Skip to contentSkip to conversation

tools: gen_eth_addr: remove getpid() operation for the random seed

PATCH New Josh Wu 3 participants 9 messages Sep 16, 2015

lore
Share & tools
lore

Josh Wu

author
#0 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

As 'time(0) | getpid()' sometimes get same value. That depends on the value of getpid().
So that is not a expected behavior. We expect different value for the seed when when run it in many times.

So this patch remove the getpid(), just use the time(0) as the seed.

---

 tools/gen_eth_addr.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/gen_eth_addr.c +1 -1
index bf9d935..53b023a 100644
@@ -15,7 +15,7 @@ main(int argc, char *argv[])
 {
     unsigned long ethaddr_low, ethaddr_high;
 
-    srand(time(0) | getpid());
+    srand(time(0));
 
     /*
      * setting the 2nd LSB in the most significant byte of
Signature · 1 line
1.9.1
#1 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

Dear Josh Wu,

Quoted 2 lines · In message <[email protected]> you wrote:

As 'time(0) | getpid()' sometimes get same value. That depends on the value of getpid().

I think removing some "random input" from the way how we compute the seed is a bad idea.

Quoted 2 lines

So that is not a expected behavior. We expect different value for the seed when when run it in many times.

What is your execution environment? In any sane OS it is higly unlikely that you will see the same or even similar PIDs for successive runs of the program - each run will start a new process, which will get a new PID.

[On a mostly idle Linux system (4.1.6 kernel) I see zero dupes in a set of 30,000 invocations of getpid().]

One can argue if ORing the values is the most clever idea, or if for example ADDing them would result in more "randomness". But completely removing the pid() is bad - any parallel runs of the program on any machines with synchronized times would predictably result in the same seeds which is definitely worse behaviour than what we have now.

Quoted 1 line

So this patch remove the getpid(), just use the time(0) as the seed.

NAK. This is a bad idea.

Best regards,

Wolfgang Denk

Signature · 4 lines
DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
I program, therefore I am.
author
#2 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

Hi, Wolfgang

Thanks for the reply.

Quoted 8 lines · Wolfgang Denk

Dear Josh Wu,

Quoted 2 lines · In message <[email protected]> you wrote:

As 'time(0) | getpid()' sometimes get same value. That depends on the value of getpid().

I think removing some "random input" from the way how we compute the seed is a bad idea.

Quoted 6 lines
Quoted 2 lines

So that is not a expected behavior. We expect different value for the seed when when run it in many times.

What is your execution environment? In any sane OS it is higly unlikely that you will see the same or even similar PIDs for successive runs of the program - each run will start a new process, which will get a new PID.

my system is Ubuntu 14.04

#uname -a
Linux melon 3.13.0-45-generic #74-Ubuntu SMP Tue Jan 13 19:36:28 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux

Following is my test history:

? tools date && ./gen_eth_addr
Wed Sep 16 14:48:53 CST 2015
4a:c3:21:45:17:b2
? tools date && ./gen_eth_addr
Wed Sep 16 14:48:56 CST 2015
a6:29:4b:0b:e6:d0
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:02 CST 2015
d2:41:66:54:64:aa
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:06 CST 2015
2a:58:1d:b0:f0:c5
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:13 CST 2015
1e:8e:6f:0e:16:b8
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:19 CST 2015
56:4f:58:67:ad:30
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:21 CST 2015
2e:53:29:97:6a:8a
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:26 CST 2015
d2:41:66:54:64:aa
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:35 CST 2015
d2:41:66:54:64:aa

? tools date && ./gen_eth_addr
Wed Sep 16 14:49:50 CST 2015
92:33:16:3f:0a:56
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:58 CST 2015
92:33:16:3f:0a:56

In above commands, I have two duplicated eth addr:
92:33:16:3f:0a:56
d2:41:66:54:64:aa

Quoted 6 lines

[On a mostly idle Linux system (4.1.6 kernel) I see zero dupes in a set of 30,000 invocations of getpid().]

One can argue if ORing the values is the most clever idea, or if for example ADDing them would result in more "randomness".

Sure. The ORing seems has big chance to get same result in my machine.

Quoted 4 lines

But completely
removing the pid() is bad - any parallel runs of the program on any machines with synchronized times would predictably result in the same seeds which is definitely worse behaviour than what we have now.

I understand your concern. My intention is make it harder to generate the duplicated result.

Maybe we can ORing the MSB of time(0)?
I'll investigate it little more.

Quoted 8 lines
Quoted 2 lines

So this patch remove the getpid(), just use the time(0) as the seed.

NAK. This is a bad idea.

Best regards,

Wolfgang Denk

Best Regards,
Josh Wu

author
#3 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

Quoted 80 lines · Josh Wu

Hi, Wolfgang

Thanks for the reply.

Quoted 9 lines · Wolfgang Denk

Dear Josh Wu,

In message <[email protected]> you

Quoted 2 lines · wrote:

As 'time(0) | getpid()' sometimes get same value. That depends on the value of getpid().

I think removing some "random input" from the way how we compute the seed is a bad idea.

Quoted 6 lines
Quoted 2 lines

So that is not a expected behavior. We expect different value for the seed when when run it in many times.

What is your execution environment? In any sane OS it is higly unlikely that you will see the same or even similar PIDs for successive runs of the program - each run will start a new process, which will get a new PID.

my system is Ubuntu 14.04

#uname -a
Linux melon 3.13.0-45-generic #74-Ubuntu SMP Tue Jan 13 19:36:28 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux

Following is my test history:

? tools date && ./gen_eth_addr
Wed Sep 16 14:48:53 CST 2015
4a:c3:21:45:17:b2
? tools date && ./gen_eth_addr
Wed Sep 16 14:48:56 CST 2015
a6:29:4b:0b:e6:d0
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:02 CST 2015
d2:41:66:54:64:aa
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:06 CST 2015
2a:58:1d:b0:f0:c5
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:13 CST 2015
1e:8e:6f:0e:16:b8
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:19 CST 2015
56:4f:58:67:ad:30
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:21 CST 2015
2e:53:29:97:6a:8a
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:26 CST 2015
d2:41:66:54:64:aa
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:35 CST 2015
d2:41:66:54:64:aa

? tools date && ./gen_eth_addr
Wed Sep 16 14:49:50 CST 2015
92:33:16:3f:0a:56
? tools date && ./gen_eth_addr
Wed Sep 16 14:49:58 CST 2015
92:33:16:3f:0a:56

In above commands, I have two duplicated eth addr:
92:33:16:3f:0a:56
d2:41:66:54:64:aa

Quoted 6 lines

[On a mostly idle Linux system (4.1.6 kernel) I see zero dupes in a set of 30,000 invocations of getpid().]

One can argue if ORing the values is the most clever idea, or if for example ADDing them would result in more "randomness".

Sure. The ORing seems has big chance to get same result in my machine.

Quoted 4 lines

But completely
removing the pid() is bad - any parallel runs of the program on any machines with synchronized times would predictably result in the same seeds which is definitely worse behaviour than what we have now.

Quoted 5 lines

I understand your concern. My intention is make it harder to generate the duplicated result.

Maybe we can ORing the MSB of time(0)?
I'll investigate it little more.

I test again, here is a duplicated case:

? tools date && ./gen_eth_addr
Wed Sep 16 15:21:30 CST 2015
pid: 0x5aa4, time: 0x55f9187a
86:77:6e:cf:6b:16
? tools date && ./gen_eth_addr
Wed Sep 16 15:21:34 CST 2015
pid: 0x5aa6, time: 0x55f9187e
86:77:6e:cf:6b:16

In above both cases, the random seed is same: (last 4 bits)
4 | a = e
6 | e = e

Quoted 4 lines
Quoted 4 lines
Quoted 3 lines

So this patch remove the getpid(), just use the time(0) as the seed.

NAK. This is a bad idea.

How about use 'time(0) + (getpid() << 8)'? my reason list in below:
1. ADD will make the result is more unlikely than OR
2. (<< 8) means larger time difference.

Best Regards,
Josh Wu

#4 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

In reply to Josh Wu

Dear Josh,

Quoted 4 lines · In message <[email protected]> you wrote:

In above commands, I have two duplicated eth addr:
92:33:16:3f:0a:56
d2:41:66:54:64:aa

Agreed. Randomness is really poor; for a sequence of 1000 invocations of gen_eth_addr in a shell loop I would only gt 124 different MAC addresses:

-> for i in {1..1000} ; do ./gen_eth_addr ; done >/tmp/0 -> sort -u /tmp/0 >/tmp/1 ; wc -l /tmp/[01]
1000 /tmp/0
124 /tmp/1

In a second run, even only 41 :-(

But without the getpid() part, it gets even worse - the same loop would produce only 15 different addresses!

Changing the '|' into a '+' would reliably generate 1000 different MAC addresses.

Quoted 2 lines

I understand your concern. My intention is make it harder to generate the duplicated result.

I understand this, and agree that we should implement such a fix.

Quoted 2 lines

Maybe we can ORing the MSB of time(0)?
I'll investigate it little more.

The following patch appears to work fine for me. Maybe you can test it?

tools/gen_eth_addr.c +1 -1
index bf9d935..834163a 100644
@@ -15,7 +15,7 @@ main(int argc, char *argv[])
 {
     unsigned long ethaddr_low, ethaddr_high;
 
-    srand(time(0) | getpid());
+    srand(time(0) + getpid());
 
     /*
      * setting the 2nd LSB in the most significant byte of

Thanks.

Best regards,

Wolfgang Denk

Signature · 7 lines
DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
I think that all right-thinking people in this country are  sick  and
tired  of  being  told that ordinary decent people are fed up in this
country with being sick and tired. I'm certainly not.  But  I'm  sick
and tired of being told that I am.                     - Monty Python
author
#5 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

Hi, Wolfgang, Andreas

Thanks for the testing and point out the wrongness in my patch.

Quoted 24 lines · Wolfgang Denk

Dear Josh,

Quoted 3 lines · In message <[email protected]> you wrote:

In above commands, I have two duplicated eth addr:
92:33:16:3f:0a:56
d2:41:66:54:64:aa

Agreed. Randomness is really poor; for a sequence of 1000 invocations of gen_eth_addr in a shell loop I would only gt 124 different MAC addresses:

-> for i in {1..1000} ; do ./gen_eth_addr ; done >/tmp/0 -> sort -u /tmp/0 >/tmp/1 ; wc -l /tmp/[01]
1000 /tmp/0
124 /tmp/1

In a second run, even only 41 :-(

But without the getpid() part, it gets even worse - the same loop would produce only 15 different addresses!

Changing the '|' into a '+' would reliably generate 1000 different MAC addresses.

Quoted 8 lines
Quoted 2 lines

I understand your concern. My intention is make it harder to generate the duplicated result.

I understand this, and agree that we should implement such a fix.

Quoted 2 lines

Maybe we can ORing the MSB of time(0)?
I'll investigate it little more.

The following patch appears to work fine for me. Maybe you can test it?

yes, this one works well. if you don't mind I want to sent v2 patch, which is like:

- srand(time(0) | getpid());
+ srand(time(0) + (getpid() << 8));

Quoted 16 lines
tools/gen_eth_addr.c +1 -1
index bf9d935..834163a 100644
@@ -15,7 +15,7 @@ main(int argc, char *argv[])
  {
      unsigned long ethaddr_low, ethaddr_high;
 
-    srand(time(0) | getpid());
+    srand(time(0) + getpid());
 
      /*
       * setting the 2nd LSB in the most significant byte of
Quoted 6 lines

Thanks.

Best regards,

Wolfgang Denk

Best Regards,
Josh Wu

#6 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

In reply to Josh Wu

Hi Josh,

Quoted 4 lines · Josh Wu

As 'time(0) | getpid()' sometimes get same value. That depends on the value of getpid().
So that is not a expected behavior. We expect different value for the seed when when run it in many times.

I don't think your change made it better. Here is a snippet from a run of time(NULL) and getpid():

---8<---
time: 1442389450; pid: 11632; time | pid: 1442397690;
time: 1442389450; pid: 11633; time | pid: 1442397691;
time: 1442389450; pid: 11634; time | pid: 1442397690;
time: 1442389450; pid: 11635; time | pid: 1442397691;
time: 1442389450; pid: 11636; time | pid: 1442397694;
time: 1442389450; pid: 11637; time | pid: 1442397695;
time: 1442389450; pid: 11638; time | pid: 1442397694;
time: 1442389450; pid: 11639; time | pid: 1442397695;
time: 1442389450; pid: 11640; time | pid: 1442397690;
time: 1442389450; pid: 11641; time | pid: 1442397691;
time: 1442389450; pid: 11642; time | pid: 1442397690;
time: 1442389450; pid: 11643; time | pid: 1442397691; --->8---

While time(NULL) is stable, getpid() is incrementing by one. As you may expect the OR'ed value is oscillating and the values almost the same. So calling gen_eth_addr three times within the same second you will get two time the same MAC.

Quoted 1 line

So this patch remove the getpid(), just use the time(0) as the seed.

So let's see the effect of your change ...

The output of gen_eth_addr at the current ToT:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
254

With your change applied:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
10

Another approach would be to change the algorithm (OR the values) here. A short test showed that using XOR could be a solution:

---8<---
time: 1442389450; pid: 11632; time ^ pid: 1442394298;
time: 1442389450; pid: 11633; time ^ pid: 1442394299;
time: 1442389450; pid: 11634; time ^ pid: 1442394296;
time: 1442389450; pid: 11635; time ^ pid: 1442394297;
time: 1442389450; pid: 11636; time ^ pid: 1442394302;
time: 1442389450; pid: 11637; time ^ pid: 1442394303;
time: 1442389450; pid: 11638; time ^ pid: 1442394300;
time: 1442389450; pid: 11639; time ^ pid: 1442394301;
time: 1442389450; pid: 11640; time ^ pid: 1442394290;
time: 1442389450; pid: 11641; time ^ pid: 1442394291;
time: 1442389450; pid: 11642; time ^ pid: 1442394288;
time: 1442389450; pid: 11643; time ^ pid: 1442394289; --->8---

It is the same input but none of the outputs is the same value.

The XOR approach applied to gen_eth_addr:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
9988

Andreas

Quoted 20 lines

---

 tools/gen_eth_addr.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/gen_eth_addr.c +1 -1
index bf9d935..53b023a 100644
@@ -15,7 +15,7 @@ main(int argc, char *argv[])
 {
     unsigned long ethaddr_low, ethaddr_high;
 
-    srand(time(0) | getpid());
+    srand(time(0));
 
     /*
      * setting the 2nd LSB in the most significant byte of
author
#7 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

Hi, Andreas

Quoted 65 lines · Andreas Bie?mann

Hi Josh,

Quoted 4 lines · Josh Wu

As 'time(0) | getpid()' sometimes get same value. That depends on the value of getpid().
So that is not a expected behavior. We expect different value for the seed when when run it in many times.

I don't think your change made it better. Here is a snippet from a run of time(NULL) and getpid():

---8<---
time: 1442389450; pid: 11632; time | pid: 1442397690;
time: 1442389450; pid: 11633; time | pid: 1442397691;
time: 1442389450; pid: 11634; time | pid: 1442397690;
time: 1442389450; pid: 11635; time | pid: 1442397691;
time: 1442389450; pid: 11636; time | pid: 1442397694;
time: 1442389450; pid: 11637; time | pid: 1442397695;
time: 1442389450; pid: 11638; time | pid: 1442397694;
time: 1442389450; pid: 11639; time | pid: 1442397695;
time: 1442389450; pid: 11640; time | pid: 1442397690;
time: 1442389450; pid: 11641; time | pid: 1442397691;
time: 1442389450; pid: 11642; time | pid: 1442397690;
time: 1442389450; pid: 11643; time | pid: 1442397691; --->8---

While time(NULL) is stable, getpid() is incrementing by one. As you may expect the OR'ed value is oscillating and the values almost the same. So calling gen_eth_addr three times within the same second you will get two time the same MAC.

Quoted 1 line

So this patch remove the getpid(), just use the time(0) as the seed.

So let's see the effect of your change ...

The output of gen_eth_addr at the current ToT:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
254

With your change applied:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
10

Another approach would be to change the algorithm (OR the values) here. A short test showed that using XOR could be a solution:

---8<---
time: 1442389450; pid: 11632; time ^ pid: 1442394298;
time: 1442389450; pid: 11633; time ^ pid: 1442394299;
time: 1442389450; pid: 11634; time ^ pid: 1442394296;
time: 1442389450; pid: 11635; time ^ pid: 1442394297;
time: 1442389450; pid: 11636; time ^ pid: 1442394302;
time: 1442389450; pid: 11637; time ^ pid: 1442394303;
time: 1442389450; pid: 11638; time ^ pid: 1442394300;
time: 1442389450; pid: 11639; time ^ pid: 1442394301;
time: 1442389450; pid: 11640; time ^ pid: 1442394290;
time: 1442389450; pid: 11641; time ^ pid: 1442394291;
time: 1442389450; pid: 11642; time ^ pid: 1442394288;
time: 1442389450; pid: 11643; time ^ pid: 1442394289; --->8---

Quoted 8 lines

It is the same input but none of the outputs is the same value.

The XOR approach applied to gen_eth_addr:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
9988

How about using ADD ?
When I change to 'time + (pid << 8)', the result is better. And the pid change will has more impact in time range.

? tools time RUN=0; while [ $RUN -lt 10000 ]; do ./gen_eth_addr_add_shift 1; RUN=$(($RUN+1)); done | sort | uniq | wc -l 10000

The disadvantage is it might cost more time.

Best Regards,
Josh Wu

Quoted 22 lines

Andreas

Quoted 20 lines

---

  tools/gen_eth_addr.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
tools/gen_eth_addr.c +1 -1
index bf9d935..53b023a 100644
@@ -15,7 +15,7 @@ main(int argc, char *argv[])
  {
      unsigned long ethaddr_low, ethaddr_high;
 
-    srand(time(0) | getpid());
+    srand(time(0));
 
      /*
       * setting the 2nd LSB in the most significant byte of
#8 lore raw

PATCH 1/1 tools: gen_eth_addr: remove getpid() operation for the random seed

Hi Josh,

Quoted 73 lines · Josh Wu
Quoted 63 lines · Andreas Bie?mann
Quoted 4 lines · Josh Wu

As 'time(0) | getpid()' sometimes get same value. That depends on the value of getpid().
So that is not a expected behavior. We expect different value for the seed when when run it in many times.

I don't think your change made it better. Here is a snippet from a run of time(NULL) and getpid():

---8<---
time: 1442389450; pid: 11632; time | pid: 1442397690;
time: 1442389450; pid: 11633; time | pid: 1442397691;
time: 1442389450; pid: 11634; time | pid: 1442397690;
time: 1442389450; pid: 11635; time | pid: 1442397691;
time: 1442389450; pid: 11636; time | pid: 1442397694;
time: 1442389450; pid: 11637; time | pid: 1442397695;
time: 1442389450; pid: 11638; time | pid: 1442397694;
time: 1442389450; pid: 11639; time | pid: 1442397695;
time: 1442389450; pid: 11640; time | pid: 1442397690;
time: 1442389450; pid: 11641; time | pid: 1442397691;
time: 1442389450; pid: 11642; time | pid: 1442397690;
time: 1442389450; pid: 11643; time | pid: 1442397691; --->8---

While time(NULL) is stable, getpid() is incrementing by one. As you may expect the OR'ed value is oscillating and the values almost the same. So calling gen_eth_addr three times within the same second you will get two time the same MAC.

Quoted 1 line

So this patch remove the getpid(), just use the time(0) as the seed.

So let's see the effect of your change ...

The output of gen_eth_addr at the current ToT:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
254

With your change applied:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
10

Another approach would be to change the algorithm (OR the values) here. A short test showed that using XOR could be a solution:

---8<---
time: 1442389450; pid: 11632; time ^ pid: 1442394298;
time: 1442389450; pid: 11633; time ^ pid: 1442394299;
time: 1442389450; pid: 11634; time ^ pid: 1442394296;
time: 1442389450; pid: 11635; time ^ pid: 1442394297;
time: 1442389450; pid: 11636; time ^ pid: 1442394302;
time: 1442389450; pid: 11637; time ^ pid: 1442394303;
time: 1442389450; pid: 11638; time ^ pid: 1442394300;
time: 1442389450; pid: 11639; time ^ pid: 1442394301;
time: 1442389450; pid: 11640; time ^ pid: 1442394290;
time: 1442389450; pid: 11641; time ^ pid: 1442394291;
time: 1442389450; pid: 11642; time ^ pid: 1442394288;
time: 1442389450; pid: 11643; time ^ pid: 1442394289; --->8---

Quoted 8 lines

It is the same input but none of the outputs is the same value.

The XOR approach applied to gen_eth_addr:

% RUN=0; while [ $RUN -lt 10000 ]; do /tmp/build-uboot-test/tools/gen_eth_addr; RUN=$(($RUN+1)); done | sort | uniq | wc -l
9988

Quoted 7 lines

How about using ADD ?
When I change to 'time + (pid << 8)', the result is better. And the pid change will has more impact in time range.

? tools time RUN=0; while [ $RUN -lt 10000 ]; do ./gen_eth_addr_add_shift 1; RUN=$(($RUN+1)); done | sort | uniq | wc -l 10000

I'm fine with that too. All I'd like to say with my answer was that adding getpid() was for some reason and removing it is worse than having it OR'ed.

Quoted 1 line

The disadvantage is it might cost more time.

I don't think this count's here. This tool would be used to generate a temporary MAC (or more than one) for testing purposes. So getting a bunch of MAC's can last some time, but should provide unique ones.

Andreas