[ 094/110] dmaengine: imx-dma: fix lockdep issue between irqhandler and tasklet

From: Greg Kroah-Hartman
Date: Fri Oct 11 2013 - 16:18:51 EST

3.10-stable review patch. If anyone has any objections, please let me know.

------------------

From: Michael Grzeschik <m.grzeschik@xxxxxxxxxxxxxx>

commit 5a276fa6bdf82fd442046969603968c83626ce0b upstream.

The tasklet and irqhandler are using spin_lock while other routines are
using spin_lock_irqsave/restore. This leads to lockdep issues as
described bellow. This patch is changing the code to use
spinlock_irq_save/restore in both code pathes.

As imxdma_xfer_desc always gets called with spin_lock_irqsave lock held,
this patch also removes the spare call inside the routine to avoid
double locking.

[ 403.358162] =================================
[ 403.362549] [ INFO: inconsistent lock state ]
[ 403.366945] 3.10.0-20130823+ #904 Not tainted
[ 403.371331] ---------------------------------
[ 403.375721] inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
[ 403.381769] swapper/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
[ 403.386762] (&(&imxdma->lock)->rlock){?.-...}, at: [<c019d77c>] imxdma_tasklet+0x20/0x134
[ 403.395201] {IN-HARDIRQ-W} state was registered at:
[ 403.400108] [<c004b264>] mark_lock+0x2a0/0x6b4
[ 403.404798] [<c004d7c8>] __lock_acquire+0x650/0x1a64
[ 403.410004] [<c004f15c>] lock_acquire+0x94/0xa8
[ 403.414773] [<c02f74e4>] _raw_spin_lock+0x54/0x8c
[ 403.419720] [<c019d094>] dma_irq_handler+0x78/0x254
[ 403.424845] [<c0061124>] handle_irq_event_percpu+0x38/0x1b4
[ 403.430670] [<c00612e4>] handle_irq_event+0x44/0x64
[ 403.435789] [<c0063a70>] handle_level_irq+0xd8/0xf0
[ 403.440903] [<c0060a20>] generic_handle_irq+0x28/0x38
[ 403.446194] [<c0009cc4>] handle_IRQ+0x68/0x8c
[ 403.450789] [<c0008714>] avic_handle_irq+0x3c/0x48
[ 403.455811] [<c0008f84>] __irq_svc+0x44/0x74
[ 403.460314] [<c0040b04>] cpu_startup_entry+0x88/0xf4
[ 403.465525] [<c02f00d0>] rest_init+0xb8/0xe0
[ 403.470045] [<c03e07dc>] start_kernel+0x28c/0x2d4
[ 403.474986] [<a0008040>] 0xa0008040
[ 403.478709] irq event stamp: 50854
[ 403.482140] hardirqs last enabled at (50854): [<c001c6b8>] tasklet_action+0x38/0xdc
[ 403.489954] hardirqs last disabled at (50853): [<c001c6a0>] tasklet_action+0x20/0xdc
[ 403.497761] softirqs last enabled at (50850): [<c001bc64>] _local_bh_enable+0x14/0x18
[ 403.505741] softirqs last disabled at (50851): [<c001c268>] irq_exit+0x88/0xdc
[ 403.513026]
[ 403.513026] other info that might help us debug this:
[ 403.519593] Possible unsafe locking scenario:
[ 403.519593]
[ 403.525548] CPU0
[ 403.528020] ----
[ 403.530491] lock(&(&imxdma->lock)->rlock);
[ 403.534828] <Interrupt>
[ 403.537474] lock(&(&imxdma->lock)->rlock);
[ 403.541983]
[ 403.541983] *** DEADLOCK ***
[ 403.541983]
[ 403.547951] no locks held by swapper/0.
[ 403.551813]
[ 403.551813] stack backtrace:
[ 403.556222] CPU: 0 PID: 0 Comm: swapper Not tainted 3.10.0-20130823+ #904
[ 403.563039] Backtrace:
[ 403.565581] [<c000b98c>] (dump_backtrace+0x0/0x10c) from [<c000bb28>] (show_stack+0x18/0x1c)
[ 403.574054] r6:00000000 r5:c05c51d8 r4:c040bd58 r3:00200000
[ 403.579872] [<c000bb10>] (show_stack+0x0/0x1c) from [<c02f398c>] (dump_stack+0x20/0x28)
[ 403.587955] [<c02f396c>] (dump_stack+0x0/0x28) from [<c02f29c8>] (print_usage_bug.part.28+0x224/0x28c)
[ 403.597340] [<c02f27a4>] (print_usage_bug.part.28+0x0/0x28c) from [<c004b404>] (mark_lock+0x440/0x6b4)
[ 403.606682] r8:c004a41c r7:00000000 r6:c040bd58 r5:c040c040 r4:00000002
[ 403.613566] [<c004afc4>] (mark_lock+0x0/0x6b4) from [<c004d844>] (__lock_acquire+0x6cc/0x1a64)
[ 403.622244] [<c004d178>] (__lock_acquire+0x0/0x1a64) from [<c004f15c>] (lock_acquire+0x94/0xa8)
[ 403.631010] [<c004f0c8>] (lock_acquire+0x0/0xa8) from [<c02f74e4>] (_raw_spin_lock+0x54/0x8c)
[ 403.639614] [<c02f7490>] (_raw_spin_lock+0x0/0x8c) from [<c019d77c>] (imxdma_tasklet+0x20/0x134)
[ 403.648434] r6:c3847010 r5:c040e890 r4:c38470d4
[ 403.653194] [<c019d75c>] (imxdma_tasklet+0x0/0x134) from [<c001c70c>] (tasklet_action+0x8c/0xdc)
[ 403.662013] r8:c0599160 r7:00000000 r6:00000000 r5:c040e890 r4:c3847114 r3:c019d75c
[ 403.670042] [<c001c680>] (tasklet_action+0x0/0xdc) from [<c001bd4c>] (__do_softirq+0xe4/0x1f0)
[ 403.678687] r7:00000101 r6:c0402000 r5:c059919c r4:00000001
[ 403.684498] [<c001bc68>] (__do_softirq+0x0/0x1f0) from [<c001c268>] (irq_exit+0x88/0xdc)
[ 403.692652] [<c001c1e0>] (irq_exit+0x0/0xdc) from [<c0009cc8>] (handle_IRQ+0x6c/0x8c)
[ 403.700514] r4:00000030 r3:00000110
[ 403.704192] [<c0009c5c>] (handle_IRQ+0x0/0x8c) from [<c0008714>] (avic_handle_irq+0x3c/0x48)
[ 403.712664] r5:c0403f28 r4:c0593ebc
[ 403.716343] [<c00086d8>] (avic_handle_irq+0x0/0x48) from [<c0008f84>] (__irq_svc+0x44/0x74)
[ 403.724733] Exception stack(0xc0403f28 to 0xc0403f70)
[ 403.729841] 3f20: 00000001 00000004 00000000 20000013 c0402000 c04104a8
[ 403.738078] 3f40: 00000002 c0b69620 a0004000 41069264 a03fb5f4 c0403f7c c0403f40 c0403f70
[ 403.746301] 3f60: c004b92c c0009e74 20000013 ffffffff
[ 403.751383] r6:ffffffff r5:20000013 r4:c0009e74 r3:c004b92c
[ 403.757210] [<c0009e30>] (arch_cpu_idle+0x0/0x4c) from [<c0040b04>] (cpu_startup_entry+0x88/0xf4)
[ 403.766161] [<c0040a7c>] (cpu_startup_entry+0x0/0xf4) from [<c02f00d0>] (rest_init+0xb8/0xe0)
[ 403.774753] [<c02f0018>] (rest_init+0x0/0xe0) from [<c03e07dc>] (start_kernel+0x28c/0x2d4)
[ 403.783051] r6:c03fc484 r5:ffffffff r4:c040a0e0
[ 403.787797] [<c03e0550>] (start_kernel+0x0/0x2d4) from [<a0008040>] (0xa0008040)

Signed-off-by: Michael Grzeschik <m.grzeschik@xxxxxxxxxxxxxx>
Signed-off-by: Vinod Koul <vinod.koul@xxxxxxxxx>
Cc: Jonghwan Choi <jhbird.choi@xxxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>

---
drivers/dma/imx-dma.c | 19 ++++++++-----------
1 file changed, 8 insertions(+), 11 deletions(-)

--- a/drivers/dma/imx-dma.c
+++ b/drivers/dma/imx-dma.c
@@ -414,17 +414,18 @@ static void dma_irq_handle_channel(struc
struct imxdma_engine *imxdma = imxdmac->imxdma;
int chno = imxdmac->channel;
struct imxdma_desc *desc;
+ unsigned long flags;

- spin_lock(&imxdma->lock);
+ spin_lock_irqsave(&imxdma->lock, flags);
if (list_empty(&imxdmac->ld_active)) {
- spin_unlock(&imxdma->lock);
+ spin_unlock_irqrestore(&imxdma->lock, flags);
goto out;
}

desc = list_first_entry(&imxdmac->ld_active,
struct imxdma_desc,
node);
- spin_unlock(&imxdma->lock);
+ spin_unlock_irqrestore(&imxdma->lock, flags);

if (desc->sg) {
u32 tmp;
@@ -496,7 +497,6 @@ static int imxdma_xfer_desc(struct imxdm
{
struct imxdma_channel *imxdmac = to_imxdma_chan(d->desc.chan);
struct imxdma_engine *imxdma = imxdmac->imxdma;
- unsigned long flags;
int slot = -1;
int i;

@@ -504,7 +504,6 @@ static int imxdma_xfer_desc(struct imxdm
switch (d->type) {
case IMXDMA_DESC_INTERLEAVED:
/* Try to get a free 2D slot */
- spin_lock_irqsave(&imxdma->lock, flags);
for (i = 0; i < IMX_DMA_2D_SLOTS; i++) {
if ((imxdma->slots_2d[i].count > 0) &&
((imxdma->slots_2d[i].xsr != d->x) ||
@@ -514,10 +513,8 @@ static int imxdma_xfer_desc(struct imxdm
slot = i;
break;
}
- if (slot < 0) {
- spin_unlock_irqrestore(&imxdma->lock, flags);
+ if (slot < 0)
return -EBUSY;
- }

imxdma->slots_2d[slot].xsr = d->x;
imxdma->slots_2d[slot].ysr = d->y;
@@ -526,7 +523,6 @@ static int imxdma_xfer_desc(struct imxdm

imxdmac->slot_2d = slot;
imxdmac->enabled_2d = true;
- spin_unlock_irqrestore(&imxdma->lock, flags);

if (slot == IMX_DMA_2D_SLOT_A) {
d->config_mem &= ~CCR_MSEL_B;
@@ -602,8 +598,9 @@ static void imxdma_tasklet(unsigned long
struct imxdma_channel *imxdmac = (void *)data;
struct imxdma_engine *imxdma = imxdmac->imxdma;
struct imxdma_desc *desc;
+ unsigned long flags;

- spin_lock(&imxdma->lock);
+ spin_lock_irqsave(&imxdma->lock, flags);

if (list_empty(&imxdmac->ld_active)) {
/* Someone might have called terminate all */
@@ -640,7 +637,7 @@ static void imxdma_tasklet(unsigned long
__func__, imxdmac->channel);
}
out:
- spin_unlock(&imxdma->lock);
+ spin_unlock_irqrestore(&imxdma->lock, flags);
}

static int imxdma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,


--
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/