Received-SPF: pass (google.com: domain of linux-kernel+bounces-179883-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Message-ID: <8aa947c2-5373-4efd-83db-6a5f75069317@I-love.SAKURA.ne.jp>
Date: Wed, 15 May 2024 22:12:00 +0900
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [Linux kernel bug] INFO: task hung in blk_mq_get_tag
To: Sam Sun <samsun1006219@gmail.com>
Cc: Hillf Danton <hdanton@sina.com>, linux-kernel@vger.kernel.org,
        linux-block@vger.kernel.org, axboe@kernel.dk,
        syzkaller-bugs@googlegroups.com, xrivendell7@gmail.com
References: <CAEkJfYMhv8AxxHSVdPT9bCX1cJZXw39+bMFh=2N9uNOB4Hcr=w@mail.gmail.com>
 <20240514103742.3137-1-hdanton@sina.com>
 <CAEkJfYPH3SJ6J3kLSjMGqkWOzgbgKZV_f2Hq05cpZZv7RmhvOg@mail.gmail.com>
 <cfe6b902-5e2d-415d-afeb-672cafd8d0b7@I-love.SAKURA.ne.jp>
 <CAEkJfYOBnhVGNENZkkdoX1g+gE=eVNFwNRG6HwFUdCw2+Kda=Q@mail.gmail.com>
Content-Language: en-US
From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
In-Reply-To: <CAEkJfYOBnhVGNENZkkdoX1g+gE=eVNFwNRG6HwFUdCw2+Kda=Q@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

On 2024/05/15 21:46, Sam Sun wrote:
>> What happens if you disable
>>
>>   sysfd = write(sysfd, input, hash - input + 1);
>>
>> line (i.e. stop updating sg_allow_dio value) in the reproducer?
>>
> 
> I tried to change the value of /sys/module/sg/parameters/allow_dio to
> 0 and remove write() call, both still triggers task hang report and
> kernel panic. I think this write is not the call crashing the kernel.
> 

Kernel panic by general protection fault upon calling trigger_all_cpu_backtrace() is
a different bug. Please be sure to keep /proc/sys/kernel/hung_task_all_cpu_backtrace 0
while investigating this hung task problem.

This hung task problem happens without touching /sys/module/sg/parameters/allow_dio ,
doesn't it?

scsi_rescan_device() is reliably printed when this hung task problem happens,
isn't it?

Then, it is strange that scsi_rescan_device() is called despite the reproducer is
almost no-op. Maybe you can trigger scsi_rescan_device() without using the reproducer.

Can you simplify steps to reproduce? For example, doing a lot of write().