Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <6ded07cc-4a78-a509-a588-d658f720a70d@linaro.org>
Date:   Thu, 6 Jul 2023 07:13:08 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.11.0
Subject: Re: qemu-x86_64 booting with 8.0.0 stil see int3: when running LTP
 tracing testing.
Content-Language: en-US
To:     "Richard W.M. Jones" <rjones@redhat.com>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Arnd Bergmann <arnd@arndb.de>,
        Naresh Kamboju <naresh.kamboju@linaro.org>,
        Anders Roxell <anders.roxell@linaro.org>,
        =?UTF-8?Q?Daniel_D=c3=adaz?= <daniel.diaz@linaro.org>,
        Benjamin Copeland <ben.copeland@linaro.org>,
        lkml - Kernel Mailing List <linux-kernel@vger.kernel.org>,
        the arch/x86 maintainers <x86@kernel.org>,
        Paolo Bonzini <pbonzini@redhat.com>
References: <CA+G9fYsETJQm0Ue7hGsb+nbsiMikwycOV3V0DPr6WC2r61KRBQ@mail.gmail.com>
 <2d7595b1-b655-4425-85d3-423801bce644@app.fastmail.com>
 <20230621160655.GL2053369@hirez.programming.kicks-ass.net>
 <20230704074620.GA17440@redhat.com> <20230705162830.GC17440@redhat.com>
 <CAFXwXrmbpuFNf5=nQxiTteo8fpCdAbK4pEAN176Cq0yvwZcfFw@mail.gmail.com>
 <20230705163523.GJ7636@redhat.com>
 <CAFXwXrk1FEZPUO-zqNVJZ6YCHKUkgNehwmyDYuOr5fx8ff0OCA@mail.gmail.com>
 <20230705164006.GL7636@redhat.com>
From:   Richard Henderson <richard.henderson@linaro.org>
In-Reply-To: <20230705164006.GL7636@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 7/5/23 17:40, Richard W.M. Jones wrote:
> On Wed, Jul 05, 2023 at 06:36:43PM +0200, Richard Henderson wrote:
>> No, I thought it would be the fix for 8.0.0.
> 
> Thanks for the suggestion anyway.
> 
> Am I right in thinking that tb_invalidate_phys_page() ought to be
> called when the kernel self-modifies its text segment?  If there's
> some function that we expect to be called in this case then I can
> instrument it.

All such go through notdirty_write(), which has a convenient tracepoint.  Recall there are 
migration and vga dirty pages as well, which also go through here, so disabling vga if 
possible may help.

Code stuff should then go through tb_invalidate_phys_range_fast(), so not 
tb_invalidate_phys_page() at all.


r~