Received-SPF: pass (google.com: domain of linux-kernel+bounces-141765-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Subject: Re: [PATCH v2 2/4] ubi: ubi_init: Fix missed ubiblock cleanup in
 error handling path
To: Daniel Golle <daniel@makrotopia.org>
CC: <richard@nod.at>, <miquel.raynal@bootlin.com>, <ben.hutchings@mind.be>,
	<linux-mtd@lists.infradead.org>, <linux-kernel@vger.kernel.org>
References: <20240411031903.3050278-1-chengzhihao1@huawei.com>
 <20240411031903.3050278-3-chengzhihao1@huawei.com>
 <ZhgGPE_5urw2izAA@makrotopia.org>
From: Zhihao Cheng <chengzhihao1@huawei.com>
Message-ID: <fb1de5ee-46d3-7148-7b59-a5d736e00f1c@huawei.com>
Date: Fri, 12 Apr 2024 09:20:28 +0800
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:68.0) Gecko/20100101
 Thunderbird/68.5.0
Precedence: bulk
MIME-Version: 1.0
In-Reply-To: <ZhgGPE_5urw2izAA@makrotopia.org>
Content-Type: text/plain; charset="gbk"; format=flowed
Content-Transfer-Encoding: 8bit

?? 2024/4/11 23:48, Daniel Golle д??:
> On Thu, Apr 11, 2024 at 11:19:01AM +0800, Zhihao Cheng wrote:
>> The ubiblock_init called by ubi_init will register device number, but
>> device number is not released in error handling path of ubi_init when
>> ubi is loaded by inserting module (eg. attaching failure), which leads
>> to subsequent ubi_init calls failed by running out of device number
>> (dmesg shows that "__register_blkdev: failed to get major for ubiblock").
>> Since ubiblock_init() registers notifier and invokes notification
>> functions, so we can move it after ubi_init_attach() to fix the problem.
>>
>> Fixes: 927c145208b0 ("mtd: ubi: attach from device tree")
>> Signed-off-by: Zhihao Cheng <chengzhihao1@huawei.com>
>> ---
>>   drivers/mtd/ubi/build.c | 51 +++++++++++++++++++++--------------------
>>   1 file changed, 26 insertions(+), 25 deletions(-)
>>
>> diff --git a/drivers/mtd/ubi/build.c b/drivers/mtd/ubi/build.c
>> index 7f95fd7968a8..bc63fbf5e947 100644
>> --- a/drivers/mtd/ubi/build.c
>> +++ b/drivers/mtd/ubi/build.c
>> @@ -1263,9 +1263,21 @@ static struct mtd_notifier ubi_mtd_notifier = {
>>   	.remove = ubi_notify_remove,
>>   };
>>   
>> +static void detach_mtd_devs(int count)
> 
> Missing __init to avoid section missmatch.
> 
> See also: https://lore.kernel.org/oe-kbuild-all/202404112327.158HJfAw-lkp@intel.com/

I think function without '__init' attributes can be called in ubi_init, 
for example misc_register, kmem_cache_create, and I verify it by make 
W=1 in local machine. And above warning(in your link) is only detected 
in my v1 series.
After investigating the '__init' and '__exit', I understand that there 
are two independent text section for these functions, for example, 
__init text section will be removed from memory after it is finished, so 
we cannot call __exit function in __init function.
> 
>> +{
>> +	int i;
>> +
>> +	for (i = 0; i < count; i++)
>> +		if (ubi_devices[i]) {
>> +			mutex_lock(&ubi_devices_mutex);
>> +			ubi_detach_mtd_dev(ubi_devices[i]->ubi_num, 1);
>> +			mutex_unlock(&ubi_devices_mutex);
>> +		}
>> +}