Re: mmc_core: question on mmc_do_erase function

[Adrian Hunter <adrian.hunter@nokia.com>]

On 13/10/10 11:35, Dong, Chuanxiao wrote:
>> -----Original Message-----
>> From: Adrian Hunter [mailto:adrian.hunter@nokia.com]
>> Sent: Wednesday, October 13, 2010 3:56 PM
>> To: Dong, Chuanxiao
>> Cc: linux-mmc@vger.kernel.org; Gao, Yunpeng
>> Subject: Re: mmc_core: question on mmc_do_erase function
>>
>> On 12/10/10 16:11, Dong, Chuanxiao wrote:
>>> Hi adrian
>>>
>>> I got some questions about MMC driver erase function. Help you can
>>> give me some suggestion. My host controller is a kind of SDHCI host
>>> controller. And testing below code by using HD micro SD card.
>>> +	memset(&cmd, 0, sizeof(struct mmc_command));
>>> +	cmd.opcode = MMC_ERASE;
>>> +	cmd.arg = arg;
>>> +	cmd.flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC;
>>> +	mmc_set_erase_timeout(card,&cmd, arg, qty);
>>> +	err = mmc_wait_for_cmd(card->host,&cmd, 0);
>>> +	if (err) {
>>> +		printk(KERN_ERR "mmc_erase: erase error %d, status %#x\n",
>>> +		       err, cmd.resp[0]);
>>> +		err = -EIO;
>>> +		goto out;
>>> +	}
>>> As MMC 4.4 standard said, CMD38 need R1B response. So when SDHCI host
>>> controller got a SDHCI_INT_RESPONSE interrupt, driver will ignore
>>> this and keep on waiting for a SDHCI_INT_DATA_END interrupt to finish
>>> MMC_ERASE command. Then I got problems....My host controller will got
>>> a DATA_TIMEOUT interrupt if host controller cannot generate any
>>> interrupt during its waiting time, unfortunately the max waiting time
>>> for my SDHCI host controller is 5s.
>>
>> I don't know SDHCI.  Can you disable the timeout altogether and use a
>> timer?
>>
> OK, thanks. If hardware can disable the timeout interrupt, I will try
> use a timer. But if the host controllers cannot disable the their
> timeout interrupt, what should they do?

The block driver currently sets max_discard_sectors to UINT_MAX i.e.
unlimited.  This could instead be based on a value from the driver
in the same way that max_sectors is.  Refer to mmc_init_queue()
in card/queue.c

>
>>> So if any ERASE operation need
>>> more than 5s seconds to finish, my controller will not finish the
>>> ERASE operation and only generate a DATA_TIMEOUT interrupt. The ERASE
>>> will be failed. Another, MMC driver also has a 10s timer to keep
>>> watch over the interrupt, if ERASE operation need more than 10s to
>>> finish, this timer also can report a TIMEOUT error. I think
>>> cmd->erase_timeout can help to set the timer expires, but how to deal
>>> with this issue for SDHCI host controller? How about split ERASE
>>> groups to be smaller ones?
>>
>> Where is the erase coming from?  A file system?  The discard ioctl?
> Erase command comes from the discard ioctl.
>
>> It is possible to set a maximum size for the discard but it pays
>> no attention to alignment, so you can get very sub-optimal erases.
>
>