lockdep: fix smc91x
Hi,
When booting using root-nfs, I'm seeing (independently) two lockdep dumps in the smc91x driver. The patch below fixes both. Both dumps look like real locking issues.
Nico - please review and ack if you think the patch is correct.
Dump 1:
Sending DHCP requests .
=================================
[ INFO: inconsistent lock state ]
---------------------------------
inconsistent {hardirq-on-W} -> {in-hardirq-W} usage. swapper/0 [HC1[1]:SC0[0]:HE0:SE1] takes:
(&lp->lock){+-..}, at: [<c0134cf8>] smc_interrupt+0x24/0x740
{hardirq-on-W} state was registered at:
[<c005d0fc>] lock_acquire+0x80/0x94 [<c01e0f70>] _spin_lock+0x30/0x40 [<c01342bc>] smc_reset+0x2c/0x140 [<c0135d74>] smc_drv_probe+0x288/0x720 [<c0131b84>] platform_drv_probe+0x20/0x24 [<c012fbc8>] driver_probe_device+0x8c/0xd4 [<c012fd48>] __driver_attach+0x88/0xe8 [<c012f000>] bus_for_each_dev+0x50/0x84 [<c012fdc8>] driver_attach+0x20/0x28 [<c012f5b8>] bus_add_driver+0x6c/0x138 [<c01302a0>] driver_register+0x90/0xa0 [<c0131c84>] platform_driver_register+0x6c/0x88 [<c0015fe4>] smc_init+0x14/0x1c [<c002019c>] init+0x140/0x318 [<c00423b8>] do_exit+0x0/0x83c irq event stamp: 19510 hardirqs last enabled at (19509): [<c0021aa4>] default_idle+0x40/0x4c hardirqs last disabled at (19510): [<c0020a68>] __irq_svc+0x28/0xc0 softirqs last enabled at (19504): [<c00454b8>] __do_softirq+0xf4/0x10c softirqs last disabled at (19495): [<c0045798>] irq_exit+0x54/0x5c
other info that might help us debug this:
no locks held by swapper/0.
stack backtrace: [<c0024ed0>] (dump_stack+0x0/0x14) from [<c005b258>] (print_usage_bug+0x19c/0x1f0) [<c005b0bc>] (print_usage_bug+0x0/0x1f0) from [<c005b3a0>] (mark_lock+0xf4/0x5b8) [<c005b2ac>] (mark_lock+0x0/0x5b8) from [<c005c358>] (__lock_acquire+0x50c/0xdd4) r8 = C021F520 r7 = C021F7DC r6 = 00000000 r5 = C02B11F4 r4 = 00000000 [<c005be4c>] (__lock_acquire+0x0/0xdd4) from [<c005d0fc>] (lock_acquire+0x80/0x94) [<c005d07c>] (lock_acquire+0x0/0x94) from [<c01e0f70>] (_spin_lock+0x30/0x40) [<c01e0f40>] (_spin_lock+0x0/0x40) from [<c0134cf8>] (smc_interrupt+0x24/0x740) r4 = C06EFB20 [<c0134cd4>] (smc_interrupt+0x0/0x740) from [<c0065878>] (handle_IRQ_event+0x34/0x6c) [<c0065844>] (handle_IRQ_event+0x0/0x6c) from [<c0066af4>] (handle_level_irq+0xe8/0x158) r8 = C021BF48 r7 = C021D124 r6 = C06EFB20 r5 = 00000029 r4 = C021D100 [<c0066a0c>] (handle_level_irq+0x0/0x158) from [<c00217f0>] (asm_do_IRQ+0x54/0x64) r8 = C001A680 r7 = 00008000 r6 = 00000029 r5 = F1100100 r4 = C021D100 [<c002179c>] (asm_do_IRQ+0x0/0x64) from [<c0020a68>] (__irq_svc+0x28/0xc0) r4 = FFFFFFFF [<c0021a64>] (default_idle+0x0/0x4c) from [<c0021b18>] (cpu_idle+0x68/0x88) r4 = C0021A64 [<c0021ab0>] (cpu_idle+0x0/0x88) from [<c0020028>] (__init_end+0x28/0x30) r5 = C0274DB4 r4 = C03AC520 [<c0020000>] (__init_end+0x0/0x30) from [<c0008aa0>] (start_kernel+0x270/0x2dc) [<c0008830>] (start_kernel+0x0/0x2dc) from [<00008078>] (0x8078) r8 = 0001A078 r7 = C03C0328 r6 = C0220984 r5 = C0275240 r4 = 00C5387D
eth0: link up, 100Mbps, full-duplex, lpa 0x45E1
Dump 2:
Sending DHCP requests .
=================================
[ INFO: inconsistent lock state ]
---------------------------------
inconsistent {hardirq-on-W} -> {in-hardirq-W} usage. swapper/0 [HC1[1]:SC0[0]:HE0:SE1] takes:
(&lp->lock){+-..}, at: [<c0134cf8>] smc_interrupt+0x24/0x740
{hardirq-on-W} state was registered at:
[<c005d0fc>] lock_acquire+0x80/0x94 [<c01e0f70>] _spin_lock+0x30/0x40 [<c0134474>] smc_shutdown+0x20/0x74 [<c0135f74>] smc_drv_probe+0x488/0x720 [<c0131b84>] platform_drv_probe+0x20/0x24 [<c012fbc8>] driver_probe_device+0x8c/0xd4 [<c012fd48>] __driver_attach+0x88/0xe8 [<c012f000>] bus_for_each_dev+0x50/0x84 [<c012fdc8>] driver_attach+0x20/0x28 [<c012f5b8>] bus_add_driver+0x6c/0x138 [<c01302a0>] driver_register+0x90/0xa0 [<c0131c84>] platform_driver_register+0x6c/0x88 [<c0015fe4>] smc_init+0x14/0x1c [<c002019c>] init+0x140/0x318 [<c00423b8>] do_exit+0x0/0x83c irq event stamp: 19304 hardirqs last enabled at (19303): [<c0021aa4>] default_idle+0x40/0x4c hardirqs last disabled at (19304): [<c0020a68>] __irq_svc+0x28/0xc0 softirqs last enabled at (19298): [<c00454b8>] __do_softirq+0xf4/0x10c softirqs last disabled at (19289): [<c0045798>] irq_exit+0x54/0x5c
other info that might help us debug this:
no locks held by swapper/0.
stack backtrace: [<c0024ed0>] (dump_stack+0x0/0x14) from [<c005b258>] (print_usage_bug+0x19c/0x1f0) [<c005b0bc>] (print_usage_bug+0x0/0x1f0) from [<c005b3a0>] (mark_lock+0xf4/0x5b8) [<c005b2ac>] (mark_lock+0x0/0x5b8) from [<c005c358>] (__lock_acquire+0x50c/0xdd4) r8 = C021F520 r7 = C021F7DC r6 = 00000000 r5 = C02B11F4 r4 = 00000000 [<c005be4c>] (__lock_acquire+0x0/0xdd4) from [<c005d0fc>] (lock_acquire+0x80/0x94) [<c005d07c>] (lock_acquire+0x0/0x94) from [<c01e0f70>] (_spin_lock+0x30/0x40) [<c01e0f40>] (_spin_lock+0x0/0x40) from [<c0134cf8>] (smc_interrupt+0x24/0x740) r4 = C06EFB20 [<c0134cd4>] (smc_interrupt+0x0/0x740) from [<c0065878>] (handle_IRQ_event+0x34/0x6c) [<c0065844>] (handle_IRQ_event+0x0/0x6c) from [<c0066af4>] (handle_level_irq+0xe8/0x158) r8 = C021BF48 r7 = C021D124 r6 = C06EFB20 r5 = 00000029 r4 = C021D100 [<c0066a0c>] (handle_level_irq+0x0/0x158) from [<c00217f0>] (asm_do_IRQ+0x54/0x64) r8 = C001A680 r7 = 00008000 r6 = 00000029 r5 = F1100100 r4 = C021D100 [<c002179c>] (asm_do_IRQ+0x0/0x64) from [<c0020a68>] (__irq_svc+0x28/0xc0) r4 = FFFFFFFF [<c0021a64>] (default_idle+0x0/0x4c) from [<c0021b18>] (cpu_idle+0x68/0x88) r4 = C0021A64 [<c0021ab0>] (cpu_idle+0x0/0x88) from [<c0020028>] (__init_end+0x28/0x30) r5 = C0274DB4 r4 = C03AC520 [<c0020000>] (__init_end+0x0/0x30) from [<c0008aa0>] (start_kernel+0x270/0x2dc) [<c0008830>] (start_kernel+0x0/0x2dc) from [<00008078>] (0x8078) r8 = 0001A078 r7 = C03C0328 r6 = C0220984 r5 = C0275240 r4 = 00C5387D
drivers/net/smc91x.c +4 -4
@@ -321,12 +321,12 @@ static void smc_reset(struct net_deviceDBG (2 ,"%s: %s\n" , dev->name, __FUNCTION__);/* Disable all interrupts, block TX tasklet */ -spin_lock (&lp->lock); +spin_lock_irq (&lp->lock);SMC_SELECT_BANK (2 );SMC_SET_INT_MASK (0 ); pending_skb = lp->pending_tx_skb; lp->pending_tx_skb =NULL ; -spin_unlock (&lp->lock); +spin_unlock_irq (&lp->lock);/* free any pending tx skb */ if (pending_skb) {
@@ -448,12 +448,12 @@ static void smc_shutdown(struct net_deviDBG (2 ,"%s: %s\n" ,CARDNAME , __FUNCTION__);/* no more interrupts for me */ -spin_lock (&lp->lock); +spin_lock_irq (&lp->lock);SMC_SELECT_BANK (2 );SMC_SET_INT_MASK (0 ); pending_skb = lp->pending_tx_skb; lp->pending_tx_skb =NULL ; -spin_unlock (&lp->lock); +spin_unlock_irq (&lp->lock);if (pending_skb)dev_kfree_skb (pending_skb);
Signature · 3 lines
Russell King Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 Serial core
Quoted 7 lines · Russell King
Hi,
When booting using root-nfs, I'm seeing (independently) two lockdep dumps in the smc91x driver. The patch below fixes both. Both dumps look like real locking issues.
Nico - please review and ack if you think the patch is correct.
Was this ever ACK'd?
I would prefer spin_lock_irqsave(), but that's just a quibble. I would be OK with applying this.
Jeff
Quoted 10 lines · Jeff Garzik
Quoted 7 lines · Russell King
Hi,
When booting using root-nfs, I'm seeing (independently) two lockdep dumps in the smc91x driver. The patch below fixes both. Both dumps look like real locking issues.
Nico - please review and ack if you think the patch is correct.
Was this ever ACK'd?
Not afaik.
Quoted 2 lines
I would prefer
spin_lock_irqsave(), but that's just a quibble. I would be OK with applying this.
Yes, though I did review all call paths to make sure it was safe first.
Signature · 3 lines
Russell King Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 Serial core