From: Eduardo Habkost <ehabkost@redhat.com>
To: Claudio Fontana <cfontana@suse.de>
Cc: "Paul Durrant" <paul@xen.org>, "Jason Wang" <jasowang@redhat.com>,
qemu-devel@nongnu.org, "Peter Xu" <peterx@redhat.com>,
haxm-team@intel.com, "Colin Xu" <colin.xu@intel.com>,
"Olaf Hering" <ohering@suse.de>,
"Stefano Stabellini" <sstabellini@kernel.org>,
"Bruce Rogers" <brogers@suse.com>,
"Emilio G . Cota" <cota@braap.org>,
"Anthony Perard" <anthony.perard@citrix.com>,
"Philippe Mathieu-Daudé" <philmd@redhat.com>,
"Laurent Vivier" <lvivier@redhat.com>,
"Thomas Huth" <thuth@redhat.com>,
"Richard Henderson" <richard.henderson@linaro.org>,
"Cameron Esfahani" <dirty@apple.com>,
"Dario Faggioli" <dfaggioli@suse.com>,
"Roman Bolshakov" <r.bolshakov@yadro.com>,
"Sunil Muthuswamy" <sunilmut@microsoft.com>,
"Marcelo Tosatti" <mtosatti@redhat.com>,
"Wenchao Wang" <wenchao.wang@intel.com>,
"Paolo Bonzini" <pbonzini@redhat.com>
Subject: Re: [RFC v6 11/11] i386: split cpu accelerators from cpu.c, using AccelCPUClass
Date: Fri, 27 Nov 2020 13:33:58 -0500 [thread overview]
Message-ID: <20201127183358.GG2271382@habkost.net> (raw)
In-Reply-To: <588c357c-da76-25bd-61f4-c28da0982639@suse.de>
On Fri, Nov 27, 2020 at 07:16:25PM +0100, Claudio Fontana wrote:
> On 11/27/20 7:09 PM, Eduardo Habkost wrote:
> > On Fri, Nov 27, 2020 at 06:53:23PM +0100, Claudio Fontana wrote:
> >> On 11/27/20 6:41 PM, Eduardo Habkost wrote:
> >>> On Thu, Nov 26, 2020 at 11:32:18PM +0100, Claudio Fontana wrote:
> >>>> i386 is the first user of AccelCPUClass, allowing to split
> >>>> cpu.c into:
> >>>>
> >>>> cpu.c cpuid and common x86 cpu functionality
> >>>> host-cpu.c host x86 cpu functions and "host" cpu type
> >>>> kvm/cpu.c KVM x86 AccelCPUClass
> >>>> hvf/cpu.c HVF x86 AccelCPUClass
> >>>> tcg/cpu.c TCG x86 AccelCPUClass
> >>>>
> >>>> Signed-off-by: Claudio Fontana <cfontana@suse.de>
> >>>> ---
> >>> [...]
> >>>> +static void tcg_cpu_class_init(CPUClass *cc)
> >>>
> >>> Is this the only case where we need to provide an
> >>> AccelCPUClass.cpu_class_init method?
> >>>
> >>>
> >>>> +{
> >>>> + cc->do_interrupt = x86_cpu_do_interrupt;
> >>>> + cc->cpu_exec_interrupt = x86_cpu_exec_interrupt;
> >>>> + cc->synchronize_from_tb = x86_cpu_synchronize_from_tb;
> >>>> + cc->cpu_exec_enter = x86_cpu_exec_enter;
> >>>> + cc->cpu_exec_exit = x86_cpu_exec_exit;
> >>>> + cc->tcg_initialize = tcg_x86_init;
> >>>> + cc->tlb_fill = x86_cpu_tlb_fill;
> >>>> +#ifndef CONFIG_USER_ONLY
> >>>> + cc->debug_excp_handler = breakpoint_handler;
> >>>> +#endif /* !CONFIG_USER_ONLY */
> >>>
> >>> I find the need for these method overrides suspicious.
> >>
> >> These mechanisms are preexistent. My refactoring only makes them more visible.
> >>
> >>>
> >>> Comparing this with the code on qemu.git master:
> >>>
> >>> static void x86_cpu_common_class_init(ObjectClass *oc, void *data)
> >>> {
> >>> [...]
> >>> #ifdef CONFIG_TCG
> >>> cc->do_interrupt = x86_cpu_do_interrupt;
> >>> cc->cpu_exec_interrupt = x86_cpu_exec_interrupt;
> >>> #endif
> >>> [...]
> >>> cc->synchronize_from_tb = x86_cpu_synchronize_from_tb;
> >>> [...]
> >>> #if defined(CONFIG_TCG) && !defined(CONFIG_USER_ONLY)
> >>> cc->debug_excp_handler = breakpoint_handler;
> >>> #endif
> >>> cc->cpu_exec_enter = x86_cpu_exec_enter;
> >>> cc->cpu_exec_exit = x86_cpu_exec_exit;
> >>> #ifdef CONFIG_TCG
> >>> cc->tcg_initialize = tcg_x86_init;
> >>> cc->tlb_fill = x86_cpu_tlb_fill;
> >>> #endif
> >>> [...]
> >>> }
> >>>
> >>> So, we have two kinds of CPUClass fields above:
> >>> * Code that was never conditional on CONFIG_TCG, and now is
> >>> conditional (synchronize_from_tb, cpu_exec_enter,
> >>> cpu_exec_exit).
> >>
> >>
> >>
> >> synchronize_from_tb, cpu_exec_enter and cpu_exec_exit only makes sense for TCG,
> >> and their code should not be compiled in for non-TCG.
> >>
> >> This is part of the effort to separate away non-pertinent code into accelerator-specific builds (and in the future modules).
> >>
> >> The fact that they were unconditionally compiled in before was a mistake, or at least this is the assumption I am making when changing this.
> >
> > Can you clarify why that was a mistake? Patch 07/11 makes
> > existing functions TCG-specific, but doesn't explain why.
>
>
>
> They are functions called only for TCG. You don't need translation blocks and cpu_exec() for other accelerators.
> I can add this information in the commit message.
Thanks! I believe we can clarify in patch 07/11 why some methods
that were not conditional on CONFIG_TCG were moved to
tcg_cpu_common_class_init().
Do you think it would be too complicated to move those methods
from CPUClass to CpusAccel? If they are really accel-specific,
CPUClass looks like the wrong place for them.
> [...]
--
Eduardo
prev parent reply other threads:[~2020-11-27 18:35 UTC|newest]
Thread overview: 43+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-11-26 22:32 [RFC v6 00/11] i386 cleanup Claudio Fontana
2020-11-26 22:32 ` [RFC v6 01/11] i386: move kvm accel files into kvm/ Claudio Fontana
2020-11-26 22:32 ` [RFC v6 02/11] i386: move whpx accel files into whpx/ Claudio Fontana
2020-11-26 22:32 ` [RFC v6 03/11] i386: move hax accel files into hax/ Claudio Fontana
2020-11-26 22:32 ` [RFC v6 04/11] i386: hvf: remove stale MAINTAINERS entry for old hvf stubs Claudio Fontana
2020-11-26 22:32 ` [RFC v6 05/11] i386: move TCG accel files into tcg/ Claudio Fontana
2020-11-26 22:32 ` [RFC v6 06/11] i386: move cpu dump out of helper.c into cpu-dump.c Claudio Fontana
2020-11-26 22:32 ` [RFC v6 07/11] i386: move TCG cpu class initialization out of helper.c Claudio Fontana
2020-11-27 19:04 ` Eduardo Habkost
2020-11-27 19:47 ` Claudio Fontana
2020-11-27 20:43 ` Eduardo Habkost
2020-11-29 11:53 ` Claudio Fontana
2020-11-26 22:32 ` [RFC v6 08/11] accel: extend AccelState and AccelClass to user-mode Claudio Fontana
2021-01-11 18:43 ` Claudio Fontana
2021-01-12 9:23 ` Claudio Fontana
2020-11-26 22:32 ` [RFC v6 09/11] accel: replace struct CpusAccel with AccelOpsClass Claudio Fontana
2020-11-26 22:32 ` [RFC v6 10/11] accel: introduce AccelCPUClass extending CPUClass Claudio Fontana
2020-11-27 6:21 ` Paolo Bonzini
2020-11-27 8:59 ` Claudio Fontana
2020-11-27 11:22 ` Claudio Fontana
2020-11-27 11:41 ` Claudio Fontana
2020-11-27 13:31 ` Paolo Bonzini
2020-11-27 13:32 ` Claudio Fontana
2020-12-18 17:51 ` Claudio Fontana
2020-12-18 18:01 ` Paolo Bonzini
2020-12-18 18:04 ` Claudio Fontana
2020-12-18 21:55 ` Claudio Fontana
2020-12-18 22:30 ` Claudio Fontana
2020-12-18 23:00 ` Claudio Fontana
2021-01-11 16:13 ` Claudio Fontana
2021-01-11 18:08 ` Claudio Fontana
2021-01-11 22:35 ` Eduardo Habkost
2021-01-11 23:35 ` Claudio Fontana
2020-11-27 17:06 ` Eduardo Habkost
2020-11-27 17:58 ` Claudio Fontana
2020-11-27 18:13 ` Eduardo Habkost
2020-11-27 18:20 ` Claudio Fontana
2020-11-26 22:32 ` [RFC v6 11/11] i386: split cpu accelerators from cpu.c, using AccelCPUClass Claudio Fontana
2020-11-27 17:41 ` Eduardo Habkost
2020-11-27 17:53 ` Claudio Fontana
2020-11-27 18:09 ` Eduardo Habkost
2020-11-27 18:16 ` Claudio Fontana
2020-11-27 18:33 ` Eduardo Habkost [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20201127183358.GG2271382@habkost.net \
--to=ehabkost@redhat.com \
--cc=anthony.perard@citrix.com \
--cc=brogers@suse.com \
--cc=cfontana@suse.de \
--cc=colin.xu@intel.com \
--cc=cota@braap.org \
--cc=dfaggioli@suse.com \
--cc=dirty@apple.com \
--cc=haxm-team@intel.com \
--cc=jasowang@redhat.com \
--cc=lvivier@redhat.com \
--cc=mtosatti@redhat.com \
--cc=ohering@suse.de \
--cc=paul@xen.org \
--cc=pbonzini@redhat.com \
--cc=peterx@redhat.com \
--cc=philmd@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=r.bolshakov@yadro.com \
--cc=richard.henderson@linaro.org \
--cc=sstabellini@kernel.org \
--cc=sunilmut@microsoft.com \
--cc=thuth@redhat.com \
--cc=wenchao.wang@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).