Skip to contentSkip to conversation

export cpu_online_map

PATCH In review [email protected] 4 participants 12 messages Oct 26, 2005 – Oct 28, 2005 Scheduler

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 export cpu_online_map

From: Andrew Morton <[email protected]>

With CONFIG_SMP=n:

*** Warning: "cpu_online_map" [drivers/firmware/dcdbas.ko] undefined!

due to set_cpus_allowed().

Questions:

- Why isn't set_cpus_allowed() just a no-op on UP? Or some trivial thing which tests for cpu #0?

- Why does cpu_online_map even exist on CONFIG_SMP=n?

---

 kernel/sched.c |    1 +
 1 files changed, 1 insertion(+)
kernel/sched.c renamed from kernel/sched.c~export-cpu_online_map +1 -0
diff -puN kernel/sched.c~export-cpu_online_map kernel/sched.c
@@ -3879,6 +3879,7 @@ EXPORT_SYMBOL(cpu_present_map);
 
 #ifndef CONFIG_SMP
 cpumask_t cpu_online_map = CPU_MASK_ALL;
+EXPORT_SYMBOL(cpu_online_map);
 cpumask_t cpu_possible_map = CPU_MASK_ALL;
 #endif
 

_

#1 lore raw

Andrew asked:

Quoted 2 lines

- Why isn't set_cpus_allowed() just a no-op on UP? Or some trivial thing which tests for cpu #0?

I don't know.

By scanning random copies of kernels left on my drive, I can see that it changed from a trivial "return 0" to the more interesting check using cpu_online_map in one of your "linus.patch" patches in the release 2.6.11-rc1-mm2.

But I don't know how to get to the history at this point to see what happened.

It looks like Linus started the git history at 2.6.12-rc2.

Could someone clue me in on how to find Linux history BG (before-git)?

Since bk no longer works for me, I have no idea how to access any history prior to about 2.6.12-rc2. Ugh.

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401
#2 lore raw
Quoted 15 lines · Paul Jackson

Andrew asked:

Quoted 2 lines

- Why isn't set_cpus_allowed() just a no-op on UP? Or some trivial thing which tests for cpu #0?

I don't know.

By scanning random copies of kernels left on my drive, I can see that it changed from a trivial "return 0" to the more interesting check using cpu_online_map in one of your "linus.patch" patches in the release 2.6.11-rc1-mm2.

But I don't know how to get to the history at this point to see what happened.

Quoted 6 lines

It looks like Linus started the git history at 2.6.12-rc2.

Could someone clue me in on how to find Linux history BG (before-git)?

Since bk no longer works for me, I have no idea how to access any history prior to about 2.6.12-rc2. Ugh.

There's still bkbits.net:

http://linux.bkbits.net:8080/linux-2.6/diffs/include/linux/[email protected]?nav=index.html|src/|src/include|src/include/linux|hist/include/linux/sched.h

From: Rusty Russell <[email protected]>

Return EINVAL for invalid sched_setaffinity on UP. I was a little surprised that sys_sched_setaffinity for CPU 1 didn't fail on my UP box. With CONFIG_SMP it would have.

---

 25-akpm/include/linux/sched.h |    3 +++
 1 files changed, 3 insertions(+)
include/linux/sched.h renamed from include/linux/sched.h~sys_sched_setaffinity-on-up-should-fail-for-non-zero +3 -0
diff -puN include/linux/sched.h~sys_sched_setaffinity-on-up-should-fail-for-non-zero include/linux/sched.h
@@ -13,6 +13,7 @@
 #include <linux/rbtree.h>
 #include <linux/thread_info.h>
 #include <linux/cpumask.h>
+#include <linux/errno.h>
 
 #include <asm/system.h>
 #include <asm/semaphore.h>
@@ -732,6 +733,8 @@ extern int set_cpus_allowed(task_t *p, c
 #else
 static inline int set_cpus_allowed(task_t *p, cpumask_t new_mask)
 {
+	if (!cpus_intersects(new_mask, cpu_online_map))
+		return -EINVAL;
 	return 0;
 }
 #endif

_

Seems silly to use cpu_online_map to test for `1'?

#3 lore raw
Quoted 6 lines · Andrew
Quoted 2 lines

Since bk no longer works for me, I have no idea how to access any history prior to about 2.6.12-rc2. Ugh.

There's still bkbits.net:

http://linux.bkbits.net:8080/linux-2.6/...

Ok - that helps. Thanks.

There is also an hg (mercurial) web based Linux history at:

http://www.kernel.org/hg/linux-2.6/

I still don't see something I can download into a local hg or git repository -- if a lurker knows how that is done, I'd be glad to know.

Quoted 1 line

Seems silly to use cpu_online_map to test for `1'?

Yeah - agreed. The include/linux/cpumask.h stuff hardcodes the UP (NR_CPUS==1) code to plain old 0's and 1's. This macro should do so as well.

I just sent a patch.

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401
#4 lore raw
Quoted 11 lines · Paul Jackson
Quoted 6 lines · Andrew
Quoted 2 lines

Since bk no longer works for me, I have no idea how to access any history prior to about 2.6.12-rc2. Ugh.

There's still bkbits.net:

http://linux.bkbits.net:8080/linux-2.6/...

Ok - that helps. Thanks.

Quoted 8 lines

There is also an hg (mercurial) web based Linux history at:

http://www.kernel.org/hg/linux-2.6/

I still don't see something I can download into a local hg or git repository -- if a lurker knows how that is done, I'd be glad to know.

I think there are ways of getting the entire kernel history in git too.

I usually just grep my patches directory, actually. <looks>. <egad>. 27,000 of them.

Quoted 9 lines
Quoted 2 lines

Seems silly to use cpu_online_map to test for `1'?

Yeah - agreed. The include/linux/cpumask.h stuff hardcodes the UP (NR_CPUS==1) code to plain old 0's and 1's. This macro should do so as well.

I just sent a patch.

Sweet, thanks. Perhaps we can remove cpu_online_map from UP builds soon - it's really wrong to have it there.

#5 lore raw
Quoted 2 lines · Andrew

Sweet, thanks. Perhaps we can remove cpu_online_map from UP builds soon - it's really wrong to have it there.

Eh ... my gut reaction is different. Even uni-processors have
online cpus - just not very many of them (and hot unplugging one of them is frowned on). Why make special cases when it serves no purpose?

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401
#6 lore raw
Quoted 8 lines · Paul Jackson
Quoted 2 lines · Andrew

Sweet, thanks. Perhaps we can remove cpu_online_map from UP builds soon - it's really wrong to have it there.

Eh ... my gut reaction is different. Even uni-processors have
online cpus - just not very many of them (and hot unplugging one of them is frowned on).

That's daft. A uniprocessor machine has one and only one CPU and it's
always online! An online_map is only needed for MP.

Now conceptually, yes, we should be able to query and perhaps set the onlineness of a CPU. But that doesn't mean that we should have storage which idiotically remembers something which was known at compile time.

Quoted 1 line

Why make special cases when it serves no purpose?

Ths presence of cpu_online_map in UP builds _is_ a special case. The kernel's overall approach to such things is to optimise them away at compile time for !SMP builds.

#7 lore raw
Quoted 1 line

That's daft.

<grin>

Ok - guess I'll let someone else handle the patches for that change then. Looks like I am not not singing in key on this one.

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.650.933.1373, 1.925.600.0401
#8 lore raw
In reply to Andrew Morton
Quoted 1 line · Andrew

I think there are ways of getting the entire kernel history in git too.

If I knew where a history was, perhaps I could put it into git (or the one I prefer - hg).

The only 3 histories I can find are:
1) web based bk
2) web based hg
3) locked forever inside my dead bk repositories

Quoted 2 lines

I usually just grep my patches directory, actually. <looks>. <egad>. 27,000 of them.

Well - I've got my email archives from the mm-commits email list, but that only goes back to July of 2005 .. a mere 4547 of them.

Signature · 3 lines
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <[email protected]> 1.925.600.0401
#11 lore raw
In reply to [email protected]
Quoted 3 lines · [email protected] wrote:

- Why isn't set_cpus_allowed() just a no-op on UP? Or some trivial thing which tests for cpu #0?

It's still needed to weed out bogus masks that have CPU 0 turned off.

You're right that it doesn't need to check cpu_online_map though. Here is a patch to make it check for CPU 0 instead.

Cheers,

Signature · 5 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
--
include/linux/sched.h +1 -1
@@ -895,7 +895,7 @@ extern int set_cpus_allowed(task_t *p, c
 #else
 static inline int set_cpus_allowed(task_t *p, cpumask_t new_mask)
 {
-	if (!cpus_intersects(new_mask, cpu_online_map))
+	if (!cpu_isset(0, new_mask))
 		return -EINVAL;
 	return 0;
 }