linux-kernel - Re: [PATCH] uio: add ioctl callback

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] [day] [month] [year] [list]

Date:	Wed, 12 Nov 2008 17:43:40 +0100
From:	Neil Armstrong <narmstrong@...tion.com>
To:	"Hans J. Koch" <hjk@...utronix.de>
CC:	gregkh@...e.de, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] uio: add ioctl callback

Hi,

I should have specified that I work on a SoC, which does not follow
strict rules like PCI devices.
From a hardware designer point of view, this is a "good" design
especially for this kind of device. (DVB Conditionnal Access Slave
Shared Memory)
If the code was in the kernel space, we could do some processing in irq
mode like other devices of the SoC.

Thanks for your time and comments, I will see the irqcontrol solution as
an alternative.

Just forget my patch since it's only applicable to very specific devices.

Neil

Hans J. Koch a écrit :
> On Wed, Nov 12, 2008 at 04:59:56PM +0100, Neil Armstrong wrote:
>   
>> Hi Hans,
>>
>> We need an ioctl callback because we need to query some values only
>> available when the irq handler is running.
>> For example, we have 3 types of reasons why the irq is triggered and
>> these bits are no more available when the irq is cleared.
>>     
>
> Ah, that one. That's why we invented the irqcontrol hook. In case of
> such broken hardware, you need to mask the irq in the kernel without
> touching the register containing those volatile bits. On a system where
> you can be sure the irq is not shared, you can simply use disable_irq().
>
> If the irq might be shared, you need to find something else. PCI cards,
> for example, often have a register within their PCI bridge that contains
> irq mask bits (that's how it is done in uio_cif.c).
>
> Userspace can then reenable the irq by writing the value 1 as a signed
> 32-bit int to /dev/uioX. You need to implement an irqcontrol() function
> in your kernel driver that does the right thing (e.g. call enable_irq()
> in the first example).
>
>   
>> The cleanest way to have this very specific information is to have a
>> dirty old ioctl returning this data.
>>     
>
> The cleanest way would be to throw such hardware into the trash bin :-)
>
> A chip where the irq mask register cannot be written without destroying
> the irq status register is simply broken and bad design.
>
> Thanks,
> Hans
>   

Download attachment "signature.asc" of type "application/pgp-signature" (253 bytes)