Message-ID: <CAOesGMiTmR6t2h1AzkPRWQon3onXKb5db5q6yVix9D5r+ExKfA@mail.gmail.com>
Date:   Fri, 4 Jan 2019 16:57:17 -0800
From:   Olof Johansson <olof@...om.net>
To:     Faiz Abbas <faiz_abbas@...com>
Cc:     Eduardo Valentin <edubezval@...il.com>,
        Ulf Hansson <ulf.hansson@...aro.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        "linux-mmc@...r.kernel.org" <linux-mmc@...r.kernel.org>,
        Adrian Hunter <adrian.hunter@...el.com>,
        Kishon <kishon@...com>, Keerthy <j-keerthy@...com>,
        Zhang Rui <rui.zhang@...el.com>,
        Daniel Lezcano <daniel.lezcano@...aro.org>,
        Santosh Shilimkar <ssantosh@...nel.org>,
        Tony Lindgren <tony@...mide.com>
Subject: Re: [PATCH v2 2/2] mmc: sdhci-omap: Workaround errata regarding
 SDR104/HS200 tuning failures (i929)

On Wed, Jan 2, 2019 at 9:58 PM Faiz Abbas <faiz_abbas@...com> wrote:
>
> Hi Olof, Eduardo,
>
> On 03/01/19 1:26 AM, Eduardo Valentin wrote:
> > On Wed, Jan 02, 2019 at 10:29:31AM -0800, Olof Johansson wrote:
> >> Hi,
> >>
> >>
> >> On Wed, Dec 12, 2018 at 1:20 AM Ulf Hansson <ulf.hansson@...aro.org> wrote:
> >>>
> >>> + Thermal maintainers
> >>>
> >>> On Tue, 11 Dec 2018 at 15:20, Faiz Abbas <faiz_abbas@...com> wrote:
> >>>>
> >>>> Errata i929 in certain OMAP5/DRA7XX/AM57XX silicon revisions
> >>>> (SPRZ426D - November 2014 - Revised February 2018 [1]) mentions
> >>>> unexpected tuning pattern errors. A small failure band may be present
> >>>> in the tuning range which may be missed by the current algorithm.
> >>>> Furthermore, the failure bands vary with temperature leading to
> >>>> different optimum tuning values for different temperatures.
> >>>>
> >>>> As suggested in the related Application Report (SPRACA9B - October 2017
> >>>> - Revised July 2018 [2]), tuning should be done in two stages.
> >>>> In stage 1, assign the optimum ratio in the maximum pass window for the
> >>>> current temperature. In stage 2, if the chosen value is close to the
> >>>> small failure band, move away from it in the appropriate direction.
> >>>>
> >>>> References:
> >>>> [1] http://www.ti.com/lit/pdf/sprz426
> >>>> [2] http://www.ti.com/lit/pdf/SPRACA9
> >>>>
> >>>> Signed-off-by: Faiz Abbas <faiz_abbas@...com>
> >>>> Acked-by: Adrian Hunter <adrian.hunter@...el.com>
> >>>> ---
> >>>>  drivers/mmc/host/Kconfig      |  2 +
> >>>>  drivers/mmc/host/sdhci-omap.c | 90 ++++++++++++++++++++++++++++++++++-
> >>>>  2 files changed, 91 insertions(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/drivers/mmc/host/Kconfig b/drivers/mmc/host/Kconfig
> >>>> index 5fa580cec831..d8f984483ab0 100644
> >>>> --- a/drivers/mmc/host/Kconfig
> >>>> +++ b/drivers/mmc/host/Kconfig
> >>>> @@ -977,6 +977,8 @@ config MMC_SDHCI_XENON
> >>>>  config MMC_SDHCI_OMAP
> >>>>         tristate "TI SDHCI Controller Support"
> >>>>         depends on MMC_SDHCI_PLTFM && OF
> >>>> +       select THERMAL
> >>>> +       select TI_SOC_THERMAL
> >>>>         help
> >>>>           This selects the Secure Digital Host Controller Interface (SDHCI)
> >>>>           support present in TI's DRA7 SOCs. The controller supports
> >>>> diff --git a/drivers/mmc/host/sdhci-omap.c b/drivers/mmc/host/sdhci-omap.c
> >>>> index f588ab679cb0..b75c55011fcb 100644
> >>>> --- a/drivers/mmc/host/sdhci-omap.c
> >>>> +++ b/drivers/mmc/host/sdhci-omap.c
> >>>> @@ -27,6 +27,7 @@
> >>>>  #include <linux/regulator/consumer.h>
> >>>>  #include <linux/pinctrl/consumer.h>
> >>>>  #include <linux/sys_soc.h>
> >>>> +#include <linux/thermal.h>
> >>>>
> >>>>  #include "sdhci-pltfm.h"
> >>>>
> >>>> @@ -286,15 +287,19 @@ static int sdhci_omap_execute_tuning(struct mmc_host *mmc, u32 opcode)
> >>>>         struct sdhci_host *host = mmc_priv(mmc);
> >>>>         struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
> >>>>         struct sdhci_omap_host *omap_host = sdhci_pltfm_priv(pltfm_host);
> >>>> +       struct thermal_zone_device *thermal_dev;
> >>>>         struct device *dev = omap_host->dev;
> >>>>         struct mmc_ios *ios = &mmc->ios;
> >>>>         u32 start_window = 0, max_window = 0;
> >>>> +       bool single_point_failure = false;
> >>>>         bool dcrc_was_enabled = false;
> >>>>         u8 cur_match, prev_match = 0;
> >>>>         u32 length = 0, max_len = 0;
> >>>>         u32 phase_delay = 0;
> >>>> +       int temperature;
> >>>>         int ret = 0;
> >>>>         u32 reg;
> >>>> +       int i;
> >>>>
> >>>>         /* clock tuning is not needed for upto 52MHz */
> >>>>         if (ios->clock <= 52000000)
> >>>> @@ -304,6 +309,16 @@ static int sdhci_omap_execute_tuning(struct mmc_host *mmc, u32 opcode)
> >>>>         if (ios->timing == MMC_TIMING_UHS_SDR50 && !(reg & CAPA2_TSDR50))
> >>>>                 return 0;
> >>>>
> >>>> +       thermal_dev = thermal_zone_get_zone_by_name("cpu_thermal");
> >>>
> >>> I couldn't find a corresponding call to a put function, like
> >>> "thermal_zone_put()" or whatever, which made me realize that the
> >>> thermal zone API is incomplete. Or depending on how you put it, it
> >>> lacks object reference counting, unless I am missing something.
> >>>
> >>> For example, what happens if the thermal zone becomes unregistered
> >>> between this point and when you call thermal_zone_get_temp() a couple
> >>> of line below. I assume it's a known problem, but just wanted to point
> >>> it out.
> >>>
> >
> > Yes, there is no ref counting. Specially because the get zones usages
> > were too specific, and mostly used in application cases that module
> > would not really be removed. Though not a good excuse, still, not very
> > problematic. Now, if the API is getting other usages, then refcounting
> > may be necessary.
> >
> >>>> +       if (IS_ERR(thermal_dev)) {
> >>>> +               dev_err(dev, "Unable to get thermal zone for tuning\n");
> >>>> +               return PTR_ERR(thermal_dev);
> >>>> +       }
> >>>> +
> >>>> +       ret = thermal_zone_get_temp(thermal_dev, &temperature);
> >>>> +       if (ret)
> >>>> +               return ret;
> >>>> +
> >>>
> >>> [...]
> >>>
> >>> Anyway, I have applied this for next, thanks!
> >>
> >> This is throwing errors on builds of keystone_defconfig in next and mainline:
> >>
> >> http://arm-soc.lixom.net/buildlogs/next/next-20190102/buildall.arm.keystone_defconfig.log.passed
> >>
> >> WARNING: unmet direct dependencies detected for TI_SOC_THERMAL
> >>   Depends on [n]: THERMAL [=y] && (ARCH_HAS_BANDGAP [=n] ||
> >> COMPILE_TEST [=n]) && HAS_IOMEM [=y]
> >>   Selected by [y]:
> >>   - MMC_SDHCI_OMAP [=y] && MMC [=y] && MMC_SDHCI_PLTFM [=y] && OF [=y]
> >>
> >> So, thermal depends on ARCH_HAS_BANDGAP, which keystone doesn't provide.
> >>
> >> Selecting a major framework such as THERMAL from a driver config is
> >> likely not the right solution anyway, especially since THERMAL does
> >> provide stubbed out versions of the functions if it's not enabled.
> >
> > Yeah, that seams a bit up-side-down. Can you guys give a bit more of
> > context? Why do you need the cpu thermal zone ? From patch description,
> > looks like you want to have your own zone then apply different tuning
> > values based on temperature (range?). Why do you need to mess up with
> > cpu_thermal zone? Don't you have a bandgap in the mem controller for
> > this application?
> >
>
> Thats correct. We don't have a bandgap in the MMC controller and thus we
> have to use the cpu one to measure temperature.
>
> THERMAL is critical for tuning. The interface is supposed to fail if we
> can't get temperature. So IMO we should ensure that it is present.
>
> I can fix this by "select TI_SOC_THERMAL if ARCH_HAS_BANDGAP" if you
> guys agree.

Building elaborate select statements is usually fragile, once
dependencies for TI_SOC_THERMAL changes you need to come back here to
fixup the select.

Supposedly this driver works on keystone (or does it?), it doesn't
actually need TI_SOC_THERMAL for basic functionality beyond tuning?
Or, at least, it needs to fall back to a reasonable behavior if it's
unavailable on keystone.

Having the driver print a warning and refuse to tune to higher speeds
is a reasonable way to do this, I think. That would carry to all
platforms, i.e. even the ones who have TI_SOC_THERMAL and
ARCH_HAS_BANDGAP, without adding the select.


-Olof

Powered by blists - more mailing lists