mbox series

[v5,0/2] bus: mhi: host: Add lock to avoid race when ringing channel DB

Message ID 1702276972-41296-1-git-send-email-quic_qianyu@quicinc.com
Headers show
Series bus: mhi: host: Add lock to avoid race when ringing channel DB | expand

Message

Qiang Yu Dec. 11, 2023, 6:42 a.m. UTC
1. We need a write lock in mhi_gen_tre otherwise there is race of the WP
used for ringing channel DB between mhi_queue and M0 transition.
2. We can not invoke local_bh_enable() when irqs are disabled, so move
read_lock_irqsave() under the mhi_gen_tre() since we add write_lock_bh() in
mhi_gen_tre().
3. Unlock xfer_cb to prevent potential lockup

v1 -> v2:
Added write_unlock_bh(&mhi_chan->lock) in mhi_gen_tre() before return
because of error process.

v2 -> v3:
1. split protecting WP and unlocking xfer_cb into two patches
2. Add a new patch to stop processing buffer and eventof a disabled or
stopped channel.

v3 -> v4:
1. Modify commit message
2. Add unlock operation before return error

v4 -> v5:
1. Squash "protecting WP" and "Take irqsave lock" into one patch
2. Drop patch 3/4 of patch v4

Bhaumik Bhatt (1):
  bus: mhi: host: Add spinlock to protect WP access when queueing TREs

Qiang Yu (1):
  bus: mhi: host: Drop chan lock before queuing buffers

 drivers/bus/mhi/host/main.c | 26 +++++++++++++++++---------
 1 file changed, 17 insertions(+), 9 deletions(-)

Comments

Jeffrey Hugo Dec. 15, 2023, 6:23 p.m. UTC | #1
On 12/10/2023 11:42 PM, Qiang Yu wrote:
> Ensure read and write locks for the channel are not taken in succession by
> dropping the read lock from parse_xfer_event() such that a callback given
> to client can potentially queue buffers and acquire the write lock in that
> process. Any queueing of buffers should be done without channel read lock
> acquired as it can result in multiple locks and a soft lockup.
> 
> Signed-off-by: Qiang Yu <quic_qianyu@quicinc.com>

Seems to work fine for AIC100

Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Tested-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Manivannan Sadhasivam Dec. 16, 2023, 5:19 a.m. UTC | #2
On Mon, Dec 11, 2023 at 02:42:52PM +0800, Qiang Yu wrote:
> Ensure read and write locks for the channel are not taken in succession by
> dropping the read lock from parse_xfer_event() such that a callback given
> to client can potentially queue buffers and acquire the write lock in that
> process. Any queueing of buffers should be done without channel read lock
> acquired as it can result in multiple locks and a soft lockup.
> 

Cc: <stable@vger.kernel.org> # 5.7
Fixes: 1d3173a3bae7 ("bus: mhi: core: Add support for processing events from client device")

> Signed-off-by: Qiang Yu <quic_qianyu@quicinc.com>

Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>

- Mani

> ---
>  drivers/bus/mhi/host/main.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/bus/mhi/host/main.c b/drivers/bus/mhi/host/main.c
> index 32021fe..25f98d6 100644
> --- a/drivers/bus/mhi/host/main.c
> +++ b/drivers/bus/mhi/host/main.c
> @@ -642,6 +642,8 @@ static int parse_xfer_event(struct mhi_controller *mhi_cntrl,
>  			mhi_del_ring_element(mhi_cntrl, tre_ring);
>  			local_rp = tre_ring->rp;
>  
> +			read_unlock_bh(&mhi_chan->lock);
> +
>  			/* notify client */
>  			mhi_chan->xfer_cb(mhi_chan->mhi_dev, &result);
>  
> @@ -667,6 +669,8 @@ static int parse_xfer_event(struct mhi_controller *mhi_cntrl,
>  					kfree(buf_info->cb_buf);
>  				}
>  			}
> +
> +			read_lock_bh(&mhi_chan->lock);
>  		}
>  		break;
>  	} /* CC_EOT */
> -- 
> 2.7.4
> 
>
Manivannan Sadhasivam Dec. 16, 2023, 5:21 a.m. UTC | #3
On Mon, Dec 11, 2023 at 02:42:50PM +0800, Qiang Yu wrote:
> 
> 1. We need a write lock in mhi_gen_tre otherwise there is race of the WP
> used for ringing channel DB between mhi_queue and M0 transition.
> 2. We can not invoke local_bh_enable() when irqs are disabled, so move
> read_lock_irqsave() under the mhi_gen_tre() since we add write_lock_bh() in
> mhi_gen_tre().
> 3. Unlock xfer_cb to prevent potential lockup
> 

Applied to mhi-next!

- Mani

> v1 -> v2:
> Added write_unlock_bh(&mhi_chan->lock) in mhi_gen_tre() before return
> because of error process.
> 
> v2 -> v3:
> 1. split protecting WP and unlocking xfer_cb into two patches
> 2. Add a new patch to stop processing buffer and eventof a disabled or
> stopped channel.
> 
> v3 -> v4:
> 1. Modify commit message
> 2. Add unlock operation before return error
> 
> v4 -> v5:
> 1. Squash "protecting WP" and "Take irqsave lock" into one patch
> 2. Drop patch 3/4 of patch v4
> 
> Bhaumik Bhatt (1):
>   bus: mhi: host: Add spinlock to protect WP access when queueing TREs
> 
> Qiang Yu (1):
>   bus: mhi: host: Drop chan lock before queuing buffers
> 
>  drivers/bus/mhi/host/main.c | 26 +++++++++++++++++---------
>  1 file changed, 17 insertions(+), 9 deletions(-)
> 
> -- 
> 2.7.4
> 
>