Re: [EXTERNAL] Re: [PATCH v10] ASoC: tas2781: Add tas2781 driver

["Amadeusz Sławiński" <amadeuszx.slawinski@linux.intel.com>]

On 4/5/2023 1:21 PM, Ding, Shenghao wrote:
> Hi Both
> I wrote a test code to verify the be32_to_cpu & be32_to_cpup,
> static int __init lkm_init(void)
> {
> 	char test_buf[]={0x12, 0x34, 0xab, 0xbc, 0x56, 0x78, 0xef};
> 	unsigned int *k, p, q;
> 	int i;
> 
> 	printk("Hello, Shanghai!\n");
> 
> 	for (i = 0; i < 4; i ++) {
> 		k = (unsigned int *)&test_buf[i];
> 		p = be32_to_cpup((__be32 *)k);
> 		q = be32_to_cpu(test_buf[i]);
> 		printk("%d: *k = 0x%08x p = 0x%08x q = 0x%08x %ld\n",
> 			i, *k, p, q, sizeof(unsigned int));
> 	}
> 	return 0;
> }
> The output is:
> [ 9109.722548] Hello, Shanghai!
> [ 9109.726287] 0: *k = 0xbcab3412 p = 0x1234abbc q = 0x12000000 4
> [ 9109.727665] 1: *k = 0x56bcab34 p = 0x34abbc56 q = 0x34000000 4
> [ 9109.728553] 2: *k = 0x7856bcab p = 0xabbc5678 q = 0xabffffff 4
> [ 9109.729308] 3: *k = 0xef7856bc p = 0xbc5678ef q = 0xbcffffff 4
> Apparently, be32_to_cpup's output is what I expected.
> Looking forward to your comments. Thanks.
> 

It makes sense, as array access reads 8 bit value, you would need to 
access value using pointer instead. So instead of doing 
be32_to_cpu(test_buf[i]) you would need to do be32_to_cpu(*(test_buf + 
i)) if I didn't mess the pointer magic ;)

And then I'm not sure which is better, be32_to_cpup or be32_to_cpu.

> BR
> Shenghao Ding
> -----Original Message-----
> From: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
> Sent: Saturday, April 1, 2023 4:41 AM
> To: Ding, Shenghao <shenghao-ding@ti.com>; Amadeusz Slawinski <amadeuszx.slawinski@linux.intel.com>; Shenghao Ding <13916275206@139.com>; broonie@kernel.org; lgirdwood@gmail.com; perex@perex.cz; pierre-louis.bossart@linux.intel.com
> Cc: Lu, Kevin <kevin-lu@ti.com>; alsa-devel@alsa-project.org; linux-kernel@vger.kernel.org; Xu, Baojun <x1077012@ti.com>; Gupta, Peeyush <peeyush@ti.com>; Navada Kanyana, Mukund <navada@ti.com>
> Subject: [EXTERNAL] Re: [PATCH v10] ASoC: tas2781: Add tas2781 driver
> 
> Le 31/03/2023 à 04:19, Ding, Shenghao a écrit :
>> Hi Amadeusz Sławiński
>> Thanks for your comment.
>> Answer inline.
>>
>> -----Original Message-----
>> From: Amadeusz Sławiński <amadeuszx.slawinski@linux.intel.com>
>> Sent: Thursday, March 30, 2023 7:54 PM
>> To: Shenghao Ding <13916275206@139.com>; broonie@kernel.org;
>> lgirdwood@gmail.com; perex@perex.cz;
>> pierre-louis.bossart@linux.intel.com
>> Cc: Lu, Kevin <kevin-lu@ti.com>; Ding, Shenghao
>> <shenghao-ding@ti.com>; alsa-devel@alsa-project.org;
>> linux-kernel@vger.kernel.org; Xu, Baojun <x1077012@ti.com>; Gupta,
>> Peeyush <peeyush@ti.com>; Navada Kanyana, Mukund <navada@ti.com>
>> Subject: [EXTERNAL] Re: [PATCH v10] ASoC: tas2781: Add tas2781 driver
>>
>> On 3/29/2023 12:01 PM, Shenghao Ding wrote:
>>> Create tas2781 driver.
>>>
>>> Signed-off-by: Shenghao Ding <13916275206@139.com>
>>>
>>> ---
>>> Changes in v10:
>>>     - using be16_to_cpu and be32_to_cpu instead of SMS_HTONS and SMS_HTONL
>>>     - optimize and reduce the boundary checks
>>>     - Add comments on some kmemdup instead of kzalloc+memcpy
>>>     Changes to be committed:
>>> 	modified:   sound/soc/codecs/Kconfig
>>> 	modified:   sound/soc/codecs/Makefile
>>> 	new file:   sound/soc/codecs/tas2781-dsp.c
>>> 	new file:   sound/soc/codecs/tas2781-dsp.h
>>> 	new file:   sound/soc/codecs/tas2781-i2c.c
>>> 	new file:   sound/soc/codecs/tas2781.h
>>> ---
>>
>> ...
>>
>>> +
>>> +static int fw_parse_block_data_kernel(struct tasdevice_fw *tas_fmw,
>>> +	struct tasdev_blk *block, const struct firmware *fmw, int offset) {
>>> +	const unsigned char *data = fmw->data;
>>> +
>>> +	if (offset + 16 > fmw->size) {
>>> +		dev_err(tas_fmw->dev, "%s: File Size error\n", __func__);
>>> +		offset = -EINVAL;
>>> +		goto out;
>>> +	}
>>> +	block->type = be32_to_cpup((__be32 *)&data[offset]);
>>
>> Wouldn't just be32_to_cpu(data[offset]) work instead of be32_to_cpup?
>> Same in other cases.
>> [DING] data[] is a char array, the code will convert data[offset],
>> data[offset + 1], data[offset + 2] and data[offset + 3] into host instead of data[offset] only.
>>
> 
> Not sure to follow you.
> Isn't it the purpose of be32_to_cpu() to take a 32 bits word, in other words 4 x 8 bits char, and swap what if needed (little endian arch)?
> 
> It ends to __swab32() ([1] for the "constant" implementation)
> 
> 
> be32_to_cpup(&p) ends to __swab32(*p), which really looks to the same as be32_to_cpu(p).
> 
> Can you elaborate more?
> 
> CJ
> 
> 
> [1]:
> https://elixir.bootlin.com/linux/v6.3-rc3/source/include/uapi/linux/swab.h#L18
>