Re: No more read/write ioctls

From: Eric Anholt <eric@anholt.net>
To: Ben Widawsky <ben@bwidawsk.net>, intel-gfx@lists.freedesktop.org
Subject: Re: No more read/write ioctls
Date: Fri, 08 Apr 2011 07:10:14 -1000	[thread overview]
Message-ID: <87fwpsfz49.fsf@pollan.anholt.net> (raw)
In-Reply-To: <20110406213826.GA32472@snipes.kumite>

[-- Attachment #1.1: Type: text/plain, Size: 1379 bytes --]

On Wed, 6 Apr 2011 14:38:27 -0700, Ben Widawsky <ben@bwidawsk.net> wrote:
> On Fri, Apr 01, 2011 at 04:54:47PM -0700, Ben Widawsky wrote:
> 
> > So if this isn't okay by everyone, let's get down to the minimum number
> > of changes required to get this accepted so I can move on to the tools,
> > and then get back to the debug stuff.
> > 
> 
> Some discussion on IRC has led to a new proposal (well 2 new proposals,
> but the first one wasn't viable).
> 
> The interface will instead of using ioctls use debugfs. The debugfs file
> will control force wake. There will be a refcount mechanism for number
> of users of the registers in the relevant power-well, and upon opening a
> specific file in debugfs (we could have one per power-well if needed),
> the refcount will get incremented, and decremented at close.
> 
> In other words, for userspace to read/write registers:
> fd = open(/sys/kernel/debug/dri...)
> normal read write mechanism
> close(fd)
> 
> There are two side effects which everyone on IRC seems fine with:
> * root-only read access (the ioctl read was promiscuous)
> * access is only available when debugfs is mounted
> 
> As a result, you should ignore both the gpu-tools patches, as well as
> these kernel patches.

I thought previously we had issues with debugfs nodes which could wedge
the system (out of range mmio reads)?

[-- Attachment #1.2: Type: application/pgp-signature, Size: 197 bytes --]

[-- Attachment #2: Type: text/plain, Size: 159 bytes --]

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx