linux-kernel - Re: [PATCH v5 2/5] hwmon: (max31790) Fix and split pwm*

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: <9bbdc7a7-f34d-4e3a-8e64-c20810456d11@cesnet.cz>
Date:   Thu, 20 May 2021 13:29:24 +0200
From:   Jan Kundrát <jan.kundrat@...net.cz>
To:     Guenter Roeck <linux@...ck-us.net>,
        Václav Kubernát <kubernat@...net.cz>
Cc:     Jean Delvare <jdelvare@...e.com>, Jonathan Corbet <corbet@....net>,
        <linux-hwmon@...r.kernel.org>, <linux-doc@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v5 2/5] hwmon: (max31790) Fix and split pwm*_enable

> As for fan[7-12]_enable, I don't even know if those can be enabled
> separately. I see two options: Drop those attributes entirely (
> assuming that those fan inputs are always enabled if the associated
> pins are configured as inputs), or align them with fan[1-6]_enable.

I think we need to decide first who provides the initial configuration for 
this chip. There's always at least six TACH inputs, and then there's six 
more pins where each can be either a PWM output or a TACH input. Who 
decides that? Is the kernel supposed to export six knobs to the userspace? 
So far, I've assumed that this should be driven via sysfs. But perhaps you 
would you like to rely on the FW (?) to set this up properly? (On our 
board, that would be a few random calls to `i2cset` from a U-Boot boot 
script. Not pretty, but doable. Just one more place to keep track of.)

It's proabably "tricky" to do this at runtime -- and I don't expect to see 
many boards where you have such a big freedom of reconnecting the actual 
fans once manufactured, anyway. So, either some DT parameters, or an 
autodetection based on whatever is in the registers at power up, which 
would make an explicit assumption that "something" has set up the nPWM/TACH 
bits properly in the Fan Configuration Register. OK, that might work, but 
the kernel must not ever reset that chip afterwards.

There's also the Fan Fault Mask register, which controls which fans 
propagate their failures to the nFAN_FAIL output pin. This one requires a 
semi-independent control than the nPWM/TACH bit above. It's feasible that 
not all TACH inputs have an actual fan connected, and this can well vary 
between products. For example, ours has just four fan connectors, so we 
don't want "failures" of fans 5 and 6 to assert the nFAN_FAIL pin. Also, 
there should be a check which prevents unmasking these failures for those 
TACH channels which are configured as PWM outputs. Or we can once again 
ignore this one and rely on the FW.

The current kernel code in max31790_read_fan() reads beyond the end of 
data->fan_dynamics, hitting the content of `fault_status` or `tach` fields 
instead, and therefore returning garbage. Not a big deal, just a missing % 
operator I guess, but to me, that's a pretty strong suggestion that nobody 
has used or even tested monitoring more than six fans on this chip, ever. 
(And yeah, the datasheet is not clear on how it's supposed to work anyway. 
Using a modulo is just a guess.)

Neither Vaclav nor me have any way of testing this feature -- hence my 
proposal to only improve what we need, and ignore TACH channels 7-12. But I 
guess it's not OK from your point of view?

With kind regards,
Jan