Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:1 as permitted sender) client-ip=2620:137:e000::3:1;
Message-ID: <d35af5a1-12c1-4b5a-8e9e-c4bc5bda4de2@igalia.com>
Date:   Wed, 29 Nov 2023 15:56:12 -0300
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH v1] futex: Add compat_sys_futex_waitv for 32bit
 compatibility
To:     Wei Gao <wegao@suse.com>
Cc:     tglx@linutronix.de, dvhart@infradead.org,
        linux-kernel@vger.kernel.org, dave@stgolabs.net, mingo@redhat.com,
        peterz@infradead.org, Arnd Bergmann <arnd@kernel.org>
References: <20231123053140.16062-1-wegao@suse.com>
 <189e733e-7056-45c4-a5c6-3371f8219b7e@igalia.com> <ZWSIfy+QJFyw/zGC@wegao>
Content-Language: en-US
From:   =?UTF-8?Q?Andr=C3=A9_Almeida?= <andrealmeid@igalia.com>
In-Reply-To: <ZWSIfy+QJFyw/zGC@wegao>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

Hi Wei,

Em 27/11/2023 09:15, Wei Gao escreveu:
> On Thu, Nov 23, 2023 at 01:09:55PM -0300, André Almeida wrote:
>> [+CC Arnd]
>>
>> Hi Wei,
>>
>> Em 23/11/2023 02:31, Wei Gao escreveu:
>>> From: wei gao <wegao@suse.com>
>>>
>>> Current implementation lead LTP test case futex_waitv failed when compiled with
>>> -m32. This patch add new compat_sys_futex_waitv to handle m32 mode syscall.
>>>
>>> The failure reason is futex_waitv in m32 mode will deliver kernel with struct
>>> old_timespec32 timeout, but this struct type can not directly used by current
>>> sys_futex_waitv implementation.
>>>
>>> The new function copy main logic of current sys_futex_waitv, just update parameter
>>> type from "struct __kernel_timespec __user *" to "struct old_timespec32 __user *,"
>>> and use get_old_timespec32 within the new function to get timeout value.
>>>
>>
>> From, what I recall, we don't want to add new syscalls with old_timespec32,
>> giving that they will have a limited lifetime. Instead, userspace should be
>> able to come up with a 64-bit timespec implementation for -m32.
>>
>> Thanks,
>> 	André
> 
> Just a comment, I have checked the glibc latest code but do not see any implemention(*.c) on
> futex_waitv syscall. So normally you have to do syscall directly with __NR_futex_waitv from
> userspace. So i guess glibc-side can not covert this struct correctly currently. Correct me if
> any misunderstanding.
> 

futex() has no syscall wrappers in glibc. Userspace needs to figure out 
everything by themselves, including which struct they should use, and I 
don't think that glibc does any conversion. If you create manually a 
timespec64 that works in -m32, and pass this to sycall(__NR_futex_waitv, 
..., &timeout, ...), it should work correctly. You can read more about 
how glibc is planning to deal with this at [1]. Please let me know if 
now it's more clear :)

[1] https://sourceware.org/glibc/wiki/Y2038ProofnessDesign

> Thanks
> Wei Gao