2020-08-13 07:32:15

by Christoph Hellwig

Subject: Re: [PATCH] block: insert a general SMP memory barrier before wake_up_bit()

On Thu, Aug 13, 2020 at 10:44:38AM +0800, Jacob Wen wrote:
> wake_up_bit() uses waitqueue_active() that needs the explicit smp_mb().

Sounds like the barrier should go into wake_up_bit then..

>
> Signed-off-by: Jacob Wen <[email protected]>
> ---
> fs/block_dev.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/fs/block_dev.c b/fs/block_dev.c
> index 0ae656e022fd..e74980848a2a 100644
> --- a/fs/block_dev.c
> +++ b/fs/block_dev.c
> @@ -1175,6 +1175,7 @@ static void bd_clear_claiming(struct block_device *whole, void *holder)
> /* tell others that we're done */
> BUG_ON(whole->bd_claiming != holder);
> whole->bd_claiming = NULL;
> + smp_mb();
> wake_up_bit(&whole->bd_claiming, 0);
> }
>
> --
> 2.17.1
>
---end quoted text---

2020-08-13 07:50:13

by Jacob Wen

Subject: Re: [PATCH] block: insert a general SMP memory barrier before wake_up_bit()

On 8/13/20 3:31 PM, Christoph Hellwig wrote:
> On Thu, Aug 13, 2020 at 10:44:38AM +0800, Jacob Wen wrote:
>> wake_up_bit() uses waitqueue_active() that needs the explicit smp_mb().
> Sounds like the barrier should go into wake_up_bit then..

wake_up_bit() doesn't know which one to chose: smp_mb__after_atomic() or
smp_mb().

>
>> Signed-off-by: Jacob Wen <[email protected]>
>> ---
>> fs/block_dev.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/fs/block_dev.c b/fs/block_dev.c
>> index 0ae656e022fd..e74980848a2a 100644
>> --- a/fs/block_dev.c
>> +++ b/fs/block_dev.c
>> @@ -1175,6 +1175,7 @@ static void bd_clear_claiming(struct block_device *whole, void *holder)
>> /* tell others that we're done */
>> BUG_ON(whole->bd_claiming != holder);
>> whole->bd_claiming = NULL;
>> + smp_mb();
>> wake_up_bit(&whole->bd_claiming, 0);
>> }
>>
>> --
>> 2.17.1
>>
> ---end quoted text---

2020-08-13 11:43:08

by Peter Zijlstra

Subject: Re: [PATCH] block: insert a general SMP memory barrier before wake_up_bit()

On Thu, Aug 13, 2020 at 08:31:15AM +0100, Christoph Hellwig wrote:
> On Thu, Aug 13, 2020 at 10:44:38AM +0800, Jacob Wen wrote:
> > wake_up_bit() uses waitqueue_active() that needs the explicit smp_mb().
>
> Sounds like the barrier should go into wake_up_bit then..

Oh, thanks for reminding me..

https://lkml.kernel.org/r/[email protected]

I'll try and get back to that.

2020-08-13 12:04:22

by Matthew Wilcox

Subject: Re: [PATCH] block: insert a general SMP memory barrier before wake_up_bit()

On Thu, Aug 13, 2020 at 01:40:50PM +0200, [email protected] wrote:
> On Thu, Aug 13, 2020 at 08:31:15AM +0100, Christoph Hellwig wrote:
> > On Thu, Aug 13, 2020 at 10:44:38AM +0800, Jacob Wen wrote:
> > > wake_up_bit() uses waitqueue_active() that needs the explicit smp_mb().
> >
> > Sounds like the barrier should go into wake_up_bit then..
>
> Oh, thanks for reminding me..
>
> https://lkml.kernel.org/r/[email protected]
>
> I'll try and get back to that.

+++ b/drivers/bluetooth/btmtkuart.c
@@ -340,11 +340,8 @@ static int btmtkuart_recv_event(struct hci_dev *hdev, struct sk_buff *skb)

if (hdr->evt == HCI_EV_VENDOR) {
if (test_and_clear_bit(BTMTKUART_TX_WAIT_VND_EVT,
- &bdev->tx_state)) {
- /* Barrier to sync with other CPUs */
- smp_mb__after_atomic();
+ &bdev->tx_state))
wake_up_bit(&bdev->tx_state, BTMTKUART_TX_WAIT_VND_EVT);
- }
}

return 0;

It'd be nice to be able to write:

if (hdr->evt == HCI_EV_VENDOR)
test_clear_and_wake_up_bit(&bdev->tx_state,
BTMTKUART_TX_WAIT_VND_EVT);

... maybe with a better name.