On Tue, Apr 10, 2012 at 5:29 AM, Bao, Zheng <[email protected]> wrote: > I tried. It behaves as bad as wbinvd does. > > Joe > >> -----Original Message----- >> From: Andrew Goodbody [mailto:[email protected]] >> Sent: Tuesday, April 10, 2012 7:04 PM >> To: Bao, Zheng >> Cc: 'Marc Jones'; Marc Jones; '[email protected]' >> Subject: Re: [coreboot] Patch merged into coreboot/master: c35c461 >> Invalidate cache before first jump >> >> On 09/04/12 04:24, Bao, Zheng wrote: >> > Hi, >> > It is unstable. In most cases, it hangs at this wbinvd. Once it pass >> that instruction, it will hang at when AP cores are launched. >> >> Try invd instead of wbinvd. There should be no need to preserve >> anything >> from the cache so the write back is not needed. The write back may also >> need functional memory which will not be there at least from power on. >> Also note that these instructions are 486+ only. >> >> Andrew
An wbind/invd by the APs may to cause problems with CAR. I think that the coherency rules will cause a wbind on all cores, but I am not positive. Stefan, can you explain the requirement or consider how else to only do this on a processor that requires it? Marc -- http://se-eng.com -- coreboot mailing list: [email protected] http://www.coreboot.org/mailman/listinfo/coreboot
