Hi Ulf & Alex,
Below is the complete analysis of this issue.
The basic flow of send command is as below:
(1) Upper layer prepare the command parameter;
(2) Call sdhci_send_command, and sdhci_send_command will call sdhci_set_transfer_mode to set Host Register 0xC;
(3) Call sdhci_writew(host, SDHCI_MAKE_CMD(cmd->opcode, flags), SDHCI_COMMAND) to send command;
For Linux kernel mainstream (without any quirk), MMC case:
(1) It will call mmc_get_ext_csd and this is an DMA transfer, then DMA enable bit will be set to 1 (0xC[0]=1'b1).
(2) Then next Command is mmc_switch(cmd6) which is none-data command, the driver will keep DMA enable bit to 1; this will let our Host failed to execute the CMD06.
According to SD Host Spec(SD Host Controller Specification verson 4.1 page 41):
Bit0 of Transfer Mode Register(0x0C) is DMA Enable bit, this bit set to 1 means a DMA operation shall begin when Host Driver writes to the upper byte of Command Register(0x0F).
The DMA enable bit should be set to zero for this none-data command(CMD6).
Our host follow the spec. so I think it should be the sdhci driver’s issue (sdhci_set_transfer_mode).
So We tried SDHCI_QUIRK2_CLEAR_TRANSFERMODE_REG_BEFORE_CMD to fix above issue.
But it encounter new problem as Alex reported for tuning command as tuning command flow in sdhci_execute_tuning function is:
(1). cmd.data = NULL;
(2). sdhci_writew(host, SDHCI_TRNS_READ, SDHCI_TRANSFER_MODE);
(3). sdhci_send_command(host, &cmd);
As data is null, this quirk will set 0x0C[0:15] to ALL zero before send tuning command including SDHCI_TRNS_READ, and the tuning command will failed.
So, can we commit one patch as below to fix this DMA enable bit for non-data command bug if you cannot accept add new QUIRK?
Code should like below(sdhci_set_transfer_mode):
if (data == NULL) {
if (host->quirks2 &
SDHCI_QUIRK2_CLEAR_TRANSFERMODE_REG_BEFORE_CMD) {
sdhci_writew(host, 0x0, SDHCI_TRANSFER_MODE);
} else {
/* clear Auto CMD settings for no data CMDs */
mode = sdhci_readw(host, SDHCI_TRANSFER_MODE);
sdhci_writew(host, mode & ~(SDHCI_TRNS_AUTO_CMD12 | SDHCI_TRNS_DMA /*Clear DMA enable bit also for no data CMDs*/
SDHCI_TRNS_AUTO_CMD23), SDHCI_TRANSFER_MODE);
}
return;
}
BR
Peter.Guo
-----Original Message-----
From: Alex Ballas [mailto:alex@xxxxxxxxxx]
Sent: Saturday, December 19, 2015 6:56 AM
To: Peter Guo
Cc: Ulf Hansson; adam8157@xxxxxxxxx; linux-mmc@xxxxxxxxxxxxxxx
Subject: Re: 1217:8520 [Dell Latitude E7450] O2 Micro, SD/MMC Card Reader doesn't work
Hello Peter & Uffe,
I tested the patch with the latest mainline kernel and it played nicely.
@Uffe how should we move forward with this?
Kind Regards,
Alex
On Fri, Dec 18, 2015 at 12:43 PM, Alex Ballas <alex@xxxxxxxxxx> wrote:
> Hello Uffe, Peter,
>
> Thank you both. I'll try out Peter's patch to see if it's addressing
> the issue and we can work through it based on Uffe's guidelines.
>
> Kind Regards,
> Alex
>
> On Fri, Dec 18, 2015 at 11:39 AM, Peter Guo <peter.guo@xxxxxxxxxxxxxx> wrote:
>> Hi Ulf,
>>
>> Thanks for your reply.
>>
>> So currently the problem is SDHCI_TRANS_DMA bit need to be cleared
>> before send Non DMA SD/MMC command for some of our host, Could you please give me any suggestion on what kind of code modification you can accept?
>>
>>
>> BR
>> Peter.Guo
>>
>>
>> -----Original Message-----
>> From: Ulf Hansson [mailto:ulf.hansson@xxxxxxxxxx]
>> Sent: Friday, December 18, 2015 4:52 PMT
>> To: Peter Guo
>> Cc: Alex Ballas; adam8157@xxxxxxxxx; linux-mmc@xxxxxxxxxxxxxxx
>> Subject: Re: 1217:8520 [Dell Latitude E7450] O2 Micro, SD/MMC Card
>> Reader doesn't work
>>
>> On 18 December 2015 at 04:14, Peter Guo <peter.guo@xxxxxxxxxxxxxx> wrote:
>>> Hi Alex & Adam,
>>>
>>> I have send the patch. Please check it.
>>> Attachment is the patch file.
>>
>> Thanks Peter for helping out!
>>
>> Although I would appreciate if you don't send patches as attachments.
>>
>> Moreover, even if the patch can be used to check whether it solves the problem - I won't accept it as is.
>> The reason is simply that I don't allow any more quirks/callbacks to be added for SDHCI.
>>
>> Kind regards
>> Uffe
��.n��������+%������w��{.n�����{��i��)�� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
