From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christian Ehrhardt Subject: Re: [PATCH 1/3] radeonfb: Fix 64 bits resources on 32 bits archs Date: Tue, 22 Apr 2008 17:55:00 +0200 Message-ID: <480E0A54.9010203@linux.vnet.ibm.com> References: <20080422012723.BA9F2DE13F@ozlabs.org> Mime-Version: 1.0 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: In-Reply-To: <20080422012723.BA9F2DE13F@ozlabs.org> Sender: linux-kernel-owner@vger.kernel.org List-ID: Content-Type: text/plain; charset="iso-8859-1"; format="flowed" To: Benjamin Herrenschmidt Cc: linux-fbdev-devel@lists.sourceforge.net, linuxppc-dev@ozlabs.org, Andrew Morton , linux-kernel@vger.kernel.org, adaplas@gmail.com, Hollis Blanchard , Wolfgang Denk , Detlev Zundel Benjamin Herrenschmidt wrote: > This fixes radeonfb to not truncate 64 bits resources on 32 bits > platforms. Unfortunately, there are still issues with addresses > returned to userspace via struct fb_fix_screeninfo. This will > have to be dealt with separately. Thanks for this patch Benjamin, I use it together with what we have dis= cussed in the "pci issue - wrong detection of pci ressources" thread. Unfortunately I now hit exactly that issue with fb_fix_screeninfo you d= escribe. =46or everyone the fb_fix_screeninfo has two "unsigned long" vars that = need to strore a IO address. This fails in my case with a 32bit powerpc system (=3Dsizeof(long)=3D4)= which has paddr >4Gb and actually it should affect any 32bit platform = with paddr>4Gb. You see it e.g. when you try to initialize X11, the x11 radeon driver i= ssues a FBIOGET_FSCREENINFO ioctl and because our address is >4Gb it ge= t's clobbered by that unsigned long in the fb_fix_screeninfo structure = (the value comes from a resource_size_t variable which has the correct = 64bit). struct fb_fix_screeninfo { char id[16]; /* identification string eg "TT= Builtin" */ unsigned long smem_start; /* Start of frame buffer mem */ [...] unsigned long mmio_start; /* Start of Memory Mapped I/O *= / [...] I tried the stupid solution to just change the fb_fix_screeninfo struct= ure to resource_size_t, but that changes the size ioctl transports and = would require awareness in the userspace applications using that ioctl. The X11 & Framebuffer driver work on 64bit systems, so I think it's jus= t an issue of not cutting that data down to 32bit when transporting it = (has anyone already checked the x11 drivers, I hope they don't use unsi= gned long too). I wanted to ask if there are any known workarounds atm that would allow= me to use my X11 for now? --=20 Gr=FCsse / regards,=20 Christian Ehrhardt IBM Linux Technology Center, Open Virtualization