On Wed, 2008-12-17 at 11:07 +0800, Liu Yu wrote: > > -----Original Message----- > > From: kvm-ppc-owner@xxxxxxxxxxxxxxx > > [mailto:kvm-ppc-owner@xxxxxxxxxxxxxxx] On Behalf Of Hollis Blanchard > > Sent: Tuesday, December 16, 2008 11:44 PM > > To: Liu Yu-B13201 > > Cc: kvm-ppc@xxxxxxxxxxxxxxx > > Subject: RE: [PATCH 07/11] E500 TLB emulation > > > > On Tue, 2008-12-16 at 18:08 +0800, Liu Yu wrote: > > > > > > > To fix this, instead of doing error-checking in all these > > > > sites, I would > > > > just mask with 1 instead of 3. > > > > > > > > > > If the guest instruction appoint MAS0[TLBSEL] = 2 or 3, > > > I think the best way is to ignore it. > > > > Masking with 1 instead of 3 will ignore it. Or do you mean completely > > ignore the instruction? The mask is and simpler and faster, and also > > perfectly legal (it was a software error to begin with). > > > > Actually, my e500 user manual (covering v1 and v2) says the > > TLBSEL field > > is only 1 bit wide, with reserved bits on either side. So > > masking with 1 > > is clearly the right solution. > > > > I found the problem. > Old manual explain the tlbivax EA format that bit 60 is used for tlbsel. > Howerver the newest verstion tells that bit 59~60 of tlbivax EA are used > for tlbsel. > Correspondingly, the MAS0 use 34~35 bits for tlbsel now. > > http://www.freescale.com/files/32bit/doc/ref_manual/EREFRM.pdf?WT_TYPE=R > eference%20Manuals&WT_VENDOR=FREESCALE&WT_FILE_FORMAT=pdf&WT_ASSET=Docum > entation This document is strange, because it seems to be trying to describe multiple Freescale cores in conjunction with the full architecture. I have to believe that the core's user manual is more accurate, since it provides information about divergences from the architecture. The e500 UM says tlbsel is 1 bit, so that's what I believe. > I think I should check the case that tlbsel is not valid in the code. And do what if the check fails? The document you listed above says "If the specified entry does not exist, the results are undefined" for both tlbwe and tlbre, which means you get to do whatever you want. You already open-code TLBSEL extraction in a number of places: get_tlb_tlbsel() uses MAS0, tlbivax uses the effective address, and tlbsx uses MAS4. It makes me nervous because the more places you do it, the more likely it is you'll miss a check somewhere. Look at it this way: no good *at* *all* can come from having (tlbsel & 0x2) == 1. There is absolutely no reason that should ever happen. So don't allow it to happen, and do it automatically so there's never a risk of forgetting to check for that condition. -- Hollis Blanchard IBM Linux Technology Center -- To unsubscribe from this list: send the line "unsubscribe kvm-ppc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
