netdev - Re: [PATCH v2] net: fec: Use unlocked timecounter reads for saving state

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <bbe8a924-7291-14f9-1e88-802a211ca0f4@prolan.hu>
Date:   Tue, 30 Aug 2022 17:05:24 +0200
From:   Csókás Bence <csokas.bence@...lan.hu>
To:     Andrew Lunn <andrew@...n.ch>
CC:     <netdev@...r.kernel.org>,
        Richard Cochran <richardcochran@...il.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>, <qiangqing.zhang@....com>,
        <kernel@...gutronix.de>, Marc Kleine-Budde <mkl@...gutronix.de>
Subject: Re: [PATCH v2] net: fec: Use unlocked timecounter reads for saving
 state

On 2022. 08. 30. 14:29, Andrew Lunn wrote:
>> -	fec_ptp_gettime(&fep->ptp_caps, &fep->ptp_saved_state.ts_phc);
>> +	if (preempt_count_equals(0)) {
> 
> ~/linux/drivers$ grep -r preempt_count_equals *
> ~/linux/drivers$
> 
> No other driver plays games like this.
> 
> Why not unconditionally take the lock?

Because then we would be back at the original problem (see Marc's message):

| [   14.001542] BUG: sleeping function called from invalid context at 
kernel/locking/mutex.c:283 

| [   14.010604] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 
13, name: kworker/0:1 

| [   14.018737] preempt_count: 201, expected: 0

We cannot take a mutex in atomic context. However, we also don't *need 
to* take a mutex in atomic context.

> 
>      Andrew

If someone has a better solution, I'm open to suggestions. But to me, it 
seems that there are only 3 options:

1. Unconditionally taking the mutex was what I originally did, but that 
caused issues in Marc's setup.
2. Not taking the mutex at all is what I proposed in v1 of this patch. 
But as Richard pointed out, `timecounter_read()` actually does a 
Read-Modify-Write on the `FEC_ATIME_CTRL` register, that *could* get 
interrupted if not guarded by the mutex (or atomic context).
3. The final option, check if we are in an atomic or otherwise 
non-interruptible context, and if not, take a mutex. Otherwise, proceed 
normally. Which is this version of the patch.

Bence