MIME-Version: 1.0
In-Reply-To: <CAMz4kuKbo5NauhRW_UdSfL4m4q28B4=pcve=EvW5xDpZ_nyjeg@mail.gmail.com>
References: <cover.1505973912.git.baolin.wang@linaro.org> <1091b589bec6317ea686937060c0f9f9db10651a.1505973912.git.baolin.wang@linaro.org>
 <CAK8P3a1uPWYMCUnsYfTo3MjcqopYEd1TnrPVqEts=ZiEAFdr8g@mail.gmail.com> <CAMz4kuKbo5NauhRW_UdSfL4m4q28B4=pcve=EvW5xDpZ_nyjeg@mail.gmail.com>
From: Arnd Bergmann <arnd@arndb.de>
Date: Fri, 22 Sep 2017 09:57:02 +0200
Message-ID: <CAK8P3a1oyUinL5vxuysRQHh7dq2kHyNzsOHb6Mk8MJobXyNgaw@mail.gmail.com>
Subject: Re: [RFC PATCH 7/7] sound: core: Avoid using timespec for struct snd_timer_tread
To: Baolin Wang <baolin.wang@linaro.org>
Cc: Jaroslav Kysela <perex@perex.cz>, Takashi Iwai <tiwai@suse.com>,
        Liam Girdwood <lgirdwood@gmail.com>, Ingo Molnar <mingo@kernel.org>,
        Takashi Sakamoto <o-takashi@sakamocchi.jp>,
        SF Markus Elfring <elfring@users.sourceforge.net>,
        Dan Carpenter <dan.carpenter@oracle.com>, jeeja.kp@intel.com,
        Vinod Koul <vinod.koul@intel.com>, dharageswari.r@intel.com,
        guneshwor.o.singh@intel.com, Bhumika Goyal <bhumirks@gmail.com>,
        gudishax.kranthikumar@intel.com, Naveen M <naveen.m@intel.com>,
        hardik.t.shah@intel.com, Arvind Yadav <arvind.yadav.cs@gmail.com>,
        Fabian Frederick <fabf@skynet.be>, Mark Brown <broonie@kernel.org>,
        Deepa Dinamani <deepa.kernel@gmail.com>, alsa-devel@alsa-project.org,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2035
Lines: 52

On Fri, Sep 22, 2017 at 5:00 AM, Baolin Wang <baolin.wang@linaro.org> wrote:
> On 21 September 2017 at 21:09, Arnd Bergmann <arnd@arndb.de> wrote:
>> On Thu, Sep 21, 2017 at 8:18 AM, Baolin Wang <baolin.wang@linaro.org> wrote:
>>
>>> +static int snd_timer_user_tread(void __user *argp, struct snd_timer_user *tu,
>>> +                               unsigned int cmd)
>>> +{
>>> +       int __user *p = argp;
>>> +       int xarg, old_tread;
>>> +
>>> +       if (tu->timeri) /* too late */
>>> +               return -EBUSY;
>>> +       if (get_user(xarg, p))
>>> +               return -EFAULT;
>>> +
>>> +       old_tread = tu->tread;
>>> +#if __BITS_PER_LONG == 64
>>> +       tu->tread = xarg ? 2 : 0;
>>> +#ifdef IA32_EMULATION
>>> +       tu->tread = xarg ? 3 : 0;
>>> +#endif
>>> +#else
>>> +       if (cmd == SNDRV_TIMER_IOCTL_TREAD64)
>>> +               tu->tread = xarg ? 2 : 0;
>>> +       else
>>> +               tu->tread = xarg ? 1 : 0;
>>> +#endif
>>
>> The 64-bit case looks broken here:
>>
>> - The tread flag is different for compat and native mode, so you
>>    must pass a flag to identify whether you are called from
>>    __snd_timer_user_ioctl or from snd_timer_user_ioctl_compat().
>
> I have some confusion here. For 64-bit, we will set tu->tread = 2 no
> matter it is native mode or compat mode, only we will set tu->tread =
> 3 for x86_32 in compat mode, right?
> So I think we do not need to identify whether called from native mode
> or compat mode.

When we have a user space program with 32-bit time_t in compat mode
(i.e. cmd==SNDRV_TIMER_IOCTL_TREAD) on a 64-bit kernel, we want
to set tread=1, and that is different from the native mode that wants to
set tread=2.

For determining whether to use tread=2 or tread=3, we have to check
both compat mode and x32 mode. This could be done by checking for
"if (IS_ENABLED(CONFIG_IA32_EMULATION) && in_compat_syscall() &&
is_x32_task())", but the in_compat_syscall() check can be skipped when
you know that you were called from .compat_ioct().

       Arnd