Skip to contentSkip to conversation

add buddyinfo /proc entry

PATCH In review Dave Hansen 6 participants 11 messages Aug 16, 2002 – Aug 19, 2002

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 add buddyinfo /proc entry

Not _another_ proc entry!

The following patch originally by Martin Bligh exports some information about the buddy allocator.

Each column of numbers represents the number of pages of that order which are available. In this case, there are 5 chunks of 2^2*PAGE_SIZE available in ZONE_DMA, and 101 chunks of 2^4*PAGE_SIZE availble in ZONE_NORMAL, etc... This information can give you a good idea about how fragmented memory is and give you a clue as to how big of an area you can safely allocate.

Node 0, zone      DMA      0      4      5      4      4      3 ...
Node 0, zone   Normal      1      0      0      1    101      8 ...
Node 0, zone  HighMem      2      0      0      1      1      0 ...
Signature · 2 lines
Dave Hansen
[email protected]

# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher. # This patch includes the following deltas:

#	           ChangeSet	1.549   -> 1.550  
#	     mm/page_alloc.c	1.88    -> 1.89   
#	 fs/proc/proc_misc.c	1.34    -> 1.35   

#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 02/08/15 [email protected] 1.550
# This small patch creates /proc/buddyinfo, which

# shows how many of each order page groups are available.
# 
# Port from -aa.
# --------------------------------------------
#
diff -Nru a/fs/proc/proc_misc.c b/fs/proc/proc_misc.c
--- a/fs/proc/proc_misc.c	Thu Aug 15 14:56:23 2002
+++ b/fs/proc/proc_misc.c	Thu Aug 15 14:56:23 2002
@@ -190,6 +190,20 @@
 #undef K
 }

+extern struct seq_operations fragmentation_op;
+static int fragmentation_open(struct inode *inode, struct file *file) +{
+ (void)inode;
+ return seq_open(file, &fragmentation_op);
+}
+
+static struct file_operations fragmentation_file_operations = {
+ open: fragmentation_open,
+ read: seq_read,
+ llseek: seq_lseek,
+ release: seq_release,
+};
+
static int version_read_proc(char *page, char **start, off_t off,
int count, int *eof, void *data)
{
@@ -606,6 +620,7 @@
create_seq_entry("partitions", 0, &proc_partitions_operations); create_seq_entry("interrupts", 0, &proc_interrupts_operations); create_seq_entry("slabinfo",S_IWUSR|S_IRUGO,&proc_slabinfo_operations);
+ create_seq_entry("buddyinfo",S_IRUGO, &fragmentation_file_operations);
#ifdef CONFIG_MODULES
create_seq_entry("modules", 0, &proc_modules_operations); create_seq_entry("ksyms", 0, &proc_ksyms_operations);

mm/page_alloc.c +71 -0
diff -Nru a/mm/page_alloc.c b/mm/page_alloc.c
@@ -924,3 +924,74 @@
 }
 
 __setup("memfrac=", setup_mem_frac);
+
+#ifdef CONFIG_PROC_FS
+
+#include <linux/seq_file.h>
+
+static void *frag_start(struct seq_file *m, loff_t *pos)
+{
+	pg_data_t *pgdat;
+	loff_t node = *pos;
+
+	(void)m;
+
+	for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
+		--node;
+
+	return pgdat;
+}
+
+static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
+{
+	pg_data_t *pgdat = (pg_data_t *)arg;
+
+	(void)m;
+	(*pos)++;
+	return pgdat->pgdat_next;
+}
+
+static void frag_stop(struct seq_file *m, void *arg)
+{
+	(void)m;
+	(void)arg;
+}
+
+/* 
+ * This walks the freelist for each zone. Whilst this is slow, I'd rather 
+ * be slow here than slow down the fast path by keeping stats - mjbligh
+ */
+static int frag_show(struct seq_file *m, void *arg)
+{
+	pg_data_t *pgdat = (pg_data_t *)arg;
+	zone_t *zone, *node_zones = pgdat->node_zones;
+	unsigned long flags;
+	int order;
+
+	for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
+		if (!zone->size)
+			continue;
+
+		spin_lock_irqsave(&zone->lock, flags);
+		seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
+		for (order = 0; order < MAX_ORDER; ++order) {
+			unsigned long nr_bufs = 0;
+			list_t *elem;
+			list_for_each(elem, &zone->free_area[order].free_list)
+				++nr_bufs;
+			seq_printf(m, "%6lu ", nr_bufs);
+		}
+		spin_unlock_irqrestore(&zone->lock, flags);
+		seq_putc(m, '\n');
+	}
+	return 0;
+}
+
+struct seq_operations fragmentation_op = {
+	start:	frag_start,
+	next:	frag_next,
+	stop:	frag_stop,
+	show:	frag_show,
+};
+
+#endif /* CONFIG_PROC_FS */
#1 lore raw
Quoted 1 line · Dave Hansen

Not _another_ proc entry!

Yes, not another one. Why not move these to driverfs, where they belong.

(ignore the driverfs name, it should be called kfs, or some such thing, as stuff more than driver info should go there, just like these entries.)

thanks,

greg k-h
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/

#2 lore raw
Quoted 6 lines · Greg KH
Quoted 1 line · Dave Hansen

Not _another_ proc entry!

Yes, not another one. Why not move these to driverfs, where they belong.

Could you show us how this particular situation might be laid out in a driverfs/kfs/gregfs tree?

It's great that you keep suggesting this, but we have another chicken-and-egg problem.

<SOAPBOX>
The problem with driverfs today is that it isn't worth it for _me_ to
use it to just get this one, single thing. If I used driverfs right now, the only thing that I would get out of it would be ... buddyinfo! How is it worth my while to use it on a shared machine where most people probably won't be mounting driverfs, or _want_ it mounted as the default? </SOAPBOX>

Quoted 3 lines

(ignore the driverfs name, it should be called kfs, or some such thing, as stuff more than driver info should go there, just like these entries.)

If even its most ardent supporters don't like its name...

Signature · 2 lines
Dave Hansen
[email protected]
#9 lore raw
In reply to Dave Hansen

mm/page_alloc.c line 958

Quoted 7 lines · Dave Hansen

..

mm/page_alloc.c +5 -0
+static void frag_stop(struct seq_file *m, void *arg)
+{
+       (void)m;
+       (void)arg;
+}

Don't tell me the compiler warns about this now?

#3 lore raw
In reply to Dave Hansen
Quoted 10 lines · Dave Hansen
Quoted 6 lines · Greg KH
Quoted 1 line · Dave Hansen

Not _another_ proc entry!

Yes, not another one. Why not move these to driverfs, where they belong.

Could you show us how this particular situation might be laid out in a driverfs/kfs/gregfs tree?

root/vm/buddyinfo ?

Ah, a gregfs, making up the components that describe me...that's going to be a pretty ugly looking fs...

Quoted 11 lines

It's great that you keep suggesting this, but we have another chicken-and-egg problem.

<SOAPBOX>
The problem with driverfs today is that it isn't worth it for _me_ to use it to just get this one, single thing. If I used driverfs right now, the only thing that I would get out of it would be ... buddyinfo! How is it worth my while to use it on a shared machine where most people probably won't be mounting driverfs, or _want_ it mounted as the default?
</SOAPBOX>

All it takes is one line added to /etc/fstab mounting driverfs at /sys. As the code is not a .config option, you are using it if you mount it or not :) The fact that no one else will look at that mount point, shouldn't matter to you.

And yes, for just one thing (hey, why don't you move _all_ the vm stats over to it), it is worth adding that one line. And you'll eventually have to do it anyway, as these things _will_ be moving there.

Hell, tell me which machine you are using, and I'll go add it.

Quoted 5 lines
Quoted 3 lines

(ignore the driverfs name, it should be called kfs, or some such thing, as stuff more than driver info should go there, just like these entries.)

If even its most ardent supporters don't like its name...

I don't have to like the name to like what it does, and is used for, right?

thanks,

greg k-h

Quoted 10 lines
Signature · 9 lines
-- 
Dave Hansen
[email protected]

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
#4 lore raw
Quoted 10 lines · Greg KH

All it takes is one line added to /etc/fstab mounting driverfs at /sys. As the code is not a .config option, you are using it if you mount it or not :) The fact that no one else will look at that mount point, shouldn't matter to you.

And yes, for just one thing (hey, why don't you move _all_ the vm stats over to it), it is worth adding that one line. And you'll eventually have to do it anyway, as these things _will_ be moving there.

Hell, tell me which machine you are using, and I'll go add it.

How long has it been in the tree (2.4 and 2.5)? I'll add it to my machine, but I am anticipating a 3 hour conversation as I explain to the other users why they got dropped to a root prompt because driverfs isn't supported when they boot.

Signature · 2 lines
Dave Hansen
[email protected]
#5 lore raw
Quoted 2 lines · Dave Hansen

How long has it been in the tree (2.4 and 2.5)?

2.5

Quoted 3 lines

I'll add it to my machine, but I am anticipating a 3 hour conversation as I explain to the other users why they got dropped to a root prompt because driverfs isn't supported when they boot.

??? My machines don't do that, they just complain and then continue on with the boot sequence.

greg k-h

#6 lore raw
In reply to Dave Hansen
Quoted 10 lines · Dave Hansen
Quoted 9 lines · Greg KH

All it takes is one line added to /etc/fstab mounting driverfs at /sys. As the code is not a .config option, you are using it if you mount it or not :) The fact that no one else will look at that mount point, shouldn't matter to you.

And yes, for just one thing (hey, why don't you move _all_ the vm stats over to it), it is worth adding that one line. And you'll eventually have to do it anyway, as these things _will_ be moving there.

Quoted 6 lines
Quoted 1 line

Hell, tell me which machine you are using, and I'll go add it.

How long has it been in the tree (2.4 and 2.5)? I'll add it to my machine, but I am anticipating a 3 hour conversation as I explain to the other users why they got dropped to a root prompt because driverfs isn't supported when they boot.

You're making things up and spreading FUD. Why would you want to do that?

Oh right, it's because most "kernel developers" would rather bitch about that which they do not understand and cut down other developers than suck it up and actually try to learn something from someone else.

Get over it.

-pat

#7 lore raw
Quoted 1 line · Patrick Mochel

You're making things up and spreading FUD. Why would you want to do that?

Because IBM has a grand, secret plan to replace driverfs with BigBlueFS. We have brainwashed Linus into taking it and we will rule the world!

No, I'm just being selfish and I'm a bit uninformed. I understand it better now because Greg explained why you want to do it. The thing that did it for me was Linus's decree that vm tunables _will_ move to driverfs. That changes the scenario from "Greg and Patrick are bitching" to "Linus wants this" and thus it will be the wave of the future. Maybe you guys need to make this a bit more clear.

Quoted 1 line

Oh right, it's because most "kernel developers" would rather bitch about that which they do not understand and cut down other developers than suck it up and actually try to learn something from someone else.

I'm trying to understand, that's why I asked so many questions. I am mistaken about the machine not booting completely if it doesn't understand a fs-type, Greg corrected me here.

Summary:
A driverfs version of this patch is on the way.

Signature · 2 lines
Dave Hansen
[email protected]
#10 lore raw
In reply to Andrew Morton

Dave Hansen wrote:
+static void frag_stop(struct seq_file *m, void *arg) +{
+ (void)m;
+ (void)arg;
+}

Quoted 1 line · Andrew Morton

Don't tell me the compiler warns about this now?

Woops -- that's actually a wli dropping. Some (?) of my code was borrowed for this. Someone pounded -Werror into my head too hard at school or something.

Cheers,
Bill

#8 lore raw
In reply to Greg KH
Quoted 6 lines · On Fri, 16 Aug 2002 07:39:25 -0700 Greg KH <[email protected]> wrote:
Quoted 5 lines · Dave Hansen
Quoted 4 lines · Greg KH
Quoted 1 line · Dave Hansen

Not _another_ proc entry!

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

Yes, not another one. Why not move these to driverfs, where they belong.

Could you show us how this particular situation might be laid out in a driverfs/kfs/gregfs tree?

root/vm/buddyinfo ?

root/memory please. vm is a little obscure.

Rusty.

Signature · 2 lines
   there are those who do and those who hang on and you don't see too
   many doers quoting their contemporaries.  -- Larry McVoy