Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH v3] cfi: fix deadloop in cfi_cmdset_0002.c do_write_buffer
To:     Tokunori Ikegami <ikegami_to@yahoo.co.jp>,
        "'liujian (CE)'" <liujian56@huawei.com>
CC:     "'Boris Brezillon'" <boris.brezillon@collabora.com>,
        "'Tokunori Ikegami'" <ikegami.t@gmail.com>,
        "bbrezillon@kernel.org" <bbrezillon@kernel.org>,
        "ikegami@allied-telesis.co.jp" <ikegami@allied-telesis.co.jp>,
        "richard@nod.at" <richard@nod.at>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "marek.vasut@gmail.com" <marek.vasut@gmail.com>,
        "linux-mtd@lists.infradead.org" <linux-mtd@lists.infradead.org>,
        "computersforpeace@gmail.com" <computersforpeace@gmail.com>,
        "dwmw2@infradead.org" <dwmw2@infradead.org>,
        "keescook@chromium.org" <keescook@chromium.org>
References: <1551189648-58073-1-git-send-email-liujian56@huawei.com>
 <016001d4cf71$865e7b60$931b7220$@gmail.com>
 <4F88C5DDA1E80143B232E89585ACE27D0264F137@DGGEMM528-MBX.china.huawei.com>
 <20190228164228.734ede80@collabora.com>
 <005a01d4d03e$39b0e0f0$ad12a2d0$@yahoo.co.jp>
 <4b3bc01a-3632-5dde-f683-94744ee7179d@ti.com>
 <000501d4d050$38eca160$aac5e420$@gmail.com>
 <20190301184317.6bad546f@collabora.com>
 <001c01d4d057$f68572e0$e39058a0$@yahoo.co.jp>
From:   Vignesh Raghavendra <vigneshr@ti.com>
Message-ID: <de75d4f6-dfff-26e0-e191-6240ffc20223@ti.com>
Date:   Sat, 2 Mar 2019 14:27:54 +0530
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:60.0) Gecko/20100101
 Thunderbird/60.5.1
MIME-Version: 1.0
In-Reply-To: <001c01d4d057$f68572e0$e39058a0$@yahoo.co.jp>
Content-Type: text/plain; charset="utf-8"
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 01-Mar-19 11:25 PM, Tokunori Ikegami wrote:
>>>> [...]
>>>>>>>>> In function do_write_buffer(), in the for loop, there is a
>>>>>>>>> case chip_ready() returns 1 while chip_good() returns 0, so
>>>>>>>>> it never break the loop.
>>>>>>>>> To fix this, chip_good() is enough and it should timeout if
>>>>>>>>> it stay bad for a while.
>>>>>>>>>
>>>>>>>>> Fixes: dfeae1073583("mtd: cfi_cmdset_0002: Change write
>>>>>>>>> buffer to check correct value")
>>>>>>>>> Signed-off-by: Yi Huaijie <yihuaijie@huawei.com>
>>>>>>>>> Signed-off-by: Liu Jian <liujian56@huawei.com>
>>>>>>>>> Reviewed-by: Tokunori Ikegami <ikegami_to@yahoo.co.jp>
>>>>>>>>> ---
>>>>>>>>> v2->v3:
>>>>>>>>> Follow Vignesh's advice:
>>>>>>>>> add one more check for check_good() even when time_after()
>>>>>>>>> returns
>>>>>> true.
>>>>>>>>>
>>>>>>>>>  drivers/mtd/chips/cfi_cmdset_0002.c | 2 +-
>>>>>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>>>>>
>>>>>>>>> diff --git a/drivers/mtd/chips/cfi_cmdset_0002.c
>>>>>>>>> b/drivers/mtd/chips/cfi_cmdset_0002.c
>>>>>>>>> index 72428b6..3da2376 100644
>>>>>>>>> --- a/drivers/mtd/chips/cfi_cmdset_0002.c
>>>>>>>>> +++ b/drivers/mtd/chips/cfi_cmdset_0002.c
>>>>>>>>> @@ -1876,7 +1876,7 @@ static int __xipram
>>>>>>>>> do_write_buffer(struct map_info *map, struct flchip *chip,
>>>>>>>>>  			continue;
>>>>>>>>>  		}
>>>>>>>>>
>>>>>>>>> -		if (time_after(jiffies, timeo)
>>>>>>>>> && !chip_ready(map, adr))
>>>>>>>>> +		if (time_after(jiffies, timeo)
>>>>>>>>> && !chip_good(map, adr, datum))
>>>>>>>>
>>>>>>>>   Just another idea to understand easily.
>>>>>>>>
>>>>>>>>     unsigned long now = jiffies;
>>>>>>>>
>>>>>>>>     if (chip_good(map, adr, datum)) {
>>>>>>>>         xip_enable(map, chip, adr);
>>>>>>>>         goto op_done;
>>>>>>>>     }
>>>>>>>>
>>>>>>>>     if (time_after(now, timeo) {
>>>>>>>>         break;
>>>>>>>>     }
>>>>>>>>
>>>>>>>
>>>>>>> Thank you~. It is more easier to understand！
>>>>>>> If there are no other comments, I will send new patch again ):
>>>>>>
>>>>>> Except this version no longer does what Vignesh suggested. See
>>>>>> how you no longer test if chip_good() is true if time_after()
>>>>>> returns true. So, imagine the thread entering this function is
>>>>>> preempted just after the first chip_good() test, and resumed a
>>>>>> few ms later. time_after() will return true, but chip_good()
>>>>>> might also return true, and you ignore it.
>>>>>
>>>>> I think that the following 3 versions will be worked for
>>>>> time_after()
>>>> as a same result and follow the Vignesh-san suggestion.
>>>>>
>>>>
>>>> As Boris explained above version 3 does not really follow my
>>>> suggestion... Please see below
>>>>
>>>>> 1. Original Vignesh-san suggestion
>>>>>
>>>>> 	if (chip_good(map, adr, datum)) {
>>>>> 		xip_enable(map, chip, adr);
>>>>> 		goto op_done;
>>>>> 	}
>>>>>
>>>>> 	if (time_after(jiffies, timeo)) {
>>>>> 		/* Test chip_good() if TRUE incorrectly again so
>>>>> write
>>>> failure by time_after() can be avoided. */
>>>>> 		if (chip_good(map, adr, datum)) {
>>>>> 			xip_enable(map, chip, adr);
>>>>> 			goto op_done;
>>>>> 		}
>>>>> 		break;
>>>>> 	}
>>>>>
>>>>
>>>>
>>>> Right, here we check chip_good() once _even when time_after() is
>>>> true_ to avoid _spurious_ timeout
>>>>
>>>>> 2. Liujian-san v3 patch
>>>>>
>>>>> 	/* Test chip_good() if FALSE correctly so write failure
>>>>> by
>>>> time_after() can be avoided. */
>>>>> 	if (time_after(jiffies, timeo) && !chip_good(map, adr))
>>>>> 		break;
>>>>>
>>>>> 	if (chip_good(map, adr, datum)) {
>>>>> 		xip_enable(map, chip, adr);
>>>>> 		goto op_done;
>>>>> 	}
>>>>>
>>>>
>>>> This is a better version of 1
>>>>
>>>>> 3. My idea
>>>>>
>>>>> 	/* Save current jiffies value before chip_good() to avoid
>>>>> write
>>>> failure by time_after() without testing chip_good() again. */
>>>>> 	unsigned long now = jiffies;
>>>>>
>>>>> 	if (chip_good(map, adr, datum)) {
>>>>> 		xip_enable(map, chip, adr);
>>>>> 		goto op_done;
>>>>> 	}
>>>>>
>>>>
>>>> What if thread gets pre-empted at this point and is re-scheduled
>>>> exactly after timeo jiffies have elapsed? Below check would be true
>>>> and exit loop
>>>
>>>   I think that the jiffies value now is save before chip_good() so
>>> below check would be false and not exit loop.
>>

Ok, I get it now.

>> True, I overlooked that part, and so Vignesh did. This proves one
>> thing: code is not easier to follow with your version. IMO, if we want
>> to make things clear, we should add a comment to Liujian's explaining
>> why the extra test after time_after(jiffies, timeo) is needed.
> 
> I see so I am okay with the change of Liujian-san v3 patch.
> Also agree with the comment to be added.
> 

Right, I like the current patch from Liujian, because its more
consistent with the existing code in this file.

Liujian, Could you re-post with a comment added as suggested above?

Regards
Vignesh