Skip to contentSkip to conversation

x86: Add portable getcpu call

PATCH Heiko Carstens 2 participants 4 messages Sep 27, 2006 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 10 lines · Linux Kernel Mailing List

commit 3cfc348bf90ffaa777c188652aa297f04eb94de8
tree 8908d6a5a61e54ab422ec7f4800d6ac591695423
parent c08c820508233b424deab3302bc404bbecc6493a author Andi Kleen <[email protected]> 1159260748 +0200 committer Andi Kleen <[email protected]> 1159260748 +0200

[PATCH] x86: Add portable getcpu call

For NUMA optimization and some other algorithms it is useful to have a fast to get the current CPU and node numbers in user space.

Hmm.. just realized that there is a new system call.

Quoted 19 lines
(quoted diff) +19 -0
+asmlinkage long sys_getcpu(unsigned __user *cpup, unsigned __user *nodep,
+	   		   struct getcpu_cache __user *cache)
+{
+	int err = 0;
+	int cpu = raw_smp_processor_id();
+	if (cpup)
+		err |= put_user(cpu, cpup);
+	if (nodep)
+		err |= put_user(cpu_to_node(cpu), nodep);
+	if (cache) {
+		/*
+		 * The cache is not needed for this implementation,
+		 * but make sure user programs pass something
+		 * valid. vsyscall implementations can instead make
+		 * good use of the cache. Only use t0 and t1 because
+		 * these are available in both 32bit and 64bit ABI (no
+		 * need for a compat_getcpu). 32bit has enough
+		 * padding
+		 */

Review

Quoted 10 lines
(quoted diff) +10 -0
+		unsigned long t0, t1;
+		get_user(t0, &cache->t0);
+		get_user(t1, &cache->t1);
+		t0++;
+		t1++;
+		put_user(t0, &cache->t0);
+		put_user(t1, &cache->t1);
+	}
+	return err ? -EFAULT : 0;
+}

In include/linux/getcpu.h we have

/* Cache for getcpu() to speed it up. Results might be upto a jiffie
out of date, but will be faster.
User programs should not refer to the contents of this structure. It is only a cache for vgetcpu(). It might change in future kernels. The user program must store this information per thread (__thread) If you want 100% accurate information pass NULL instead. */

struct getcpu_cache {
	unsigned long t0;
	unsigned long t1;
	unsigned long res[4];
};

So this means that the contents of getcpu_cache will look completely different if a process runs in 32bit mode or 64bit mode. Even if you're saying "user programs should not..." this looks odd to me. Is this really on purpose and do you really think that no user space application will ever rely on the format of getcpu_cache?

#2 lore raw
Quoted 7 lines

So this means that the contents of getcpu_cache will look completely different if a process runs in 32bit mode or 64bit mode. Even if you're saying "user programs should not..." this looks odd to me. Is this really on purpose and do you really think that no user space application will ever rely on the format of getcpu_cache?

The vsyscalls do, but if anything else does it deserves breaking. In the user headers it will also be just a array blob.

I was considering to xor it with a random value to bring the point across more strongly, but then didn't do it. Do you think I should?

-Andi

#3 lore raw
Quoted 8 lines
Quoted 5 lines

So this means that the contents of getcpu_cache will look completely different if a process runs in 32bit mode or 64bit mode. Even if you're saying "user programs should not..." this looks odd to me. Is this really on purpose and do you really think that no user space application will ever rely on the format of getcpu_cache?

The vsyscalls do, but if anything else does it deserves breaking. In the user headers it will also be just a array blob.

Ah, ok. The blob thing is the part I missed then.

Quoted 2 lines

I was considering to xor it with a random value to bring the point across more strongly, but then didn't do it. Do you think I should?

No, I don't think that will help much.

#4 lore raw
Quoted 10 lines · Heiko Carstens
Quoted 8 lines
Quoted 5 lines

So this means that the contents of getcpu_cache will look completely different if a process runs in 32bit mode or 64bit mode. Even if you're saying "user programs should not..." this looks odd to me. Is this really on purpose and do you really think that no user space application will ever rely on the format of getcpu_cache?

The vsyscalls do, but if anything else does it deserves breaking. In the user headers it will also be just a array blob.

Ah, ok. The blob thing is the part I missed then.

Hmm, perhaps it's better to do that in the standard kernel headers.

-Andi