Re: [PATCH] net: via-velocity.c fix sleep-with-spinlock bug duringMTU change
From: Séguier Régis
Date: Wed Jun 18 2008 - 12:56:36 EST
Séguier Régis a écrit :
Francois Romieu a écrit :
Séguier Régis <rseguier@xxxxxxxxxxxxxx> :
[...]
The mtu change doesn't work :
Jun 16 23:22:34 apollo kernel: ------------[ cut here ]------------
Jun 16 23:22:34 apollo kernel: WARNING: at
arch/x86/kernel/pci-dma.c:482 dma_free_coherent+0x3a/0x9c()
Jun 16 23:22:34 apollo kernel: Pid: 1527, comm: ip Tainted: G
W 2.6.26-rc6EPIA_SN_VB7001 #3
Jun 16 23:22:34 apollo kernel: [<c0115fe5>] warn_on_slowpath+0x3b/0x5f
Jun 16 23:22:34 apollo kernel: [<c0133940>]
get_page_from_freelist+0x24a/0x36f
Jun 16 23:22:34 apollo kernel: [<c012d86b>] handle_IRQ_event+0x1a/0x3f
Jun 16 23:22:34 apollo kernel: [<c0241193>]
velocity_free_rd_ring+0xa5/0xb6
Jun 16 23:22:34 apollo kernel: [<c0144752>] kfree+0x6a/0x72
Jun 16 23:22:34 apollo kernel: [<c0241193>]
velocity_free_rd_ring+0xa5/0xb6
Jun 16 23:22:34 apollo kernel: [<c0105816>]
dma_free_coherent+0x3a/0x9c
Jun 16 23:22:34 apollo kernel: [<c02411eb>]
velocity_free_dma_rings+0x47/0x4d
Jun 16 23:22:34 apollo kernel: [<c024225b>]
velocity_change_mtu+0xf6/0x157
Jun 16 23:22:34 apollo kernel: [<c02856c4>] dev_set_mtu+0x2a/0x4f
Jun 16 23:22:34 apollo kernel: [<c0286642>] dev_ioctl+0x4ab/0x530
Jun 16 23:22:34 apollo kernel: [<c012e4cb>]
handle_fasteoi_irq+0x74/0x77
Jun 16 23:22:34 apollo kernel: [<c010479b>] do_IRQ+0x50/0x60
Jun 16 23:22:34 apollo kernel: [<c027d785>] sock_ioctl+0x0/0x177
Does the patch below make a difference ?
diff --git a/drivers/net/via-velocity.c b/drivers/net/via-velocity.c
index 71a5133..fa303da 100644
--- a/drivers/net/via-velocity.c
+++ b/drivers/net/via-velocity.c
@@ -1274,7 +1274,7 @@ static void velocity_free_rd_ring(struct
velocity_info *vptr)
PCI_DMA_FROMDEVICE);
rd_info->skb_dma = (dma_addr_t) NULL;
- dev_kfree_skb(rd_info->skb);
+ dev_kfree_skb_any(rd_info->skb);
rd_info->skb = NULL;
}
@@ -1336,7 +1336,7 @@ static void velocity_free_td_ring_entry(struct
velocity_info *vptr,
td_info->skb_dma[i] = (dma_addr_t) NULL;
}
}
- dev_kfree_skb(td_info->skb);
+ dev_kfree_skb_any(td_info->skb);
td_info->skb = NULL;
}
}
Idem.
Jun 18 01:30:15 apollo kernel: ------------[ cut here ]------------
Jun 18 01:30:15 apollo kernel: WARNING: at
arch/x86/kernel/pci-dma.c:482 dma_free_coherent+0x3a/0x9c()
Jun 18 01:30:15 apollo kernel: Pid: 1401, comm: ip Not tainted
2.6.26-rc6EPIA_SN_VB7001 #4
Jun 18 01:30:15 apollo kernel: [<c0115fe5>] warn_on_slowpath+0x3b/0x5f
Jun 18 01:30:15 apollo kernel: [<c0133b19>]
__alloc_pages_internal+0xb4/0x349
Jun 18 01:30:15 apollo kernel: [<c0112bfd>] hrtick_start_fair+0x67/0xfd
Jun 18 01:30:15 apollo kernel: [<c011414f>]
check_preempt_wakeup+0x9d/0xbb
Jun 18 01:30:15 apollo kernel: [<c0144752>] kfree+0x6a/0x72
Jun 18 01:30:15 apollo kernel: [<c0241193>]
velocity_free_rd_ring+0xa5/0xb6
Jun 18 01:30:15 apollo kernel: [<c0105816>] dma_free_coherent+0x3a/0x9c
Jun 18 01:30:15 apollo kernel: [<c02411eb>]
velocity_free_dma_rings+0x47/0x4d
Jun 18 01:30:15 apollo kernel: [<c024225b>]
velocity_change_mtu+0xf6/0x157
Jun 18 01:30:15 apollo kernel: [<c02856c4>] dev_set_mtu+0x2a/0x4f
Jun 18 01:30:15 apollo kernel: [<c0286642>] dev_ioctl+0x4ab/0x530
Jun 18 01:30:15 apollo kernel: [<c013964a>] __do_fault+0x256/0x28e
Jun 18 01:30:15 apollo kernel: [<c027d8d7>] sock_ioctl+0x152/0x177
Jun 18 01:30:15 apollo kernel: [<c027d785>] sock_ioctl+0x0/0x177
Jun 18 01:30:15 apollo kernel: [<c0150b4a>] vfs_ioctl+0x16/0x48
Jun 18 01:30:15 apollo kernel: [<c0150d5a>] do_vfs_ioctl+0x1de/0x1f1
Jun 18 01:30:15 apollo kernel: [<c0150dae>] sys_ioctl+0x41/0x5b
Jun 18 01:30:15 apollo kernel: [<c010291d>] sysenter_past_esp+0x6a/0x91
Jun 18 01:30:15 apollo kernel: =======================
Jun 18 01:30:15 apollo kernel: ---[ end trace abc2c54f7fac91dc ]---
This correction seems to work.
I read somewhere pci_free_consistent (in velocity_free_dma_rings)
couldn't be use in a spin_lock.
When I move velocity_free_rings outside the spin_lock, error desappeare
but data transfert doesn't work after the mtu change.
When I add velocity_give_many_rx_descs all work.
I will make some more tests.
--- drivers/net/via-velocity.c 2008-06-18 20:17:18.000000000 +0200
+++ ../linux-2.6.26-rc6-mtu/drivers/net/via-velocity.c 2008-06-18
20:28:14.000000000 +0200
@@ -1968,6 +1968,8 @@ static int velocity_change_mtu(struct ne
if (dev->mtu != new_mtu) {
struct velocity_info *tmp_vptr;
unsigned long flags;
+ struct tx_info tx;
+ struct rx_info rx;
tmp_vptr = kzalloc(sizeof(*tmp_vptr), GFP_KERNEL);
if (!tmp_vptr) {
@@ -1989,13 +1991,16 @@ static int velocity_change_mtu(struct ne
netif_stop_queue(dev);
velocity_shutdown(vptr);
- velocity_free_rings(vptr);
-
+ rx=vptr->rx;
+ tx=vptr->tx;
vptr->rx = tmp_vptr->rx;
vptr->tx = tmp_vptr->tx;
+ tmp_vptr->rx=rx;
+ tmp_vptr->tx=tx;
dev->mtu = new_mtu;
+ velocity_give_many_rx_descs(vptr);
velocity_init_registers(vptr, VELOCITY_INIT_COLD);
mac_enable_int(vptr->mac_regs);
@@ -2003,6 +2008,7 @@ static int velocity_change_mtu(struct ne
spin_unlock_irqrestore(&vptr->lock, flags);
+ velocity_free_rings(tmp_vptr);
out_free_tmp_vptr_1:
kfree(tmp_vptr);
}
--
Régis
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/