Re: Linux-next 20190218: am57xx-evm: mmc1: ADMA error

From: Faiz Abbas
Date: Wed Feb 27 2019 - 04:42:56 EST

Hi,

On 26/02/19 6:36 PM, Ming Lei wrote:
> On Tue, Feb 26, 2019 at 05:04:40PM +0530, Faiz Abbas wrote:
>> Hi,
>>
>> On 26/02/19 3:36 PM, Ming Lei wrote:
>>> On Tue, Feb 26, 2019 at 2:47 PM Faiz Abbas <faiz_abbas@xxxxxx> wrote:
>>>>
>>>> Hi Ming Lei,
>>>>
>>>> On 26/02/19 7:11 AM, Ming Lei wrote:
>>>>> On Mon, Feb 25, 2019 at 9:14 PM Faiz Abbas <faiz_abbas@xxxxxx> wrote:
>>>>>>
>>>>>> Hi Naresh,
>>>>>>
>>>>>> + Commit authors.
>>>>>>
>>>>>> On 19/02/19 6:38 PM, Faiz Abbas wrote:
>>>>>>> Hi Naresh,
>>>>>>>
...
>> It looks like it only happens for some requests. More verbose log with
>> dma descriptor entries:
>>
>> [ 14.840865] mmc0: ADMA error
>> [ 14.840869] mmc0: sdhci: ============ SDHCI REGISTER DUMP ===========
>> [ 14.840874] mmc0: sdhci: Sys addr: 0x00000000 | Version: 0x00003302
>> [ 14.840879] mmc0: sdhci: Blk size: 0x00000200 | Blk cnt: 0x00000000
>> [ 14.840884] mmc0: sdhci: Argument: 0x00000200 | Trn mode: 0x00000033
>> [ 14.840889] mmc0: sdhci: Present: 0x00000000 | Host ctl: 0x00000012
>> [ 14.840893] mmc0: sdhci: Power: 0x0000000f | Blk gap: 0x00000000
>> [ 14.840898] mmc0: sdhci: Wake-up: 0x00000000 | Clock: 0x00000107
>> [ 14.840903] mmc0: sdhci: Timeout: 0x0000000a | Int stat: 0x00000000
>> [ 14.840908] mmc0: sdhci: Int enab: 0x027f000b | Sig enab: 0x027f000b
>> [ 14.840912] mmc0: sdhci: ACmd stat: 0x00000000 | Slot int: 0x00000000
>> [ 14.840917] mmc0: sdhci: Caps: 0x25e90080 | Caps_1: 0x00000f77
>> [ 14.840922] mmc0: sdhci: Cmd: 0x0000123a | Max curr: 0x00000000
>> [ 14.840926] mmc0: sdhci: Resp[0]: 0x00000900 | Resp[1]: 0x20050044
>> [ 14.840931] mmc0: sdhci: Resp[2]: 0x53445531 | Resp[3]: 0x744a6055
>> [ 14.840935] mmc0: sdhci: Host ctl2: 0x00000000
>> [ 14.840939] mmc0: sdhci: ADMA Err: 0x00000007 | ADMA Ptr: 0xae857300
>> [ 14.840943] mmc0: sdhci: ============================================
>> [ 14.840950] mmc0: sdhci: be2c9004: DMA 0xab1bd000, LEN 0x1000, Attr=0x21
>> [ 14.840956] mmc0: sdhci: 92173e21: DMA 0xab1bc000, LEN 0x1000, Attr=0x21
>> [ 14.840962] mmc0: sdhci: c8a0cde4: DMA 0xab1bb000, LEN 0x1000, Attr=0x21
>> [ 14.840967] mmc0: sdhci: 4bb03017: DMA 0xab1ba000, LEN 0x1000, Attr=0x21
>> [ 14.840972] mmc0: sdhci: 2fb0d59e: DMA 0xab1b9000, LEN 0x1000, Attr=0x21
>> [ 14.840978] mmc0: sdhci: c3024ff2: DMA 0xab1b8000, LEN 0x1000, Attr=0x21
>> [ 14.840983] mmc0: sdhci: 0738188d: DMA 0xab179000, LEN 0x1000, Attr=0x21
>> [ 14.840989] mmc0: sdhci: 78ecca83: DMA 0xab178000, LEN 0x1000, Attr=0x21
>> [ 14.840994] mmc0: sdhci: 1432e5a9: DMA 0xab0d7000, LEN 0x1000, Attr=0x21
>> [ 14.840999] mmc0: sdhci: 8a36c77c: DMA 0xab0d6000, LEN 0x1000, Attr=0x21
>> [ 14.841005] mmc0: sdhci: b7196410: DMA 0xab0d5000, LEN 0x1000, Attr=0x21
>> [ 14.841010] mmc0: sdhci: dcb25259: DMA 0xab0d4000, LEN 0x1000, Attr=0x21
>> [ 14.841015] mmc0: sdhci: ef1e5d32: DMA 0xab0d3000, LEN 0x1000, Attr=0x21
>> [ 14.841020] mmc0: sdhci: 0319c66c: DMA 0xab0d2000, LEN 0x1000, Attr=0x21
>> [ 14.841026] mmc0: sdhci: 2e6b85d9: DMA 0xab0d1000, LEN 0x1000, Attr=0x21
>> [ 14.841031] mmc0: sdhci: d4dd19da: DMA 0xab0d0000, LEN 0x1000, Attr=0x21
>> [ 14.841036] mmc0: sdhci: 55cdc0f6: DMA 0xab27f000, LEN 0x1000, Attr=0x21
>> [ 14.841041] mmc0: sdhci: a172f4f3: DMA 0xab27e000, LEN 0x1000, Attr=0x21
>> [ 14.841046] mmc0: sdhci: ed27e53e: DMA 0xab27d000, LEN 0x1000, Attr=0x21
>> [ 14.841051] mmc0: sdhci: c04971ce: DMA 0xab27c000, LEN 0x1000, Attr=0x21
>> [ 14.841057] mmc0: sdhci: f43985d3: DMA 0xab27b000, LEN 0x1000, Attr=0x21
>> [ 14.841062] mmc0: sdhci: b977bd17: DMA 0xab27a000, LEN 0x1000, Attr=0x21
>> [ 14.841067] mmc0: sdhci: 8b74ee6f: DMA 0xab279000, LEN 0x1000, Attr=0x21
>> [ 14.841072] mmc0: sdhci: 12e52bc8: DMA 0xab30d000, LEN 0xffff, Attr=0x21
>> [ 14.841077] mmc0: sdhci: b39efa31: DMA 0xae857000, LEN 0x0001, Attr=0x21
>> [ 14.841082] mmc0: sdhci: bc4b71f0: DMA 0xab31d000, LEN 0x3000, Attr=0x21
>> [ 14.841087] mmc0: sdhci: 4cb5aa08: DMA 0xab2a8000, LEN 0x2000, Attr=0x21
>> [ 14.841092] mmc0: sdhci: 5e717781: DMA 0xab12a000, LEN 0x2000, Attr=0x21
>> [ 14.841098] mmc0: sdhci: 125d82b5: DMA 0xab2b4000, LEN 0x4000, Attr=0x21
>> [ 14.841103] mmc0: sdhci: b33874b9: DMA 0xab148000, LEN 0x4000, Attr=0x21
>> [ 14.841108] mmc0: sdhci: 9b0e47a5: DMA 0xab218000, LEN 0x8000, Attr=0x21
>> [ 14.841113] mmc0: sdhci: 47ce17da: DMA 0xab2a0000, LEN 0x2000, Attr=0x21
>> [ 14.841118] mmc0: sdhci: 97ea0d9f: DMA 0x00000000, LEN 0x0000, Attr=0x03
>>
>> There is a big transfer of 0xffff length followed by a smaller transfer
>> of 0x1 (at address 0xae857000 above) and that is where it fails. This is
>> the same signature every time it happens.
>
> Thanks for the investigation, and that is very helpful!
>
> Then I guess it is caused by bad segment size, see sdhci_setup_host():
>
> if (host->flags & SDHCI_USE_ADMA) {
> if (host->quirks & SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC)
> mmc->max_seg_size = 65535;
> else
> mmc->max_seg_size = 65536;
> } else {
> mmc->max_seg_size = mmc->max_req_size;
> }
>
> Could you confirm it by collecting the following log?
>
> (cd /sys/block/mmcblk0/queue && find . -type f -exec grep -aH . {} \;)

The max_segment_size is 65535.

root@dra7xx-evm:~# (cd /sys/block/mmcblk0/queue && find . -type f -exec
grep -aH . {} \;)
./hw_sector_size:512
./max_discard_segments:1
./max_segment_size:65535
./physical_block_size:512
./discard_max_bytes:4194304
./rotational:0
./iosched/fifo_batch:16
./iosched/read_expire:500
./iosched/writes_starved:2
./iosched/write_expire:5000
./iosched/front_merges:1
./write_same_max_bytes:0
./zoned:none
./max_sectors_kb:512
./discard_zeroes_data:0
./read_ahead_kb:128
./discard_max_hw_bytes:4194304
./nomerges:0
./max_segments:128
./rq_affinity:1
./iostats:1
./dax:0
./minimum_io_size:512
./chunk_sectors:0
./io_poll:0
./write_zeroes_max_bytes:0
./max_hw_sectors_kb:512
./add_random:0
./optimal_io_size:0
./nr_requests:128
./scheduler:[mq-deadline] kyber none
./io_timeout:60000
./discard_granularity:4194304
./logical_block_size:512
./nr_zones:0
./fua:1
./io_poll_delay:-1
./max_integrity_segments:0
./write_cache:write back
root@dra7xx-evm:~#

>
> If 'max_segment_size' is 65535, we may need the following patch:
>
> diff --git a/block/blk-settings.c b/block/blk-settings.c
> index 6375afaedcec..6fb7a312b4ea 100644
> --- a/block/blk-settings.c
> +++ b/block/blk-settings.c
> @@ -309,7 +309,7 @@ void blk_queue_max_segment_size(struct request_queue *q, unsigned int max_size)
> __func__, max_size);
> }
>
> - q->limits.max_segment_size = max_size;
> + q->limits.max_segment_size = round_down(max_size, 512);
> }
> EXPORT_SYMBOL(blk_queue_max_segment_size);
>

This patch fixes it for me. Thanks!

Regards,
Faiz