b54af3bdbd4e3e
author: Jacob Moody <moody@posixcafe.org>
date: Sat Jan 4 19:26:52 EST 2025
kprof: don't downsample the pc Kprof rounded the pc to 8 byte boundaries in order to save memory. With x86 instruction rounding this can lead to some pretty misleading results. With the old rounding kpdata was ~680kb for a pc64 kernel, with this change we use ~5.4M (8x) and considering we only allocate when we first attach this seems reasonable.
--- a/sys/man/3/kprof
+++ b/sys/man/3/kprof
@@ -24,10 +24,9 @@
The first count
holds the total number of clock ticks during profiling;
the second the number of ticks that occurred while the kernel
-was running. The rest each hold the number of ticks
-the kernel program counter was within the
-corresponding 8-byte range of kernel text, starting from the base
-of kernel text.
+was running. The rest each hold the number of total ticks
+the kernel spent at any one program counter, indexed by
+their offset from the minimum kernel text address.
.PP
The file
.B kpctl
--- a/sys/src/9/port/devkprof.c
+++ b/sys/src/9/port/devkprof.c
@@ -6,7 +6,6 @@
#include "../port/error.h"
-#define LRES 3 /* log of PC resolution */
#define SZ 4 /* sizeof of count cell; well known as 4 */
struct
@@ -46,7 +45,6 @@
kprof.buf[0] += TK2MS(1);
if(kprof.minpc<=pc && pc<kprof.maxpc){pc -= kprof.minpc;
- pc >>= LRES;
kprof.buf[pc] += TK2MS(1);
}else
kprof.buf[1] += TK2MS(1);
@@ -68,7 +66,7 @@
/* allocate when first used */
kprof.minpc = KTZERO;
kprof.maxpc = (uintptr)etext;
- kprof.nbuf = (kprof.maxpc-kprof.minpc) >> LRES;
+ kprof.nbuf = kprof.maxpc-kprof.minpc;
n = kprof.nbuf*SZ;
if(kprof.buf == 0) {kprof.buf = xalloc(n);
--- a/sys/src/cmd/kprof.c
+++ b/sys/src/cmd/kprof.c
@@ -3,8 +3,6 @@
#include <bio.h>
#include <mach.h>
-#define PCRES 8
-
struct COUNTER
{char *name; /* function name */
@@ -82,7 +80,7 @@
data = malloc(d->length);
if(data == 0)
error(1, "malloc");
- if(read(fd, data, d->length) < 0)
+ if(readn(fd, data, d->length) != d->length)
error(1, "text read");
close(fd);
for(i=0; i<n; i++)
@@ -112,7 +110,6 @@
if (!textsym(&s, i)) /* get next symbol */
break;
s.value -= tbase;
- s.value /= PCRES;
sum = 0;
while (j < n && j < s.value)
sum += data[j++];
--
⑨