Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1740340 > unrolled thread

Re: percpu allocation failures

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: percpu allocation failures Dennis Zhou <dennisszhou@gmail.com> - 2017-09-27 04:00 +0200
    Re: percpu allocation failures Luis Henriques <lhenriques@suse.com> - 2017-09-27 11:10 +0200

#1740340 — Re: percpu allocation failures

Hi Luis,

This seems to be an issue with the reserved chunk being unable to
allocate memory when loading kernel modules. Unfortunately, I have not 
been successful in reproducing this with the reserved chunk allocation
path exposed or by inserting the nft_meta module.

Could you please send me the output when ran with the following patch
and the output of the percpu memory statistics file before and after
inserting the module (PERCPU_STATS)? The stats are in
/sys/kernel/debug/percpu_stats.

Thanks,
Dennis

---
 mm/percpu.c | 32 ++++++++++++++++++++++++++++++--
 1 file changed, 30 insertions(+), 2 deletions(-)

diff --git a/mm/percpu.c b/mm/percpu.c
index 59d44d6..031fd91 100644
--- a/mm/percpu.c
+++ b/mm/percpu.c
@@ -1335,6 +1335,7 @@ static void __percpu *pcpu_alloc(size_t size, size_t align, bool reserved,
 {
 	static int warn_limit = 10;
 	struct pcpu_chunk *chunk;
+	struct pcpu_block_md *block;
 	const char *err;
 	bool is_atomic = (gfp & GFP_KERNEL) != GFP_KERNEL;
 	int slot, off, cpu, ret;
@@ -1371,17 +1372,43 @@ static void __percpu *pcpu_alloc(size_t size, size_t align, bool reserved,
 	if (reserved && pcpu_reserved_chunk) {
 		chunk = pcpu_reserved_chunk;
 
+		printk(KERN_DEBUG "percpu: reserved chunk: %d, %d, %d, %d, %d, %d, %d",
+		       chunk->free_bytes, chunk->contig_bits,
+		       chunk->contig_bits_start, chunk->first_bit,
+		       chunk->start_offset, chunk->end_offset,
+		       chunk->nr_pages);
+
+		printk(KERN_DEBUG "percpu: rchunk md blocks");
+		for (block = chunk->md_blocks;
+		     block < chunk->md_blocks + pcpu_chunk_nr_blocks(chunk);
+		     block++) {
+			printk(KERN_DEBUG "   percpu: %d, %d, %d, %d, %d",
+			       block->contig_hint,
+			       block->contig_hint_start,
+			       block->left_free,
+			       block->right_free,
+			       block->first_free);
+		}
+
 		off = pcpu_find_block_fit(chunk, bits, bit_align, is_atomic);
+
+		printk(KERN_DEBUG "percpu: pcpu_find_block_fit: %d, %zu, %zu",
+		       off, bits, bit_align);
+
 		if (off < 0) {
-			err = "alloc from reserved chunk failed";
+			err = "alloc from reserved chunk failed to find fit";
 			goto fail_unlock;
 		}
 
 		off = pcpu_alloc_area(chunk, bits, bit_align, off);
+
+		printk(KERN_DEBUG "percpu: pcpu_alloc_area: %d, %zu, %zu",
+		       off, bits, bit_align);
+
 		if (off >= 0)
 			goto area_found;
 
-		err = "alloc from reserved chunk failed";
+		err = "alloc from reserved chunk failed to alloc area";
 		goto fail_unlock;
 	}
 
@@ -1547,6 +1574,7 @@ void __percpu *__alloc_reserved_percpu(size_t size, size_t align)
 {
 	return pcpu_alloc(size, align, true, GFP_KERNEL);
 }
+EXPORT_SYMBOL_GPL(__alloc_reserved_percpu);
 
 /**
  * pcpu_balance_workfn - manage the amount of free chunks and populated pages
-- 
1.8.3.1

[toc] | [next] | [standalone]

#1740524

[Multipart message — attachments visible in raw view] — view raw

Dennis Zhou <dennisszhou@gmail.com> writes:

> Hi Luis,
>
> This seems to be an issue with the reserved chunk being unable to
> allocate memory when loading kernel modules. Unfortunately, I have not 
> been successful in reproducing this with the reserved chunk allocation
> path exposed or by inserting the nft_meta module.
>
> Could you please send me the output when ran with the following patch
> and the output of the percpu memory statistics file before and after
> inserting the module (PERCPU_STATS)? The stats are in
> /sys/kernel/debug/percpu_stats.

Please find attached all the info you requested.  Hope it helps.

Also, here's the nft script I'm using to trigger the issue locally:

#!/usr/sbin/nft -f

flush ruleset

table inet filter {
	chain input {
		type filter hook input priority 0;
		iif lo accept
		ct state established,related accept
		tcp dport { 22 } ct state new accept
		ip6 nexthdr icmpv6 icmpv6 type { nd-neighbor-solicit,  nd-router-advert, nd-neighbor-advert } accept
		counter drop
	}
}

Cheers,
-- 
Luis

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web