On 26/02/19 5:23 AM, BOUGH CHEN wrote:
> After system suspend, CQE is in cqhci_off state, which set the HALT bit, make
> CQE in HALT state. If the SoC do not power down the USDHC module, then when
> system resume back, this bit keep the same, still set. Though there is a
> sdhci reset during sdhci_resume_host(), but this reset do not impact the
> CQE part, so need to clear this bit when enable CQE, otherwise CQE will
> stuck in the first CMDQ request after system resume back.
>
> Find this issue on NXP i.MX845s-mek board
>
> [ 105.919862] mmc2: cqhci: timeout for tag 6
> [ 105.923965] mmc2: cqhci: ============ CQHCI REGISTER DUMP ===========
> [ 105.930407] mmc2: cqhci: Caps: 0x0000310a | Version: 0x00000510
> [ 105.936847] mmc2: cqhci: Config: 0x00001001 | Control: 0x00000001
> [ 105.943286] mmc2: cqhci: Int stat: 0x00000000 | Int enab: 0x00000006
> [ 105.949725] mmc2: cqhci: Int sig: 0x00000006 | Int Coal: 0x00000000
> [ 105.956164] mmc2: cqhci: TDL base: 0x7809b000 | TDL up32: 0x00000000
> [ 105.962604] mmc2: cqhci: Doorbell: 0x00000040 | TCN: 0x00000000
> [ 105.969043] mmc2: cqhci: Dev queue: 0x00000000 | Dev Pend: 0x00000000
> [ 105.975483] mmc2: cqhci: Task clr: 0x00000000 | SSC1: 0x00011000
> [ 105.981922] mmc2: cqhci: SSC2: 0x00000001 | DCMD rsp: 0x00000000
> [ 105.988362] mmc2: cqhci: RED mask: 0xfdf9a080 | TERRI: 0x00000000
> [ 105.994801] mmc2: cqhci: Resp idx: 0x00000000 | Resp arg: 0x00000000
> [ 106.001240] mmc2: sdhci: ============ SDHCI REGISTER DUMP ===========
> [ 106.007680] mmc2: sdhci: Sys addr: 0xb2b37800 | Version: 0x00000002
> [ 106.014120] mmc2: sdhci: Blk size: 0x00000200 | Blk cnt: 0x00000001
> [ 106.020560] mmc2: sdhci: Argument: 0x00010000 | Trn mode: 0x00000013
> [ 106.026999] mmc2: sdhci: Present: 0x01f88008 | Host ctl: 0x00000030
> [ 106.033439] mmc2: sdhci: Power: 0x00000002 | Blk gap: 0x00000080
> [ 106.039878] mmc2: sdhci: Wake-up: 0x00000008 | Clock: 0x0000000f
> [ 106.046318] mmc2: sdhci: Timeout: 0x0000008f | Int stat: 0x00000000
> [ 106.052757] mmc2: sdhci: Int enab: 0x107f4000 | Sig enab: 0x107f4000
> [ 106.059196] mmc2: sdhci: AC12 err: 0x00000000 | Slot int: 0x00000502
> [ 106.065635] mmc2: sdhci: Caps: 0x07eb0000 | Caps_1: 0x8000b407
> [ 106.072075] mmc2: sdhci: Cmd: 0x00000d1a | Max curr: 0x00ffffff
> [ 106.078514] mmc2: sdhci: Resp[0]: 0x00000900 | Resp[1]: 0x31360181
> [ 106.084954] mmc2: sdhci: Resp[2]: 0x44473430 | Resp[3]: 0x00450100
> [ 106.091392] mmc2: sdhci: Host ctl2: 0x00000008
> [ 106.095836] mmc2: sdhci: ADMA Err: 0x00000000 | ADMA Ptr: 0x7804b208
> [ 106.102274] mmc2: sdhci: ============================================
> [ 106.108785] mmc2: running CQE recovery
>
> Signed-off-by: Haibo Chen <haibo.chen@xxxxxxx>
Minor comment below, nevertheless:
Acked-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
>
> ---
> Changes for V2
> - move the clear HALT behavior from common code __cqhci_enable to platform
> specific function esdhc_cqe_enable
> ---
> drivers/mmc/host/sdhci-esdhc-imx.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git a/drivers/mmc/host/sdhci-esdhc-imx.c b/drivers/mmc/host/sdhci-esdhc-imx.c
> index c7215f5321d7..e2523beb495d 100644
> --- a/drivers/mmc/host/sdhci-esdhc-imx.c
> +++ b/drivers/mmc/host/sdhci-esdhc-imx.c
> @@ -1245,6 +1245,7 @@ static void sdhci_esdhc_imx_hwinit(struct sdhci_host *host)
> static void esdhc_cqe_enable(struct mmc_host *mmc)
> {
> struct sdhci_host *host = mmc_priv(mmc);
> + struct cqhci_host *cq_host = mmc->cqe_private;
> u32 reg;
> u16 mode;
> int count = 10;
> @@ -1277,6 +1278,18 @@ static void esdhc_cqe_enable(struct mmc_host *mmc)
> mode |= SDHCI_TRNS_BLK_CNT_EN;
> sdhci_writew(host, mode, SDHCI_TRANSFER_MODE);
>
> + /*
> + * Though Runtime resume reset the entire host controller,
> + * but do not impact the CQHCI side, need to clear the
> + * HALT bit and clear all tasks, avoid CQHCI stuck in the
This is not clearing all tasks, so this comment is misleading.
> + * first request when system resume back.
> + */
> + cqhci_writel(cq_host, 0, CQHCI_CTL);
> + if (cqhci_readl(cq_host, CQHCI_CTL) && CQHCI_HALT)
> + dev_err(mmc_dev(host->mmc),
> + "failed to exit halt state when enable CQE\n");
> +
> +
> sdhci_cqe_enable(mmc);
> }
>
>
