[patch] drivers/iommu: Fix iova DEBUG_PREEMPT splat

[patch] drivers/iommu: Fix iova DEBUG_PREEMPT splat

[Mike Galbraith]

[  646.138847] BUG: using smp_processor_id() in preemptible [00000000] code: usb-storage/309
[  646.138852] caller is debug_smp_processor_id+0x17/0x20
[  646.138854] CPU: 0 PID: 309 Comm: usb-storage Not tainted 4.7.0-master #855
[  646.138855] Hardware name: MEDION MS-7848/MS-7848, BIOS M7848W08.20C 09/23/2013
[  646.138856]  0000000000000000 ffff8803f784ba48 ffffffff812da34c 0000000000000000
[  646.138859]  ffffffff81a18ad8 ffff8803f784ba78 ffffffff812f4c1c ffffffff81a0dd4e
[  646.138860]  000000000001db40 ffff8803f9100318 0000000000000001 ffff8803f784ba88
[  646.138862] Call Trace:
[  646.138866]  [<ffffffff812da34c>] dump_stack+0x65/0x89
[  646.138868]  [<ffffffff812f4c1c>] check_preemption_disabled+0xdc/0xf0
[  646.138869]  [<ffffffff812f4c47>] debug_smp_processor_id+0x17/0x20
[  646.138873]  [<ffffffff813e9c5b>] alloc_iova_fast+0xbb/0x1e0
[  646.138874]  [<ffffffff813ed1cf>] intel_alloc_iova+0x9f/0xd0
[  646.138876]  [<ffffffff813f118f>] intel_map_sg+0xbf/0x220
[  646.138878]  [<ffffffff8147cc84>] usb_hcd_map_urb_for_dma+0x2d4/0x4e0
[  646.138879]  [<ffffffff8147cf37>] usb_hcd_submit_urb+0xa7/0xa90
[  646.138881]  [<ffffffff8162717a>] ? __schedule+0x27a/0x750
[  646.138882]  [<ffffffff81627699>] ? schedule+0x49/0xa0
[  646.138884]  [<ffffffff8162a510>] ? schedule_timeout+0x210/0x290
[  646.138885]  [<ffffffff8147e869>] usb_submit_urb.part.6+0x299/0x530
[  646.138887]  [<ffffffff8147eb34>] usb_submit_urb+0x34/0x70
[  646.138888]  [<ffffffff8147fc48>] usb_sg_wait+0x88/0x180
[  646.138891]  [<ffffffffa008d58e>] usb_stor_bulk_transfer_sglist.part.4+0x7e/0xe0 [usb_storage]
[  646.138893]  [<ffffffffa008d63f>] usb_stor_bulk_srb+0x4f/0x60 [usb_storage]
[  646.138895]  [<ffffffffa008d773>] usb_stor_Bulk_transport+0x123/0x3a0 [usb_storage]
[  646.138896]  [<ffffffff81627699>] ? schedule+0x49/0xa0
[  646.138898]  [<ffffffffa008dfac>] usb_stor_invoke_transport+0x2c/0x4c0 [usb_storage]
[  646.138900]  [<ffffffff81085b35>] ? wake_up_process+0x15/0x20
[  646.138903]  [<ffffffff81062ff5>] ? wakeup_softirqd+0x35/0x40
[  646.138904]  [<ffffffff81082363>] ? preempt_count_add+0xa3/0xc0
[  646.138906]  [<ffffffff8162af8d>] ? _raw_spin_unlock_irq+0x1d/0x40
[  646.138907]  [<ffffffff816288d8>] ? wait_for_completion_interruptible+0x128/0x150
[  646.138909]  [<ffffffffa008cf20>] usb_stor_pad12_command+0x40/0x50 [usb_storage]
[  646.138910]  [<ffffffffa008f456>] usb_stor_control_thread+0x146/0x240 [usb_storage]
[  646.138912]  [<ffffffffa008f310>] ? fill_inquiry_response+0x20/0x20 [usb_storage]
[  646.138914]  [<ffffffffa008f310>] ? fill_inquiry_response+0x20/0x20 [usb_storage]
[  646.138916]  [<ffffffff8107bb79>] kthread+0xc9/0xe0
[  646.138918]  [<ffffffff8162b79f>] ret_from_fork+0x1f/0x40
[  646.138919]  [<ffffffff8107bab0>] ? kthread_park+0x60/0x60

Signed-off-by: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Joerg Roedel <joro@8bytes.org>
---
 drivers/iommu/iova.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

--- a/drivers/iommu/iova.c
+++ b/drivers/iommu/iova.c
@@ -749,8 +749,9 @@ static bool __iova_rcache_insert(struct
 	bool can_insert = false;
 	unsigned long flags;
 
+	local_irq_save(flags);
 	cpu_rcache = this_cpu_ptr(rcache->cpu_rcaches);
-	spin_lock_irqsave(&cpu_rcache->lock, flags);
+	spin_lock(&cpu_rcache->lock);
 
 	if (!iova_magazine_full(cpu_rcache->loaded)) {
 		can_insert = true;
@@ -812,8 +813,9 @@ static unsigned long __iova_rcache_get(s
 	bool has_pfn = false;
 	unsigned long flags;
 
+	local_irq_save(flags);
 	cpu_rcache = this_cpu_ptr(rcache->cpu_rcaches);
-	spin_lock_irqsave(&cpu_rcache->lock, flags);
+	spin_lock(&cpu_rcache->lock);
 
 	if (!iova_magazine_empty(cpu_rcache->loaded)) {
 		has_pfn = true;

Re: [patch] drivers/iommu: Fix iova DEBUG_PREEMPT splat

[Cc: Joerg Roedel]

On Mon, Jun 13, 2016 at 02:56:36PM +0200, Mike Galbraith wrote:
> 
> Signed-off-by: Mike Galbraith <umgwanakikbuti@gmail.com>
> Cc: Joerg Roedel <joro@8bytes.org>
> ---
>  drivers/iommu/iova.c |    6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> --- a/drivers/iommu/iova.c
> +++ b/drivers/iommu/iova.c
> @@ -749,8 +749,9 @@ static bool __iova_rcache_insert(struct
>  	bool can_insert = false;
>  	unsigned long flags;
>  
> +	local_irq_save(flags);
>  	cpu_rcache = this_cpu_ptr(rcache->cpu_rcaches);
> -	spin_lock_irqsave(&cpu_rcache->lock, flags);
> +	spin_lock(&cpu_rcache->lock);
>  
>  	if (!iova_magazine_full(cpu_rcache->loaded)) {
>  		can_insert = true;
> @@ -812,8 +813,9 @@ static unsigned long __iova_rcache_get(s
>  	bool has_pfn = false;
>  	unsigned long flags;
>  
> +	local_irq_save(flags);
>  	cpu_rcache = this_cpu_ptr(rcache->cpu_rcaches);
> -	spin_lock_irqsave(&cpu_rcache->lock, flags);
> +	spin_lock(&cpu_rcache->lock);
>  
>  	if (!iova_magazine_empty(cpu_rcache->loaded)) {
>  		has_pfn = true;

>From an earlier patch submitted by Sagar (Cc'ed) I think the right fix
is to just make the code use get_cpu_ptr/put_cpu_ptr instead of
this_cpu_ptr.


	Joerg