On 27/11/19 10:25 AM, BOUGH CHEN wrote:
>
>> -----Original Message-----
>> From: linux-mmc-owner@xxxxxxxxxxxxxxx <linux-mmc-owner@xxxxxxxxxxxxxxx>
>> On Behalf Of Adrian Hunter
>> Sent: 2019年11月26日 19:46
>> To: BOUGH CHEN <haibo.chen@xxxxxxx>; ulf.hansson@xxxxxxxxxx
>> Cc: linux-mmc@xxxxxxxxxxxxxxx; dl-linux-imx <linux-imx@xxxxxxx>
>> Subject: Re: [PATCH 03/14] mmc: sdhci: correct the DMA setting for IOMMU
>>
>> On 15/11/19 1:07 PM, haibo.chen@xxxxxxx wrote:
>>> From: Haibo Chen <haibo.chen@xxxxxxx>
>>>
>>> The default max segment size of IOMMU is 64KB, which exceed the ADMA
>>> limitation if ADMA only support max to 65535, 64KB - 1Byte. IOMMU will
>>> optimize the segments it received, merge the little segment into one
>>> big segment. If we use the default IOMMU config, then ADMA will get
>>> some segments which it's size is 64KB. Then ADMA error will shows up.
>>>
>>> Signed-off-by: Haibo Chen <haibo.chen@xxxxxxx>
>>> ---
>>> drivers/mmc/host/sdhci.c | 16 ++++++++++++++--
>>> 1 file changed, 14 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c index
>>> 1436cc9c5f82..3a8093de26c7 100644
>>> --- a/drivers/mmc/host/sdhci.c
>>> +++ b/drivers/mmc/host/sdhci.c
>>> @@ -3743,6 +3743,7 @@ static inline bool sdhci_can_64bit_dma(struct
>>> sdhci_host *host) int sdhci_setup_host(struct sdhci_host *host) {
>>> struct mmc_host *mmc;
>>> + struct device *dev;
>>> u32 max_current_caps;
>>> unsigned int ocr_avail;
>>> unsigned int override_timeout_clk;
>>> @@ -3754,6 +3755,7 @@ int sdhci_setup_host(struct sdhci_host *host)
>>> return -EINVAL;
>>>
>>> mmc = host->mmc;
>>> + dev = mmc_dev(mmc);
>>>
>>> /*
>>> * If there are external regulators, get them. Note this must be
>>> done @@ -4224,10 +4226,20 @@ int sdhci_setup_host(struct sdhci_host
>> *host)
>>> * be larger than 64 KiB though.
>>> */
>>> if (host->flags & SDHCI_USE_ADMA) {
>>> - if (host->quirks & SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC)
>>> + if (host->quirks & SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC) {
>>> mmc->max_seg_size = 65535;
>>> - else
>>> +
>>> + /*
>>> + * send the ADMA limitation to IOMMU. In default,
>>> + * the max segment size of IOMMU is 64KB, this exceed
>>> + * the ADMA max segment limitation, which is 65535.
>>> + */
>>> + dev->dma_parms = devm_kzalloc(dev,
>>> + sizeof(*dev->dma_parms), GFP_KERNEL);
>>> + dma_set_max_seg_size(dev, SZ_64K - 1);
>>
>> Doesn't mmc_setup_queue() already do this?
>
> mmc_setup_queue do call the function dma_set_max_seg_size(), but it do not give value to dev->dma_parms.
> So in dma_set_max_seg_size(), it do nothing, just return -EIO.
> Should I fix this in mmc_setup_queue()?
>
> 725 static inline int dma_set_max_seg_size(struct device *dev, unsigned int size)
> 726 {
> 727 if (dev->dma_parms) {
> 728 dev->dma_parms->max_segment_size = size;
> 729 return 0;
> 730 }
> 731 return -EIO;
> 732 }
I don't know where is the best place to create dev->dma_parms. Ask some DMA
people.
>
>>
>>> + } else {
>>> mmc->max_seg_size = 65536;
>>> + }
>>> } else {
>>> mmc->max_seg_size = mmc->max_req_size;
>>> }
>>>
>
