From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:46007) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1axX4J-0004hB-6Y for qemu-devel@nongnu.org; Tue, 03 May 2016 05:56:45 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1axX47-0001an-9g for qemu-devel@nongnu.org; Tue, 03 May 2016 05:56:33 -0400 Received: from mail-lf0-x22a.google.com ([2a00:1450:4010:c07::22a]:36282) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1axX45-0001NP-Kp for qemu-devel@nongnu.org; Tue, 03 May 2016 05:56:27 -0400 Received: by mail-lf0-x22a.google.com with SMTP id u64so15554133lff.3 for ; Tue, 03 May 2016 02:56:11 -0700 (PDT) References: <5727B088.1090400@gmail.com> <5727B623.2070901@gmail.com> From: Sergey Fedorov Message-ID: <572875B6.5090705@gmail.com> Date: Tue, 3 May 2016 12:56:06 +0300 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] tcg: How 'CPUState::current_tb' is used? List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Peter Maydell Cc: QEMU Developers , Blue Swirl , Paolo Bonzini , Riku Voipio , =?UTF-8?Q?Alex_Benn=c3=a9e?= , Richard Henderson On 03/05/16 03:02, Peter Maydell wrote: > On 2 May 2016 at 21:18, Sergey Fedorov wrote: >> On 02/05/16 22:54, Sergey Fedorov wrote: >> >> Hi, >> >> I can't figure out how this field is used. The comment says it's "Currently >> executing TB", but actually it's the first TB in a chain of TBs executed. >> Grep shows the only place it is really checked is >> tb_invalidate_phys_page_range(). That code seems to be introduced long ago >> in: >> >> commit ea1c18022edd0e2c45552d6fc2da6e15a3486b33 >> Author: bellard >> Date: Mon Jun 14 18:56:36 2004 +0000 >> >> fixed self modifying code in case of asynchronous interrupt >> >> >> I suspect it's only related to user emulation. But I would appreciate if >> someone could give me an idea of how this really works :) >> >> >> UPD: 'CPUState::current_tb' was used in that version of QEMU by this code: >> >> /* mask must never be zero, except for A20 change call */ >> void cpu_interrupt(CPUState *env, int mask) >> { >> TranslationBlock *tb; >> static int interrupt_lock; >> >> env->interrupt_request |= mask; >> /* if the cpu is currently executing code, we must unlink it and >> all the potentially executing TB */ >> tb = env->current_tb; >> if (tb && !testandset(&interrupt_lock)) { >> env->current_tb = NULL; >> tb_reset_jump_recursive(tb); >> interrupt_lock = 0; >> } >> } >> >> >> cpu_interrupt() has changed almost completely since that time. I'm wondering >> if checking 'cpu->current_tb' by this code in >> tb_invalidate_phys_page_range() still makes any sense: >> >> if (cpu->interrupt_request && cpu->current_tb) { >> cpu_interrupt(cpu, cpu->interrupt_request); >> } >> >> >> BTW, I'm not sure about the purpose of this piece of code either :) > I think it's now obsolete. When cpu_interrupt() worked > by unlinking the TB being executed and all the ones that it > chained to, then (as you see in the code you quote) cpu_interrupt() > only did actual work if env->current_tb was set. The code in > tb_invalidate_phys_page_range() doesn't want that work to happen > while it's in tb_phys_invalidate() [it would have tried to > modify the TB graph in the signal handler in the middle of > tb_phys_invalidate also modifying the graph and corrupted it], > so it sets cpu->current_tb to NULL to suppress this. However > that then meant that if we had an asynchronous interrupt > (ie executed cpu_interrupt() in a signal handler) it would > have done nothing, so the tb_invalidate_phys_page_range() > code now has to say "if we did get an interrupt, do the work > now" after it restores the current_tb pointer. > > Since cpu_interrupt() no longer does complicated TB graph > modification it now does it unconditionally, so the work > done by tb_invalidate_phys_page_range() to clear cpu->current_tb > is unnecessary and so is the extra call to cpu_interrupt() > afterwards. > > So I think the current_tb field can be deleted, and so > can the code fragments > /* we need to do that to handle the case where a signal > occurs while doing tb_phys_invalidate() */ > saved_tb = NULL; > if (cpu != NULL) { > saved_tb = cpu->current_tb; > cpu->current_tb = NULL; > } > > and > if (cpu != NULL) { > cpu->current_tb = saved_tb; > if (cpu->interrupt_request && cpu->current_tb) { > cpu_interrupt(cpu, cpu->interrupt_request); > } > } > > because with our current code a signal and resulting > call to cpu_interrupt() is perfectly safe even if it > happens while we're executing tb_phys_invalidate(). Thank you for the detailed explanation. Now, in the morning, I can see it :) I'll prepare a patch for this. Kind regards, Sergey