Date: Thu, 30 Mar 2017 20:54:45 -0700 From: Mark Millard <markmi@dsl-only.net> To: freebsd-hackers@freebsd.org Cc: freebsd-arm <freebsd-arm@freebsd.org> Subject: Is this procstat -v output valid/expected? Explanation? Message-ID: <D526FEE8-E234-4219-8850-70EB7C689946@dsl-only.net>
next in thread | raw e-mail | index | archive | help
The following is based on a test-case program that:
A) Allocates lots of 14KiByte "regions" with malloc,
initializing each byte of each region to a
never-zero pattern of bytes. "Lots" uses up
most of 256 MiBytes across the regions. Once
initialized none of these bytes are written
again (not even by the later child process).
B) Tests the byte patterns (SIGABRT if the pattern
test fails).
C) Forks.
D) The parent waits for child; child sleeps 60 sec.
Note: In the full test forcing swapping of
both is involved during the sleep but
that is not being done here.
E) The child checks the byte patterns and exits
(SIGABRT if the pattern test fails). The child
does not write to any of the allocation regions.
F) The (former) parent checks the byte patterns
and exits (SIGABRT if the pattern test fails).
It does not write to any of the allocated
regions during this activity.
[The context happens to be arm64.]
Note the two instances of "67306" in the below from
what will become the parent process:
# procstat -v 6310
PID START END PRT RES PRES REF SHD FLAG =
TP PATH
6310 0x10000 0x11000 r-- 1 51 3 1 CN-- =
vn /root/c_tests/swaptesting2
6310 0x20000 0x21000 r-x 1 51 3 1 CN-- =
vn /root/c_tests/swaptesting2
6310 0x30000 0x40000 rw- 16 0 1 0 C--- =
vn /root/c_tests/swaptesting2
6310 0x40000 0x41000 r-- 1 38 2 0 ---- =
df=20
6310 0x41000 0x75000 rw- 37 38 2 0 ---- =
df=20
6310 0x40030000 0x4004b000 r-x 27 29 59 27 CN-- =
vn /libexec/ld-elf.so.1
6310 0x4004b000 0x40052000 rw- 7 7 1 0 ---- =
df=20
6310 0x4005a000 0x4005b000 rw- 1 0 1 0 C--- =
vn /libexec/ld-elf.so.1
6310 0x4005b000 0x4005c000 rw- 1 1 1 0 ---- =
df=20
6310 0x4005c000 0x401b4000 r-x 344 376 59 27 CN-- =
vn /lib/libc.so.7
6310 0x401b4000 0x401c3000 --- 0 0 1 0 ---- =
df=20
6310 0x401c3000 0x401cf000 rw- 12 0 1 0 C--- =
vn /lib/libc.so.7
6310 0x401cf000 0x40202000 rw- 22 67306 2 0 ---- =
df=20
6310 0x40400000 0x50e00000 rw- 67284 67306 2 0 ---- =
df=20
6310 0xfffffffdf000 0xfffffffff000 rw- 3 3 1 0 ---D =
df=20
6310 0xfffffffff000 0x1000000000000 r-x 1 1 35 0 ---- =
ph=20
Later after the fork (so child sleeping and parent waiting)
it as turned into:
# procstat -v 6310
PID START END PRT RES PRES REF SHD FLAG =
TP PATH
6310 0x10000 0x11000 r-- 1 51 5 1 CN-- =
vn /root/c_tests/swaptesting2
6310 0x20000 0x21000 r-x 1 51 5 1 CN-- =
vn /root/c_tests/swaptesting2
6310 0x30000 0x40000 rw- 16 0 1 0 C--- =
vn /root/c_tests/swaptesting2
6310 0x40000 0x41000 r-- 1 1 2 0 CN-- =
df=20
6310 0x41000 0x75000 rw- 37 37 2 0 CN-- =
df=20
6310 0x40030000 0x4004b000 r-x 27 29 60 27 CN-- =
vn /libexec/ld-elf.so.1
6310 0x4004b000 0x40052000 rw- 7 0 1 0 C--- =
df=20
6310 0x4005a000 0x4005b000 rw- 1 0 2 0 CN-- =
vn /libexec/ld-elf.so.1
6310 0x4005b000 0x4005c000 rw- 1 0 1 0 C--- =
df=20
6310 0x4005c000 0x401b4000 r-x 344 376 60 27 CN-- =
vn /lib/libc.so.7
6310 0x401b4000 0x401c3000 --- 0 0 2 0 CN-- =
df=20
6310 0x401c3000 0x401cf000 rw- 12 0 2 0 CN-- =
vn /lib/libc.so.7
6310 0x401cf000 0x40202000 rw- 22 22 2 0 CN-- =
df=20
6310 0x40400000 0x50e00000 rw- 67284 67284 2 0 CN-- =
df=20
6310 0xfffffffdf000 0xfffffffff000 rw- 3 0 1 0 C--D =
df=20
6310 0xfffffffff000 0x1000000000000 r-x 1 1 36 0 ---- =
ph=20
The child never shows the large PRES figure for the range:
0x401cf000 0x40202000
But for the size of that range the earlier PRES=3D=3D67306 seems odd,
as if it spans the following:
0x40400000 0x50e0000
In fact 22+67284=3D=3D67306.
Another point that I noticed that the I found SHD
stays zero on the memory area spanning the allocations
(0x40400000 0x50e00000) and more:
(This was during the child's sleep.)
# procstat -v 6313
PID START END PRT RES PRES REF SHD FLAG =
TP PATH
6313 0x10000 0x11000 r-- 1 51 5 1 CN-- =
vn /root/c_tests/swaptesting2
6313 0x20000 0x21000 r-x 1 51 5 1 CN-- =
vn /root/c_tests/swaptesting2
6313 0x30000 0x40000 rw- 16 0 1 0 C--- =
vn /root/c_tests/swaptesting2
6313 0x40000 0x41000 r-- 1 1 2 0 CN-- =
df=20
6313 0x41000 0x75000 rw- 37 37 2 0 CN-- =
df=20
6313 0x40030000 0x4004b000 r-x 27 29 60 27 CN-- =
vn /libexec/ld-elf.so.1
6313 0x4004b000 0x40052000 rw- 7 0 1 0 C--- =
df=20
6313 0x4005a000 0x4005b000 rw- 1 0 2 0 CN-- =
vn /libexec/ld-elf.so.1
6313 0x4005b000 0x4005c000 rw- 1 0 1 0 C--- =
df=20
6313 0x4005c000 0x401b4000 r-x 344 376 60 27 CN-- =
vn /lib/libc.so.7
6313 0x401b4000 0x401c3000 --- 0 0 2 0 CN-- =
df=20
6313 0x401c3000 0x401cf000 rw- 12 0 2 0 CN-- =
vn /lib/libc.so.7
6313 0x401cf000 0x40202000 rw- 22 22 2 0 CN-- =
df=20
6313 0x40400000 0x50e00000 rw- 67284 67284 2 0 CN-- =
df=20
6313 0xfffffffdf000 0xfffffffff000 rw- 3 0 1 0 C--D =
df=20
6313 0xfffffffff000 0x1000000000000 r-x 1 1 36 0 ---- =
ph=20
For:
0x40400000 0x50e00000
(and more)
my first thought was that forking would shadow for copy-on-write
and so the shadow page count would be non-zero in one or both
of the parent vs. child. But Ive never seen procstat -v report
such a figure for the range holding the allocations.
The REF=3D=3D2 also seems odd: it lasts from before the fork through
after it as well, both parent and child processes still existing.
It would seem that the REF's are not per-process.
Context details:
# uname -paKU
FreeBSD pine64 12.0-CURRENT FreeBSD 12.0-CURRENT r315914M arm64 =
aarch64 1200027 1200027
FYI: the source code is. . .
(Ignore comments tied to swapping and
its/the "problem" for this question.)
# more swap_testing2.c
// swap_testing2.c
// Built via (c++ was clang++ 4.0 in my case):
//
// cc -g -std=3Dc11 -Wpedantic -o swaptesting2 swap_testing2.c
// -O0 and -O2 also gets the problem.
#include <unistd.h> // for fork(), sleep(.)
#include <sys/types.h> // for pid_t
#include <sys/wait.h> // for wait(.)
#include <signal.h> // for raise(.), SIGABRT
extern void test_setup(void); // Sets up the memory byte =
patterns.
extern void test_check(void); // Tests the memory byte patterns.
extern void partial_test_check(void); // Tests just [0] of =
dyn_regions[0]
int main(void) {
test_setup();
test_check(); // Before fork() [passes]
pid_t pid =3D fork();
int wait_status =3D 0;;
// After fork; before waitsleep/swap-out.
//if (0=3D=3Dpid) partial_test_check();
// Even the above is sufficient by
// itself to prevent failure for
// region_size 1u through
// 4u*1024u!
// But 4u*1024u+1u and above fail
// with this access to memory.
// The failing test is of
// (*dyn_regions[0]).array[4096u].
// This test never fails here.
if (0<pid) partial_test_check(); // This never prevents
// later failures (and
// never fails here).
if (0<pid) { wait(&wait_status); }
if (-1!=3Dwait_status && 0<=3Dpid) {
if (0=3D=3Dpid) {
sleep(60);
// During this manually force this process to
// swap out. I use something like:
// stress -m 1 --vm-bytes 1800M
// in another shell and ^C'ing it after top
// shows the swapped status desired. 1800M
// just happened to work on the Pine64+ 2GB
// that I was using. I watch with top -PCwaopid .
}
test_check(); // After wait/sleep [fails for small-enough =
region_sizes]
// raise(SIGABRT);
}
}
// The memory and test code follows.
#include <stddef.h> // for size_t, NULL
#include <stdlib.h> // for malloc(.), free(.)
#define region_size (14u*1024u)
// Bad dyn_regions patterns, parent and child
// processes:
// 256u, 2u*1024u, 4u*1024u, 8u*1024u,
// 9u*1024u, 12u*1024u, 14u*1024u
// (but see the partial_test_check() call
// notes above).
// Works:
// 14u*1024u+1u, 15u*1024u, 16u*1024u,
// 32u*1024u, 256u*1024u*1024u
#define num_regions (256u*1024u*1024u/region_size)
typedef volatile unsigned char value_type;
struct region_struct { value_type array[region_size]; };
typedef struct region_struct region;
static region * volatile dyn_regions[num_regions] =3D {NULL,};
static value_type value(size_t v) { return (value_type)((v&0xFEu)|0x1u); =
}
// value now avoids the zero value since the failures
// are zeros.
void test_setup(void) {
for(size_t i=3D0u; i<num_regions; i++) {
dyn_regions[i] =3D malloc(sizeof(region));
if (!dyn_regions[i]) raise(SIGABRT);
for(size_t j=3D0u; j<region_size; j++) {
(*dyn_regions[i]).array[j] =3D value(j);
}
}
}
void partial_test_check(void) {
if (value(0u)!=3D(*dyn_regions[0u]).array[0]) raise(SIGABRT);
}
static volatile size_t first_failure_idx =3D 0u;
static volatile size_t first_failure_pos =3D 0u;
static volatile size_t after_bad_idx =3D 0u;
static volatile size_t after_bad_pos =3D 0u;
void test_check(void) {
first_failure_idx =3D first_failure_pos =3D 0u;
while (first_failure_idx < num_regions) {
while ( first_failure_pos < region_size
&& ( value(first_failure_pos)
=3D=3D =
(*dyn_regions[first_failure_idx]).array[first_failure_pos]
)
) {
first_failure_pos++;
}
if (region_size !=3D first_failure_pos) break;
first_failure_idx++;
first_failure_pos =3D 0u;
}
if (num_regions =3D=3D first_failure_idx) return;
after_bad_idx =3D first_failure_idx;
after_bad_pos =3D first_failure_pos;
while (after_bad_idx < num_regions) {
while ( after_bad_pos < region_size
&& ( value(after_bad_pos)
!=3D (*dyn_regions[after_bad_idx]).array[after_bad_pos]
)
) {
after_bad_pos++;
}
if(region_size !=3D after_bad_pos) break;
after_bad_idx++;
after_bad_pos =3D 0u;
}
raise(SIGABRT);
}
=3D=3D=3D
Mark Millard
markmi at dsl-only.net
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?D526FEE8-E234-4219-8850-70EB7C689946>