On Fri, 21 Jul 2023 at 12:14, Victor Shih <victorshihgli@xxxxxxxxx> wrote:
>
> From: Victor Shih <victor.shih@xxxxxxxxxxxxxxxxxxx>
>
> Embed UHS-II access/control functionality into the MMC request
> processing flow.
This deserves to be extended a bit. There is quite some code being
added in the $subject patch.
>
> Signed-off-by: Ulf Hansson <ulf.hansson@xxxxxxxxxx>
> Signed-off-by: Jason Lai <jason.lai@xxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Victor Shih <victor.shih@xxxxxxxxxxxxxxxxxxx>
> ---
>
> Updates in V8:
> - Add MMC_UHS2_SUPPORT to be cleared in sd_uhs2_detect().
> - Modify return value in sd_uhs2_attach().
>
> Updates in V7:
> - Add mmc_uhs2_card_prepare_cmd helper function in sd_ops.h.
> - Drop uhs2_state in favor of ios->timing.
> - Remove unnecessary functions.
>
> ---
>
> drivers/mmc/core/block.c | 18 +-
> drivers/mmc/core/core.c | 8 +
> drivers/mmc/core/mmc_ops.c | 25 +-
> drivers/mmc/core/mmc_ops.h | 1 +
> drivers/mmc/core/sd.c | 13 +-
> drivers/mmc/core/sd.h | 4 +
> drivers/mmc/core/sd_ops.c | 11 +
> drivers/mmc/core/sd_ops.h | 18 +
> drivers/mmc/core/sd_uhs2.c | 1137 +++++++++++++++++++++++++++++++++++-
> 9 files changed, 1176 insertions(+), 59 deletions(-)
>
> diff --git a/drivers/mmc/core/block.c b/drivers/mmc/core/block.c
> index f701efb1fa78..6617ae9fc840 100644
> --- a/drivers/mmc/core/block.c
> +++ b/drivers/mmc/core/block.c
> @@ -918,15 +918,9 @@ static int mmc_sd_num_wr_blocks(struct mmc_card *card, u32 *written_blocks)
>
> struct scatterlist sg;
>
> - cmd.opcode = MMC_APP_CMD;
> - cmd.arg = card->rca << 16;
> - cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC;
> -
> - err = mmc_wait_for_cmd(card->host, &cmd, 0);
> - if (err)
> - return err;
> - if (!mmc_host_is_spi(card->host) && !(cmd.resp[0] & R1_APP_CMD))
> - return -EIO;
> + err = mmc_app_cmd(card->host, card);
> + if (err)
> + return err;
>
> memset(&cmd, 0, sizeof(struct mmc_command));
The entire chunk of change above deserves its own separate
cleanup-patch. If you want to send it separately I can apply
immediately - or if you decide to make it part of the series then it
should precede the $subject patch.
Note that, after the cleanup above, the call to memset() can be dropped too.
>
> @@ -1612,6 +1606,9 @@ static void mmc_blk_rw_rq_prep(struct mmc_queue_req *mqrq,
I commented on the changes in mmc_blk_rw_rq_prep() already in version
6 [1] - but it seems like you haven't addressed my comments yet.
I have therefore copied the similar comment again, see below.
> struct request *req = mmc_queue_req_to_req(mqrq);
> struct mmc_blk_data *md = mq->blkdata;
> bool do_rel_wr, do_data_tag;
> + bool do_multi;
> +
> + do_multi = (card->host->flags & MMC_UHS2_SD_TRAN) ? true : false;
>
> mmc_blk_data_prep(mq, mqrq, recovery_mode, &do_rel_wr, &do_data_tag);
>
> @@ -1622,7 +1619,7 @@ static void mmc_blk_rw_rq_prep(struct mmc_queue_req *mqrq,
> brq->cmd.arg <<= 9;
> brq->cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_ADTC;
>
> - if (brq->data.blocks > 1 || do_rel_wr) {
> + if (brq->data.blocks > 1 || do_rel_wr || do_multi) {
This looks wrong to me. UHS2 can use single block read/writes too. Right?
> /* SPI multiblock writes terminate using a special
> * token, not a STOP_TRANSMISSION request.
> */
> @@ -1635,6 +1632,7 @@ static void mmc_blk_rw_rq_prep(struct mmc_queue_req *mqrq,
> brq->mrq.stop = NULL;
> readcmd = MMC_READ_SINGLE_BLOCK;
> writecmd = MMC_WRITE_BLOCK;
> + brq->cmd.uhs2_tmode0_flag = 1;
> }
As "do_multi" is always set for UHS2, setting this flag here seems to
be wrong/redundant.
Anyway, if I understand correctly, the flag is intended to be used to
inform the host driver whether the so-called 2L_HD_mode (half-duplex
or full-duplex) should be used for the I/O request or not. Did I
understand this correctly?
To fix the above behaviour, I suggest we try to move the entire
control of the flag into mmc_uhs2_prepare_cmd(). It seems like we need
the flag to be set for multi block read/writes (CMD18 and CMD25), but
only if the host and card supports the 2L_HD_mode too, right?
According to my earlier suggestions, we should also be able to check
the 2L_HD_mode via the bits we have set in the ios->timing, no?
Moreover, by making mmc_uhs2_prepare_cmd() responsible for setting the
flag, we can move the definition of the flag into the struct
uhs2_command instead. While at it, I suggest we also rename the flag
into "tmode_half_duplex", to better describe its purpose. Note that,
this also means the interpretation of the flag becomes inverted.
> brq->cmd.opcode = rq_data_dir(req) == READ ? readcmd : writecmd;
>
Until we have agreed on how to move forward with the above, I am
temporarily pausing further review.
[...]
Kind regards
Uffe
[1]
https://lore.kernel.org/linux-mmc/CAPDyKFoV3Ch-xzXxiT2RnDeLvsO454Pwq1vQL_bdNLptM+amAg@xxxxxxxxxxxxxx/
