* Re: [PATCH] IPv6: Cleanup of net/ipv6/reassambly.c
From: Ingo Oeser @ 2006-03-12 7:17 UTC (permalink / raw)
To: YOSHIFUJI Hideaki / 吉藤英明
Cc: David S. Miller, netdev, linux-kernel
In-Reply-To: <200603120049.49294.ioe-lkml@rameria.de>
Hi,
On Sunday, 12. March 2006 00:49, Ingo Oeser wrote:
> From: Ingo Oeser <ioe-lkml@rameria.de>
>
> Two minor cleanups:
>
> 1. Using kzalloc() in fraq_alloc_queue()
> saves the memset() in ipv6_frag_create().
>
> 2. Invert sense of if-statements to streamline code.
> Inverts the comment, too.
>
These are against net-2.6.17 of course.
I also compile tested this and my other kzalloc() changes.
Forgot to mention this yesterday...
Regards
Ingo Oeser
^ permalink raw reply
* Re: Linux v2.6.16-rc6
From: David S. Miller @ 2006-03-12 2:39 UTC (permalink / raw)
To: michal.k.k.piotrowski; +Cc: torvalds, linux-kernel, netdev
In-Reply-To: <6bffcb0e0603111751i1ed30794s@mail.gmail.com>
From: "Michal Piotrowski" <michal.k.k.piotrowski@gmail.com>
Date: Sun, 12 Mar 2006 02:51:40 +0100
> I have noticed this warnings
> TCP: Treason uncloaked! Peer 82.113.55.2:11759/50967 shrinks window
> 148470938:148470943. Repaired.
> TCP: Treason uncloaked! Peer 82.113.55.2:11759/50967 shrinks window
> 148470938:148470943. Repaired.
> TCP: Treason uncloaked! Peer 82.113.55.2:11759/59768 shrinks window
> 1124211698:1124211703. Repaired.
> TCP: Treason uncloaked! Peer 82.113.55.2:11759/59768 shrinks window
> 1124211698:1124211703. Repaired.
>
> It maybe problem with ktorrent.
It is a problem with the remote TCP implementation, it is
illegally advertising a smaller window that it previously
did.
^ permalink raw reply
* Re: Linux v2.6.16-rc6
From: Michal Piotrowski @ 2006-03-12 1:51 UTC (permalink / raw)
To: Linus Torvalds; +Cc: Linux Kernel Mailing List, netdev
In-Reply-To: <Pine.LNX.4.64.0603111551330.18022@g5.osdl.org>
Hi,
On 12/03/06, Linus Torvalds <torvalds@osdl.org> wrote:
>
> Ok, we're getting closer, although the 2.6.16 release certainly seems to
> drag out more than it should have.
>
I have noticed this warnings
TCP: Treason uncloaked! Peer 82.113.55.2:11759/50967 shrinks window
148470938:148470943. Repaired.
TCP: Treason uncloaked! Peer 82.113.55.2:11759/50967 shrinks window
148470938:148470943. Repaired.
TCP: Treason uncloaked! Peer 82.113.55.2:11759/59768 shrinks window
1124211698:1124211703. Repaired.
TCP: Treason uncloaked! Peer 82.113.55.2:11759/59768 shrinks window
1124211698:1124211703. Repaired.
It maybe problem with ktorrent.
Here is config http://www.stardust.webpages.pl/files/linux/2.6.16-rc6/config
Here is dmesg http://www.stardust.webpages.pl/files/linux/2.6.16-rc6/dmesg
Regards,
Michal
--
Michal K. K. Piotrowski
LTG - Linux Testers Group
(http://www.stardust.webpages.pl/ltg/wiki/)
^ permalink raw reply
* [PATCH] IPv6: Cleanup of net/ipv6/reassambly.c
From: Ingo Oeser @ 2006-03-11 23:49 UTC (permalink / raw)
To: YOSHIFUJI Hideaki / 吉藤英明
Cc: David S. Miller, netdev, linux-kernel
From: Ingo Oeser <ioe-lkml@rameria.de>
Two minor cleanups:
1. Using kzalloc() in fraq_alloc_queue()
saves the memset() in ipv6_frag_create().
2. Invert sense of if-statements to streamline code.
Inverts the comment, too.
Signed-off-by: Ingo Oeser <ioe-lkml@rameria.de>
---
Hi,
I first did the kzalloc cleanup, but then decided,
that I can do this change, too. If you consider it worse,
please tell me and I'll just do the kzalloc() one.
Regards
Ingo Oeser
diff --git a/net/ipv6/reassembly.c b/net/ipv6/reassembly.c
index 15e1456..6d8c9bb 100644
--- a/net/ipv6/reassembly.c
+++ b/net/ipv6/reassembly.c
@@ -203,7 +203,7 @@ static inline void frag_free_queue(struc
static inline struct frag_queue *frag_alloc_queue(void)
{
- struct frag_queue *fq = kmalloc(sizeof(struct frag_queue), GFP_ATOMIC);
+ struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
if(!fq)
return NULL;
@@ -288,6 +288,7 @@ static void ip6_evictor(void)
static void ip6_frag_expire(unsigned long data)
{
struct frag_queue *fq = (struct frag_queue *) data;
+ struct net_device *dev;
spin_lock(&fq->lock);
@@ -299,22 +300,22 @@ static void ip6_frag_expire(unsigned lon
IP6_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
IP6_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
- /* Send error only if the first segment arrived. */
- if (fq->last_in&FIRST_IN && fq->fragments) {
- struct net_device *dev = dev_get_by_index(fq->iif);
-
- /*
- But use as source device on which LAST ARRIVED
- segment was received. And do not use fq->dev
- pointer directly, device might already disappeared.
- */
- if (dev) {
- fq->fragments->dev = dev;
- icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0,
- dev);
- dev_put(dev);
- }
- }
+ /* Don't send error if the first segment did not arrive. */
+ if (!(fq->last_in&FIRST_IN) || !fq->fragments)
+ goto out;
+
+ dev = dev_get_by_index(fq->iif);
+ if (!dev)
+ goto out;
+
+ /*
+ But use as source device on which LAST ARRIVED
+ segment was received. And do not use fq->dev
+ pointer directly, device might already disappeared.
+ */
+ fq->fragments->dev = dev;
+ icmpv6_send(fq->fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
+ dev_put(dev);
out:
spin_unlock(&fq->lock);
fq_put(fq, NULL);
@@ -368,8 +369,6 @@ ip6_frag_create(unsigned int hash, u32 i
if ((fq = frag_alloc_queue()) == NULL)
goto oom;
- memset(fq, 0, sizeof(struct frag_queue));
-
fq->id = id;
ipv6_addr_copy(&fq->saddr, src);
ipv6_addr_copy(&fq->daddr, dst);
^ permalink raw reply related
* [PATCH] Nearly complete kzalloc cleanup for net/ipv6
From: Ingo Oeser @ 2006-03-11 20:36 UTC (permalink / raw)
To: YOSHIFUJI Hideaki / 吉藤英明
Cc: David Miller, linux-kernel, netdev
From: Ingo Oeser <ioe-lkml@rameria.de>
Stupidly use kzalloc() instead of kmalloc()/memset()
everywhere where this is possible in net/ipv6/*.c .
Signed-off-by: Ingo Oeser <ioe-lkml@rameria.de>
---
Hi,
I'm going to refactor some of the simple cases to reduce code
duplication which will depend on this patch.
net/ipv6/reassambly.c will get a separate patch,
since I refactored it a little bit.
The netfilter part is NOT included, because Harald should see these, too.
Patch is against net-2.6.17
Regards
Ingo Oeser
diff --git a/net/ipv6/ah6.c b/net/ipv6/ah6.c
index 8ff7147..b876a10 100644
--- a/net/ipv6/ah6.c
+++ b/net/ipv6/ah6.c
@@ -354,12 +354,10 @@ static int ah6_init_state(struct xfrm_st
if (x->encap)
goto error;
- ahp = kmalloc(sizeof(*ahp), GFP_KERNEL);
+ ahp = kzalloc(sizeof(*ahp), GFP_KERNEL);
if (ahp == NULL)
return -ENOMEM;
- memset(ahp, 0, sizeof(*ahp));
-
ahp->key = x->aalg->alg_key;
ahp->key_len = (x->aalg->alg_key_len+7)/8;
ahp->tfm = crypto_alloc_tfm(x->aalg->alg_name, 0);
diff --git a/net/ipv6/anycast.c b/net/ipv6/anycast.c
index 840a33d..39ec528 100644
--- a/net/ipv6/anycast.c
+++ b/net/ipv6/anycast.c
@@ -308,7 +308,7 @@ int ipv6_dev_ac_inc(struct net_device *d
* not found: create a new one.
*/
- aca = kmalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
+ aca = kzalloc(sizeof(struct ifacaddr6), GFP_ATOMIC);
if (aca == NULL) {
err = -ENOMEM;
@@ -322,8 +322,6 @@ int ipv6_dev_ac_inc(struct net_device *d
goto out;
}
- memset(aca, 0, sizeof(struct ifacaddr6));
-
ipv6_addr_copy(&aca->aca_addr, addr);
aca->aca_idev = idev;
aca->aca_rt = rt;
@@ -550,7 +548,7 @@ static int ac6_seq_open(struct inode *in
{
struct seq_file *seq;
int rc = -ENOMEM;
- struct ac6_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
+ struct ac6_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
if (!s)
goto out;
@@ -561,7 +559,6 @@ static int ac6_seq_open(struct inode *in
seq = file->private_data;
seq->private = s;
- memset(s, 0, sizeof(*s));
out:
return rc;
out_kfree:
diff --git a/net/ipv6/esp6.c b/net/ipv6/esp6.c
index aa7f100..3dcaac7 100644
--- a/net/ipv6/esp6.c
+++ b/net/ipv6/esp6.c
@@ -305,12 +305,10 @@ static int esp6_init_state(struct xfrm_s
if (x->encap)
goto error;
- esp = kmalloc(sizeof(*esp), GFP_KERNEL);
+ esp = kzalloc(sizeof(*esp), GFP_KERNEL);
if (esp == NULL)
return -ENOMEM;
- memset(esp, 0, sizeof(*esp));
-
if (x->aalg) {
struct xfrm_algo_desc *aalg_desc;
diff --git a/net/ipv6/ip6_flowlabel.c b/net/ipv6/ip6_flowlabel.c
index 69cbe8a..f9ca639 100644
--- a/net/ipv6/ip6_flowlabel.c
+++ b/net/ipv6/ip6_flowlabel.c
@@ -287,10 +287,9 @@ fl_create(struct in6_flowlabel_req *freq
int err;
err = -ENOMEM;
- fl = kmalloc(sizeof(*fl), GFP_KERNEL);
+ fl = kzalloc(sizeof(*fl), GFP_KERNEL);
if (fl == NULL)
goto done;
- memset(fl, 0, sizeof(*fl));
olen = optlen - CMSG_ALIGN(sizeof(*freq));
if (olen > 0) {
@@ -663,7 +662,7 @@ static int ip6fl_seq_open(struct inode *
{
struct seq_file *seq;
int rc = -ENOMEM;
- struct ip6fl_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
+ struct ip6fl_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
if (!s)
goto out;
@@ -674,7 +673,6 @@ static int ip6fl_seq_open(struct inode *
seq = file->private_data;
seq->private = s;
- memset(s, 0, sizeof(*s));
out:
return rc;
out_kfree:
diff --git a/net/ipv6/ipcomp6.c b/net/ipv6/ipcomp6.c
index 3c7b324..028b636 100644
--- a/net/ipv6/ipcomp6.c
+++ b/net/ipv6/ipcomp6.c
@@ -428,11 +428,10 @@ static int ipcomp6_init_state(struct xfr
goto out;
err = -ENOMEM;
- ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
+ ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
if (!ipcd)
goto out;
- memset(ipcd, 0, sizeof(*ipcd));
x->props.header_len = 0;
if (x->props.mode)
x->props.header_len += sizeof(struct ipv6hdr);
diff --git a/net/ipv6/mcast.c b/net/ipv6/mcast.c
index 807c021..6e871af 100644
--- a/net/ipv6/mcast.c
+++ b/net/ipv6/mcast.c
@@ -767,10 +767,10 @@ static void mld_add_delrec(struct inet6_
* for deleted items allows change reports to use common code with
* non-deleted or query-response MCA's.
*/
- pmc = kmalloc(sizeof(*pmc), GFP_ATOMIC);
+ pmc = kzalloc(sizeof(*pmc), GFP_ATOMIC);
if (!pmc)
return;
- memset(pmc, 0, sizeof(*pmc));
+
spin_lock_bh(&im->mca_lock);
spin_lock_init(&pmc->mca_lock);
pmc->idev = im->idev;
@@ -893,7 +893,7 @@ int ipv6_dev_mc_inc(struct net_device *d
* not found: create a new one.
*/
- mc = kmalloc(sizeof(struct ifmcaddr6), GFP_ATOMIC);
+ mc = kzalloc(sizeof(struct ifmcaddr6), GFP_ATOMIC);
if (mc == NULL) {
write_unlock_bh(&idev->lock);
@@ -901,7 +901,6 @@ int ipv6_dev_mc_inc(struct net_device *d
return -ENOMEM;
}
- memset(mc, 0, sizeof(struct ifmcaddr6));
init_timer(&mc->mca_timer);
mc->mca_timer.function = igmp6_timer_handler;
mc->mca_timer.data = (unsigned long) mc;
@@ -1934,10 +1933,10 @@ static int ip6_mc_add1_src(struct ifmcad
psf_prev = psf;
}
if (!psf) {
- psf = kmalloc(sizeof(*psf), GFP_ATOMIC);
+ psf = kzalloc(sizeof(*psf), GFP_ATOMIC);
if (!psf)
return -ENOBUFS;
- memset(psf, 0, sizeof(*psf));
+
psf->sf_addr = *psfsrc;
if (psf_prev) {
psf_prev->sf_next = psf;
@@ -2431,7 +2430,7 @@ static int igmp6_mc_seq_open(struct inod
{
struct seq_file *seq;
int rc = -ENOMEM;
- struct igmp6_mc_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
+ struct igmp6_mc_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
if (!s)
goto out;
@@ -2442,7 +2441,6 @@ static int igmp6_mc_seq_open(struct inod
seq = file->private_data;
seq->private = s;
- memset(s, 0, sizeof(*s));
out:
return rc;
out_kfree:
@@ -2606,7 +2604,7 @@ static int igmp6_mcf_seq_open(struct ino
{
struct seq_file *seq;
int rc = -ENOMEM;
- struct igmp6_mcf_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
+ struct igmp6_mcf_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
if (!s)
goto out;
@@ -2617,7 +2615,6 @@ static int igmp6_mcf_seq_open(struct ino
seq = file->private_data;
seq->private = s;
- memset(s, 0, sizeof(*s));
out:
return rc;
out_kfree:
diff --git a/net/ipv6/raw.c b/net/ipv6/raw.c
index 0f63078..fa1ce0a 100644
--- a/net/ipv6/raw.c
+++ b/net/ipv6/raw.c
@@ -1198,7 +1198,7 @@ static int raw6_seq_open(struct inode *i
{
struct seq_file *seq;
int rc = -ENOMEM;
- struct raw6_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
+ struct raw6_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
if (!s)
goto out;
rc = seq_open(file, &raw6_seq_ops);
@@ -1206,7 +1206,6 @@ static int raw6_seq_open(struct inode *i
goto out_kfree;
seq = file->private_data;
seq->private = s;
- memset(s, 0, sizeof(*s));
out:
return rc;
out_kfree:
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index 074c49f..abcc7b2 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -1877,11 +1877,10 @@ int inet6_dump_fib(struct sk_buff *skb,
/*
* 2. allocate and initialize walker.
*/
- w = kmalloc(sizeof(*w), GFP_ATOMIC);
+ w = kzalloc(sizeof(*w), GFP_ATOMIC);
if (w == NULL)
return -ENOMEM;
RT6_TRACE("dump<%p", w);
- memset(w, 0, sizeof(*w));
w->root = &ip6_routing_table;
w->func = fib6_dump_node;
w->args = &arg;
^ permalink raw reply related
* [git patches] net driver fixes
From: Jeff Garzik @ 2006-03-11 18:40 UTC (permalink / raw)
To: Andrew Morton, Linus Torvalds; +Cc: netdev, linux-kernel
Please pull from 'upstream-fixes' branch of
master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/netdev-2.6.git
to receive the following updates:
drivers/net/3c509.c | 13 ++++++++-----
drivers/net/Kconfig | 3 ++-
drivers/net/de620.c | 2 +-
drivers/net/dl2k.c | 25 +++++++++++++++----------
drivers/net/e1000/e1000_main.c | 2 +-
drivers/net/sky2.c | 32 +++++++++++++++++++++-----------
drivers/net/tulip/de2104x.c | 26 +++++++++++++-------------
7 files changed, 61 insertions(+), 42 deletions(-)
Andrew Morton:
3c509: bus registration fix
David S. Miller:
Wrong return value corrupts free object in e1000 driver
Francois Romieu:
de2104x: prevent interrupt before the interrupt handler is registered
de2104x: fix the TX watchdog
Jesper Juhl:
NE2000 Kconfig help entry improvement
Jon Mason:
dl2k: DMA freeing error
Sam Ravnborg:
de620: fix section mismatch warning
Stephen Hemminger:
sky2: not random enough
sky2: force early transmit interrupts
sky2: truncate oversize packets
diff --git a/drivers/net/3c509.c b/drivers/net/3c509.c
index 824e430..830528d 100644
--- a/drivers/net/3c509.c
+++ b/drivers/net/3c509.c
@@ -1574,6 +1574,7 @@ MODULE_LICENSE("GPL");
static int __init el3_init_module(void)
{
+ int ret = 0;
el3_cards = 0;
if (debug >= 0)
@@ -1589,14 +1590,16 @@ static int __init el3_init_module(void)
}
#ifdef CONFIG_EISA
- if (eisa_driver_register (&el3_eisa_driver) < 0) {
- eisa_driver_unregister (&el3_eisa_driver);
- }
+ ret = eisa_driver_register(&el3_eisa_driver);
#endif
#ifdef CONFIG_MCA
- mca_register_driver(&el3_mca_driver);
+ {
+ int err = mca_register_driver(&el3_mca_driver);
+ if (ret == 0)
+ ret = err;
+ }
#endif
- return 0;
+ return ret;
}
static void __exit el3_cleanup_module(void)
diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
index e45a8f9..aa633fa 100644
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -1087,7 +1087,8 @@ config NE2000
without a specific driver are compatible with NE2000.
If you have a PCI NE2000 card however, say N here and Y to "PCI
- NE2000 support", above. If you have a NE2000 card and are running on
+ NE2000 and clone support" under "EISA, VLB, PCI and on board
+ controllers" below. If you have a NE2000 card and are running on
an MCA system (a bus system used on some IBM PS/2 computers and
laptops), say N here and Y to "NE/2 (ne2000 MCA version) support",
below.
diff --git a/drivers/net/de620.c b/drivers/net/de620.c
index 0069f5f..22fc5b8 100644
--- a/drivers/net/de620.c
+++ b/drivers/net/de620.c
@@ -1012,7 +1012,7 @@ static int __init read_eeprom(struct net
#ifdef MODULE
static struct net_device *de620_dev;
-int init_module(void)
+int __init init_module(void)
{
de620_dev = de620_probe(-1);
if (IS_ERR(de620_dev))
diff --git a/drivers/net/dl2k.c b/drivers/net/dl2k.c
index 430c628..fb9dae3 100644
--- a/drivers/net/dl2k.c
+++ b/drivers/net/dl2k.c
@@ -50,8 +50,8 @@
*/
#define DRV_NAME "D-Link DL2000-based linux driver"
-#define DRV_VERSION "v1.17a"
-#define DRV_RELDATE "2002/10/04"
+#define DRV_VERSION "v1.17b"
+#define DRV_RELDATE "2006/03/10"
#include "dl2k.h"
static char version[] __devinitdata =
@@ -765,7 +765,7 @@ rio_free_tx (struct net_device *dev, int
break;
skb = np->tx_skbuff[entry];
pci_unmap_single (np->pdev,
- np->tx_ring[entry].fraginfo,
+ np->tx_ring[entry].fraginfo & 0xffffffffffff,
skb->len, PCI_DMA_TODEVICE);
if (irq)
dev_kfree_skb_irq (skb);
@@ -892,14 +892,16 @@ receive_packet (struct net_device *dev)
/* Small skbuffs for short packets */
if (pkt_len > copy_thresh) {
- pci_unmap_single (np->pdev, desc->fraginfo,
+ pci_unmap_single (np->pdev,
+ desc->fraginfo & 0xffffffffffff,
np->rx_buf_sz,
PCI_DMA_FROMDEVICE);
skb_put (skb = np->rx_skbuff[entry], pkt_len);
np->rx_skbuff[entry] = NULL;
} else if ((skb = dev_alloc_skb (pkt_len + 2)) != NULL) {
pci_dma_sync_single_for_cpu(np->pdev,
- desc->fraginfo,
+ desc->fraginfo &
+ 0xffffffffffff,
np->rx_buf_sz,
PCI_DMA_FROMDEVICE);
skb->dev = dev;
@@ -910,7 +912,8 @@ receive_packet (struct net_device *dev)
pkt_len, 0);
skb_put (skb, pkt_len);
pci_dma_sync_single_for_device(np->pdev,
- desc->fraginfo,
+ desc->fraginfo &
+ 0xffffffffffff,
np->rx_buf_sz,
PCI_DMA_FROMDEVICE);
}
@@ -1796,8 +1799,9 @@ rio_close (struct net_device *dev)
np->rx_ring[i].fraginfo = 0;
skb = np->rx_skbuff[i];
if (skb) {
- pci_unmap_single (np->pdev, np->rx_ring[i].fraginfo,
- skb->len, PCI_DMA_FROMDEVICE);
+ pci_unmap_single(np->pdev,
+ np->rx_ring[i].fraginfo & 0xffffffffffff,
+ skb->len, PCI_DMA_FROMDEVICE);
dev_kfree_skb (skb);
np->rx_skbuff[i] = NULL;
}
@@ -1805,8 +1809,9 @@ rio_close (struct net_device *dev)
for (i = 0; i < TX_RING_SIZE; i++) {
skb = np->tx_skbuff[i];
if (skb) {
- pci_unmap_single (np->pdev, np->tx_ring[i].fraginfo,
- skb->len, PCI_DMA_TODEVICE);
+ pci_unmap_single(np->pdev,
+ np->tx_ring[i].fraginfo & 0xffffffffffff,
+ skb->len, PCI_DMA_TODEVICE);
dev_kfree_skb (skb);
np->tx_skbuff[i] = NULL;
}
diff --git a/drivers/net/e1000/e1000_main.c b/drivers/net/e1000/e1000_main.c
index 5b7d0f4..4c4db96 100644
--- a/drivers/net/e1000/e1000_main.c
+++ b/drivers/net/e1000/e1000_main.c
@@ -2917,7 +2917,7 @@ e1000_xmit_frame(struct sk_buff *skb, st
if (!__pskb_pull_tail(skb, pull_size)) {
printk(KERN_ERR "__pskb_pull_tail failed.\n");
dev_kfree_skb_any(skb);
- return -EFAULT;
+ return NETDEV_TX_OK;
}
len = skb->len - skb->data_len;
}
diff --git a/drivers/net/sky2.c b/drivers/net/sky2.c
index 72c1630..7326036 100644
--- a/drivers/net/sky2.c
+++ b/drivers/net/sky2.c
@@ -74,7 +74,7 @@
#define TX_RING_SIZE 512
#define TX_DEF_PENDING (TX_RING_SIZE - 1)
#define TX_MIN_PENDING 64
-#define MAX_SKB_TX_LE (4 + 2*MAX_SKB_FRAGS)
+#define MAX_SKB_TX_LE (4 + (sizeof(dma_addr_t)/sizeof(u32))*MAX_SKB_FRAGS)
#define STATUS_RING_SIZE 2048 /* 2 ports * (TX + 2*RX) */
#define STATUS_LE_BYTES (STATUS_RING_SIZE*sizeof(struct sky2_status_le))
@@ -622,8 +622,8 @@ static void sky2_mac_init(struct sky2_hw
/* Configure Rx MAC FIFO */
sky2_write8(hw, SK_REG(port, RX_GMF_CTRL_T), GMF_RST_CLR);
- sky2_write16(hw, SK_REG(port, RX_GMF_CTRL_T),
- GMF_RX_CTRL_DEF);
+ sky2_write32(hw, SK_REG(port, RX_GMF_CTRL_T),
+ GMF_OPER_ON | GMF_RX_F_FL_ON);
/* Flush Rx MAC FIFO on any flow control or error */
sky2_write16(hw, SK_REG(port, RX_GMF_FL_MSK), GMR_FS_ANY_ERR);
@@ -995,6 +995,10 @@ static int sky2_rx_start(struct sky2_por
sky2_rx_add(sky2, re->mapaddr);
}
+ /* Truncate oversize frames */
+ sky2_write16(hw, SK_REG(sky2->port, RX_GMF_TR_THR), sky2->rx_bufsize - 8);
+ sky2_write32(hw, SK_REG(sky2->port, RX_GMF_CTRL_T), RX_TRUNC_ON);
+
/* Tell chip about available buffers */
sky2_write16(hw, Y2_QADDR(rxq, PREF_UNIT_PUT_IDX), sky2->rx_put);
sky2->rx_last_put = sky2_read16(hw, Y2_QADDR(rxq, PREF_UNIT_PUT_IDX));
@@ -1145,6 +1149,7 @@ static int sky2_xmit_frame(struct sk_buf
struct sky2_tx_le *le = NULL;
struct tx_ring_info *re;
unsigned i, len;
+ int avail;
dma_addr_t mapping;
u32 addr64;
u16 mss;
@@ -1287,12 +1292,16 @@ static int sky2_xmit_frame(struct sk_buf
re->idx = sky2->tx_prod;
le->ctrl |= EOP;
+ avail = tx_avail(sky2);
+ if (mss != 0 || avail < TX_MIN_PENDING) {
+ le->ctrl |= FRC_STAT;
+ if (avail <= MAX_SKB_TX_LE)
+ netif_stop_queue(dev);
+ }
+
sky2_put_idx(hw, txqaddr[sky2->port], sky2->tx_prod,
&sky2->tx_last_put, TX_RING_SIZE);
- if (tx_avail(sky2) <= MAX_SKB_TX_LE)
- netif_stop_queue(dev);
-
out_unlock:
spin_unlock(&sky2->tx_lock);
@@ -1707,10 +1716,12 @@ static void sky2_tx_timeout(struct net_d
#define roundup(x, y) ((((x)+((y)-1))/(y))*(y))
-/* Want receive buffer size to be multiple of 64 bits, and incl room for vlan */
+/* Want receive buffer size to be multiple of 64 bits
+ * and incl room for vlan and truncation
+ */
static inline unsigned sky2_buf_size(int mtu)
{
- return roundup(mtu + ETH_HLEN + 4, 8);
+ return roundup(mtu + ETH_HLEN + VLAN_HLEN, 8) + 8;
}
static int sky2_change_mtu(struct net_device *dev, int new_mtu)
@@ -1793,7 +1804,7 @@ static struct sk_buff *sky2_receive(stru
if (!(status & GMR_FS_RX_OK))
goto resubmit;
- if ((status >> 16) != length || length > sky2->rx_bufsize)
+ if (length > sky2->netdev->mtu + ETH_HLEN)
goto oversize;
if (length < copybreak) {
@@ -3243,8 +3254,7 @@ static int __devinit sky2_probe(struct p
}
}
- err = request_irq(pdev->irq, sky2_intr, SA_SHIRQ | SA_SAMPLE_RANDOM,
- DRV_NAME, hw);
+ err = request_irq(pdev->irq, sky2_intr, SA_SHIRQ, DRV_NAME, hw);
if (err) {
printk(KERN_ERR PFX "%s: cannot assign irq %d\n",
pci_name(pdev), pdev->irq);
diff --git a/drivers/net/tulip/de2104x.c b/drivers/net/tulip/de2104x.c
index d7fb3ff..2d0cfbc 100644
--- a/drivers/net/tulip/de2104x.c
+++ b/drivers/net/tulip/de2104x.c
@@ -1362,7 +1362,6 @@ static int de_open (struct net_device *d
{
struct de_private *de = dev->priv;
int rc;
- unsigned long flags;
if (netif_msg_ifup(de))
printk(KERN_DEBUG "%s: enabling interface\n", dev->name);
@@ -1376,18 +1375,20 @@ static int de_open (struct net_device *d
return rc;
}
- rc = de_init_hw(de);
- if (rc) {
- printk(KERN_ERR "%s: h/w init failure, err=%d\n",
- dev->name, rc);
- goto err_out_free;
- }
+ dw32(IntrMask, 0);
rc = request_irq(dev->irq, de_interrupt, SA_SHIRQ, dev->name, dev);
if (rc) {
printk(KERN_ERR "%s: IRQ %d request failure, err=%d\n",
dev->name, dev->irq, rc);
- goto err_out_hw;
+ goto err_out_free;
+ }
+
+ rc = de_init_hw(de);
+ if (rc) {
+ printk(KERN_ERR "%s: h/w init failure, err=%d\n",
+ dev->name, rc);
+ goto err_out_free_irq;
}
netif_start_queue(dev);
@@ -1395,11 +1396,8 @@ static int de_open (struct net_device *d
return 0;
-err_out_hw:
- spin_lock_irqsave(&de->lock, flags);
- de_stop_hw(de);
- spin_unlock_irqrestore(&de->lock, flags);
-
+err_out_free_irq:
+ free_irq(dev->irq, dev);
err_out_free:
de_free_rings(de);
return rc;
@@ -1455,6 +1453,8 @@ static void de_tx_timeout (struct net_de
synchronize_irq(dev->irq);
de_clean_rings(de);
+ de_init_rings(de);
+
de_init_hw(de);
netif_wake_queue(dev);
^ permalink raw reply related
* Re: [UPDATED PATCH] Re: Re: [Patch 7/7] Generic netlink interface (delay accounting)
From: jamal @ 2006-03-11 13:30 UTC (permalink / raw)
To: balbir; +Cc: Shailabh Nagar, netdev, linux-kernel, lse-tech
In-Reply-To: <20060310163927.GA6537@in.ibm.com>
On Fri, 2006-10-03 at 22:09 +0530, Balbir Singh wrote:
> On Fri, Mar 10, 2006 at 09:53:53AM -0500, jamal wrote:
> > On kernel->user (in the case of response to #a or async notifiation as
> > in #b) you really dont need to specify the TG/PID since they appear in
> > the STATS etc.
>
> I see your point now. I am looking at other users of netlink like
> rtnetlink and I see the classical usage.
>
> We can implement TLV's in our code, but for the most part the data we exchange
> between the user <-> kernel has all the TLV's listed in the enum above.
>
> The major differnece is the type (pid/tgid). Hence we created a structure
> (taskstats) instead of using TLV's.
Something to remember:
1) TLVs are essentially giving you the flexibility to send optionally
appearing elements. It is up to the receiver (in the kernel or user
space) to check for the presence of mandatory elements or execute things
depending on the presence of certain TLVs. Example in your case:
if the tgid TLV appears then the user is requesting for that TLV
if the pid appears then they are requesting for that
if both appear then it is the && of the two.
You should always ignore TLVs you dont understand - to allow for forward
compatibility.
2) The "T" part is essentially also encoding (semantically) what size
the value is; the "L" part is useful for validation. So the receiver
will always know what the size of the TLV is by definition and uses the
L to make sure it is the right size. Reject what is of the wrong size.
cheers,
jamal
-------------------------------------------------------
This SF.Net email is sponsored by xPML, a groundbreaking scripting language
that extends applications into web and mobile media. Attend the live webcast
and join the prime developer group breaking into this new coding territory!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=110944&bid=241720&dat=121642
^ permalink raw reply
* Re: [PATCH 8/8] [I/OAT] TCP recv offload to I/OAT
From: Andrew Morton @ 2006-03-11 9:41 UTC (permalink / raw)
To: Chris Leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060311022936.3950.86896.stgit@gitlost.site>
Chris Leech <christopher.leech@intel.com> wrote:
>
> Locks down user pages and sets up for DMA in tcp_recvmsg, then calls
> dma_async_try_early_copy in tcp_v4_do_rcv
All those ifdefs are still there. They really do put a maintenance burden
on, of all places, net/ipv4/tcp.c. Please find a way of abstracting out
the conceptual additions which IOAT makes to TCP and to wrap that into
inline functions, etc. See how tidily NF_HOOK() has inserted things into
the net stack.
Also, for something which is billed as an performance enhancement,
benchmark numbers are the key thing which we need to judge the desirability
of these changes. What is the plan there?
Thanks.
^ permalink raw reply
* Re: [PATCH 3/8] [I/OAT] Setup the networking subsystem as a DMA client
From: Andrew Morton @ 2006-03-11 9:00 UTC (permalink / raw)
To: Chris Leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060311022924.3950.33580.stgit@gitlost.site>
Chris Leech <christopher.leech@intel.com> wrote:
>
> +#ifdef CONFIG_NET_DMA
> +#include <linux/dmaengine.h>
> +#endif
There are still a number of instances of this in the patch series. Did you
decide to keep the ifdefs in there for some reason?
^ permalink raw reply
* Re: [PATCH 3/8] [I/OAT] Setup the networking subsystem as a DMA client
From: Andrew Morton @ 2006-03-11 8:56 UTC (permalink / raw)
To: Chris Leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060311022924.3950.33580.stgit@gitlost.site>
Chris Leech <christopher.leech@intel.com> wrote:
>
There seems to be a small race here.
> +static void net_dma_rebalance(void)
> +{
> + unsigned int cpu, i, n;
> + struct dma_chan *chan;
> +
> + lock_cpu_hotplug();
> +
> + if (net_dma_count == 0) {
> + for_each_online_cpu(cpu)
> + rcu_assign_pointer(per_cpu(softnet_data.net_dma, cpu), NULL);
> + unlock_cpu_hotplug();
> + return;
> + }
If some other CPU does netdev_dma_event(DMA_RESOURCE_REMOVED) now
> + i = 0;
> + cpu = first_cpu(cpu_online_map);
> +
> + rcu_read_lock();
> + list_for_each_entry(chan, &net_dma_client->channels, client_node) {
> + n = ((num_online_cpus() / net_dma_count)
> + + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
This will get a divide-by-zero.
> + while(n) {
> + per_cpu(softnet_data.net_dma, cpu) = chan;
> + cpu = next_cpu(cpu, cpu_online_map);
> + n--;
> + }
> + i++;
> + }
> + rcu_read_unlock();
> +
> + unlock_cpu_hotplug();
> +}
> +
> +/**
> + * netdev_dma_event - event callback for the net_dma_client
> + * @client: should always be net_dma_client
> + * @chan:
> + * @event:
> + */
> +static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
> + enum dma_event event)
> +{
> + switch (event) {
> + case DMA_RESOURCE_ADDED:
> + net_dma_count++;
> + net_dma_rebalance();
> + break;
> + case DMA_RESOURCE_REMOVED:
> + net_dma_count--;
> + net_dma_rebalance();
> + break;
> + default:
> + break;
> + }
> +}
^ permalink raw reply
* Re: [PATCH 1/8] [I/OAT] DMA memcpy subsystem
From: Andrew Morton @ 2006-03-11 8:53 UTC (permalink / raw)
To: Chris Leech; +Cc: linux-kernel, netdev
In-Reply-To: <20060311022919.3950.43835.stgit@gitlost.site>
Chris Leech <christopher.leech@intel.com> wrote:
>
> +void dma_async_device_cleanup(struct kref *kref);
>
Declarations go in header files, please. Or give it static scope.
^ permalink raw reply
* RE: [PATCH 4/6 v2] IB: address translation to map IP toIB addresses (GIDs)
From: Sean Hefty @ 2006-03-11 6:10 UTC (permalink / raw)
To: 'Roland Dreier'; +Cc: linux-kernel, netdev, openib-general
In-Reply-To: <ada1wx9hjfe.fsf@cisco.com>
>The ib_addr module depends on CONFIG_INET, because it uses symbols
>like arp_tbl, which are only exported if INET is enabled.
>
>I fixed this up by creating a new (non-user-visible) config symbol to
>control when ib_addr is built -- I put the following diff on top of
>your patch in my tree:
Thanks!
-Sean
^ permalink raw reply
* [2.6 patch] drivers/net/wireless/ipw2200.c: fix an array overun
From: Adrian Bunk @ 2006-03-11 3:42 UTC (permalink / raw)
To: yi.zhu, jketreno; +Cc: netdev, linville, linux-kernel
This patch fixes a big array overun found by the Coverity checker.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
--- linux-2.6.16-rc5-mm3-full/drivers/net/wireless/ipw2200.c.old 2006-03-11 02:41:23.000000000 +0100
+++ linux-2.6.16-rc5-mm3-full/drivers/net/wireless/ipw2200.c 2006-03-11 02:42:04.000000000 +0100
@@ -9956,9 +9956,8 @@ static int ipw_ethtool_set_eeprom(struct
return -EINVAL;
mutex_lock(&p->mutex);
memcpy(&p->eeprom[eeprom->offset], bytes, eeprom->len);
- for (i = IPW_EEPROM_DATA;
- i < IPW_EEPROM_DATA + IPW_EEPROM_IMAGE_SIZE; i++)
- ipw_write8(p, i, p->eeprom[i]);
+ for (i = 0; i < IPW_EEPROM_IMAGE_SIZE; i++)
+ ipw_write8(p, i + IPW_EEPROM_DATA, p->eeprom[i]);
mutex_unlock(&p->mutex);
return 0;
}
^ permalink raw reply
* Re: [2.6 patch] net/decnet/dn_route.c: fix inconsequent NULL checking
From: Herbert Xu @ 2006-03-11 2:51 UTC (permalink / raw)
To: Adrian Bunk; +Cc: patrick, linux-decnet-user, netdev, linux-kernel
In-Reply-To: <20060310230233.GB21864@stusta.de>
Adrian Bunk <bunk@stusta.de> wrote:
> The Coverity checker noted this inconsequent NULL checking in
> dnrt_drop().
>
> Since all callers ensure that NULL isn't passed, we can simply remove
> the check.
Ack.
In fact it's pointless even if the caller didn't check as dst_release
checks it anyway.
Cheers,
--
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
^ permalink raw reply
* [PATCH 2/8] [I/OAT] Driver for the Intel(R) I/OAT DMA engine
From: Leech, Christopher @ 2006-03-11 2:29 UTC (permalink / raw)
To: linux-kernel, netdev; +Cc: Leech, Christopher
[-- Attachment #1: Type: text/plain, Size: 680 bytes --]
From: Chris Leech [mailto:christopher.leech@intel.com]
Sent: Friday, March 10, 2006 6:29 PM
To:
Subject: [PATCH 2/8] [I/OAT] Driver for the Intel(R) I/OAT DMA engine
Adds a new ioatdma driver
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
drivers/dma/Kconfig | 9
drivers/dma/Makefile | 1
drivers/dma/ioatdma.c | 783
+++++++++++++++++++++++++++++++++++++++
drivers/dma/ioatdma.h | 127 ++++++
drivers/dma/ioatdma_hw.h | 52 +++
drivers/dma/ioatdma_io.h | 104 +++++
drivers/dma/ioatdma_registers.h | 128 ++++++
7 files changed, 1204 insertions(+), 0 deletions(-)
[-- Attachment #2: ioatdma_driver.gz --]
[-- Type: application/x-gzip, Size: 9132 bytes --]
^ permalink raw reply
* [PATCH 8/8] [I/OAT] TCP recv offload to I/OAT
From: Chris Leech @ 2006-03-11 2:29 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060311022759.3950.58788.stgit@gitlost.site>
Locks down user pages and sets up for DMA in tcp_recvmsg, then calls
dma_async_try_early_copy in tcp_v4_do_rcv
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
include/net/netdma.h | 1
net/ipv4/tcp.c | 110 +++++++++++++++++++++++++++++++++++++++++++++-----
net/ipv4/tcp_input.c | 74 ++++++++++++++++++++++++++++++----
net/ipv4/tcp_ipv4.c | 18 ++++++++
net/ipv6/tcp_ipv6.c | 12 +++++
5 files changed, 193 insertions(+), 22 deletions(-)
diff --git a/include/net/netdma.h b/include/net/netdma.h
index feb499f..3d9c222 100644
--- a/include/net/netdma.h
+++ b/include/net/netdma.h
@@ -38,6 +38,7 @@ static inline struct dma_chan *get_softn
int dma_skb_copy_datagram_iovec(struct dma_chan* chan,
const struct sk_buff *skb, int offset, struct iovec *to,
size_t len, struct dma_pinned_list *pinned_list);
+int dma_async_try_early_copy(struct sock *sk, struct sk_buff *skb, int hlen);
#endif /* CONFIG_NET_DMA */
#endif /* NETDMA_H */
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 9122520..a277398 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -262,7 +262,7 @@
#include <net/tcp.h>
#include <net/xfrm.h>
#include <net/ip.h>
-
+#include <net/netdma.h>
#include <asm/uaccess.h>
#include <asm/ioctls.h>
@@ -1109,6 +1109,7 @@ int tcp_recvmsg(struct kiocb *iocb, stru
int target; /* Read at least this many bytes */
long timeo;
struct task_struct *user_recv = NULL;
+ int copied_early = 0;
lock_sock(sk);
@@ -1132,6 +1133,12 @@ int tcp_recvmsg(struct kiocb *iocb, stru
target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
+#ifdef CONFIG_NET_DMA
+ tp->ucopy.dma_chan = NULL;
+ if ((len > sysctl_tcp_dma_copybreak) && !(flags & MSG_PEEK) && !sysctl_tcp_low_latency && __get_cpu_var(softnet_data.net_dma))
+ tp->ucopy.pinned_list = dma_pin_iovec_pages(msg->msg_iov, len);
+#endif
+
do {
struct sk_buff *skb;
u32 offset;
@@ -1273,6 +1280,10 @@ int tcp_recvmsg(struct kiocb *iocb, stru
} else
sk_wait_data(sk, &timeo);
+#ifdef CONFIG_NET_DMA
+ tp->ucopy.wakeup = 0;
+#endif
+
if (user_recv) {
int chunk;
@@ -1328,13 +1339,39 @@ do_prequeue:
}
if (!(flags & MSG_TRUNC)) {
- err = skb_copy_datagram_iovec(skb, offset,
- msg->msg_iov, used);
- if (err) {
- /* Exception. Bailout! */
- if (!copied)
- copied = -EFAULT;
- break;
+#ifdef CONFIG_NET_DMA
+ if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
+ tp->ucopy.dma_chan = get_softnet_dma();
+
+ if (tp->ucopy.dma_chan) {
+ tp->ucopy.dma_cookie = dma_skb_copy_datagram_iovec(
+ tp->ucopy.dma_chan, skb, offset,
+ msg->msg_iov, used,
+ tp->ucopy.pinned_list);
+
+ if (tp->ucopy.dma_cookie < 0) {
+
+ printk(KERN_ALERT "dma_cookie < 0\n");
+
+ /* Exception. Bailout! */
+ if (!copied)
+ copied = -EFAULT;
+ break;
+ }
+ if ((offset + used) == skb->len)
+ copied_early = 1;
+
+ } else
+#endif
+ {
+ err = skb_copy_datagram_iovec(skb, offset,
+ msg->msg_iov, used);
+ if (err) {
+ /* Exception. Bailout! */
+ if (!copied)
+ copied = -EFAULT;
+ break;
+ }
}
}
@@ -1354,15 +1391,33 @@ skip_copy:
if (skb->h.th->fin)
goto found_fin_ok;
- if (!(flags & MSG_PEEK))
- sk_eat_skb(sk, skb);
+ if (!(flags & MSG_PEEK)) {
+ if (!copied_early)
+ sk_eat_skb(sk, skb);
+#ifdef CONFIG_NET_DMA
+ else {
+ __skb_unlink(skb, &sk->sk_receive_queue);
+ __skb_queue_tail(&sk->sk_async_wait_queue, skb);
+ copied_early = 0;
+ }
+#endif
+ }
continue;
found_fin_ok:
/* Process the FIN. */
++*seq;
- if (!(flags & MSG_PEEK))
- sk_eat_skb(sk, skb);
+ if (!(flags & MSG_PEEK)) {
+ if (!copied_early)
+ sk_eat_skb(sk, skb);
+#ifdef CONFIG_NET_DMA
+ else {
+ __skb_unlink(skb, &sk->sk_receive_queue);
+ __skb_queue_tail(&sk->sk_async_wait_queue, skb);
+ copied_early = 0;
+ }
+#endif
+ }
break;
} while (len > 0);
@@ -1385,6 +1440,34 @@ skip_copy:
tp->ucopy.len = 0;
}
+#ifdef CONFIG_NET_DMA
+ if (tp->ucopy.dma_chan) {
+ struct sk_buff *skb;
+ dma_cookie_t done, used;
+
+ dma_async_memcpy_issue_pending(tp->ucopy.dma_chan);
+
+ while (dma_async_memcpy_complete(tp->ucopy.dma_chan,
+ tp->ucopy.dma_cookie, &done,
+ &used) == DMA_IN_PROGRESS) {
+ /* do partial cleanup of sk_async_wait_queue */
+ while ((skb = skb_peek(&sk->sk_async_wait_queue)) &&
+ (dma_async_is_complete(skb->dma_cookie, done,
+ used) == DMA_SUCCESS)) {
+ __skb_dequeue(&sk->sk_async_wait_queue);
+ kfree_skb(skb);
+ }
+ }
+
+ /* Safe to free early-copied skbs now */
+ __skb_queue_purge(&sk->sk_async_wait_queue);
+ dma_unpin_iovec_pages(tp->ucopy.pinned_list);
+ dma_chan_put(tp->ucopy.dma_chan);
+ tp->ucopy.dma_chan = NULL;
+ tp->ucopy.pinned_list = NULL;
+ }
+#endif
+
/* According to UNIX98, msg_name/msg_namelen are ignored
* on connected socket. I was just happy when found this 8) --ANK
*/
@@ -1652,6 +1735,9 @@ int tcp_disconnect(struct sock *sk, int
__skb_queue_purge(&sk->sk_receive_queue);
sk_stream_writequeue_purge(sk);
__skb_queue_purge(&tp->out_of_order_queue);
+#ifdef CONFIG_NET_DMA
+ __skb_queue_purge(&sk->sk_async_wait_queue);
+#endif
inet->dport = 0;
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 7625eaf..5307e17 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -71,6 +71,7 @@
#include <net/inet_common.h>
#include <linux/ipsec.h>
#include <asm/unaligned.h>
+#include <net/netdma.h>
int sysctl_tcp_timestamps = 1;
int sysctl_tcp_window_scaling = 1;
@@ -3901,14 +3902,23 @@ int tcp_rcv_established(struct sock *sk,
}
} else {
int eaten = 0;
+ int copied_early = 0;
- if (tp->ucopy.task == current &&
- tp->copied_seq == tp->rcv_nxt &&
- len - tcp_header_len <= tp->ucopy.len &&
- sock_owned_by_user(sk)) {
- __set_current_state(TASK_RUNNING);
+ if (tp->copied_seq == tp->rcv_nxt &&
+ len - tcp_header_len <= tp->ucopy.len) {
+#ifdef CONFIG_NET_DMA
+ if (tcp_dma_try_early_copy(sk, skb, tcp_header_len)) {
+ copied_early = 1;
+ eaten = 1;
+ }
+#endif
+ if (tp->ucopy.task == current && sock_owned_by_user(sk) && !copied_early) {
+ __set_current_state(TASK_RUNNING);
- if (!tcp_copy_to_iovec(sk, skb, tcp_header_len)) {
+ if (!tcp_copy_to_iovec(sk, skb, tcp_header_len))
+ eaten = 1;
+ }
+ if (eaten) {
/* Predicted packet is in window by definition.
* seq == rcv_nxt and rcv_wup <= rcv_nxt.
* Hence, check seq<=rcv_wup reduces to:
@@ -3924,8 +3934,9 @@ int tcp_rcv_established(struct sock *sk,
__skb_pull(skb, tcp_header_len);
tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
NET_INC_STATS_BH(LINUX_MIB_TCPHPHITSTOUSER);
- eaten = 1;
}
+ if (copied_early)
+ tcp_cleanup_rbuf(sk, skb->len);
}
if (!eaten) {
if (tcp_checksum_complete_user(sk, skb))
@@ -3966,6 +3977,11 @@ int tcp_rcv_established(struct sock *sk,
__tcp_ack_snd_check(sk, 0);
no_ack:
+#ifdef CONFIG_NET_DMA
+ if (copied_early)
+ __skb_queue_tail(&sk->sk_async_wait_queue, skb);
+ else
+#endif
if (eaten)
__kfree_skb(skb);
else
@@ -4049,6 +4065,50 @@ discard:
return 0;
}
+#ifdef CONFIG_NET_DMA
+int tcp_dma_try_early_copy(struct sock *sk, struct sk_buff *skb, int hlen)
+{
+ struct tcp_sock *tp = tcp_sk(sk);
+ int chunk = skb->len - hlen;
+ int dma_cookie;
+ int copied_early = 0;
+
+ if (tp->ucopy.wakeup)
+ return 0;
+
+ if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
+ tp->ucopy.dma_chan = get_softnet_dma();
+
+ if (tp->ucopy.dma_chan && skb->ip_summed == CHECKSUM_UNNECESSARY) {
+
+ dma_cookie = dma_skb_copy_datagram_iovec(tp->ucopy.dma_chan,
+ skb, hlen, tp->ucopy.iov, chunk, tp->ucopy.pinned_list);
+
+ if (dma_cookie < 0)
+ goto out;
+
+ tp->ucopy.dma_cookie = dma_cookie;
+ copied_early = 1;
+
+ tp->ucopy.len -= chunk;
+ tp->copied_seq += chunk;
+ tcp_rcv_space_adjust(sk);
+
+ if ((tp->ucopy.len == 0) ||
+ (tcp_flag_word(skb->h.th) & TCP_FLAG_PSH) ||
+ (atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1))) {
+ tp->ucopy.wakeup = 1;
+ sk->sk_data_ready(sk, 0);
+ }
+ } else if (chunk > 0) {
+ tp->ucopy.wakeup = 1;
+ sk->sk_data_ready(sk, 0);
+ }
+out:
+ return copied_early;
+}
+#endif /* CONFIG_NET_DMA */
+
static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
struct tcphdr *th, unsigned len)
{
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index 9e85c04..5ed065f 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -71,6 +71,7 @@
#include <net/inet_common.h>
#include <net/timewait_sock.h>
#include <net/xfrm.h>
+#include <net/netdma.h>
#include <linux/inet.h>
#include <linux/ipv6.h>
@@ -1091,8 +1092,18 @@ process:
bh_lock_sock(sk);
ret = 0;
if (!sock_owned_by_user(sk)) {
- if (!tcp_prequeue(sk, skb))
+#ifdef CONFIG_NET_DMA
+ struct tcp_sock *tp = tcp_sk(sk);
+ if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
+ tp->ucopy.dma_chan = get_softnet_dma();
+ if (tp->ucopy.dma_chan)
ret = tcp_v4_do_rcv(sk, skb);
+ else
+#endif
+ {
+ if (!tcp_prequeue(sk, skb))
+ ret = tcp_v4_do_rcv(sk, skb);
+ }
} else
sk_add_backlog(sk, skb);
bh_unlock_sock(sk);
@@ -1296,6 +1307,11 @@ int tcp_v4_destroy_sock(struct sock *sk)
/* Cleans up our, hopefully empty, out_of_order_queue. */
__skb_queue_purge(&tp->out_of_order_queue);
+#ifdef CONFIG_NET_DMA
+ /* Cleans up our sk_async_wait_queue */
+ __skb_queue_purge(&sk->sk_async_wait_queue);
+#endif
+
/* Clean prequeue, it must be empty really */
__skb_queue_purge(&tp->ucopy.prequeue);
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 301eee7..a50eb30 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1218,8 +1218,16 @@ process:
bh_lock_sock(sk);
ret = 0;
if (!sock_owned_by_user(sk)) {
- if (!tcp_prequeue(sk, skb))
- ret = tcp_v6_do_rcv(sk, skb);
+#ifdef CONFIG_NET_DMA
+ struct tcp_sock *tp = tcp_sk(sk);
+ if (tp->ucopy.dma_chan)
+ ret = tcp_v6_do_rcv(sk, skb);
+ else
+#endif
+ {
+ if (!tcp_prequeue(sk, skb))
+ ret = tcp_v6_do_rcv(sk, skb);
+ }
} else
sk_add_backlog(sk, skb);
bh_unlock_sock(sk);
^ permalink raw reply related
* [PATCH 7/8] [I/OAT] Add a sysctl for tuning the I/OAT offloaded I/O threshold
From: Chris Leech @ 2006-03-11 2:29 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060311022759.3950.58788.stgit@gitlost.site>
Any socket recv of less than this ammount will not be offloaded
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
include/linux/sysctl.h | 1 +
include/net/tcp.h | 1 +
net/core/user_dma.c | 4 ++++
net/ipv4/sysctl_net_ipv4.c | 10 ++++++++++
4 files changed, 16 insertions(+), 0 deletions(-)
diff --git a/include/linux/sysctl.h b/include/linux/sysctl.h
index 76eaeff..cd9e7c0 100644
--- a/include/linux/sysctl.h
+++ b/include/linux/sysctl.h
@@ -403,6 +403,7 @@ enum
NET_TCP_MTU_PROBING=113,
NET_TCP_BASE_MSS=114,
NET_IPV4_TCP_WORKAROUND_SIGNED_WINDOWS=115,
+ NET_TCP_DMA_COPYBREAK=116,
};
enum {
diff --git a/include/net/tcp.h b/include/net/tcp.h
index afc4b8a..f319368 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -221,6 +221,7 @@ extern int sysctl_tcp_adv_win_scale;
extern int sysctl_tcp_tw_reuse;
extern int sysctl_tcp_frto;
extern int sysctl_tcp_low_latency;
+extern int sysctl_tcp_dma_copybreak;
extern int sysctl_tcp_nometrics_save;
extern int sysctl_tcp_moderate_rcvbuf;
extern int sysctl_tcp_tso_win_divisor;
diff --git a/net/core/user_dma.c b/net/core/user_dma.c
index 24e51eb..a85d1f1 100644
--- a/net/core/user_dma.c
+++ b/net/core/user_dma.c
@@ -33,6 +33,10 @@ file called LICENSE.
#ifdef CONFIG_NET_DMA
+#define NET_DMA_DEFAULT_COPYBREAK 1024
+
+int sysctl_tcp_dma_copybreak = NET_DMA_DEFAULT_COPYBREAK;
+
/**
* dma_skb_copy_datagram_iovec - Copy a datagram to an iovec.
* @skb - buffer to copy
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 6b6c3ad..6a6aa53 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -688,6 +688,16 @@ ctl_table ipv4_table[] = {
.mode = 0644,
.proc_handler = &proc_dointvec
},
+#ifdef CONFIG_NET_DMA
+ {
+ .ctl_name = NET_TCP_DMA_COPYBREAK,
+ .procname = "tcp_dma_copybreak",
+ .data = &sysctl_tcp_dma_copybreak,
+ .maxlen = sizeof(int),
+ .mode = 0644,
+ .proc_handler = &proc_dointvec
+ },
+#endif
{ .ctl_name = 0 }
};
^ permalink raw reply related
* [PATCH 6/8] [I/OAT] Rename cleanup_rbuf to tcp_cleanup_rbuf and make non-static
From: Chris Leech @ 2006-03-11 2:29 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060311022759.3950.58788.stgit@gitlost.site>
Needed to be able to call tcp_cleanup_rbuf in tcp_input.c for I/OAT
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
include/net/tcp.h | 2 ++
net/ipv4/tcp.c | 10 +++++-----
2 files changed, 7 insertions(+), 5 deletions(-)
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 610f66b..afc4b8a 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -296,6 +296,8 @@ extern int tcp_rcv_established(struct
extern void tcp_rcv_space_adjust(struct sock *sk);
+extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
+
extern int tcp_twsk_unique(struct sock *sk,
struct sock *sktw, void *twp);
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 4b0272c..9122520 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -936,7 +936,7 @@ static int tcp_recv_urg(struct sock *sk,
* calculation of whether or not we must ACK for the sake of
* a window update.
*/
-static void cleanup_rbuf(struct sock *sk, int copied)
+void tcp_cleanup_rbuf(struct sock *sk, int copied)
{
struct tcp_sock *tp = tcp_sk(sk);
int time_to_ack = 0;
@@ -1085,7 +1085,7 @@ int tcp_read_sock(struct sock *sk, read_
/* Clean up data we have read: This will do ACK frames. */
if (copied)
- cleanup_rbuf(sk, copied);
+ tcp_cleanup_rbuf(sk, copied);
return copied;
}
@@ -1219,7 +1219,7 @@ int tcp_recvmsg(struct kiocb *iocb, stru
}
}
- cleanup_rbuf(sk, copied);
+ tcp_cleanup_rbuf(sk, copied);
if (!sysctl_tcp_low_latency && tp->ucopy.task == user_recv) {
/* Install new reader */
@@ -1390,7 +1390,7 @@ skip_copy:
*/
/* Clean up data we have read: This will do ACK frames. */
- cleanup_rbuf(sk, copied);
+ tcp_cleanup_rbuf(sk, copied);
TCP_CHECK_TIMER(sk);
release_sock(sk);
@@ -1852,7 +1852,7 @@ static int do_tcp_setsockopt(struct sock
(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT) &&
inet_csk_ack_scheduled(sk)) {
icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;
- cleanup_rbuf(sk, 1);
+ tcp_cleanup_rbuf(sk, 1);
if (!(val & 1))
icsk->icsk_ack.pingpong = 1;
}
^ permalink raw reply related
* [PATCH 5/8] [I/OAT] Structure changes for TCP recv offload to I/OAT
From: Chris Leech @ 2006-03-11 2:29 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060311022759.3950.58788.stgit@gitlost.site>
Adds an async_wait_queue and some additional fields to tcp_sock, and a
dma_cookie_t to sk_buff.
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
include/linux/skbuff.h | 6 ++++++
include/linux/tcp.h | 10 ++++++++++
include/net/sock.h | 2 ++
include/net/tcp.h | 9 +++++++++
net/core/sock.c | 6 ++++++
5 files changed, 33 insertions(+), 0 deletions(-)
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 613b951..91b1081 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -29,6 +29,9 @@
#include <linux/net.h>
#include <linux/textsearch.h>
#include <net/checksum.h>
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+#endif
#define HAVE_ALLOC_SKB /* For the drivers to know */
#define HAVE_ALIGNABLE_SKB /* Ditto 8) */
@@ -285,6 +288,9 @@ struct sk_buff {
__u16 tc_verd; /* traffic control verdict */
#endif
#endif
+#ifdef CONFIG_NET_DMA
+ dma_cookie_t dma_cookie;
+#endif
/* These elements must be at the end, see alloc_skb() for details. */
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 542d395..3bc1d37 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -18,6 +18,9 @@
#define _LINUX_TCP_H
#include <linux/types.h>
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+#endif
#include <asm/byteorder.h>
struct tcphdr {
@@ -233,6 +236,13 @@ struct tcp_sock {
struct iovec *iov;
int memory;
int len;
+#ifdef CONFIG_NET_DMA
+ /* members for async copy */
+ struct dma_chan *dma_chan;
+ int wakeup;
+ struct dma_pinned_list *pinned_list;
+ dma_cookie_t dma_cookie;
+#endif
} ucopy;
__u32 snd_wl1; /* Sequence for window update */
diff --git a/include/net/sock.h b/include/net/sock.h
index 6e133f7..8c75642 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -132,6 +132,7 @@ struct sock_common {
* @sk_receive_queue: incoming packets
* @sk_wmem_alloc: transmit queue bytes committed
* @sk_write_queue: Packet sending queue
+ * @sk_async_wait_queue: DMA copied packets
* @sk_omem_alloc: "o" is "option" or "other"
* @sk_wmem_queued: persistent queue size
* @sk_forward_alloc: space allocated forward
@@ -205,6 +206,7 @@ struct sock {
atomic_t sk_omem_alloc;
struct sk_buff_head sk_receive_queue;
struct sk_buff_head sk_write_queue;
+ struct sk_buff_head sk_async_wait_queue;
int sk_wmem_queued;
int sk_forward_alloc;
gfp_t sk_allocation;
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 9418f4d..610f66b 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -28,6 +28,9 @@
#include <linux/cache.h>
#include <linux/percpu.h>
#include <linux/skbuff.h>
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+#endif
#include <net/inet_connection_sock.h>
#include <net/inet_timewait_sock.h>
@@ -820,6 +823,12 @@ static inline void tcp_prequeue_init(str
tp->ucopy.len = 0;
tp->ucopy.memory = 0;
skb_queue_head_init(&tp->ucopy.prequeue);
+#ifdef CONFIG_NET_DMA
+ tp->ucopy.dma_chan = NULL;
+ tp->ucopy.wakeup = 0;
+ tp->ucopy.pinned_list = NULL;
+ tp->ucopy.dma_cookie = 0;
+#endif
}
/* Packet is added to VJ-style prequeue for processing in process
diff --git a/net/core/sock.c b/net/core/sock.c
index e1c1a36..d90e1bd 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -735,6 +735,9 @@ struct sock *sk_clone(const struct sock
atomic_set(&newsk->sk_omem_alloc, 0);
skb_queue_head_init(&newsk->sk_receive_queue);
skb_queue_head_init(&newsk->sk_write_queue);
+#ifdef CONFIG_NET_DMA
+ skb_queue_head_init(&newsk->sk_async_wait_queue);
+#endif
rwlock_init(&newsk->sk_dst_lock);
rwlock_init(&newsk->sk_callback_lock);
@@ -1286,6 +1289,9 @@ void sock_init_data(struct socket *sock,
skb_queue_head_init(&sk->sk_receive_queue);
skb_queue_head_init(&sk->sk_write_queue);
skb_queue_head_init(&sk->sk_error_queue);
+#ifdef CONFIG_NET_DMA
+ skb_queue_head_init(&sk->sk_async_wait_queue);
+#endif
sk->sk_send_head = NULL;
^ permalink raw reply related
* [PATCH 4/8] [I/OAT] Utility functions for offloading sk_buff to iovec copies
From: Chris Leech @ 2006-03-11 2:29 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060311022759.3950.58788.stgit@gitlost.site>
Provides for pinning user space pages in memory, copying to iovecs,
and copying from sk_buffs including fragmented and chained sk_buffs.
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
drivers/dma/Makefile | 3
drivers/dma/iovlock.c | 301 +++++++++++++++++++++++++++++++++++++++++++++
include/linux/dmaengine.h | 22 +++
include/net/netdma.h | 6 +
net/core/Makefile | 3
net/core/user_dma.c | 141 +++++++++++++++++++++
6 files changed, 474 insertions(+), 2 deletions(-)
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index c8a5f56..bdcfdbd 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -1,2 +1,3 @@
-obj-y += dmaengine.o
+obj-$(CONFIG_DMA_ENGINE) += dmaengine.o
+obj-$(CONFIG_NET_DMA) += iovlock.o
obj-$(CONFIG_INTEL_IOATDMA) += ioatdma.o
diff --git a/drivers/dma/iovlock.c b/drivers/dma/iovlock.c
new file mode 100644
index 0000000..18d54ce
--- /dev/null
+++ b/drivers/dma/iovlock.c
@@ -0,0 +1,301 @@
+/*****************************************************************************
+Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
+Portions based on net/core/datagram.c and copyrighted by their authors.
+
+This program is free software; you can redistribute it and/or modify it
+under the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 2 of the License, or (at your option)
+any later version.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59
+Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+
+The full GNU General Public License is included in this distribution in the
+file called LICENSE.
+*****************************************************************************/
+
+/*
+ * This code allows the net stack to make use of a DMA engine for
+ * skb to iovec copies.
+ */
+
+#include <linux/dmaengine.h>
+#include <linux/pagemap.h>
+#include <net/tcp.h> /* for memcpy_toiovec */
+#include <asm/io.h>
+#include <asm/uaccess.h>
+
+int num_pages_spanned(struct iovec *iov)
+{
+ return
+ ((PAGE_ALIGN((unsigned long)iov->iov_base + iov->iov_len) -
+ ((unsigned long)iov->iov_base & PAGE_MASK)) >> PAGE_SHIFT);
+}
+
+/*
+ * Pin down all the iovec pages needed for len bytes.
+ * Return a struct dma_pinned_list to keep track of pages pinned down.
+ *
+ * We are allocating a single chunk of memory, and then carving it up into
+ * 3 sections, the latter 2 whose size depends on the number of iovecs and the
+ * total number of pages, respectively.
+ */
+struct dma_pinned_list *dma_pin_iovec_pages(struct iovec *iov, size_t len)
+{
+ struct dma_pinned_list *local_list;
+ struct page **pages;
+ int i;
+ int ret;
+ int nr_iovecs = 0;
+ int iovec_len_used = 0;
+ int iovec_pages_used = 0;
+ long err;
+
+ /* don't pin down non-user-based iovecs */
+ if (segment_eq(get_fs(), KERNEL_DS))
+ return NULL;
+
+ /* determine how many iovecs/pages there are, up front */
+ do {
+ iovec_len_used += iov[nr_iovecs].iov_len;
+ iovec_pages_used += num_pages_spanned(&iov[nr_iovecs]);
+ nr_iovecs++;
+ } while (iovec_len_used < len);
+
+ /* single kmalloc for pinned list, page_list[], and the page arrays */
+ local_list = kmalloc(sizeof(*local_list)
+ + (nr_iovecs * sizeof (struct dma_page_list))
+ + (iovec_pages_used * sizeof (struct page*)), GFP_KERNEL);
+ if (!local_list) {
+ err = -ENOMEM;
+ goto out;
+ }
+
+ /* list of pages starts right after the page list array */
+ pages = (struct page **) &local_list->page_list[nr_iovecs];
+
+ for (i = 0; i < nr_iovecs; i++) {
+ struct dma_page_list *page_list = &local_list->page_list[i];
+
+ len -= iov[i].iov_len;
+
+ if (!access_ok(VERIFY_WRITE, iov[i].iov_base, iov[i].iov_len)) {
+ err = -EFAULT;
+ goto unpin;
+ }
+
+ page_list->nr_pages = num_pages_spanned(&iov[i]);
+ page_list->base_address = iov[i].iov_base;
+
+ page_list->pages = pages;
+ pages += page_list->nr_pages;
+
+ /* pin pages down */
+ down_read(¤t->mm->mmap_sem);
+ ret = get_user_pages(
+ current,
+ current->mm,
+ (unsigned long) iov[i].iov_base,
+ page_list->nr_pages,
+ 1, /* write */
+ 0, /* force */
+ page_list->pages,
+ NULL);
+ up_read(¤t->mm->mmap_sem);
+
+ if (ret != page_list->nr_pages) {
+ err = -ENOMEM;
+ goto unpin;
+ }
+
+ local_list->nr_iovecs = i + 1;
+ }
+
+ return local_list;
+
+unpin:
+ dma_unpin_iovec_pages(local_list);
+out:
+ return ERR_PTR(err);
+}
+
+void dma_unpin_iovec_pages(struct dma_pinned_list *pinned_list)
+{
+ int i, j;
+
+ if (!pinned_list)
+ return;
+
+ for (i = 0; i < pinned_list->nr_iovecs; i++) {
+ struct dma_page_list *page_list = &pinned_list->page_list[i];
+ for (j = 0; j < page_list->nr_pages; j++) {
+ set_page_dirty_lock(page_list->pages[j]);
+ page_cache_release(page_list->pages[j]);
+ }
+ }
+
+ kfree(pinned_list);
+}
+
+static dma_cookie_t dma_memcpy_to_kernel_iovec(struct dma_chan *chan, struct
+ iovec *iov, unsigned char *kdata, size_t len)
+{
+ dma_cookie_t dma_cookie = 0;
+
+ while (len > 0) {
+ if (iov->iov_len) {
+ int copy = min_t(unsigned int, iov->iov_len, len);
+ dma_cookie = dma_async_memcpy_buf_to_buf(
+ chan,
+ iov->iov_base,
+ kdata,
+ copy);
+ kdata += copy;
+ len -= copy;
+ iov->iov_len -= copy;
+ iov->iov_base += copy;
+ }
+ iov++;
+ }
+
+ return dma_cookie;
+}
+
+/*
+ * We have already pinned down the pages we will be using in the iovecs.
+ * Each entry in iov array has corresponding entry in pinned_list->page_list.
+ * Using array indexing to keep iov[] and page_list[] in sync.
+ * Initial elements in iov array's iov->iov_len will be 0 if already copied into
+ * by another call.
+ * iov array length remaining guaranteed to be bigger than len.
+ */
+dma_cookie_t dma_memcpy_to_iovec(struct dma_chan *chan, struct iovec *iov,
+ struct dma_pinned_list *pinned_list, unsigned char *kdata, size_t len)
+{
+ int iov_byte_offset;
+ int copy;
+ dma_cookie_t dma_cookie = 0;
+ int iovec_idx;
+ int page_idx;
+
+ if (!chan)
+ return memcpy_toiovec(iov, kdata, len);
+
+ /* -> kernel copies (e.g. smbfs) */
+ if (!pinned_list)
+ return dma_memcpy_to_kernel_iovec(chan, iov, kdata, len);
+
+ iovec_idx = 0;
+ while (iovec_idx < pinned_list->nr_iovecs) {
+ struct dma_page_list *page_list;
+
+ /* skip already used-up iovecs */
+ while (!iov[iovec_idx].iov_len)
+ iovec_idx++;
+
+ page_list = &pinned_list->page_list[iovec_idx];
+
+ iov_byte_offset = ((unsigned long)iov[iovec_idx].iov_base & ~PAGE_MASK);
+ page_idx = (((unsigned long)iov[iovec_idx].iov_base & PAGE_MASK)
+ - ((unsigned long)page_list->base_address & PAGE_MASK)) >> PAGE_SHIFT;
+
+ /* break up copies to not cross page boundary */
+ while (iov[iovec_idx].iov_len) {
+ copy = min_t(int, PAGE_SIZE - iov_byte_offset, len);
+ copy = min_t(int, copy, iov[iovec_idx].iov_len);
+
+ dma_cookie = dma_async_memcpy_buf_to_pg(chan,
+ page_list->pages[page_idx],
+ iov_byte_offset,
+ kdata,
+ copy);
+
+ len -= copy;
+ iov[iovec_idx].iov_len -= copy;
+ iov[iovec_idx].iov_base += copy;
+
+ if (!len)
+ return dma_cookie;
+
+ kdata += copy;
+ iov_byte_offset = 0;
+ page_idx++;
+ }
+ iovec_idx++;
+ }
+
+ /* really bad if we ever run out of iovecs */
+ BUG();
+ return -EFAULT;
+}
+
+dma_cookie_t dma_memcpy_pg_to_iovec(struct dma_chan *chan, struct iovec *iov,
+ struct dma_pinned_list *pinned_list, struct page *page,
+ unsigned int offset, size_t len)
+{
+ int iov_byte_offset;
+ int copy;
+ dma_cookie_t dma_cookie = 0;
+ int iovec_idx;
+ int page_idx;
+ int err;
+
+ /* this needs as-yet-unimplemented buf-to-buff, so punt. */
+ /* TODO: use dma for this */
+ if (!chan || !pinned_list) {
+ u8 *vaddr = kmap(page);
+ err = memcpy_toiovec(iov, vaddr + offset, len);
+ kunmap(page);
+ return err;
+ }
+
+ iovec_idx = 0;
+ while (iovec_idx < pinned_list->nr_iovecs) {
+ struct dma_page_list *page_list;
+
+ /* skip already used-up iovecs */
+ while (!iov[iovec_idx].iov_len)
+ iovec_idx++;
+
+ page_list = &pinned_list->page_list[iovec_idx];
+
+ iov_byte_offset = ((unsigned long)iov[iovec_idx].iov_base & ~PAGE_MASK);
+ page_idx = (((unsigned long)iov[iovec_idx].iov_base & PAGE_MASK)
+ - ((unsigned long)page_list->base_address & PAGE_MASK)) >> PAGE_SHIFT;
+
+ /* break up copies to not cross page boundary */
+ while (iov[iovec_idx].iov_len) {
+ copy = min_t(int, PAGE_SIZE - iov_byte_offset, len);
+ copy = min_t(int, copy, iov[iovec_idx].iov_len);
+
+ dma_cookie = dma_async_memcpy_pg_to_pg(chan,
+ page_list->pages[page_idx],
+ iov_byte_offset,
+ page,
+ offset,
+ copy);
+
+ len -= copy;
+ iov[iovec_idx].iov_len -= copy;
+ iov[iovec_idx].iov_base += copy;
+
+ if (!len)
+ return dma_cookie;
+
+ offset += copy;
+ iov_byte_offset = 0;
+ page_idx++;
+ }
+ iovec_idx++;
+ }
+
+ /* really bad if we ever run out of iovecs */
+ BUG();
+ return -EFAULT;
+}
diff --git a/include/linux/dmaengine.h b/include/linux/dmaengine.h
index ac3bff9..a2ea457 100644
--- a/include/linux/dmaengine.h
+++ b/include/linux/dmaengine.h
@@ -319,5 +319,27 @@ static inline enum dma_status dma_async_
int dma_async_device_register(struct dma_device *device);
void dma_async_device_unregister(struct dma_device *device);
+/* --- Helper iov-locking functions --- */
+
+struct dma_page_list {
+ char *base_address;
+ int nr_pages;
+ struct page **pages;
+};
+
+struct dma_pinned_list {
+ int nr_iovecs;
+ struct dma_page_list page_list[0];
+};
+
+struct dma_pinned_list *dma_pin_iovec_pages(struct iovec *iov, size_t len);
+void dma_unpin_iovec_pages(struct dma_pinned_list* pinned_list);
+
+dma_cookie_t dma_memcpy_to_iovec(struct dma_chan *chan, struct iovec *iov,
+ struct dma_pinned_list *pinned_list, unsigned char *kdata, size_t len);
+dma_cookie_t dma_memcpy_pg_to_iovec(struct dma_chan *chan, struct iovec *iov,
+ struct dma_pinned_list *pinned_list, struct page *page,
+ unsigned int offset, size_t len);
+
#endif /* CONFIG_DMA_ENGINE */
#endif /* DMAENGINE_H */
diff --git a/include/net/netdma.h b/include/net/netdma.h
index 6435aef..feb499f 100644
--- a/include/net/netdma.h
+++ b/include/net/netdma.h
@@ -22,6 +22,7 @@ file called LICENSE.
#define NETDMA_H
#ifdef CONFIG_NET_DMA
#include <linux/dmaengine.h>
+#include <linux/skbuff.h>
static inline struct dma_chan *get_softnet_dma(void)
{
@@ -33,5 +34,10 @@ static inline struct dma_chan *get_softn
rcu_read_unlock();
return chan;
}
+
+int dma_skb_copy_datagram_iovec(struct dma_chan* chan,
+ const struct sk_buff *skb, int offset, struct iovec *to,
+ size_t len, struct dma_pinned_list *pinned_list);
+
#endif /* CONFIG_NET_DMA */
#endif /* NETDMA_H */
diff --git a/net/core/Makefile b/net/core/Makefile
index 630da0f..d02132b 100644
--- a/net/core/Makefile
+++ b/net/core/Makefile
@@ -8,7 +8,8 @@ obj-y := sock.o request_sock.o skbuff.o
obj-$(CONFIG_SYSCTL) += sysctl_net_core.o
obj-y += dev.o ethtool.o dev_mcast.o dst.o \
- neighbour.o rtnetlink.o utils.o link_watch.o filter.o
+ neighbour.o rtnetlink.o utils.o link_watch.o filter.o \
+ user_dma.o
obj-$(CONFIG_XFRM) += flow.o
obj-$(CONFIG_SYSFS) += net-sysfs.o
diff --git a/net/core/user_dma.c b/net/core/user_dma.c
new file mode 100644
index 0000000..24e51eb
--- /dev/null
+++ b/net/core/user_dma.c
@@ -0,0 +1,141 @@
+/*****************************************************************************
+Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
+Portions based on net/core/datagram.c and copyrighted by their authors.
+
+This program is free software; you can redistribute it and/or modify it
+under the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 2 of the License, or (at your option)
+any later version.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59
+Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+
+The full GNU General Public License is included in this distribution in the
+file called LICENSE.
+*****************************************************************************/
+
+/*
+ * This code allows the net stack to make use of a DMA engine for
+ * skb to iovec copies.
+ */
+
+#include <linux/dmaengine.h>
+#include <linux/socket.h>
+#include <linux/rtnetlink.h> /* for BUG_TRAP */
+#include <net/tcp.h>
+
+
+#ifdef CONFIG_NET_DMA
+
+/**
+ * dma_skb_copy_datagram_iovec - Copy a datagram to an iovec.
+ * @skb - buffer to copy
+ * @offset - offset in the buffer to start copying from
+ * @iovec - io vector to copy to
+ * @len - amount of data to copy from buffer to iovec
+ * @pinned_list - locked iovec buffer data
+ *
+ * Note: the iovec is modified during the copy.
+ */
+int dma_skb_copy_datagram_iovec(struct dma_chan *chan,
+ struct sk_buff *skb, int offset, struct iovec *to,
+ size_t len, struct dma_pinned_list *pinned_list)
+{
+ int start = skb_headlen(skb);
+ int i, copy = start - offset;
+ dma_cookie_t cookie = 0;
+
+ /* Copy header. */
+ if (copy > 0) {
+ if (copy > len)
+ copy = len;
+ cookie = dma_memcpy_to_iovec(chan, to, pinned_list,
+ skb->data + offset, copy);
+ if (cookie < 0)
+ goto fault;
+ len -= copy;
+ if (len == 0)
+ goto end;
+ offset += copy;
+ }
+
+ /* Copy paged appendix. Hmm... why does this look so complicated? */
+ for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+ int end;
+
+ BUG_TRAP(start <= offset + len);
+
+ end = start + skb_shinfo(skb)->frags[i].size;
+ copy = end - offset;
+ if ((copy = end - offset) > 0) {
+ skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+ struct page *page = frag->page;
+
+ if (copy > len)
+ copy = len;
+
+ cookie = dma_memcpy_pg_to_iovec(chan, to, pinned_list, page,
+ frag->page_offset + offset - start, copy);
+ if (cookie < 0)
+ goto fault;
+ len -= copy;
+ if (len == 0)
+ goto end;
+ offset += copy;
+ }
+ start = end;
+ }
+
+ if (skb_shinfo(skb)->frag_list) {
+ struct sk_buff *list = skb_shinfo(skb)->frag_list;
+
+ for (; list; list = list->next) {
+ int end;
+
+ BUG_TRAP(start <= offset + len);
+
+ end = start + list->len;
+ copy = end - offset;
+ if (copy > 0) {
+ if (copy > len)
+ copy = len;
+ cookie = dma_skb_copy_datagram_iovec(chan, list,
+ offset - start, to, copy,
+ pinned_list);
+ if (cookie < 0)
+ goto fault;
+ len -= copy;
+ if (len == 0)
+ goto end;
+ offset += copy;
+ }
+ start = end;
+ }
+ }
+
+end:
+ if (!len) {
+ skb->dma_cookie = cookie;
+ return cookie;
+ }
+
+fault:
+ return -EFAULT;
+}
+
+#else
+
+int dma_skb_copy_datagram_iovec(struct dma_chan *chan,
+ const struct sk_buff *skb, int offset, struct iovec *to,
+ size_t len, struct dma_pinned_list *pinned_list)
+{
+ return skb_copy_datagram_iovec(skb, offset, to, len);
+}
+
+#endif
^ permalink raw reply related
* [PATCH 3/8] [I/OAT] Setup the networking subsystem as a DMA client
From: Chris Leech @ 2006-03-11 2:29 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060311022759.3950.58788.stgit@gitlost.site>
Attempts to allocate per-CPU DMA channels
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
drivers/dma/Kconfig | 12 +++++
include/linux/netdevice.h | 6 +++
include/net/netdma.h | 37 +++++++++++++++++
net/core/dev.c | 100 +++++++++++++++++++++++++++++++++++++++++++++
4 files changed, 155 insertions(+), 0 deletions(-)
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index 0f15e76..30d021d 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -10,6 +10,18 @@ config DMA_ENGINE
DMA engines offload copy operations from the CPU to dedicated
hardware, allowing the copies to happen asynchronously.
+comment "DMA Clients"
+
+config NET_DMA
+ bool "Network: TCP receive copy offload"
+ depends on DMA_ENGINE && NET
+ default y
+ ---help---
+ This enables the use of DMA engines in the network stack to
+ offload receive copy-to-user operations, freeing CPU cycles.
+ Since this is the main user of the DMA engine, it should be enabled;
+ say Y here.
+
comment "DMA Devices"
config INTEL_IOATDMA
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 950dc55..25d8610 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -37,6 +37,9 @@
#include <linux/config.h>
#include <linux/device.h>
#include <linux/percpu.h>
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+#endif
struct divert_blk;
struct vlan_group;
@@ -592,6 +595,9 @@ struct softnet_data
struct sk_buff *completion_queue;
struct net_device backlog_dev; /* Sorry. 8) */
+#ifdef CONFIG_NET_DMA
+ struct dma_chan *net_dma;
+#endif
};
DECLARE_PER_CPU(struct softnet_data,softnet_data);
diff --git a/include/net/netdma.h b/include/net/netdma.h
new file mode 100644
index 0000000..6435aef
--- /dev/null
+++ b/include/net/netdma.h
@@ -0,0 +1,37 @@
+/*****************************************************************************
+Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
+
+This program is free software; you can redistribute it and/or modify it
+under the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 2 of the License, or (at your option)
+any later version.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59
+Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+
+The full GNU General Public License is included in this distribution in the
+file called LICENSE.
+*****************************************************************************/
+#ifndef NETDMA_H
+#define NETDMA_H
+#ifdef CONFIG_NET_DMA
+#include <linux/dmaengine.h>
+
+static inline struct dma_chan *get_softnet_dma(void)
+{
+ struct dma_chan *chan;
+ rcu_read_lock();
+ chan = rcu_dereference(__get_cpu_var(softnet_data.net_dma));
+ if (chan)
+ dma_chan_get(chan);
+ rcu_read_unlock();
+ return chan;
+}
+#endif /* CONFIG_NET_DMA */
+#endif /* NETDMA_H */
diff --git a/net/core/dev.c b/net/core/dev.c
index f7f6f99..d7e61b4 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -115,6 +115,7 @@
#include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
#include <net/iw_handler.h>
#endif /* CONFIG_NET_RADIO */
+#include <linux/dmaengine.h>
#include <asm/current.h>
/*
@@ -149,6 +150,11 @@ static DEFINE_SPINLOCK(ptype_lock);
static struct list_head ptype_base[16]; /* 16 way hashed list */
static struct list_head ptype_all; /* Taps */
+#ifdef CONFIG_NET_DMA
+static struct dma_client *net_dma_client;
+static unsigned int net_dma_count;
+#endif
+
/*
* The @dev_base list is protected by @dev_base_lock and the rtln
* semaphore.
@@ -1750,6 +1756,19 @@ static void net_rx_action(struct softirq
}
}
out:
+#ifdef CONFIG_NET_DMA
+ /*
+ * There may not be any more sk_buffs coming right now, so push
+ * any pending DMA copies to hardware
+ */
+ if (net_dma_client) {
+ struct dma_chan *chan;
+ rcu_read_lock();
+ list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
+ dma_async_memcpy_issue_pending(chan);
+ rcu_read_unlock();
+ }
+#endif
local_irq_enable();
return;
@@ -3205,6 +3224,85 @@ static int dev_cpu_callback(struct notif
}
#endif /* CONFIG_HOTPLUG_CPU */
+#ifdef CONFIG_NET_DMA
+/**
+ * net_dma_rebalance -
+ * This is called when the number of channels allocated to the net_dma_client
+ * changes. The net_dma_client tries to have one DMA channel per CPU.
+ */
+static void net_dma_rebalance(void)
+{
+ unsigned int cpu, i, n;
+ struct dma_chan *chan;
+
+ lock_cpu_hotplug();
+
+ if (net_dma_count == 0) {
+ for_each_online_cpu(cpu)
+ rcu_assign_pointer(per_cpu(softnet_data.net_dma, cpu), NULL);
+ unlock_cpu_hotplug();
+ return;
+ }
+
+ i = 0;
+ cpu = first_cpu(cpu_online_map);
+
+ rcu_read_lock();
+ list_for_each_entry(chan, &net_dma_client->channels, client_node) {
+ n = ((num_online_cpus() / net_dma_count)
+ + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
+
+ while(n) {
+ per_cpu(softnet_data.net_dma, cpu) = chan;
+ cpu = next_cpu(cpu, cpu_online_map);
+ n--;
+ }
+ i++;
+ }
+ rcu_read_unlock();
+
+ unlock_cpu_hotplug();
+}
+
+/**
+ * netdev_dma_event - event callback for the net_dma_client
+ * @client: should always be net_dma_client
+ * @chan:
+ * @event:
+ */
+static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
+ enum dma_event event)
+{
+ switch (event) {
+ case DMA_RESOURCE_ADDED:
+ net_dma_count++;
+ net_dma_rebalance();
+ break;
+ case DMA_RESOURCE_REMOVED:
+ net_dma_count--;
+ net_dma_rebalance();
+ break;
+ default:
+ break;
+ }
+}
+
+/**
+ * netdev_dma_regiser - register the networking subsystem as a DMA client
+ */
+static int __init netdev_dma_register(void)
+{
+ net_dma_client = dma_async_client_register(netdev_dma_event);
+ if (net_dma_client == NULL)
+ return -ENOMEM;
+
+ dma_async_client_chan_request(net_dma_client, num_online_cpus());
+ return 0;
+}
+
+#else
+static int __init netdev_dma_register(void) { return -ENODEV; }
+#endif /* CONFIG_NET_DMA */
/*
* Initialize the DEV module. At boot time this walks the device list and
@@ -3258,6 +3356,8 @@ static int __init net_dev_init(void)
atomic_set(&queue->backlog_dev.refcnt, 1);
}
+ netdev_dma_register();
+
dev_boot_phase = 0;
open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
^ permalink raw reply related
* [PATCH 1/8] [I/OAT] DMA memcpy subsystem
From: Chris Leech @ 2006-03-11 2:29 UTC (permalink / raw)
To: linux-kernel, netdev
In-Reply-To: <20060311022759.3950.58788.stgit@gitlost.site>
Provides an API for offloading memory copies to DMA devices
Signed-off-by: Chris Leech <christopher.leech@intel.com>
---
drivers/Kconfig | 2
drivers/Makefile | 1
drivers/dma/Kconfig | 13 ++
drivers/dma/Makefile | 1
drivers/dma/dmaengine.c | 360 +++++++++++++++++++++++++++++++++++++++++++++
include/linux/dmaengine.h | 323 ++++++++++++++++++++++++++++++++++++++++
6 files changed, 700 insertions(+), 0 deletions(-)
diff --git a/drivers/Kconfig b/drivers/Kconfig
index bddf431..ce7ffa7 100644
--- a/drivers/Kconfig
+++ b/drivers/Kconfig
@@ -70,4 +70,6 @@ source "drivers/sn/Kconfig"
source "drivers/edac/Kconfig"
+source "drivers/dma/Kconfig"
+
endmenu
diff --git a/drivers/Makefile b/drivers/Makefile
index 5c69b86..516ba5e 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -73,3 +73,4 @@ obj-$(CONFIG_SGI_SN) += sn/
obj-y += firmware/
obj-$(CONFIG_CRYPTO) += crypto/
obj-$(CONFIG_SUPERH) += sh/
+obj-$(CONFIG_DMA_ENGINE) += dma/
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
new file mode 100644
index 0000000..f9ac4bc
--- /dev/null
+++ b/drivers/dma/Kconfig
@@ -0,0 +1,13 @@
+#
+# DMA engine configuration
+#
+
+menu "DMA Engine support"
+
+config DMA_ENGINE
+ bool "Support for DMA engines"
+ ---help---
+ DMA engines offload copy operations from the CPU to dedicated
+ hardware, allowing the copies to happen asynchronously.
+
+endmenu
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
new file mode 100644
index 0000000..10b7391
--- /dev/null
+++ b/drivers/dma/Makefile
@@ -0,0 +1 @@
+obj-y += dmaengine.o
diff --git a/drivers/dma/dmaengine.c b/drivers/dma/dmaengine.c
new file mode 100644
index 0000000..35a63d8
--- /dev/null
+++ b/drivers/dma/dmaengine.c
@@ -0,0 +1,360 @@
+/*****************************************************************************
+Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
+
+This program is free software; you can redistribute it and/or modify it
+under the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 2 of the License, or (at your option)
+any later version.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59
+Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+
+The full GNU General Public License is included in this distribution in the
+file called LICENSE.
+*****************************************************************************/
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/dmaengine.h>
+#include <linux/hardirq.h>
+#include <linux/spinlock.h>
+#include <linux/percpu.h>
+#include <linux/rcupdate.h>
+
+static DEFINE_SPINLOCK(dma_list_lock);
+static LIST_HEAD(dma_device_list);
+static LIST_HEAD(dma_client_list);
+
+/* --- sysfs implementation --- */
+
+static ssize_t show_memcpy_count(struct class_device *cd, char *buf)
+{
+ struct dma_chan *chan = container_of(cd, struct dma_chan, class_dev);
+ unsigned long count = 0;
+ int i;
+
+ for_each_cpu(i)
+ count += per_cpu_ptr(chan->local, i)->memcpy_count;
+
+ return sprintf(buf, "%lu\n", count);
+}
+
+static ssize_t show_bytes_transferred(struct class_device *cd, char *buf)
+{
+ struct dma_chan *chan = container_of(cd, struct dma_chan, class_dev);
+ unsigned long count = 0;
+ int i;
+
+ for_each_cpu(i)
+ count += per_cpu_ptr(chan->local, i)->bytes_transferred;
+
+ return sprintf(buf, "%lu\n", count);
+}
+
+static ssize_t show_in_use(struct class_device *cd, char *buf)
+{
+ struct dma_chan *chan = container_of(cd, struct dma_chan, class_dev);
+
+ return sprintf(buf, "%d\n", (chan->client ? 1 : 0));
+}
+
+static struct class_device_attribute dma_class_attrs[] = {
+ __ATTR(memcpy_count, S_IRUGO, show_memcpy_count, NULL),
+ __ATTR(bytes_transferred, S_IRUGO, show_bytes_transferred, NULL),
+ __ATTR(in_use, S_IRUGO, show_in_use, NULL),
+ __ATTR_NULL
+};
+
+static void dma_async_device_cleanup(struct kref *kref);
+
+static void dma_class_dev_release(struct class_device *cd)
+{
+ struct dma_chan *chan = container_of(cd, struct dma_chan, class_dev);
+ kref_put(&chan->device->refcount, dma_async_device_cleanup);
+}
+
+static struct class dma_devclass = {
+ .name = "dma",
+ .class_dev_attrs = dma_class_attrs,
+ .release = dma_class_dev_release,
+};
+
+/* --- client and device registration --- */
+
+/**
+ * dma_client_chan_alloc - try to allocate a channel to a client
+ * @client: &dma_client
+ *
+ * Called with dma_list_lock held.
+ */
+static struct dma_chan *dma_client_chan_alloc(struct dma_client *client)
+{
+ struct dma_device *device;
+ struct dma_chan *chan;
+ unsigned long flags;
+
+ /* Find a channel, any DMA engine will do */
+ list_for_each_entry(device, &dma_device_list, global_node) {
+ list_for_each_entry(chan, &device->channels, device_node) {
+ if (chan->client)
+ continue;
+
+ if (chan->device->device_alloc_chan_resources(chan) >= 0) {
+ kref_get(&device->refcount);
+ kref_init(&chan->refcount);
+ chan->slow_ref = 0;
+ INIT_RCU_HEAD(&chan->rcu);
+ chan->client = client;
+ spin_lock_irqsave(&client->lock, flags);
+ list_add_tail_rcu(&chan->client_node, &client->channels);
+ spin_unlock_irqrestore(&client->lock, flags);
+ return chan;
+ }
+ }
+ }
+
+ return NULL;
+}
+
+/**
+ * dma_client_chan_free - release a DMA channel
+ * @chan: &dma_chan
+ */
+void dma_async_device_cleanup(struct kref *kref);
+void dma_chan_cleanup(struct kref *kref)
+{
+ struct dma_chan *chan = container_of(kref, struct dma_chan, refcount);
+ chan->device->device_free_chan_resources(chan);
+ chan->client = NULL;
+ kref_put(&chan->device->refcount, dma_async_device_cleanup);
+}
+
+static void dma_chan_free_rcu(struct rcu_head *rcu)
+{
+ struct dma_chan *chan = container_of(rcu, struct dma_chan, rcu);
+ int bias = 0x7FFFFFFF;
+ int i;
+ for_each_cpu(i)
+ bias -= local_read(&per_cpu_ptr(chan->local, i)->refcount);
+ atomic_sub(bias, &chan->refcount.refcount);
+ kref_put(&chan->refcount, dma_chan_cleanup);
+}
+
+static void dma_client_chan_free(struct dma_chan *chan)
+{
+ atomic_add(0x7FFFFFFF, &chan->refcount.refcount);
+ chan->slow_ref = 1;
+ call_rcu(&chan->rcu, dma_chan_free_rcu);
+}
+
+/**
+ * dma_chans_rebalance - reallocate channels to clients
+ *
+ * When the number of DMA channel in the system changes,
+ * channels need to be rebalanced among clients
+ */
+static void dma_chans_rebalance(void)
+{
+ struct dma_client *client;
+ struct dma_chan *chan;
+ unsigned long flags;
+
+ spin_lock(&dma_list_lock);
+ list_for_each_entry(client, &dma_client_list, global_node) {
+
+ while (client->chans_desired > client->chan_count) {
+ chan = dma_client_chan_alloc(client);
+ if (!chan)
+ break;
+
+ client->chan_count++;
+ client->event_callback(client, chan, DMA_RESOURCE_ADDED);
+ }
+
+ while (client->chans_desired < client->chan_count) {
+ spin_lock_irqsave(&client->lock, flags);
+ chan = list_entry(client->channels.next, struct dma_chan, client_node);
+ list_del_rcu(&chan->client_node);
+ spin_unlock_irqrestore(&client->lock, flags);
+ client->chan_count--;
+ client->event_callback(client, chan, DMA_RESOURCE_REMOVED);
+ dma_client_chan_free(chan);
+ }
+ }
+ spin_unlock(&dma_list_lock);
+}
+
+/**
+ * dma_async_client_register - allocate and register a &dma_client
+ * @event_callback: callback for notification of channel addition/removal
+ */
+struct dma_client *dma_async_client_register(dma_event_callback event_callback)
+{
+ struct dma_client *client;
+
+ client = kzalloc(sizeof(*client), GFP_KERNEL);
+ if (!client)
+ return NULL;
+
+ INIT_LIST_HEAD(&client->channels);
+ spin_lock_init(&client->lock);
+
+ client->chans_desired = 0;
+ client->chan_count = 0;
+ client->event_callback = event_callback;
+
+ spin_lock(&dma_list_lock);
+ list_add_tail(&client->global_node, &dma_client_list);
+ spin_unlock(&dma_list_lock);
+
+ return client;
+}
+
+/**
+ * dma_async_client_unregister - unregister a client and free the &dma_client
+ * @client:
+ *
+ * Force frees any allocated DMA channels, frees the &dma_client memory
+ */
+void dma_async_client_unregister(struct dma_client *client)
+{
+ struct dma_chan *chan;
+
+ if (!client)
+ return;
+
+ rcu_read_lock();
+ list_for_each_entry_rcu(chan, &client->channels, client_node) {
+ dma_client_chan_free(chan);
+ }
+ rcu_read_unlock();
+
+ spin_lock(&dma_list_lock);
+ list_del(&client->global_node);
+ spin_unlock(&dma_list_lock);
+
+ kfree(client);
+ dma_chans_rebalance();
+}
+
+/**
+ * dma_async_client_chan_request - request DMA channels
+ * @client: &dma_client
+ * @number: count of DMA channels requested
+ *
+ * Clients call dma_async_client_chan_request() to specify how many
+ * DMA channels they need, 0 to free all currently allocated.
+ * The resulting allocations/frees are indicated to the client via the
+ * event callback.
+ */
+void dma_async_client_chan_request(struct dma_client *client,
+ unsigned int number)
+{
+ client->chans_desired = number;
+ dma_chans_rebalance();
+}
+
+/**
+ * dma_async_device_register -
+ * @device: &dma_device
+ */
+int dma_async_device_register(struct dma_device *device)
+{
+ static int id;
+ int chancnt = 0;
+ struct dma_chan* chan;
+
+ if (!device)
+ return -ENODEV;
+
+ init_completion(&device->done);
+ kref_init(&device->refcount);
+ device->dev_id = id++;
+
+ /* represent channels in sysfs. Probably want devs too */
+ list_for_each_entry(chan, &device->channels, device_node) {
+ chan->local = alloc_percpu(typeof(*chan->local));
+ if (chan->local == NULL)
+ continue;
+
+ chan->chan_id = chancnt++;
+ chan->class_dev.class = &dma_devclass;
+ chan->class_dev.dev = NULL;
+ snprintf(chan->class_dev.class_id, BUS_ID_SIZE, "dma%dchan%d",
+ device->dev_id, chan->chan_id);
+
+ kref_get(&device->refcount);
+ class_device_register(&chan->class_dev);
+ }
+
+ spin_lock(&dma_list_lock);
+ list_add_tail(&device->global_node, &dma_device_list);
+ spin_unlock(&dma_list_lock);
+
+ dma_chans_rebalance();
+
+ return 0;
+}
+
+/**
+ * dma_async_device_unregister -
+ * @device: &dma_device
+ */
+static void dma_async_device_cleanup(struct kref *kref)
+{
+ struct dma_device *device = container_of(kref, struct dma_device, refcount);
+ complete(&device->done);
+}
+
+void dma_async_device_unregister(struct dma_device* device)
+{
+ struct dma_chan *chan;
+ unsigned long flags;
+
+ spin_lock(&dma_list_lock);
+ list_del(&device->global_node);
+ spin_unlock(&dma_list_lock);
+
+ list_for_each_entry(chan, &device->channels, device_node) {
+ if (chan->client) {
+ spin_lock_irqsave(&chan->client->lock, flags);
+ list_del(&chan->client_node);
+ chan->client->chan_count--;
+ spin_unlock_irqrestore(&chan->client->lock, flags);
+ chan->client->event_callback(chan->client, chan, DMA_RESOURCE_REMOVED);
+ dma_client_chan_free(chan);
+ }
+ class_device_unregister(&chan->class_dev);
+ }
+
+ dma_chans_rebalance();
+
+ kref_put(&device->refcount, dma_async_device_cleanup);
+ wait_for_completion(&device->done);
+}
+
+static int __init dma_bus_init(void)
+{
+ spin_lock_init(&dma_list_lock);
+
+ return class_register(&dma_devclass);
+}
+
+subsys_initcall(dma_bus_init);
+
+EXPORT_SYMBOL(dma_async_client_register);
+EXPORT_SYMBOL(dma_async_client_unregister);
+EXPORT_SYMBOL(dma_async_client_chan_request);
+EXPORT_SYMBOL(dma_async_memcpy_buf_to_buf);
+EXPORT_SYMBOL(dma_async_memcpy_buf_to_pg);
+EXPORT_SYMBOL(dma_async_memcpy_pg_to_pg);
+EXPORT_SYMBOL(dma_async_memcpy_complete);
+EXPORT_SYMBOL(dma_async_memcpy_issue_pending);
+EXPORT_SYMBOL(dma_async_device_register);
+EXPORT_SYMBOL(dma_async_device_unregister);
diff --git a/include/linux/dmaengine.h b/include/linux/dmaengine.h
new file mode 100644
index 0000000..ac3bff9
--- /dev/null
+++ b/include/linux/dmaengine.h
@@ -0,0 +1,323 @@
+/*****************************************************************************
+Copyright(c) 2004 - 2006 Intel Corporation. All rights reserved.
+
+This program is free software; you can redistribute it and/or modify it
+under the terms of the GNU General Public License as published by the Free
+Software Foundation; either version 2 of the License, or (at your option)
+any later version.
+
+This program is distributed in the hope that it will be useful, but WITHOUT
+ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+more details.
+
+You should have received a copy of the GNU General Public License along with
+this program; if not, write to the Free Software Foundation, Inc., 59
+Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+
+The full GNU General Public License is included in this distribution in the
+file called LICENSE.
+*****************************************************************************/
+#ifndef DMAENGINE_H
+#define DMAENGINE_H
+#ifdef CONFIG_DMA_ENGINE
+
+#include <linux/device.h>
+#include <linux/uio.h>
+#include <linux/kref.h>
+#include <linux/completion.h>
+#include <linux/rcupdate.h>
+
+/**
+ * enum dma_event - resource PNP/power managment events
+ * @DMA_RESOURCE_SUSPEND: DMA device going into low power state
+ * @DMA_RESOURCE_RESUME: DMA device returning to full power
+ * @DMA_RESOURCE_ADDED: DMA device added to the system
+ * @DMA_RESOURCE_REMOVED: DMA device removed from the system
+ */
+enum dma_event {
+ DMA_RESOURCE_SUSPEND,
+ DMA_RESOURCE_RESUME,
+ DMA_RESOURCE_ADDED,
+ DMA_RESOURCE_REMOVED,
+};
+
+/**
+ * typedef dma_cookie_t
+ *
+ * if dma_cookie_t is >0 it's a DMA request cookie, <0 it's an error code
+ */
+typedef s32 dma_cookie_t;
+
+#define dma_submit_error(cookie) ((cookie) < 0 ? 1 : 0)
+
+/**
+ * enum dma_status - DMA transaction status
+ * @DMA_SUCCESS: transaction completed successfully
+ * @DMA_IN_PROGRESS: transaction not yet processed
+ * @DMA_ERROR: transaction failed
+ */
+enum dma_status {
+ DMA_SUCCESS,
+ DMA_IN_PROGRESS,
+ DMA_ERROR,
+};
+
+struct dma_chan_percpu {
+ local_t refcount;
+ /* stats */
+ unsigned long memcpy_count;
+ unsigned long bytes_transferred;
+};
+
+/**
+ * struct dma_chan - devices supply DMA channels, clients use them
+ * @client: ptr to the client user of this chan, will be NULL when unused
+ * @device: ptr to the dma device who supplies this channel, always !NULL
+ * @cookie: last cookie value returned to client
+ * @chan_id:
+ * @class_dev:
+ * @client_node: used to add this to the client chan list
+ * @device_node: used to add this to the device chan list
+ */
+struct dma_chan {
+ struct dma_client *client;
+ struct dma_device *device;
+ dma_cookie_t cookie;
+
+ /* sysfs */
+ int chan_id;
+ struct class_device class_dev;
+
+ struct kref refcount;
+ int slow_ref;
+ struct rcu_head rcu;
+
+ struct list_head client_node;
+ struct list_head device_node;
+ struct dma_chan_percpu *local;
+};
+
+void dma_chan_cleanup(struct kref *kref);
+
+static inline void dma_chan_get(struct dma_chan *chan)
+{
+ if (unlikely(chan->slow_ref))
+ kref_get(&chan->refcount);
+ else {
+ local_inc(&(per_cpu_ptr(chan->local, get_cpu())->refcount));
+ put_cpu();
+ }
+}
+
+static inline void dma_chan_put(struct dma_chan *chan)
+{
+ if (unlikely(chan->slow_ref))
+ kref_put(&chan->refcount, dma_chan_cleanup);
+ else {
+ local_dec(&(per_cpu_ptr(chan->local, get_cpu())->refcount));
+ put_cpu();
+ }
+}
+
+/*
+ * typedef dma_event_callback - function pointer to a DMA event callback
+ */
+typedef void (*dma_event_callback) (struct dma_client *client,
+ struct dma_chan *chan, enum dma_event event);
+
+/**
+ * struct dma_client - info on the entity making use of DMA services
+ * @event_callback: func ptr to call when something happens
+ * @chan_count: number of chans allocated
+ * @chans_desired: number of chans requested. Can be +/- chan_count
+ * @lock: protects access to the channels list
+ * @channels: the list of DMA channels allocated
+ * @global_node: list_head for global dma_client_list
+ */
+struct dma_client {
+ dma_event_callback event_callback;
+ unsigned int chan_count;
+ unsigned int chans_desired;
+
+ spinlock_t lock;
+ struct list_head channels;
+ struct list_head global_node;
+};
+
+/**
+ * struct dma_device - info on the entity supplying DMA services
+ * @chancnt: how many DMA channels are supported
+ * @channels: the list of struct dma_chan
+ * @global_node: list_head for global dma_device_list
+ * @dev_id:
+ * Other func ptrs: used to make use of this device's capabilities
+ */
+struct dma_device {
+
+ unsigned int chancnt;
+ struct list_head channels;
+ struct list_head global_node;
+
+ struct kref refcount;
+ struct completion done;
+
+ int dev_id;
+
+ int (*device_alloc_chan_resources)(struct dma_chan *chan);
+ void (*device_free_chan_resources)(struct dma_chan *chan);
+ dma_cookie_t (*device_memcpy_buf_to_buf)(struct dma_chan *chan,
+ void *dest, void *src, size_t len);
+ dma_cookie_t (*device_memcpy_buf_to_pg)(struct dma_chan *chan,
+ struct page *page, unsigned int offset, void *kdata,
+ size_t len);
+ dma_cookie_t (*device_memcpy_pg_to_pg)(struct dma_chan *chan,
+ struct page *dest_pg, unsigned int dest_off,
+ struct page *src_pg, unsigned int src_off, size_t len);
+ enum dma_status (*device_memcpy_complete)(struct dma_chan *chan,
+ dma_cookie_t cookie, dma_cookie_t *last,
+ dma_cookie_t *used);
+ void (*device_memcpy_issue_pending)(struct dma_chan *chan);
+};
+
+/* --- public DMA engine API --- */
+
+struct dma_client *dma_async_client_register(dma_event_callback event_callback);
+void dma_async_client_unregister(struct dma_client *client);
+void dma_async_client_chan_request(struct dma_client *client,
+ unsigned int number);
+
+/**
+ * dma_async_memcpy_buf_to_buf - offloaded copy between virtual addresses
+ * @chan: DMA channel to offload copy to
+ * @dest: destination address (virtual)
+ * @src: source address (virtual)
+ * @len: length
+ *
+ * Both @dest and @src must be mappable to a bus address according to the
+ * DMA mapping API rules for streaming mappings.
+ * Both @dest and @src must stay memory resident (kernel memory or locked
+ * user space pages)
+ */
+static inline dma_cookie_t dma_async_memcpy_buf_to_buf(struct dma_chan *chan,
+ void *dest, void *src, size_t len)
+{
+ int cpu = get_cpu();
+ per_cpu_ptr(chan->local, cpu)->bytes_transferred += len;
+ per_cpu_ptr(chan->local, cpu)->memcpy_count++;
+ put_cpu();
+
+ return chan->device->device_memcpy_buf_to_buf(chan, dest, src, len);
+}
+
+/**
+ * dma_async_memcpy_buf_to_pg - offloaded copy
+ * @chan: DMA channel to offload copy to
+ * @page: destination page
+ * @offset: offset in page to copy to
+ * @kdata: source address (virtual)
+ * @len: length
+ *
+ * Both @page/@offset and @kdata must be mappable to a bus address according
+ * to the DMA mapping API rules for streaming mappings.
+ * Both @page/@offset and @kdata must stay memory resident (kernel memory or
+ * locked user space pages)
+ */
+static inline dma_cookie_t dma_async_memcpy_buf_to_pg(struct dma_chan *chan,
+ struct page *page, unsigned int offset, void *kdata, size_t len)
+{
+ int cpu = get_cpu();
+ per_cpu_ptr(chan->local, cpu)->bytes_transferred += len;
+ per_cpu_ptr(chan->local, cpu)->memcpy_count++;
+ put_cpu();
+
+ return chan->device->device_memcpy_buf_to_pg(chan, page, offset,
+ kdata, len);
+}
+
+/**
+ * dma_async_memcpy_buf_to_pg - offloaded copy
+ * @chan: DMA channel to offload copy to
+ * @dest_page: destination page
+ * @dest_off: offset in page to copy to
+ * @src_page: source page
+ * @src_off: offset in page to copy from
+ * @len: length
+ *
+ * Both @dest_page/@dest_off and @src_page/@src_off must be mappable to a bus
+ * address according to the DMA mapping API rules for streaming mappings.
+ * Both @dest_page/@dest_off and @src_page/@src_off must stay memory resident
+ * (kernel memory or locked user space pages)
+ */
+static inline dma_cookie_t dma_async_memcpy_pg_to_pg(struct dma_chan *chan,
+ struct page *dest_pg, unsigned int dest_off, struct page *src_pg,
+ unsigned int src_off, size_t len)
+{
+ int cpu = get_cpu();
+ per_cpu_ptr(chan->local, cpu)->bytes_transferred += len;
+ per_cpu_ptr(chan->local, cpu)->memcpy_count++;
+ put_cpu();
+
+ return chan->device->device_memcpy_pg_to_pg(chan, dest_pg, dest_off,
+ src_pg, src_off, len);
+}
+
+/**
+ * dma_async_memcpy_issue_pending - flush pending copies to HW
+ * @chan:
+ *
+ * This allows drivers to push copies to HW in batches,
+ * reducing MMIO writes where possible.
+ */
+static inline void dma_async_memcpy_issue_pending(struct dma_chan *chan)
+{
+ return chan->device->device_memcpy_issue_pending(chan);
+}
+
+/**
+ * dma_async_memcpy_complete - poll for transaction completion
+ * @chan: DMA channel
+ * @cookie: transaction identifier to check status of
+ * @last: returns last completed cookie, can be NULL
+ * @used: returns last issued cookie, can be NULL
+ *
+ * If @last and @used are passed in, upon return they reflect the driver
+ * internal state and can be used with dma_async_is_complete() to check
+ * the status of multiple cookies without re-checking hardware state.
+ */
+static inline enum dma_status dma_async_memcpy_complete(struct dma_chan *chan,
+ dma_cookie_t cookie, dma_cookie_t *last, dma_cookie_t *used)
+{
+ return chan->device->device_memcpy_complete(chan, cookie, last, used);
+}
+
+/**
+ * dma_async_is_complete - test a cookie against chan state
+ * @cookie: transaction identifier to test status of
+ * @last_complete: last know completed transaction
+ * @last_used: last cookie value handed out
+ *
+ * dma_async_is_complete() is used in dma_async_memcpy_complete()
+ * the test logic is seperated for lightweight testing of multiple cookies
+ */
+static inline enum dma_status dma_async_is_complete(dma_cookie_t cookie,
+ dma_cookie_t last_complete, dma_cookie_t last_used)
+{
+ if (last_complete <= last_used) {
+ if ((cookie <= last_complete) || (cookie > last_used))
+ return DMA_SUCCESS;
+ } else {
+ if ((cookie <= last_complete) && (cookie > last_used))
+ return DMA_SUCCESS;
+ }
+ return DMA_IN_PROGRESS;
+}
+
+
+/* --- DMA device --- */
+
+int dma_async_device_register(struct dma_device *device);
+void dma_async_device_unregister(struct dma_device *device);
+
+#endif /* CONFIG_DMA_ENGINE */
+#endif /* DMAENGINE_H */
^ permalink raw reply related
* [PATCH 0/8] Intel I/O Acceleration Technology (I/OAT)
From: Chris Leech @ 2006-03-11 2:27 UTC (permalink / raw)
To: linux-kernel, netdev
This patch series is the a full release of the Intel(R) I/O
Acceleration Technology (I/OAT) for Linux. It includes an in kernel API
for offloading memory copies to hardware, a driver for the I/OAT DMA memcpy
engine, and changes to the TCP stack to offload copies of received
networking data to application space.
Changes from last weeks posting:
fixed return value from sysfs show functions as suggested by Joe Perches
code style fixes suggested by Andrew Morton, David Miller, and others
renamed anything related to pinning pages from lock/locked to pin/pinned
renamed ioatdma register read/write functions with less generic names
return a pinned list from dma_pin_iovec_pages instead of passing in a
**dma_pinned_list
replaced all cb/CB symbol prefixes in ioatdma with ioat/IOAT,
CB was an abbreviation of an early code name
use set_page_dirty_lock instead of SetPageDirty pointed out by Andrew Morton
rename dma_async_try_early_copy to tcp_dma_try_early_copy and stop exporting
I'll be focusing on reducing ifdefs and adding much needed comments, with
another release early next week.
These changes apply to DaveM's net-2.6.17 tree as of commit
32639ad6b7e3da27f233c0516471f0747f1178f5 ([SPARC]: Fixup SO_*SEC values on 32-bit sparc.)
They are available to pull from
git://198.78.49.142/~cleech/linux-2.6 ioat-2.6.17
There are 8 patches in the series:
1) The memcpy offload APIs and class code
2) The Intel I/OAT DMA driver (ioatdma)
3) Core networking code to setup networking as a DMA memcpy client
4) Utility functions for sk_buff to iovec offloaded copy
5) Structure changes needed for TCP receive offload
6) Rename cleanup_rbuf to tcp_cleanup_rbuf
7) Add a sysctl to tune the minimum offloaded I/O size for TCP
8) The main TCP receive offload changes
--
Chris Leech <christopher.leech@intel.com>
I/O Acceleration Technology Software Development
LAN Access Division / Digital Enterprise Group
^ permalink raw reply
* drivers/net/chelsio/sge.c: two array overflows
From: Adrian Bunk @ 2006-03-11 1:37 UTC (permalink / raw)
To: maintainers; +Cc: jgarzik, netdev, linux-kernel
The Coverity checker spotted the following two array overflows in
drivers/net/chelsio/sge.c (in both cases, the arrays contain 3 elements):
<-- snip -->
...
static void restart_tx_queues(struct sge *sge)
{
...
sge->stats.cmdQ_restarted[3]++;
...
static int t1_sge_tx(struct sk_buff *skb, struct adapter *adapter,
unsigned int qid, struct net_device *dev)
{
...
sge->stats.cmdQ_full[3]++;
...
<-- snip -->
cu
Adrian
--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
^ permalink raw reply
* Re: [PATCH 4/6 v2] IB: address translation to map IP toIB addresses (GIDs)
From: Roland Dreier @ 2006-03-11 1:14 UTC (permalink / raw)
To: Sean Hefty; +Cc: netdev, linux-kernel, openib-general
In-Reply-To: <ORSMSX401FRaqbC8wSA0000000d@orsmsx401.amr.corp.intel.com>
The ib_addr module depends on CONFIG_INET, because it uses symbols
like arp_tbl, which are only exported if INET is enabled.
I fixed this up by creating a new (non-user-visible) config symbol to
control when ib_addr is built -- I put the following diff on top of
your patch in my tree:
diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index bdf0891..48c8bb5 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -29,6 +29,11 @@ config INFINIBAND_USER_ACCESS
libibverbs, libibcm and a hardware driver library from
<http://www.openib.org>.
+config INFINIBAND_ADDR_TRANS
+ tristate
+ depends on INFINIBAND && INET
+ default y
+
source "drivers/infiniband/hw/mthca/Kconfig"
source "drivers/infiniband/ulp/ipoib/Kconfig"
diff --git a/drivers/infiniband/core/Makefile b/drivers/infiniband/core/Makefile
index 2393e9d..935851d 100644
--- a/drivers/infiniband/core/Makefile
+++ b/drivers/infiniband/core/Makefile
@@ -1,7 +1,8 @@
obj-$(CONFIG_INFINIBAND) += ib_core.o ib_mad.o ib_sa.o \
- ib_cm.o ib_addr.o
+ ib_cm.o
obj-$(CONFIG_INFINIBAND_USER_MAD) += ib_umad.o
obj-$(CONFIG_INFINIBAND_USER_ACCESS) += ib_uverbs.o ib_ucm.o
+obj-$(CONFIG_INFINIBAND_ADDR_TRANS) += ib_addr.o
ib_core-y := packer.o ud_header.o verbs.o sysfs.o \
device.o fmr_pool.o cache.o
^ permalink raw reply related
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox