On 2021-02-24 02:25 AM, Manivannan Sadhasivam wrote:
On Wed, Feb 24, 2021 at 11:18:50AM +0100, Loic Poulain wrote:
A recent change created a dedicated workqueue for the state-change
work
with WQ_HIGHPRI (no strong reason for that) and WQ_MEM_RECLAIM flags,
but the state-change work (mhi_pm_st_worker) does not guarantee
forward
progress under memory pressure, and will even wait on various memory
allocations when e.g. creating devices, loading firmware, etc... The
work is then not part of a memory reclaim path...
Moreover, this causes a warning in check_flush_dependency() since we
end
up in code that flushes a non-reclaim workqueue:
[ 40.969601] workqueue: WQ_MEM_RECLAIM
mhi_hiprio_wq:mhi_pm_st_worker [mhi] is flushing !WQ_MEM_RECLAIM
events_highpri:flush_backlog
[ 40.969612] WARNING: CPU: 4 PID: 158 at kernel/workqueue.c:2607
check_flush_dependency+0x11c/0x140
[ 40.969733] Call Trace:
[ 40.969740] __flush_work+0x97/0x1d0
[ 40.969745] ? wake_up_process+0x15/0x20
[ 40.969749] ? insert_work+0x70/0x80
[ 40.969750] ? __queue_work+0x14a/0x3e0
[ 40.969753] flush_work+0x10/0x20
[ 40.969756] rollback_registered_many+0x1c9/0x510
[ 40.969759] unregister_netdevice_queue+0x94/0x120
[ 40.969761] unregister_netdev+0x1d/0x30
[ 40.969765] mhi_net_remove+0x1a/0x40 [mhi_net]
[ 40.969770] mhi_driver_remove+0x124/0x250 [mhi]
[ 40.969776] device_release_driver_internal+0xf0/0x1d0
[ 40.969778] device_release_driver+0x12/0x20
[ 40.969782] bus_remove_device+0xe1/0x150
[ 40.969786] device_del+0x17b/0x3e0
[ 40.969791] mhi_destroy_device+0x9a/0x100 [mhi]
[ 40.969796] ? mhi_unmap_single_use_bb+0x50/0x50 [mhi]
[ 40.969799] device_for_each_child+0x5e/0xa0
[ 40.969804] mhi_pm_st_worker+0x921/0xf50 [mhi]
Fixes: 8f7039787687 ("bus: mhi: core: Move to using high priority
workqueue")
Signed-off-by: Loic Poulain <loic.poulain@xxxxxxxxxx>
Reviewed-by: Bhaumik Bhatt <bbhatt@xxxxxxxxxxxxxx>
Fix looks good to me but I want Bhaumik to share his review since he
authored the offending patch.
We have seen this internally as well. I agree this patch needs to go in.
We had previously seen issues using global workqueue hence decided to
move to a
dedicated one with WQ_HIGHPRI in order to speed up execution of the
worker when
a certain task is queued. For example, handling SBL or power down needs
to be
done promptly.
Thanks,
Mani
---
drivers/bus/mhi/core/init.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/bus/mhi/core/init.c b/drivers/bus/mhi/core/init.c
index 32eb90f..03ddd6e 100644
--- a/drivers/bus/mhi/core/init.c
+++ b/drivers/bus/mhi/core/init.c
@@ -890,8 +890,7 @@ int mhi_register_controller(struct mhi_controller
*mhi_cntrl,
INIT_WORK(&mhi_cntrl->st_worker, mhi_pm_st_worker);
init_waitqueue_head(&mhi_cntrl->state_event);
- mhi_cntrl->hiprio_wq = alloc_ordered_workqueue
- ("mhi_hiprio_wq", WQ_MEM_RECLAIM | WQ_HIGHPRI);
+ mhi_cntrl->hiprio_wq = alloc_ordered_workqueue("mhi_hiprio_wq",
WQ_HIGHPRI);
if (!mhi_cntrl->hiprio_wq) {
dev_err(mhi_cntrl->cntrl_dev, "Failed to allocate workqueue\n");
ret = -ENOMEM;
--
2.7.4
Thanks,
Bhaumik
---
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora
Forum,
a Linux Foundation Collaborative Project
