Re: [PATCH v2 04/12] mtd: rawnand: stm32_fmc2: manage all errors cases at probe time

From: Marek Vasut
Date: Mon Apr 27 2020 - 16:10:43 EST


On 4/27/20 10:08 PM, Miquel Raynal wrote:
> Hi Marek,
>
> Marek Vasut <marex@xxxxxxx> wrote on Mon, 27 Apr 2020 21:46:44 +0200:
>
>> On 4/27/20 8:08 PM, Miquel Raynal wrote:
>> [...]
>>>>>> /* FMC2 init routine */
>>>>>> stm32_fmc2_init(fmc2);
>>>>>> @@ -1997,7 +2001,7 @@ static int stm32_fmc2_probe(struct platform_device *pdev)
>>>>>> /* Scan to find existence of the device */
>>>>>> ret = nand_scan(chip, nand->ncs);
>>>>>> if (ret)
>>>>>> - goto err_scan;
>>>>>> + goto err_dma_setup;
>>>>>>
>>>>>> ret = mtd_device_register(mtd, NULL, 0);
>>>>>> if (ret)
>>>>>> @@ -2010,7 +2014,7 @@ static int stm32_fmc2_probe(struct platform_device *pdev)
>>>>>> err_device_register:
>>>>>> nand_cleanup(chip);
>>>>>>
>>>>>> -err_scan:
>>>>>> +err_dma_setup:
>>>>>> if (fmc2->dma_ecc_ch)
>>>>>> dma_release_channel(fmc2->dma_ecc_ch);
>>>>>> if (fmc2->dma_tx_ch)
>>>>>> @@ -2021,6 +2025,7 @@ static int stm32_fmc2_probe(struct platform_device *pdev)
>>>>>> sg_free_table(&fmc2->dma_data_sg);
>>>>>> sg_free_table(&fmc2->dma_ecc_sg);
>>>>>>
>>>>>> +err_clk_disable:
>>>>>> clk_disable_unprepare(fmc2->clk);
>>>>>>
>>>>>> return ret;
>>>>>
>>>>> I didn't spot it during my earlier reviews but I really prefer using
>>>>> labels explaining what you do than having the same name of the function
>>>>> which failed. This way you don't have to rework the error path when
>>>>> you handle an additional error.
>>>>>
>>>>> So, would you mind doing this in two steps:
>>>>>
>>>>> 1/
>>>>> Replace
>>>>>
>>>>> err_scan:
>>>>>
>>>>> with, eg.
>>>>>
>>>>> release_dma_objs:
>>>>
>>>> The ^err_ prefix in failpath labels is useful, since it's easily
>>>> possible to match on it with regexes ; not so much on arbitrary label name.
>>>
>>> I guess so, but is it actually useful to catch labels in a regex? (real
>>> question)
>>
>> I find it useful to have a unified way to find those labels, e.g.
>> err_because_foo:
>> err_because_bar:
>> err_last_one:
>> is much nicer than:
>> foo_failed:
>> bar_also_failed:
>> its_total_randomness:
>
> My point being, Christophe, you can use err_ as a prefix but I think
> it's better to use:
>
> err_do_this_cleanup
>
> than
>
> err_this_failed

That's fine either way.

>>> Any way I suppose catching ":\n" is already a good approximation to
>>> find labels?
>>
>> Not very practical with git grep (^err.*: works nicely though)
>
> I suppose ^.*:$ would work the same ;)

Try and see how much other irrelevant stuff that sucks in ;-)