Re: [RFC][PATCH 0/2] Export platform_pci_set_power_state() and make radeonfb use it

From: "Rafael J. Wysocki" <rjw@sisk.pl>
To: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>,
	Linux PCI <linux-pci@vger.kernel.org>,
	pm list <linux-pm@lists.linux-foundation.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Alex Deucher <alexdeucher@gmail.com>
Subject: Re: [RFC][PATCH 0/2] Export platform_pci_set_power_state() and make radeonfb use it
Date: Tue, 24 Mar 2009 23:25:22 +0100	[thread overview]
Message-ID: <200903242325.23483.rjw@sisk.pl> (raw)
In-Reply-To: <1237929175.25062.727.camel@pasglop>

On Tuesday 24 March 2009, Benjamin Herrenschmidt wrote:
> On Tue, 2009-03-24 at 12:00 +0100, Rafael J. Wysocki wrote:
> 
> > In fact I have yet another idea.  If we use the "retransmission with exponential
> > backoff" algorithm in pci_raw_set_power_state(), we won't have to add any
> > extra parameters to pci_set_power_state() and the radeon case will be covered
> > automatically.  That should also cover any other devices having similar
> > problems IMO.
> > 
> > A patch to implement that is appended, please tell me what you think.
> 
> This is crazy....
> 
> Most devices don't need that and those who are a bit "touchy" may well
> puke on being written to too fast while they are in the middle of
> the transition.
> 
> Let's wait to see if Alex from ATI has an answer here, and if not, I
> would suggest keeping the dirt inside radeonfb.

BTW, whatever you do inside of radeonfb will be based on specific assumptions
regarding how pci_set_power_state() works internally, which makes that
extremely fragile.  This way you'll also make the core depend on the radeonfb
driver to some extent, which I don't think is desirable.

Thanks,
Rafael