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: <d402a6a0a11d84906ecba3909f4c8f880298dc0f.camel@linux.ibm.com>
Subject: Re: [GIT PULL] Please pull powerpc/linux.git powerpc-6.1-1 tag
From:   Andrew Donnellan <ajd@linux.ibm.com>
To:     "Jason A. Donenfeld" <Jason@zx2c4.com>,
        Michael Ellerman <mpe@ellerman.id.au>, rmclure@linux.ibm.com
Cc:     Linus Torvalds <torvalds@linux-foundation.org>,
        aneesh.kumar@linux.ibm.com, atrajeev@linux.vnet.ibm.com,
        christophe.leroy@csgroup.eu, cuigaosheng1@huawei.com,
        david@redhat.com, farosas@linux.ibm.com, geoff@infradead.org,
        gustavoars@kernel.org, haren@linux.ibm.com, hbathini@linux.ibm.com,
        joel@jms.id.au, lihuafei1@huawei.com, linux-kernel@vger.kernel.org,
        linux@roeck-us.net, linuxppc-dev@lists.ozlabs.org,
        lukas.bulwahn@gmail.com, mikey@neuling.org, nathan@kernel.org,
        nathanl@linux.ibm.com, nicholas@linux.ibm.com, npiggin@gmail.com,
        pali@kernel.org, paul@paul-moore.com, ruscur@russell.cc,
        windhl@126.com, wsa+renesas@sang-engineering.com,
        ye.xingchen@zte.com.cn, yuanjilin@cdjrlc.com,
        zhengyongjun3@huawei.com
Date:   Tue, 11 Oct 2022 11:00:15 +1100
In-Reply-To: <Y0ScAhqysKK6Hrks@zx2c4.com>
References: <87edvhntv0.fsf@mpe.ellerman.id.au> <Y0RxpSFyn9m68zIb@zx2c4.com>
         <Y0R6fcvgGA858TQA@zx2c4.com> <Y0ScAhqysKK6Hrks@zx2c4.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.38.3-1 
Content-Transfer-Encoding: 8bit
MIME-Version: 1.0
Precedence: bulk

On Mon, 2022-10-10 at 16:26 -0600, Jason A. Donenfeld wrote:
> 
> Bisected:
> 
> 7e92e01b724526b98cbc7f03dd4afa0295780d56 is the first bad commit
> commit 7e92e01b724526b98cbc7f03dd4afa0295780d56
> Author: Rohan McLure <rmclure@linux.ibm.com>
> Date:   Wed Sep 21 16:56:01 2022 +1000
> 
>     powerpc: Provide syscall wrapper
> 
>     Implement syscall wrapper as per s390, x86, arm64. When enabled
>     cause handlers to accept parameters from a stack frame rather
> than
>     from user scratch register state. This allows for user registers
> to be
>     safely cleared in order to reduce caller influence on speculation
>     within syscall routine. The wrapper is a macro that emits syscall
>     handler symbols that call into the target handler, obtaining its
>     parameters from a struct pt_regs on the stack.
> 
>     As registers are already saved to the stack prior to calling
>     system_call_exception, it appears that this function is executed
> more
>     efficiently with the new stack-pointer convention than with
> parameters
>     passed by registers, avoiding the allocation of a stack frame for
> this
>     method. On a 32-bit system, we see >20% performance increases on
> the
>     null_syscall microbenchmark, and on a Power 8 the performance
> gains
>     amortise the cost of clearing and restoring registers which is
>     implemented at the end of this series, seeing final result of
> ~5.6%
>     performance improvement on null_syscall.
> 
>     Syscalls are wrapped in this fashion on all platforms except for
> the
>     Cell processor as this commit does not provide SPU support. This
> can be
>     quickly fixed in a successive patch, but requires
> spu_sys_callback to
>     allocate a pt_regs structure to satisfy the wrapped calling
> convention.
> 
>     Co-developed-by: Andrew Donnellan <ajd@linux.ibm.com>
>     Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
>     Signed-off-by: Rohan McLure <rmclure@linux.ibm.com>
>     Reviewed-by: Nicholas Piggin <npiggin@gmai.com>
>     [mpe: Make incompatible with COMPAT to retain clearing of high
> bits of args]
>     Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
>     Link:
> https://lore.kernel.org/r/20220921065605.1051927-22-rmclure@linux.ibm.com

Thanks for bisecting, this is interesting! Could you provide your
.config and the environment you're running in? Your reproducer doesn't
seem to trigger it on my baremetal POWER8 pseries_le_defconfig.

-- 
Andrew Donnellan    OzLabs, ADL Canberra
ajd@linux.ibm.com   IBM Australia Limited