Re: [PATCH RFC v4 1/1] scsi: ufs: Fix ufs power down/on specs violation

From: Can Guo
Date: Mon Jan 04 2021 - 20:40:58 EST

Next message: Yang Zhong: "[PATCH 0/2] Enumerate and expose AVX_VNNI feature"
Previous message: David Woodhouse: "[PATCH] iommu/amd: Stop irq_remapping_select() matching when remapping is disabled"
In reply to: Bjorn Andersson: "Re: [PATCH RFC v4 1/1] scsi: ufs: Fix ufs power down/on specs violation"
Next in thread: Bjorn Andersson: "Re: [PATCH RFC v4 1/1] scsi: ufs: Fix ufs power down/on specs violation"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On 2021-01-05 02:57, Bjorn Andersson wrote:

On Mon 28 Dec 19:18 CST 2020, Can Guo wrote:

On 2020-12-29 01:55, Bjorn Andersson wrote:
> On Tue 22 Dec 07:49 CST 2020, Ziqi Chen wrote:
>
> > As per specs, e.g, JESD220E chapter 7.2, while powering
> > off/on the ufs device, RST_N signal and REF_CLK signal
> > should be between VSS(Ground) and VCCQ/VCCQ2.
> >
> > To flexibly control device reset line, refactor the function
> > ufschd_vops_device_reset(sturct ufs_hba *hba) to ufshcd_
> > vops_device_reset(sturct ufs_hba *hba, bool asserted). The
> > new parameter "bool asserted" is used to separate device reset
> > line pulling down from pulling up.
> >
> > Cc: Kiwoong Kim <kwmad.kim@xxxxxxxxxxx>
> > Cc: Stanley Chu <stanley.chu@xxxxxxxxxxxx>
> > Signed-off-by: Ziqi Chen <ziqichen@xxxxxxxxxxxxxx>
> > ---
> > drivers/scsi/ufs/ufs-mediatek.c | 32 ++++++++++++++++----------------
> > drivers/scsi/ufs/ufs-qcom.c | 24 +++++++++++++++---------
> > drivers/scsi/ufs/ufshcd.c | 36
> > +++++++++++++++++++++++++-----------
> > drivers/scsi/ufs/ufshcd.h | 8 ++++----
> > 4 files changed, 60 insertions(+), 40 deletions(-)
> >
> > diff --git a/drivers/scsi/ufs/ufs-mediatek.c
> > b/drivers/scsi/ufs/ufs-mediatek.c
> > index 80618af..072f4db 100644
> > --- a/drivers/scsi/ufs/ufs-mediatek.c
> > +++ b/drivers/scsi/ufs/ufs-mediatek.c
> > @@ -841,27 +841,27 @@ static int ufs_mtk_link_startup_notify(struct
> > ufs_hba *hba,
> > return ret;
> > }
> >
> > -static int ufs_mtk_device_reset(struct ufs_hba *hba)
> > +static int ufs_mtk_device_reset(struct ufs_hba *hba, bool asserted)
> > {
> > struct arm_smccc_res res;
> >
> > - ufs_mtk_device_reset_ctrl(0, res);
> > + if (asserted) {
> > + ufs_mtk_device_reset_ctrl(0, res);
> >
> > - /*
> > - * The reset signal is active low. UFS devices shall detect
> > - * more than or equal to 1us of positive or negative RST_n
> > - * pulse width.
> > - *
> > - * To be on safe side, keep the reset low for at least 10us.
> > - */
> > - usleep_range(10, 15);
> > -
> > - ufs_mtk_device_reset_ctrl(1, res);
> > -
> > - /* Some devices may need time to respond to rst_n */
> > - usleep_range(10000, 15000);
> > + /*
> > + * The reset signal is active low. UFS devices shall detect
> > + * more than or equal to 1us of positive or negative RST_n
> > + * pulse width.
> > + *
> > + * To be on safe side, keep the reset low for at least 10us.
> > + */
> > + usleep_range(10, 15);
>
> I see no point in allowing vendors to "tweak" the 1us->10us adjustment.
> The specification says 1us and we all agree that 10us gives us good
> enough slack. I.e. this is common code.

Hi Bjron,

We tried, but Samsung fellows wanted 5us. We couldn't get a agreement
on this delay in short term, so we chose to leave it in vops.

I'm not able to find the code you're referring to.

static void exynos_ufs_dev_hw_reset(struct ufs_hba *hba)
{
struct exynos_ufs *ufs = ufshcd_get_variant(hba);

hci_writel(ufs, 0 << 0, HCI_GPIO_OUT);
udelay(5);
hci_writel(ufs, 1 << 0, HCI_GPIO_OUT);
}

>
> > + } else {
> > + ufs_mtk_device_reset_ctrl(1, res);
> >
> > - dev_info(hba->dev, "device reset done\n");
> > + /* Some devices may need time to respond to rst_n */
> > + usleep_range(10000, 15000);
>
> The comment in both the Qualcomm and Mediatek drivers claim that this is
> sleep relates to the UFS device (not host), so why should it be
> different?
>
> What happens if I take the device that Mediatek see a need for a 10ms
> delay and hook that up to a Qualcomm host? This really should go in the
> common code.
>

Agree, but Qualcomm host didn't have any problems with 10us yet, so if we
put
the 10ms delay to common code, Qualcomm host would suffer longer delay when
device reset happens - both bootup and resume(xpm_lvl = 5/6) latency would
be increased.

Okay, for the resume case I accept that this is a measurable difference.
I still believe this is a property of the device and not the platform
though.

True, I also wanted to make the codes simpler and universal for everyone...

Regards,
Bjorn

Regards,
Can Guo.

>
>
> As such I really would prefer to see these delays in the common code!
> You really shouldn't write code based on a speculation that one day
> there might come someone who need other values - when that day come we
> can just change the code, and if it never comes we're better off with
> the cleaner implementation.
>
> Regards,
> Bjorn
>
> > + }
> >
> > return 0;
> > }
> > diff --git a/drivers/scsi/ufs/ufs-qcom.c b/drivers/scsi/ufs/ufs-qcom.c
> > index 2206b1e..fed10e5 100644
> > --- a/drivers/scsi/ufs/ufs-qcom.c
> > +++ b/drivers/scsi/ufs/ufs-qcom.c
> > @@ -1406,10 +1406,11 @@ static void ufs_qcom_dump_dbg_regs(struct
> > ufs_hba *hba)
> > /**
> > * ufs_qcom_device_reset() - toggle the (optional) device reset line
> > * @hba: per-adapter instance
> > + * @asserted: assert or deassert device reset line
> > *
> > * Toggles the (optional) reset line to reset the attached device.
> > */
> > -static int ufs_qcom_device_reset(struct ufs_hba *hba)
> > +static int ufs_qcom_device_reset(struct ufs_hba *hba, bool asserted)
> > {
> > struct ufs_qcom_host *host = ufshcd_get_variant(hba);
> >
> > @@ -1417,15 +1418,20 @@ static int ufs_qcom_device_reset(struct
> > ufs_hba *hba)
> > if (!host->device_reset)
> > return -EOPNOTSUPP;
> >
> > - /*
> > - * The UFS device shall detect reset pulses of 1us, sleep for 10us
> > to
> > - * be on the safe side.
> > - */
> > - gpiod_set_value_cansleep(host->device_reset, 1);
> > - usleep_range(10, 15);
> > + if (asserted) {
> > + gpiod_set_value_cansleep(host->device_reset, 1);
> >
> > - gpiod_set_value_cansleep(host->device_reset, 0);
> > - usleep_range(10, 15);
> > + /*
> > + * The UFS device shall detect reset pulses of 1us, sleep for
> > 10us to
> > + * be on the safe side.
> > + */
> > + usleep_range(10, 15);
> > + } else {
> > + gpiod_set_value_cansleep(host->device_reset, 0);
> > +
> > + /* Some devices may need time to respond to rst_n */
> > + usleep_range(10, 15);
> > + }
> >
> > return 0;
> > }
> > diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
> > index e221add..f2daac2 100644
> > --- a/drivers/scsi/ufs/ufshcd.c
> > +++ b/drivers/scsi/ufs/ufshcd.c
> > @@ -585,7 +585,13 @@ static void ufshcd_device_reset(struct ufs_hba
> > *hba)
> > {
> > int err;
> >
> > - err = ufshcd_vops_device_reset(hba);
> > + err = ufshcd_vops_device_reset(hba, true);
> > + if (err) {
> > + dev_err(hba->dev, "asserting device reset failed: %d\n", err);
> > + return;
> > + }
> > +
> > + err = ufshcd_vops_device_reset(hba, false);
> >
> > if (!err) {
> > ufshcd_set_ufs_dev_active(hba);
> > @@ -593,7 +599,11 @@ static void ufshcd_device_reset(struct ufs_hba
> > *hba)
> > hba->wb_enabled = false;
> > hba->wb_buf_flush_enabled = false;
> > }
> > + dev_dbg(hba->dev, "device reset done\n");
> > + } else {
> > + dev_err(hba->dev, "deasserting device reset failed: %d\n", err);
> > }
> > +
> > if (err != -EOPNOTSUPP)
> > ufshcd_update_evt_hist(hba, UFS_EVT_DEV_RESET, err);
> > }
> > @@ -8686,8 +8696,6 @@ static int ufshcd_suspend(struct ufs_hba *hba,
> > enum ufs_pm_op pm_op)
> > if (ret)
> > goto set_dev_active;
> >
> > - ufshcd_vreg_set_lpm(hba);
> > -
> > disable_clks:
> > /*
> > * Call vendor specific suspend callback. As these callbacks may
> > access
> > @@ -8703,6 +8711,9 @@ static int ufshcd_suspend(struct ufs_hba *hba,
> > enum ufs_pm_op pm_op)
> > */
> > ufshcd_disable_irq(hba);
> >
> > + if (ufshcd_is_link_off(hba))
> > + ufshcd_vops_device_reset(hba, true);
> > +
> > ufshcd_setup_clocks(hba, false);
> >
> > if (ufshcd_is_clkgating_allowed(hba)) {
> > @@ -8711,6 +8722,8 @@ static int ufshcd_suspend(struct ufs_hba *hba,
> > enum ufs_pm_op pm_op)
> > hba->clk_gating.state);
> > }
> >
> > + ufshcd_vreg_set_lpm(hba);
> > +
> > /* Put the host controller in low power mode if possible */
> > ufshcd_hba_vreg_set_lpm(hba);
> > goto out;
> > @@ -8778,18 +8791,19 @@ static int ufshcd_resume(struct ufs_hba
> > *hba, enum ufs_pm_op pm_op)
> > old_link_state = hba->uic_link_state;
> >
> > ufshcd_hba_vreg_set_hpm(hba);
> > +
> > + ret = ufshcd_vreg_set_hpm(hba);
> > + if (ret)
> > + goto out;
> > +
> > /* Make sure clocks are enabled before accessing controller */
> > ret = ufshcd_setup_clocks(hba, true);
> > if (ret)
> > - goto out;
> > + goto disable_vreg;
> >
> > /* enable the host irq as host controller would be active soon */
> > ufshcd_enable_irq(hba);
> >
> > - ret = ufshcd_vreg_set_hpm(hba);
> > - if (ret)
> > - goto disable_irq_and_vops_clks;
> > -
> > /*
> > * Call vendor specific resume callback. As these callbacks may
> > access
> > * vendor specific host controller register space call them when the
> > @@ -8797,7 +8811,7 @@ static int ufshcd_resume(struct ufs_hba *hba,
> > enum ufs_pm_op pm_op)
> > */
> > ret = ufshcd_vops_resume(hba, pm_op);
> > if (ret)
> > - goto disable_vreg;
> > + goto disable_irq_and_vops_clks;
> >
> > /* For DeepSleep, the only supported option is to have the link
> > off */
> > WARN_ON(ufshcd_is_ufs_dev_deepsleep(hba) &&
> > !ufshcd_is_link_off(hba));
> > @@ -8864,8 +8878,6 @@ static int ufshcd_resume(struct ufs_hba *hba,
> > enum ufs_pm_op pm_op)
> > ufshcd_link_state_transition(hba, old_link_state, 0);
> > vendor_suspend:
> > ufshcd_vops_suspend(hba, pm_op);
> > -disable_vreg:
> > - ufshcd_vreg_set_lpm(hba);
> > disable_irq_and_vops_clks:
> > ufshcd_disable_irq(hba);
> > if (hba->clk_scaling.is_allowed)
> > @@ -8876,6 +8888,8 @@ static int ufshcd_resume(struct ufs_hba *hba,
> > enum ufs_pm_op pm_op)
> > trace_ufshcd_clk_gating(dev_name(hba->dev),
> > hba->clk_gating.state);
> > }
> > +disable_vreg:
> > + ufshcd_vreg_set_lpm(hba);
> > out:
> > hba->pm_op_in_progress = 0;
> > if (ret)
> > diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h
> > index 9bb5f0e..d5fbaba 100644
> > --- a/drivers/scsi/ufs/ufshcd.h
> > +++ b/drivers/scsi/ufs/ufshcd.h
> > @@ -319,7 +319,7 @@ struct ufs_pwr_mode_info {
> > * @resume: called during host controller PM callback
> > * @dbg_register_dump: used to dump controller debug information
> > * @phy_initialization: used to initialize phys
> > - * @device_reset: called to issue a reset pulse on the UFS device
> > + * @device_reset: called to assert or deassert device reset line
> > * @program_key: program or evict an inline encryption key
> > * @event_notify: called to notify important events
> > */
> > @@ -350,7 +350,7 @@ struct ufs_hba_variant_ops {
> > int (*resume)(struct ufs_hba *, enum ufs_pm_op);
> > void (*dbg_register_dump)(struct ufs_hba *hba);
> > int (*phy_initialization)(struct ufs_hba *);
> > - int (*device_reset)(struct ufs_hba *hba);
> > + int (*device_reset)(struct ufs_hba *hba, bool asserted);
> > void (*config_scaling_param)(struct ufs_hba *hba,
> > struct devfreq_dev_profile *profile,
> > void *data);
> > @@ -1216,10 +1216,10 @@ static inline void
> > ufshcd_vops_dbg_register_dump(struct ufs_hba *hba)
> > hba->vops->dbg_register_dump(hba);
> > }
> >
> > -static inline int ufshcd_vops_device_reset(struct ufs_hba *hba)
> > +static inline int ufshcd_vops_device_reset(struct ufs_hba *hba,
> > bool asserted)
> > {
> > if (hba->vops && hba->vops->device_reset)
> > - return hba->vops->device_reset(hba);
> > + return hba->vops->device_reset(hba, asserted);
> >
> > return -EOPNOTSUPP;
> > }
> > --
> > The Qualcomm Innovation Center, Inc. is a member of the Code Aurora
> > Forum,
> > a Linux Foundation Collaborative Project
> >

Next message: Yang Zhong: "[PATCH 0/2] Enumerate and expose AVX_VNNI feature"
Previous message: David Woodhouse: "[PATCH] iommu/amd: Stop irq_remapping_select() matching when remapping is disabled"
In reply to: Bjorn Andersson: "Re: [PATCH RFC v4 1/1] scsi: ufs: Fix ufs power down/on specs violation"
Next in thread: Bjorn Andersson: "Re: [PATCH RFC v4 1/1] scsi: ufs: Fix ufs power down/on specs violation"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]