2008-02-08 12:20:09 +00:00
|
|
|
/* Copyright (c) 2007 Coraid, Inc. See COPYING for GPL terms. */
|
2005-04-16 22:20:36 +00:00
|
|
|
/*
|
|
|
|
* aoedev.c
|
|
|
|
* AoE device utility functions; maintains device list.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/hdreg.h>
|
|
|
|
#include <linux/blkdev.h>
|
|
|
|
#include <linux/netdevice.h>
|
aoe: dynamically allocate a capped number of skbs when necessary
What this Patch Does
Even before this recent series of 12 patches to 2.6.22-rc4, the aoe
driver was reusing a small set of skbs that were allocated once and
were only used for outbound AoE commands.
The network layer cannot be allowed to put_page on the data that is
still associated with a bio we haven't returned to the block layer,
so the aoe driver (even before the patch under discussion) is still
the owner of skbs that have been handed to the network layer for
transmission. We need to keep track of these skbs so that we can
free them, but by tracking them, we can also easily re-use them.
The new patch was a response to the behavior of certain network
drivers. We cannot reuse an skb that the network driver still has
in its transmit ring. Network drivers can defer transmit ring
cleanup and then use the state in the skb to determine how many data
segments to clean up in its transmit ring. The tg3 driver is one
driver that behaves in this way.
When the network driver defers cleanup of its transmit ring, the aoe
driver can find itself in a situation where it would like to send an
AoE command, and the AoE target is ready for more work, but the
network driver still has all of the pre-allocated skbs. In that
case, the new patch just calls alloc_skb, as you'd expect.
We don't want to get carried away, though. We try not to do
excessive allocation in the write path, so we cap the number of skbs
we dynamically allocate.
Probably calling it a "dynamic pool" is misleading. We were already
trying to use a small fixed-size set of pre-allocated skbs before
this patch, and this patch just provides a little headroom (with a
ceiling, though) to accomodate network drivers that hang onto skbs,
by allocating when needed. The d->skbpool_hd list of allocated skbs
is necessary so that we can free them later.
We didn't notice the need for this headroom until AoE targets got
fast enough.
Alternatives
If the network layer never did a put_page on the pages in the bio's
we get from the block layer, then it would be possible for us to
hand skbs to the network layer and forget about them, allowing the
network layer to free skbs itself (and thereby calling our own
skb->destructor callback function if we needed that). In that case
we could get rid of the pre-allocated skbs and also the
d->skbpool_hd, instead just calling alloc_skb every time we wanted
to transmit a packet. The slab allocator would effectively maintain
the list of skbs.
Besides a loss of CPU cache locality, the main concern with that
approach the danger that it would increase the likelihood of
deadlock when VM is trying to free pages by writing dirty data from
the page cache through the aoe driver out to persistent storage on
an AoE device. Right now we have a situation where we have
pre-allocation that corresponds to how much we use, which seems
ideal.
Of course, there's still the separate issue of receiving the packets
that tell us that a write has successfully completed on the AoE
target. When memory is low and VM is using AoE to flush dirty data
to free up pages, it would be perfect if there were a way for us to
register a fast callback that could recognize write command
completion responses. But I don't think the current problems with
the receive side of the situation are a justification for
exacerbating the problem on the transmit side.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:05 +00:00
|
|
|
#include <linux/delay.h>
|
2005-04-16 22:20:36 +00:00
|
|
|
#include "aoe.h"
|
|
|
|
|
2008-02-08 12:20:03 +00:00
|
|
|
static void dummy_timer(ulong);
|
|
|
|
static void aoedev_freedev(struct aoedev *);
|
aoe: dynamically allocate a capped number of skbs when necessary
What this Patch Does
Even before this recent series of 12 patches to 2.6.22-rc4, the aoe
driver was reusing a small set of skbs that were allocated once and
were only used for outbound AoE commands.
The network layer cannot be allowed to put_page on the data that is
still associated with a bio we haven't returned to the block layer,
so the aoe driver (even before the patch under discussion) is still
the owner of skbs that have been handed to the network layer for
transmission. We need to keep track of these skbs so that we can
free them, but by tracking them, we can also easily re-use them.
The new patch was a response to the behavior of certain network
drivers. We cannot reuse an skb that the network driver still has
in its transmit ring. Network drivers can defer transmit ring
cleanup and then use the state in the skb to determine how many data
segments to clean up in its transmit ring. The tg3 driver is one
driver that behaves in this way.
When the network driver defers cleanup of its transmit ring, the aoe
driver can find itself in a situation where it would like to send an
AoE command, and the AoE target is ready for more work, but the
network driver still has all of the pre-allocated skbs. In that
case, the new patch just calls alloc_skb, as you'd expect.
We don't want to get carried away, though. We try not to do
excessive allocation in the write path, so we cap the number of skbs
we dynamically allocate.
Probably calling it a "dynamic pool" is misleading. We were already
trying to use a small fixed-size set of pre-allocated skbs before
this patch, and this patch just provides a little headroom (with a
ceiling, though) to accomodate network drivers that hang onto skbs,
by allocating when needed. The d->skbpool_hd list of allocated skbs
is necessary so that we can free them later.
We didn't notice the need for this headroom until AoE targets got
fast enough.
Alternatives
If the network layer never did a put_page on the pages in the bio's
we get from the block layer, then it would be possible for us to
hand skbs to the network layer and forget about them, allowing the
network layer to free skbs itself (and thereby calling our own
skb->destructor callback function if we needed that). In that case
we could get rid of the pre-allocated skbs and also the
d->skbpool_hd, instead just calling alloc_skb every time we wanted
to transmit a packet. The slab allocator would effectively maintain
the list of skbs.
Besides a loss of CPU cache locality, the main concern with that
approach the danger that it would increase the likelihood of
deadlock when VM is trying to free pages by writing dirty data from
the page cache through the aoe driver out to persistent storage on
an AoE device. Right now we have a situation where we have
pre-allocation that corresponds to how much we use, which seems
ideal.
Of course, there's still the separate issue of receiving the packets
that tell us that a write has successfully completed on the AoE
target. When memory is low and VM is using AoE to flush dirty data
to free up pages, it would be perfect if there were a way for us to
register a fast callback that could recognize write command
completion responses. But I don't think the current problems with
the receive side of the situation are a justification for
exacerbating the problem on the transmit side.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:05 +00:00
|
|
|
static void freetgt(struct aoedev *d, struct aoetgt *t);
|
|
|
|
static void skbpoolfree(struct aoedev *d);
|
2008-02-08 12:20:03 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static struct aoedev *devlist;
|
2008-02-08 12:20:10 +00:00
|
|
|
static DEFINE_SPINLOCK(devlist_lock);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
struct aoedev *
|
2005-04-19 05:00:18 +00:00
|
|
|
aoedev_by_aoeaddr(int maj, int min)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct aoedev *d;
|
|
|
|
ulong flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&devlist_lock, flags);
|
|
|
|
|
|
|
|
for (d=devlist; d; d=d->next)
|
2005-04-19 05:00:18 +00:00
|
|
|
if (d->aoemajor == maj && d->aoeminor == min)
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
|
|
|
|
spin_unlock_irqrestore(&devlist_lock, flags);
|
|
|
|
return d;
|
|
|
|
}
|
|
|
|
|
2006-01-19 18:46:19 +00:00
|
|
|
static void
|
|
|
|
dummy_timer(ulong vp)
|
|
|
|
{
|
|
|
|
struct aoedev *d;
|
|
|
|
|
|
|
|
d = (struct aoedev *)vp;
|
|
|
|
if (d->flags & DEVFL_TKILL)
|
|
|
|
return;
|
|
|
|
d->timer.expires = jiffies + HZ;
|
|
|
|
add_timer(&d->timer);
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
void
|
|
|
|
aoedev_downdev(struct aoedev *d)
|
|
|
|
{
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
struct aoetgt **t, **te;
|
2005-04-16 22:20:36 +00:00
|
|
|
struct frame *f, *e;
|
|
|
|
struct buf *buf;
|
|
|
|
struct bio *bio;
|
|
|
|
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
t = d->targets;
|
|
|
|
te = t + NTARGETS;
|
|
|
|
for (; t < te && *t; t++) {
|
|
|
|
f = (*t)->frames;
|
|
|
|
e = f + (*t)->nframes;
|
|
|
|
for (; f < e; f->tag = FREETAG, f->buf = NULL, f++) {
|
|
|
|
if (f->tag == FREETAG || f->buf == NULL)
|
|
|
|
continue;
|
|
|
|
buf = f->buf;
|
|
|
|
bio = buf->bio;
|
|
|
|
if (--buf->nframesout == 0
|
|
|
|
&& buf != d->inprocess) {
|
|
|
|
mempool_free(buf, d->bufpool);
|
|
|
|
bio_endio(bio, -EIO);
|
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
(*t)->maxout = (*t)->nframes;
|
|
|
|
(*t)->nout = 0;
|
|
|
|
}
|
|
|
|
buf = d->inprocess;
|
|
|
|
if (buf) {
|
|
|
|
bio = buf->bio;
|
|
|
|
mempool_free(buf, d->bufpool);
|
|
|
|
bio_endio(bio, -EIO);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
d->inprocess = NULL;
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
d->htgt = NULL;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
while (!list_empty(&d->bufq)) {
|
|
|
|
buf = container_of(d->bufq.next, struct buf, bufs);
|
|
|
|
list_del(d->bufq.next);
|
|
|
|
bio = buf->bio;
|
|
|
|
mempool_free(buf, d->bufpool);
|
2007-09-27 10:47:43 +00:00
|
|
|
bio_endio(bio, -EIO);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
if (d->gd)
|
|
|
|
d->gd->capacity = 0;
|
|
|
|
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
d->flags &= ~DEVFL_UP;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
2008-02-08 12:20:03 +00:00
|
|
|
static void
|
|
|
|
aoedev_freedev(struct aoedev *d)
|
|
|
|
{
|
|
|
|
struct aoetgt **t, **e;
|
|
|
|
|
|
|
|
if (d->gd) {
|
|
|
|
aoedisk_rm_sysfs(d);
|
|
|
|
del_gendisk(d->gd);
|
|
|
|
put_disk(d->gd);
|
|
|
|
}
|
|
|
|
t = d->targets;
|
|
|
|
e = t + NTARGETS;
|
|
|
|
for (; t < e && *t; t++)
|
aoe: dynamically allocate a capped number of skbs when necessary
What this Patch Does
Even before this recent series of 12 patches to 2.6.22-rc4, the aoe
driver was reusing a small set of skbs that were allocated once and
were only used for outbound AoE commands.
The network layer cannot be allowed to put_page on the data that is
still associated with a bio we haven't returned to the block layer,
so the aoe driver (even before the patch under discussion) is still
the owner of skbs that have been handed to the network layer for
transmission. We need to keep track of these skbs so that we can
free them, but by tracking them, we can also easily re-use them.
The new patch was a response to the behavior of certain network
drivers. We cannot reuse an skb that the network driver still has
in its transmit ring. Network drivers can defer transmit ring
cleanup and then use the state in the skb to determine how many data
segments to clean up in its transmit ring. The tg3 driver is one
driver that behaves in this way.
When the network driver defers cleanup of its transmit ring, the aoe
driver can find itself in a situation where it would like to send an
AoE command, and the AoE target is ready for more work, but the
network driver still has all of the pre-allocated skbs. In that
case, the new patch just calls alloc_skb, as you'd expect.
We don't want to get carried away, though. We try not to do
excessive allocation in the write path, so we cap the number of skbs
we dynamically allocate.
Probably calling it a "dynamic pool" is misleading. We were already
trying to use a small fixed-size set of pre-allocated skbs before
this patch, and this patch just provides a little headroom (with a
ceiling, though) to accomodate network drivers that hang onto skbs,
by allocating when needed. The d->skbpool_hd list of allocated skbs
is necessary so that we can free them later.
We didn't notice the need for this headroom until AoE targets got
fast enough.
Alternatives
If the network layer never did a put_page on the pages in the bio's
we get from the block layer, then it would be possible for us to
hand skbs to the network layer and forget about them, allowing the
network layer to free skbs itself (and thereby calling our own
skb->destructor callback function if we needed that). In that case
we could get rid of the pre-allocated skbs and also the
d->skbpool_hd, instead just calling alloc_skb every time we wanted
to transmit a packet. The slab allocator would effectively maintain
the list of skbs.
Besides a loss of CPU cache locality, the main concern with that
approach the danger that it would increase the likelihood of
deadlock when VM is trying to free pages by writing dirty data from
the page cache through the aoe driver out to persistent storage on
an AoE device. Right now we have a situation where we have
pre-allocation that corresponds to how much we use, which seems
ideal.
Of course, there's still the separate issue of receiving the packets
that tell us that a write has successfully completed on the AoE
target. When memory is low and VM is using AoE to flush dirty data
to free up pages, it would be perfect if there were a way for us to
register a fast callback that could recognize write command
completion responses. But I don't think the current problems with
the receive side of the situation are a justification for
exacerbating the problem on the transmit side.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:05 +00:00
|
|
|
freetgt(d, *t);
|
2008-02-08 12:20:03 +00:00
|
|
|
if (d->bufpool)
|
|
|
|
mempool_destroy(d->bufpool);
|
aoe: dynamically allocate a capped number of skbs when necessary
What this Patch Does
Even before this recent series of 12 patches to 2.6.22-rc4, the aoe
driver was reusing a small set of skbs that were allocated once and
were only used for outbound AoE commands.
The network layer cannot be allowed to put_page on the data that is
still associated with a bio we haven't returned to the block layer,
so the aoe driver (even before the patch under discussion) is still
the owner of skbs that have been handed to the network layer for
transmission. We need to keep track of these skbs so that we can
free them, but by tracking them, we can also easily re-use them.
The new patch was a response to the behavior of certain network
drivers. We cannot reuse an skb that the network driver still has
in its transmit ring. Network drivers can defer transmit ring
cleanup and then use the state in the skb to determine how many data
segments to clean up in its transmit ring. The tg3 driver is one
driver that behaves in this way.
When the network driver defers cleanup of its transmit ring, the aoe
driver can find itself in a situation where it would like to send an
AoE command, and the AoE target is ready for more work, but the
network driver still has all of the pre-allocated skbs. In that
case, the new patch just calls alloc_skb, as you'd expect.
We don't want to get carried away, though. We try not to do
excessive allocation in the write path, so we cap the number of skbs
we dynamically allocate.
Probably calling it a "dynamic pool" is misleading. We were already
trying to use a small fixed-size set of pre-allocated skbs before
this patch, and this patch just provides a little headroom (with a
ceiling, though) to accomodate network drivers that hang onto skbs,
by allocating when needed. The d->skbpool_hd list of allocated skbs
is necessary so that we can free them later.
We didn't notice the need for this headroom until AoE targets got
fast enough.
Alternatives
If the network layer never did a put_page on the pages in the bio's
we get from the block layer, then it would be possible for us to
hand skbs to the network layer and forget about them, allowing the
network layer to free skbs itself (and thereby calling our own
skb->destructor callback function if we needed that). In that case
we could get rid of the pre-allocated skbs and also the
d->skbpool_hd, instead just calling alloc_skb every time we wanted
to transmit a packet. The slab allocator would effectively maintain
the list of skbs.
Besides a loss of CPU cache locality, the main concern with that
approach the danger that it would increase the likelihood of
deadlock when VM is trying to free pages by writing dirty data from
the page cache through the aoe driver out to persistent storage on
an AoE device. Right now we have a situation where we have
pre-allocation that corresponds to how much we use, which seems
ideal.
Of course, there's still the separate issue of receiving the packets
that tell us that a write has successfully completed on the AoE
target. When memory is low and VM is using AoE to flush dirty data
to free up pages, it would be perfect if there were a way for us to
register a fast callback that could recognize write command
completion responses. But I don't think the current problems with
the receive side of the situation are a justification for
exacerbating the problem on the transmit side.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:05 +00:00
|
|
|
skbpoolfree(d);
|
2008-02-08 12:20:03 +00:00
|
|
|
kfree(d);
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
aoedev_flush(const char __user *str, size_t cnt)
|
|
|
|
{
|
|
|
|
ulong flags;
|
|
|
|
struct aoedev *d, **dd;
|
|
|
|
struct aoedev *rmd = NULL;
|
|
|
|
char buf[16];
|
|
|
|
int all = 0;
|
|
|
|
|
|
|
|
if (cnt >= 3) {
|
|
|
|
if (cnt > sizeof buf)
|
|
|
|
cnt = sizeof buf;
|
|
|
|
if (copy_from_user(buf, str, cnt))
|
|
|
|
return -EFAULT;
|
|
|
|
all = !strncmp(buf, "all", 3);
|
|
|
|
}
|
|
|
|
|
|
|
|
flush_scheduled_work();
|
|
|
|
spin_lock_irqsave(&devlist_lock, flags);
|
|
|
|
dd = &devlist;
|
|
|
|
while ((d = *dd)) {
|
|
|
|
spin_lock(&d->lock);
|
|
|
|
if ((!all && (d->flags & DEVFL_UP))
|
|
|
|
|| (d->flags & (DEVFL_GDALLOC|DEVFL_NEWSIZE))
|
|
|
|
|| d->nopen) {
|
|
|
|
spin_unlock(&d->lock);
|
|
|
|
dd = &d->next;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
*dd = d->next;
|
|
|
|
aoedev_downdev(d);
|
|
|
|
d->flags |= DEVFL_TKILL;
|
|
|
|
spin_unlock(&d->lock);
|
|
|
|
d->next = rmd;
|
|
|
|
rmd = d;
|
|
|
|
}
|
|
|
|
spin_unlock_irqrestore(&devlist_lock, flags);
|
|
|
|
while ((d = rmd)) {
|
|
|
|
rmd = d->next;
|
|
|
|
del_timer_sync(&d->timer);
|
|
|
|
aoedev_freedev(d); /* must be able to sleep */
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
aoe: dynamically allocate a capped number of skbs when necessary
What this Patch Does
Even before this recent series of 12 patches to 2.6.22-rc4, the aoe
driver was reusing a small set of skbs that were allocated once and
were only used for outbound AoE commands.
The network layer cannot be allowed to put_page on the data that is
still associated with a bio we haven't returned to the block layer,
so the aoe driver (even before the patch under discussion) is still
the owner of skbs that have been handed to the network layer for
transmission. We need to keep track of these skbs so that we can
free them, but by tracking them, we can also easily re-use them.
The new patch was a response to the behavior of certain network
drivers. We cannot reuse an skb that the network driver still has
in its transmit ring. Network drivers can defer transmit ring
cleanup and then use the state in the skb to determine how many data
segments to clean up in its transmit ring. The tg3 driver is one
driver that behaves in this way.
When the network driver defers cleanup of its transmit ring, the aoe
driver can find itself in a situation where it would like to send an
AoE command, and the AoE target is ready for more work, but the
network driver still has all of the pre-allocated skbs. In that
case, the new patch just calls alloc_skb, as you'd expect.
We don't want to get carried away, though. We try not to do
excessive allocation in the write path, so we cap the number of skbs
we dynamically allocate.
Probably calling it a "dynamic pool" is misleading. We were already
trying to use a small fixed-size set of pre-allocated skbs before
this patch, and this patch just provides a little headroom (with a
ceiling, though) to accomodate network drivers that hang onto skbs,
by allocating when needed. The d->skbpool_hd list of allocated skbs
is necessary so that we can free them later.
We didn't notice the need for this headroom until AoE targets got
fast enough.
Alternatives
If the network layer never did a put_page on the pages in the bio's
we get from the block layer, then it would be possible for us to
hand skbs to the network layer and forget about them, allowing the
network layer to free skbs itself (and thereby calling our own
skb->destructor callback function if we needed that). In that case
we could get rid of the pre-allocated skbs and also the
d->skbpool_hd, instead just calling alloc_skb every time we wanted
to transmit a packet. The slab allocator would effectively maintain
the list of skbs.
Besides a loss of CPU cache locality, the main concern with that
approach the danger that it would increase the likelihood of
deadlock when VM is trying to free pages by writing dirty data from
the page cache through the aoe driver out to persistent storage on
an AoE device. Right now we have a situation where we have
pre-allocation that corresponds to how much we use, which seems
ideal.
Of course, there's still the separate issue of receiving the packets
that tell us that a write has successfully completed on the AoE
target. When memory is low and VM is using AoE to flush dirty data
to free up pages, it would be perfect if there were a way for us to
register a fast callback that could recognize write command
completion responses. But I don't think the current problems with
the receive side of the situation are a justification for
exacerbating the problem on the transmit side.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:05 +00:00
|
|
|
/* I'm not really sure that this is a realistic problem, but if the
|
|
|
|
network driver goes gonzo let's just leak memory after complaining. */
|
|
|
|
static void
|
|
|
|
skbfree(struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
enum { Sms = 100, Tms = 3*1000};
|
|
|
|
int i = Tms / Sms;
|
|
|
|
|
|
|
|
if (skb == NULL)
|
|
|
|
return;
|
|
|
|
while (atomic_read(&skb_shinfo(skb)->dataref) != 1 && i-- > 0)
|
|
|
|
msleep(Sms);
|
|
|
|
if (i <= 0) {
|
|
|
|
printk(KERN_ERR
|
|
|
|
"aoe: %s holds ref: %s\n",
|
|
|
|
skb->dev ? skb->dev->name : "netif",
|
|
|
|
"cannot free skb -- memory leaked.");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
skb_shinfo(skb)->nr_frags = skb->data_len = 0;
|
|
|
|
skb_trim(skb, 0);
|
|
|
|
dev_kfree_skb(skb);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
skbpoolfree(struct aoedev *d)
|
|
|
|
{
|
|
|
|
struct sk_buff *skb;
|
|
|
|
|
|
|
|
while ((skb = d->skbpool_hd)) {
|
|
|
|
d->skbpool_hd = skb->next;
|
|
|
|
skb->next = NULL;
|
|
|
|
skbfree(skb);
|
|
|
|
}
|
|
|
|
d->skbpool_tl = NULL;
|
|
|
|
}
|
|
|
|
|
2006-01-19 18:46:19 +00:00
|
|
|
/* find it or malloc it */
|
2005-04-16 22:20:36 +00:00
|
|
|
struct aoedev *
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
aoedev_by_sysminor_m(ulong sysminor)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
struct aoedev *d;
|
|
|
|
ulong flags;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&devlist_lock, flags);
|
|
|
|
|
|
|
|
for (d=devlist; d; d=d->next)
|
2005-04-29 14:24:22 +00:00
|
|
|
if (d->sysminor == sysminor)
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
if (d)
|
|
|
|
goto out;
|
|
|
|
d = kcalloc(1, sizeof *d, GFP_ATOMIC);
|
|
|
|
if (!d)
|
|
|
|
goto out;
|
|
|
|
INIT_WORK(&d->work, aoecmd_sleepwork);
|
|
|
|
spin_lock_init(&d->lock);
|
|
|
|
init_timer(&d->timer);
|
|
|
|
d->timer.data = (ulong) d;
|
|
|
|
d->timer.function = dummy_timer;
|
|
|
|
d->timer.expires = jiffies + HZ;
|
|
|
|
add_timer(&d->timer);
|
|
|
|
d->bufpool = NULL; /* defer to aoeblk_gdalloc */
|
|
|
|
d->tgt = d->targets;
|
|
|
|
INIT_LIST_HEAD(&d->bufq);
|
|
|
|
d->sysminor = sysminor;
|
|
|
|
d->aoemajor = AOEMAJOR(sysminor);
|
|
|
|
d->aoeminor = AOEMINOR(sysminor);
|
|
|
|
d->mintimer = MINTIMER;
|
|
|
|
d->next = devlist;
|
|
|
|
devlist = d;
|
|
|
|
out:
|
2006-01-19 18:46:19 +00:00
|
|
|
spin_unlock_irqrestore(&devlist_lock, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
return d;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
aoe: dynamically allocate a capped number of skbs when necessary
What this Patch Does
Even before this recent series of 12 patches to 2.6.22-rc4, the aoe
driver was reusing a small set of skbs that were allocated once and
were only used for outbound AoE commands.
The network layer cannot be allowed to put_page on the data that is
still associated with a bio we haven't returned to the block layer,
so the aoe driver (even before the patch under discussion) is still
the owner of skbs that have been handed to the network layer for
transmission. We need to keep track of these skbs so that we can
free them, but by tracking them, we can also easily re-use them.
The new patch was a response to the behavior of certain network
drivers. We cannot reuse an skb that the network driver still has
in its transmit ring. Network drivers can defer transmit ring
cleanup and then use the state in the skb to determine how many data
segments to clean up in its transmit ring. The tg3 driver is one
driver that behaves in this way.
When the network driver defers cleanup of its transmit ring, the aoe
driver can find itself in a situation where it would like to send an
AoE command, and the AoE target is ready for more work, but the
network driver still has all of the pre-allocated skbs. In that
case, the new patch just calls alloc_skb, as you'd expect.
We don't want to get carried away, though. We try not to do
excessive allocation in the write path, so we cap the number of skbs
we dynamically allocate.
Probably calling it a "dynamic pool" is misleading. We were already
trying to use a small fixed-size set of pre-allocated skbs before
this patch, and this patch just provides a little headroom (with a
ceiling, though) to accomodate network drivers that hang onto skbs,
by allocating when needed. The d->skbpool_hd list of allocated skbs
is necessary so that we can free them later.
We didn't notice the need for this headroom until AoE targets got
fast enough.
Alternatives
If the network layer never did a put_page on the pages in the bio's
we get from the block layer, then it would be possible for us to
hand skbs to the network layer and forget about them, allowing the
network layer to free skbs itself (and thereby calling our own
skb->destructor callback function if we needed that). In that case
we could get rid of the pre-allocated skbs and also the
d->skbpool_hd, instead just calling alloc_skb every time we wanted
to transmit a packet. The slab allocator would effectively maintain
the list of skbs.
Besides a loss of CPU cache locality, the main concern with that
approach the danger that it would increase the likelihood of
deadlock when VM is trying to free pages by writing dirty data from
the page cache through the aoe driver out to persistent storage on
an AoE device. Right now we have a situation where we have
pre-allocation that corresponds to how much we use, which seems
ideal.
Of course, there's still the separate issue of receiving the packets
that tell us that a write has successfully completed on the AoE
target. When memory is low and VM is using AoE to flush dirty data
to free up pages, it would be perfect if there were a way for us to
register a fast callback that could recognize write command
completion responses. But I don't think the current problems with
the receive side of the situation are a justification for
exacerbating the problem on the transmit side.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:05 +00:00
|
|
|
freetgt(struct aoedev *d, struct aoetgt *t)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2006-09-20 18:36:49 +00:00
|
|
|
struct frame *f, *e;
|
|
|
|
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
f = t->frames;
|
|
|
|
e = f + t->nframes;
|
aoe: dynamically allocate a capped number of skbs when necessary
What this Patch Does
Even before this recent series of 12 patches to 2.6.22-rc4, the aoe
driver was reusing a small set of skbs that were allocated once and
were only used for outbound AoE commands.
The network layer cannot be allowed to put_page on the data that is
still associated with a bio we haven't returned to the block layer,
so the aoe driver (even before the patch under discussion) is still
the owner of skbs that have been handed to the network layer for
transmission. We need to keep track of these skbs so that we can
free them, but by tracking them, we can also easily re-use them.
The new patch was a response to the behavior of certain network
drivers. We cannot reuse an skb that the network driver still has
in its transmit ring. Network drivers can defer transmit ring
cleanup and then use the state in the skb to determine how many data
segments to clean up in its transmit ring. The tg3 driver is one
driver that behaves in this way.
When the network driver defers cleanup of its transmit ring, the aoe
driver can find itself in a situation where it would like to send an
AoE command, and the AoE target is ready for more work, but the
network driver still has all of the pre-allocated skbs. In that
case, the new patch just calls alloc_skb, as you'd expect.
We don't want to get carried away, though. We try not to do
excessive allocation in the write path, so we cap the number of skbs
we dynamically allocate.
Probably calling it a "dynamic pool" is misleading. We were already
trying to use a small fixed-size set of pre-allocated skbs before
this patch, and this patch just provides a little headroom (with a
ceiling, though) to accomodate network drivers that hang onto skbs,
by allocating when needed. The d->skbpool_hd list of allocated skbs
is necessary so that we can free them later.
We didn't notice the need for this headroom until AoE targets got
fast enough.
Alternatives
If the network layer never did a put_page on the pages in the bio's
we get from the block layer, then it would be possible for us to
hand skbs to the network layer and forget about them, allowing the
network layer to free skbs itself (and thereby calling our own
skb->destructor callback function if we needed that). In that case
we could get rid of the pre-allocated skbs and also the
d->skbpool_hd, instead just calling alloc_skb every time we wanted
to transmit a packet. The slab allocator would effectively maintain
the list of skbs.
Besides a loss of CPU cache locality, the main concern with that
approach the danger that it would increase the likelihood of
deadlock when VM is trying to free pages by writing dirty data from
the page cache through the aoe driver out to persistent storage on
an AoE device. Right now we have a situation where we have
pre-allocation that corresponds to how much we use, which seems
ideal.
Of course, there's still the separate issue of receiving the packets
that tell us that a write has successfully completed on the AoE
target. When memory is low and VM is using AoE to flush dirty data
to free up pages, it would be perfect if there were a way for us to
register a fast callback that could recognize write command
completion responses. But I don't think the current problems with
the receive side of the situation are a justification for
exacerbating the problem on the transmit side.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:05 +00:00
|
|
|
for (; f < e; f++)
|
|
|
|
skbfree(f->skb);
|
aoe: handle multiple network paths to AoE device
A remote AoE device is something can process ATA commands and is identified by
an AoE shelf number and an AoE slot number. Such a device might have more
than one network interface, and it might be reachable by more than one local
network interface. This patch tracks the available network paths available to
each AoE device, allowing them to be used more efficiently.
Andrew Morton asked about the call to msleep_interruptible in the revalidate
function. Yes, if a signal is pending, then msleep_interruptible will not
return 0. That means we will not loop but will call aoenet_xmit with a NULL
skb, which is a noop. If the system is too low on memory or the aoe driver is
too low on frames, then the user can hit control-C to interrupt the attempt to
do a revalidate. I have added a comment to the code summarizing that.
Andrew Morton asked whether the allocation performed inside addtgt could use a
more relaxed allocation like GFP_KERNEL, but addtgt is called when the aoedev
lock has been locked with spin_lock_irqsave. It would be nice to allocate the
memory under fewer restrictions, but targets are only added when the device is
being discovered, and if the target can't be added right now, we can try again
in a minute when then next AoE config query broadcast goes out.
Andrew Morton pointed out that the "too many targets" message could be printed
for failing GFP_ATOMIC allocations. The last patch in this series makes the
messages more specific.
Signed-off-by: Ed L. Cashin <ecashin@coraid.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:20:00 +00:00
|
|
|
kfree(t->frames);
|
|
|
|
kfree(t);
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
void
|
|
|
|
aoedev_exit(void)
|
|
|
|
{
|
|
|
|
struct aoedev *d;
|
|
|
|
ulong flags;
|
|
|
|
|
|
|
|
flush_scheduled_work();
|
|
|
|
|
|
|
|
while ((d = devlist)) {
|
|
|
|
devlist = d->next;
|
|
|
|
|
|
|
|
spin_lock_irqsave(&d->lock, flags);
|
|
|
|
aoedev_downdev(d);
|
2006-01-19 18:46:19 +00:00
|
|
|
d->flags |= DEVFL_TKILL;
|
2005-04-16 22:20:36 +00:00
|
|
|
spin_unlock_irqrestore(&d->lock, flags);
|
|
|
|
|
|
|
|
del_timer_sync(&d->timer);
|
|
|
|
aoedev_freedev(d);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
int __init
|
|
|
|
aoedev_init(void)
|
|
|
|
{
|
|
|
|
return 0;
|
|
|
|
}
|