* [PATCH] b44: netpoll locking fix @ 2007-05-29 21:14 Francois Romieu 2007-05-29 21:27 ` Stephen Hemminger 0 siblings, 1 reply; 5+ messages in thread From: Francois Romieu @ 2007-05-29 21:14 UTC (permalink / raw) To: Gary Zambrano; +Cc: jgarzik, akpm, netdev The irq handling thread (b44_interrupt) uses the same lock as the NAPI thread. This change should prevent a deadlock if something interrupts the b44 NAPI thread and tries to printk through netconsole. Signed-off-by: Francois Romieu <romieu@fr.zoreil.com> --- drivers/net/b44.c | 8 ++++---- 1 files changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/net/b44.c b/drivers/net/b44.c index 879a2ff..64d75e4 100644 --- a/drivers/net/b44.c +++ b/drivers/net/b44.c @@ -853,16 +853,18 @@ static int b44_rx(struct b44 *bp, int budget) static int b44_poll(struct net_device *netdev, int *budget) { struct b44 *bp = netdev_priv(netdev); + unsigned long flags; int done; - spin_lock_irq(&bp->lock); + spin_lock_irqsave(&bp->lock, flags); if (bp->istat & (ISTAT_TX | ISTAT_TO)) { /* spin_lock(&bp->tx_lock); */ b44_tx(bp); /* spin_unlock(&bp->tx_lock); */ } - spin_unlock_irq(&bp->lock); + + spin_unlock_irqrestore(&bp->lock, flags); done = 1; if (bp->istat & ISTAT_RX) { @@ -882,8 +884,6 @@ static int b44_poll(struct net_device *netdev, int *budget) } if (bp->istat & ISTAT_ERRORS) { - unsigned long flags; - spin_lock_irqsave(&bp->lock, flags); b44_halt(bp); b44_init_rings(bp); -- 1.5.2 ^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH] b44: netpoll locking fix 2007-05-29 21:14 [PATCH] b44: netpoll locking fix Francois Romieu @ 2007-05-29 21:27 ` Stephen Hemminger 2007-05-29 22:20 ` Francois Romieu 0 siblings, 1 reply; 5+ messages in thread From: Stephen Hemminger @ 2007-05-29 21:27 UTC (permalink / raw) To: Francois Romieu; +Cc: Gary Zambrano, jgarzik, akpm, netdev On Tue, 29 May 2007 23:14:58 +0200 Francois Romieu <romieu@fr.zoreil.com> wrote: > The irq handling thread (b44_interrupt) uses the same lock as the NAPI > thread. This change should prevent a deadlock if something interrupts > the b44 NAPI thread and tries to printk through netconsole. > > Signed-off-by: Francois Romieu <romieu@fr.zoreil.com> Better to just get rid of using the lock as a transmit lock and use netif_tx_lock instead. --- a/drivers/net/b44.c 2007-05-29 09:51:43.000000000 -0700 +++ b/drivers/net/b44.c 2007-05-29 14:26:03.000000000 -0700 @@ -607,6 +607,7 @@ static void b44_tx(struct b44 *bp) { u32 cur, cons; + netif_tx_lock(bp->dev); cur = br32(bp, B44_DMATX_STAT) & DMATX_STAT_CDMASK; cur /= sizeof(struct dma_desc); @@ -622,7 +623,7 @@ static void b44_tx(struct b44 *bp) skb->len, PCI_DMA_TODEVICE); rp->skb = NULL; - dev_kfree_skb_irq(skb); + dev_kfree_skb_any(skb); } bp->tx_cons = cons; @@ -631,6 +632,7 @@ static void b44_tx(struct b44 *bp) netif_wake_queue(bp->dev); bw32(bp, B44_GPTIMER, 0); + netif_tx_unlock(bp->dev); } /* Works like this. This chip writes a 'struct rx_header" 30 bytes @@ -855,14 +857,8 @@ static int b44_poll(struct net_device *n struct b44 *bp = netdev_priv(netdev); int done; - spin_lock_irq(&bp->lock); - - if (bp->istat & (ISTAT_TX | ISTAT_TO)) { - /* spin_lock(&bp->tx_lock); */ + if (bp->istat & (ISTAT_TX | ISTAT_TO)) b44_tx(bp); - /* spin_unlock(&bp->tx_lock); */ - } - spin_unlock_irq(&bp->lock); done = 1; if (bp->istat & ISTAT_RX) { @@ -970,21 +966,19 @@ static int b44_start_xmit(struct sk_buff { struct b44 *bp = netdev_priv(dev); struct sk_buff *bounce_skb; - int rc = NETDEV_TX_OK; dma_addr_t mapping; u32 len, entry, ctrl; - len = skb->len; - spin_lock_irq(&bp->lock); - - /* This is a hard error, log it. */ if (unlikely(TX_BUFFS_AVAIL(bp) < 1)) { - netif_stop_queue(dev); - printk(KERN_ERR PFX "%s: BUG! Tx Ring full when queue awake!\n", - dev->name); - goto err_out; + if (!netif_queue_stopped(dev)) { + netif_stop_queue(dev); + printk(KERN_ERR PFX "%s: BUG! Tx Ring full when queue awake!\n", + dev->name); + } + return NETDEV_TX_BUSY; } + len = skb->len; mapping = pci_map_single(bp->pdev, skb->data, len, PCI_DMA_TODEVICE); if (dma_mapping_error(mapping) || mapping + len > DMA_30BIT_MASK) { /* Chip can't handle DMA to/from >1GB, use bounce buffer */ @@ -1044,16 +1038,14 @@ static int b44_start_xmit(struct sk_buff if (TX_BUFFS_AVAIL(bp) < 1) netif_stop_queue(dev); - dev->trans_start = jiffies; - -out_unlock: - spin_unlock_irq(&bp->lock); + mmiowb(); - return rc; + dev->trans_start = jiffies; + return NETDEV_TX_OK; err_out: - rc = NETDEV_TX_BUSY; - goto out_unlock; + dev_kfree_skb(skb); + return NETDEV_TX_OK; } static int b44_change_mtu(struct net_device *dev, int new_mtu) ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] b44: netpoll locking fix 2007-05-29 21:27 ` Stephen Hemminger @ 2007-05-29 22:20 ` Francois Romieu 2007-05-29 22:30 ` Stephen Hemminger 2007-05-29 23:13 ` John W. Linville 0 siblings, 2 replies; 5+ messages in thread From: Francois Romieu @ 2007-05-29 22:20 UTC (permalink / raw) To: Stephen Hemminger; +Cc: Gary Zambrano, jgarzik, akpm, netdev [-- Warning: decoded text below may be mangled, UTF-8 assumed --] [-- Attachment #1: Type: text/plain; charset=unknown-8bit, Size: 941 bytes --] Stephen Hemminger <shemminger@linux-foundation.org> : ⅜...] > Better to just get rid of using the lock as a transmit lock and > use netif_tx_lock instead. > --- a/drivers/net/b44.c 2007-05-29 09:51:43.000000000 -0700 > +++ b/drivers/net/b44.c 2007-05-29 14:26:03.000000000 -0700 > @@ -607,6 +607,7 @@ static void b44_tx(struct b44 *bp) > { > u32 cur, cons; > > + netif_tx_lock(bp->dev); > cur = br32(bp, B44_DMATX_STAT) & DMATX_STAT_CDMASK; > cur /= sizeof(struct dma_desc); > (damn, you are quick) I am not completely convinced. 1. netpoll_send_skb (calls netif_tx_trylock(dev)) -> netpoll_poll(np) -> poll_napi(np) -> np->dev->poll(np->dev, &budget) ( == b44_poll) -> b44_tx -> netif_tx_lock(bp->dev) *deadlock* 2. Hunk #5 clashes with John Linville's wireless-dev#master 3. Moderately appropriate for 2.6.22-rc (imho, fwiw, etc.) -- Ueimor ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] b44: netpoll locking fix 2007-05-29 22:20 ` Francois Romieu @ 2007-05-29 22:30 ` Stephen Hemminger 2007-05-29 23:13 ` John W. Linville 1 sibling, 0 replies; 5+ messages in thread From: Stephen Hemminger @ 2007-05-29 22:30 UTC (permalink / raw) To: Francois Romieu; +Cc: Gary Zambrano, jgarzik, akpm, netdev On Wed, 30 May 2007 00:20:41 +0200 Francois Romieu <romieu@fr.zoreil.com> wrote: > Stephen Hemminger <shemminger@linux-foundation.org> : > ⅜...] > > Better to just get rid of using the lock as a transmit lock and > > use netif_tx_lock instead. > > --- a/drivers/net/b44.c 2007-05-29 09:51:43.000000000 -0700 > > +++ b/drivers/net/b44.c 2007-05-29 14:26:03.000000000 -0700 > > @@ -607,6 +607,7 @@ static void b44_tx(struct b44 *bp) > > { > > u32 cur, cons; > > > > + netif_tx_lock(bp->dev); > > cur = br32(bp, B44_DMATX_STAT) & DMATX_STAT_CDMASK; > > cur /= sizeof(struct dma_desc); > > > > (damn, you are quick) > > I am not completely convinced. > > 1. netpoll_send_skb (calls netif_tx_trylock(dev)) > -> netpoll_poll(np) > -> poll_napi(np) > -> np->dev->poll(np->dev, &budget) ( == b44_poll) > -> b44_tx > -> netif_tx_lock(bp->dev) *deadlock* Netpoll needs to be fixed. (or scrapped), as is it will break drivers trying to use tx_lock in poll routine. I know sky2 would get borked. Something like this: --- a/net/core/netpoll.c 2007-05-08 14:19:32.000000000 -0700 +++ b/net/core/netpoll.c 2007-05-29 15:28:22.000000000 -0700 @@ -250,22 +250,23 @@ static void netpoll_send_skb(struct netp unsigned long flags; local_irq_save(flags); - if (netif_tx_trylock(dev)) { - /* try until next clock tick */ - for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; - tries > 0; --tries) { + /* try until next clock tick */ + for (tries = jiffies_to_usecs(1)/USEC_PER_POLL; + tries > 0; --tries) { + if (netif_tx_trylock(dev)) { if (!netif_queue_stopped(dev)) status = dev->hard_start_xmit(skb, dev); + netif_tx_unlock(dev); if (status == NETDEV_TX_OK) break; - /* tickle device maybe there is some cleanup */ - netpoll_poll(np); - - udelay(USEC_PER_POLL); } - netif_tx_unlock(dev); + + /* tickle device maybe there is some cleanup */ + netpoll_poll(np); + + udelay(USEC_PER_POLL); } local_irq_restore(flags); } ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH] b44: netpoll locking fix 2007-05-29 22:20 ` Francois Romieu 2007-05-29 22:30 ` Stephen Hemminger @ 2007-05-29 23:13 ` John W. Linville 1 sibling, 0 replies; 5+ messages in thread From: John W. Linville @ 2007-05-29 23:13 UTC (permalink / raw) To: Francois Romieu; +Cc: Stephen Hemminger, Gary Zambrano, jgarzik, akpm, netdev On Wed, May 30, 2007 at 12:20:41AM +0200, Francois Romieu wrote: > 2. Hunk #5 clashes with John Linville's wireless-dev#master Let's not worry too much about this fact -- I'll (make Michael) fix-up things in my tree as appropriate. :-) John -- John W. Linville linville@tuxdriver.com ^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2007-05-29 23:26 UTC | newest] Thread overview: 5+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2007-05-29 21:14 [PATCH] b44: netpoll locking fix Francois Romieu 2007-05-29 21:27 ` Stephen Hemminger 2007-05-29 22:20 ` Francois Romieu 2007-05-29 22:30 ` Stephen Hemminger 2007-05-29 23:13 ` John W. Linville
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).