On 14/12/16 20:13, Gwendal Grignou wrote:
> On Tue, Dec 13, 2016 at 11:19 PM, Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
>> On 13/12/16 21:20, Gwendal Grignou wrote:
>>> Working with a SDHCI controller and a Toshiba 16GB THGBMFG7C2LBAIL.
>>> The Toshiba part has a large erase group size and a long erase timeout.
>>> It barely fits within the controller timeout, but mmc reports a
>>> 512B discard size to the block layer instead of the expected 4MB:
>>> /sys/block/mmcblk0/queue/discard_max_bytes = 512
>>>
>>> The host caps[0] is 0x...b2, so host->max_busy_timeout is set to
>>> 2684.35ms: (1 << 27 / ((0xb2 & 0x3f) * 1000).
>>>
>>> In Toshiba EXT_CSD, HC_ERASE_GRP_SIZE is set to 7 (2100ms erase timeout)
>>> HC_ERASE_GRP_SIZE is set to 8, (8192 512B sectors erase size).
>>>
>>> In mmc_do_calc_max_discard(), as a safety, for eMMC device, number of
>>> erase groups that could safely be erased with the host timeout (qty) was
>>> reduced by 1.
>>
>> It was not for safety - it was to allow for crossing an erase group boundary.
> Got it.
>>
>> But the current code is quite different now, so please have a look at that
>> because this code does not apply anymore.
> Indeed, looking at the block layer, it is paying attention not to
> cross boundary, issuing an extra erase command when necessary.
> in blkdev_issue_discard(), we use bdev_discard_alignment to be sure
> erasing within a partition still start on a erase group boundary.
> It has been addressed with c6e66634 (block: split discard into aligned
> requests) and following changes.
I think you are misreading that patch. It won't split a discard unless it
exceeds the maximum discard size. So a discard of 2 sectors crossing an
erase group boundary is not split unless the max discard is 1. The mmc core
now does a split since commit 642c28ab86f7 ("mmc: core: Optimize case for
exactly one erase-group budget")
--
To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
