Re: [PATCH] mhi: pci_generic: Remove WQ_MEM_RECLAIM flag from state workqueue

From: Bhaumik Bhatt <bbhatt@codeaurora.org>
To: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Cc: Loic Poulain <loic.poulain@linaro.org>,
	hemantk@codeaurora.org, linux-arm-msm@vger.kernel.org
Subject: Re: [PATCH] mhi: pci_generic: Remove WQ_MEM_RECLAIM flag from state workqueue
Date: Wed, 24 Feb 2021 09:46:43 -0800	[thread overview]
Message-ID: <e6a1b0b928dbd2b9addd617e9c72e80f@codeaurora.org> (raw)
In-Reply-To: <20210224102516.GW27945@work>

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@linaro.org>

Reviewed-by: Bhaumik Bhatt <bbhatt@codeaurora.org>

> 
> 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