From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: <4BF50D90.20203@domain.hid> Date: Thu, 20 May 2010 12:23:12 +0200 From: Jan Kiszka MIME-Version: 1.0 References: <201005191125.51812.smolorz@domain.hid> <4BF50124.20801@domain.hid> <4BF50BEA.4030208@domain.hid> In-Reply-To: <4BF50BEA.4030208@domain.hid> Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Subject: Re: [Adeos-main] [BUG] Interrupt recursion with libata-sff List-Id: General discussion about Adeos List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Gilles Chanteperdrix Cc: adeos-main Gilles Chanteperdrix wrote: > Jan Kiszka wrote: >> Looks like your problem is x86-32 returning from interrupts with hard >> IRQs enabled (and therefore flushing the pipeline already over the >> interrupt handler). That should ideally be unified with x86-64 now, also >> as x86-64 may already gain better testing than 32 bit. > > I may have not understood right, but from my understanding, the big > difference between x86_32 and x86_64 is that interrupts flags are > virtualized in x86_32's entry.S. To unify x86_32 and x86_64 would > basically mean rewrite this. And the the two entry.S are separated, so > x86_32 has to be tested separately from x86_64. For sure. But there will be more synergies in arch/x86/kernel/ipipe.c then, less #ifdefs. And the complexity of entry_32.S changes may decrease. Moreover, given the same design, there is also a higher probability to catch common problems. Jan -- Siemens AG, Corporate Technology, CT T DE IT 1 Corporate Competence Center Embedded Linux