Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit e407a7f6 authored by Ed L. Cashin's avatar Ed L. Cashin Committed by Greg Kroah-Hartman
Browse files

aoe: zero copy write 1 of 2 

Avoid memory copy on writes.
(This patch depends on fixes in patch 9 to follow.)

Although skb->len should not be set when working with linear skbuffs,
the skb->tail pointer maintained by skb_put/skb_trim is not relevant
to what happens when the skb_fill_page_desc function is called.  This
issue was raised without comment in linux-kernel and netdev earlier
this month:

  http://thread.gmane.org/gmane.linux.kernel/446474/
  http://thread.gmane.org/gmane.linux.network/45444/



So until there is something analogous to skb_put that works for
zero-copy write skbuffs, we will do what the other callers of
skb_fill_page_desc are doing.

Signed-off-by: default avatar"Ed L. Cashin" <ecashin@coraid.com>
Acked-by: default avatarAlan Cox <alan@redhat.com>
Signed-off-by: default avatarGreg Kroah-Hartman <gregkh@suse.de>
parent 2fdc0ea7

drivers/block/aoe/aoe.h

+2 −5
Original line number Original line Diff line number Diff line
@@ -107,11 +107,7 @@ struct frame { 
	ulong waited;

	ulong waited;

	struct buf *buf;

	struct buf *buf;

	char *bufaddr;

	char *bufaddr;

	int writedatalen;

	struct sk_buff *skb;

	int ndata;



	/* largest possible */

	unsigned char data[sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr)];

};

};





struct aoedev {

struct aoedev {
@@ -157,6 +153,7 @@ void aoecmd_cfg(ushort aoemajor, unsigned char aoeminor); 
void aoecmd_ata_rsp(struct sk_buff *);

void aoecmd_ata_rsp(struct sk_buff *);

void aoecmd_cfg_rsp(struct sk_buff *);

void aoecmd_cfg_rsp(struct sk_buff *);

void aoecmd_sleepwork(void *vp);

void aoecmd_sleepwork(void *vp);

struct sk_buff *new_skb(ulong);





int aoedev_init(void);

int aoedev_init(void);

void aoedev_exit(void);

void aoedev_exit(void);

drivers/block/aoe/aoecmd.c

+37 −57
Original line number Original line Diff line number Diff line
@@ -17,15 +17,14 @@ 
#define MAXTIMER (HZ << 1)

#define MAXTIMER (HZ << 1)

#define MAXWAIT (60 * 3)	/* After MAXWAIT seconds, give up and fail dev */

#define MAXWAIT (60 * 3)	/* After MAXWAIT seconds, give up and fail dev */





static struct sk_buff *

struct sk_buff *

new_skb(struct net_device *if_dev, ulong len)

new_skb(ulong len)

{

{

	struct sk_buff *skb;

	struct sk_buff *skb;





	skb = alloc_skb(len, GFP_ATOMIC);

	skb = alloc_skb(len, GFP_ATOMIC);

	if (skb) {

	if (skb) {

		skb->nh.raw = skb->mac.raw = skb->data;

		skb->nh.raw = skb->mac.raw = skb->data;

		skb->dev = if_dev;

		skb->protocol = __constant_htons(ETH_P_AOE);

		skb->protocol = __constant_htons(ETH_P_AOE);

		skb->priority = 0;

		skb->priority = 0;

		skb_put(skb, len);

		skb_put(skb, len);
@@ -40,29 +39,6 @@ new_skb(struct net_device *if_dev, ulong len) 
	return skb;

	return skb;

}

}





static struct sk_buff *

skb_prepare(struct aoedev *d, struct frame *f)

{

	struct sk_buff *skb;

	char *p;



	skb = new_skb(d->ifp, f->ndata + f->writedatalen);

	if (!skb) {

		printk(KERN_INFO "aoe: skb_prepare: failure to allocate skb\n");

		return NULL;

	}



	p = skb->mac.raw;

	memcpy(p, f->data, f->ndata);



	if (f->writedatalen) {

		p += sizeof(struct aoe_hdr) + sizeof(struct aoe_atahdr);

		memcpy(p, f->bufaddr, f->writedatalen);

	}



	return skb;

}



static struct frame *

static struct frame *

getframe(struct aoedev *d, int tag)

getframe(struct aoedev *d, int tag)

{

{
@@ -129,10 +105,11 @@ aoecmd_ata_rw(struct aoedev *d, struct frame *f) 
		bcnt = MAXATADATA;

		bcnt = MAXATADATA;





	/* initialize the headers & frame */

	/* initialize the headers & frame */

	h = (struct aoe_hdr *) f->data;

	skb = f->skb;

	h = (struct aoe_hdr *) skb->mac.raw;

	ah = (struct aoe_atahdr *) (h+1);

	ah = (struct aoe_atahdr *) (h+1);

	f->ndata = sizeof *h + sizeof *ah;

	skb->len = sizeof *h + sizeof *ah;

	memset(h, 0, f->ndata);

	memset(h, 0, skb->len);

	f->tag = aoehdr_atainit(d, h);

	f->tag = aoehdr_atainit(d, h);

	f->waited = 0;

	f->waited = 0;

	f->buf = buf;

	f->buf = buf;
@@ -155,11 +132,13 @@ aoecmd_ata_rw(struct aoedev *d, struct frame *f) 
	}

	}





	if (bio_data_dir(buf->bio) == WRITE) {

	if (bio_data_dir(buf->bio) == WRITE) {

		skb_fill_page_desc(skb, 0, virt_to_page(f->bufaddr),

			offset_in_page(f->bufaddr), bcnt);

		ah->aflags |= AOEAFL_WRITE;

		ah->aflags |= AOEAFL_WRITE;

		f->writedatalen = bcnt;

	} else {

	} else {

		skb_shinfo(skb)->nr_frags = 0;

		skb->len = ETH_ZLEN;

		writebit = 0;

		writebit = 0;

		f->writedatalen = 0;

	}

	}





	ah->cmdstat = WIN_READ | writebit | extbit;

	ah->cmdstat = WIN_READ | writebit | extbit;
@@ -179,8 +158,8 @@ aoecmd_ata_rw(struct aoedev *d, struct frame *f) 
		buf->bufaddr = page_address(buf->bv->bv_page) + buf->bv->bv_offset;

		buf->bufaddr = page_address(buf->bv->bv_page) + buf->bv->bv_offset;

	}

	}





	skb = skb_prepare(d, f);

	skb->dev = d->ifp;

	if (skb) {

	skb_get(skb);

	skb->next = NULL;

	skb->next = NULL;

	if (d->sendq_hd)

	if (d->sendq_hd)

		d->sendq_tl->next = skb;

		d->sendq_tl->next = skb;
@@ -188,7 +167,6 @@ aoecmd_ata_rw(struct aoedev *d, struct frame *f) 
		d->sendq_hd = skb;

		d->sendq_hd = skb;

	d->sendq_tl = skb;

	d->sendq_tl = skb;

}

}

}





/* some callers cannot sleep, and they can call this function,

/* some callers cannot sleep, and they can call this function,

 * transmitting the packets later, when interrupts are on

 * transmitting the packets later, when interrupts are on
@@ -209,11 +187,12 @@ aoecmd_cfg_pkts(ushort aoemajor, unsigned char aoeminor, struct sk_buff **tail) 
		if (!is_aoe_netif(ifp))

		if (!is_aoe_netif(ifp))

			continue;

			continue;





		skb = new_skb(ifp, sizeof *h + sizeof *ch);

		skb = new_skb(sizeof *h + sizeof *ch);

		if (skb == NULL) {

		if (skb == NULL) {

			printk(KERN_INFO "aoe: aoecmd_cfg: skb alloc failure\n");

			printk(KERN_INFO "aoe: aoecmd_cfg: skb alloc failure\n");

			continue;

			continue;

		}

		}

		skb->dev = ifp;

		if (sl_tail == NULL)

		if (sl_tail == NULL)

			sl_tail = skb;

			sl_tail = skb;

		h = (struct aoe_hdr *) skb->mac.raw;

		h = (struct aoe_hdr *) skb->mac.raw;
@@ -283,14 +262,15 @@ rexmit(struct aoedev *d, struct frame *f) 
		d->aoemajor, d->aoeminor, f->tag, jiffies, n);

		d->aoemajor, d->aoeminor, f->tag, jiffies, n);

	aoechr_error(buf);

	aoechr_error(buf);





	h = (struct aoe_hdr *) f->data;

	skb = f->skb;

	h = (struct aoe_hdr *) skb->mac.raw;

	f->tag = n;

	f->tag = n;

	h->tag = cpu_to_be32(n);

	h->tag = cpu_to_be32(n);

	memcpy(h->dst, d->addr, sizeof h->dst);

	memcpy(h->dst, d->addr, sizeof h->dst);

	memcpy(h->src, d->ifp->dev_addr, sizeof h->src);

	memcpy(h->src, d->ifp->dev_addr, sizeof h->src);





	skb = skb_prepare(d, f);

	skb->dev = d->ifp;

	if (skb) {

	skb_get(skb);

	skb->next = NULL;

	skb->next = NULL;

	if (d->sendq_hd)

	if (d->sendq_hd)

		d->sendq_tl->next = skb;

		d->sendq_tl->next = skb;
@@ -298,7 +278,6 @@ rexmit(struct aoedev *d, struct frame *f) 
		d->sendq_hd = skb;

		d->sendq_hd = skb;

	d->sendq_tl = skb;

	d->sendq_tl = skb;

}

}

}





static int

static int

tsince(int tag)

tsince(int tag)
@@ -514,7 +493,7 @@ aoecmd_ata_rsp(struct sk_buff *skb) 
	calc_rttavg(d, tsince(f->tag));

	calc_rttavg(d, tsince(f->tag));





	ahin = (struct aoe_atahdr *) (hin+1);

	ahin = (struct aoe_atahdr *) (hin+1);

	ahout = (struct aoe_atahdr *) (f->data + sizeof(struct aoe_hdr));

	ahout = (struct aoe_atahdr *) (f->skb->mac.raw + sizeof(struct aoe_hdr));

	buf = f->buf;

	buf = f->buf;





	if (ahout->cmdstat == WIN_IDENTIFY)

	if (ahout->cmdstat == WIN_IDENTIFY)
@@ -620,20 +599,21 @@ aoecmd_ata_id(struct aoedev *d) 
	}

	}





	/* initialize the headers & frame */

	/* initialize the headers & frame */

	h = (struct aoe_hdr *) f->data;

	skb = f->skb;

	h = (struct aoe_hdr *) skb->mac.raw;

	ah = (struct aoe_atahdr *) (h+1);

	ah = (struct aoe_atahdr *) (h+1);

	f->ndata = sizeof *h + sizeof *ah;

	skb->len = sizeof *h + sizeof *ah;

	memset(h, 0, f->ndata);

	memset(h, 0, skb->len);

	f->tag = aoehdr_atainit(d, h);

	f->tag = aoehdr_atainit(d, h);

	f->waited = 0;

	f->waited = 0;

	f->writedatalen = 0;





	/* set up ata header */

	/* set up ata header */

	ah->scnt = 1;

	ah->scnt = 1;

	ah->cmdstat = WIN_IDENTIFY;

	ah->cmdstat = WIN_IDENTIFY;

	ah->lba3 = 0xa0;

	ah->lba3 = 0xa0;





	skb = skb_prepare(d, f);

	skb->dev = d->ifp;

	skb_get(skb);





	d->rttavg = MAXTIMER;

	d->rttavg = MAXTIMER;

	d->timer.function = rexmit_timer;

	d->timer.function = rexmit_timer;

drivers/block/aoe/aoedev.c

+30 −12
Original line number Original line Diff line number Diff line
@@ -63,22 +63,32 @@ aoedev_newdev(ulong nframes) 
	struct frame *f, *e;

	struct frame *f, *e;





	d = kzalloc(sizeof *d, GFP_ATOMIC);

	d = kzalloc(sizeof *d, GFP_ATOMIC);

	if (d == NULL)

		return NULL;

	f = kcalloc(nframes, sizeof *f, GFP_ATOMIC);

	f = kcalloc(nframes, sizeof *f, GFP_ATOMIC);

	if (f == NULL) {

 	switch (!d || !f) {

		kfree(d);

 	case 0:

		return NULL;

	}



	INIT_WORK(&d->work, aoecmd_sleepwork, d);



 		d->nframes = nframes;

 		d->nframes = nframes;

 		d->frames = f;

 		d->frames = f;

 		e = f + nframes;

 		e = f + nframes;

	for (; f<e; f++)

 		for (; f<e; f++) {

 			f->tag = FREETAG;

 			f->tag = FREETAG;



 			f->skb = new_skb(ETH_ZLEN);

 			if (!f->skb)

 				break;

 		}

 		if (f == e)

 			break;

 		while (f > d->frames) {

 			f--;

 			dev_kfree_skb(f->skb);

 		}

 	default:

 		if (f)

 			kfree(f);

 		if (d)

 			kfree(d);

		return NULL;

	}

	INIT_WORK(&d->work, aoecmd_sleepwork, d);

	spin_lock_init(&d->lock);

	spin_lock_init(&d->lock);

	init_timer(&d->timer);

	init_timer(&d->timer);

	d->timer.data = (ulong) d;

	d->timer.data = (ulong) d;
@@ -160,11 +170,19 @@ aoedev_by_sysminor_m(ulong sysminor, ulong bufcnt) 
static void

static void

aoedev_freedev(struct aoedev *d)

aoedev_freedev(struct aoedev *d)

{

{

	struct frame *f, *e;



	if (d->gd) {

	if (d->gd) {

		aoedisk_rm_sysfs(d);

		aoedisk_rm_sysfs(d);

		del_gendisk(d->gd);

		del_gendisk(d->gd);

		put_disk(d->gd);

		put_disk(d->gd);

	}

	}

	f = d->frames;

	e = f + d->nframes;

	for (; f<e; f++) {

		skb_shinfo(f->skb)->nr_frags = 0;

		dev_kfree_skb(f->skb);

	}

	kfree(d->frames);

	kfree(d->frames);

	if (d->bufpool)

	if (d->bufpool)

		mempool_destroy(d->bufpool);

		mempool_destroy(d->bufpool);