From mboxrd@z Thu Jan 1 00:00:00 1970 From: Benjamin Herrenschmidt Subject: Re: Question about removing memslots Date: Thu, 29 Mar 2012 08:04:18 +1100 Message-ID: <1332968658.2882.108.camel@pasglop> References: <1332919486.2882.88.camel@pasglop> <4F72DBE2.6060909@redhat.com> <1332928770.2882.98.camel@pasglop> <4F72E274.2090007@redhat.com> <1332929849.2882.101.camel@pasglop> <4F72ED38.5000007@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: kvm@vger.kernel.org, Alexander Graf , kvm-ppc@vger.kernel.org, Scott Wood , Alex Williamson To: Avi Kivity Return-path: In-Reply-To: <4F72ED38.5000007@redhat.com> Sender: kvm-ppc-owner@vger.kernel.org List-Id: kvm.vger.kernel.org On Wed, 2012-03-28 at 12:51 +0200, Avi Kivity wrote: > Ah, then it's not a guest problem, it's how the chip was designed.=20 > Newer chips do allow a workaround (GR31 bit 6): >=20 > 6 System Source Location (Revision A): If this bit is =E2=80=981=E2=80= =99, the CL-GD546X > responds to write accesses at 000BC000h=E2=80=93000BFFFFh for color-e= xpand > BitBLTs. This frees the linear address apertures for other, concurren= t > accesses. If this bit is =E2=80=980=E2=80=99, the CL-GD546X uses the = linear aperture for > BitBLTs (compatible with CL-GD543X/=E2=80=994X). > System Source Location (Revision B): If this bit is =E2=80=981=E2=80= =99, > system-to-screen BitBLTs use the second 16-Mbyte window specified in > PCI10. This allows direct frame buffer accesses in the first window t= o > be mixed with system-to-screen writes in the second window without > restrictions. >=20 > If a system-to-screen BitBLT requiring data is not active, writes to = the > second window complete in the minimum time and the data is discarded. > Writes to the first window are ignored by the BitBLT engine (but are > taken as direct writes to the frame buffer). >=20 > If this bit is =E2=80=980=E2=80=99, system-to-screen BitBLTs use the = first 16-Mbyte > window (compatible with CL-GD543X/=E2=80=994X). >=20 > But it looks like this refers to 546x, even though I found it in the > 5446 manual. Right. The first option uses legacy VGA memory which I don't always hav= e access to so that's not really an option for cirrusfb in general (in fact I made it not use IO either, it's doing full MMIO for configuring the CRTCs as well). I could (I have patches to do so) open a ISA/VGA memory window on the bridge, in fact I need that for the X driver to work for other reasons, but I'd rather not have cirrusfb deal with that. The funny thing here is we have "clean" APIs to let userspace map (when available) and access legacy VGA memory & IO ports, but we don't have a good in-kernel API to do the same thing :-) On x86 things are somewhat easy because it's just all hard coded and there's really only one PCI d= o main but on anything else it's a mess. In any case, X seems to avoid it, maybe nobody does color expansion nowadays (I suppose modern desktops don't, maybe using ancient X apps might trigger that path). So no biggie. I'll have to fix KVM powerpc dealing with memslot changes anyways. Thanks for your help, Cheers, Ben.