Re: [PATCH v2 2/4] MdeModulePkg/SdMmcPciHcDxe: Add UhsSignaling to SdMmcOverride protocol

[Marcin Wojtas <mw@semihalf.com>]

pon., 8 paź 2018 o 15:43 Ard Biesheuvel <ard.biesheuvel@linaro.org> napisał(a):
>
> On 8 October 2018 at 15:37, Marcin Wojtas <mw@semihalf.com> wrote:
> > pon., 8 paź 2018 o 15:27 Ard Biesheuvel <ard.biesheuvel@linaro.org> napisał(a):
> >>
> >> On 8 October 2018 at 15:17, Marcin Wojtas <mw@semihalf.com> wrote:
> >> > pon., 8 paź 2018 o 15:07 Ard Biesheuvel <ard.biesheuvel@linaro.org> napisał(a):
> >> >>
> >> >> On 8 October 2018 at 14:59, Marcin Wojtas <mw@semihalf.com> wrote:
> >> >> > Hi Ard,
> >> >> >
> >> >> > pon., 8 paź 2018 o 14:41 Ard Biesheuvel <ard.biesheuvel@linaro.org> napisał(a):
> >> >> >>
> >> >> >> (add MdeModulePkg maintainers)
> >> >> >>
> >> >> >> On 5 October 2018 at 15:25, Marcin Wojtas <mw@semihalf.com> wrote:
> >> >> >> > From: Tomasz Michalec <tm@semihalf.com>
> >> >> >> >
> >> >> >> > Some SD Host Controlers use different values in Host Control 2 Register
> >> >> >> > to select UHS Mode. This patch adds a new UhsSignaling type routine to
> >> >> >> > the NotifyPhase of the SdMmcOverride protocol.
> >> >> >> >
> >> >> >> > UHS signaling configuration is moved to a common, default routine
> >> >> >> > (SdMmcHcUhsSignaling), which is called when SdMmcOverride does not
> >> >> >> > cover this functionality.
> >> >> >> >
> >> >> >> > Contributed-under: TianoCore Contribution Agreement 1.1
> >> >> >> > Signed-off-by: Marcin Wojtas <mw@semihalf.com>
> >> >> >> > ---
> >> >> >> >  MdeModulePkg/Bus/Pci/SdMmcPciHcDxe/SdMmcPciHci.h |  50 +++++++
> >> >> >> >  MdeModulePkg/Include/Protocol/SdMmcOverride.h    |   2 +
> >> >> >> >  MdeModulePkg/Bus/Pci/SdMmcPciHcDxe/EmmcDevice.c  | 153 ++++++++++++--------
> >> >> >> >  MdeModulePkg/Bus/Pci/SdMmcPciHcDxe/SdDevice.c    |  37 +++--
> >> >> >> >  MdeModulePkg/Bus/Pci/SdMmcPciHcDxe/SdMmcPciHci.c |  69 +++++++++
> >> >> >> >  5 files changed, 243 insertions(+), 68 deletions(-)
> >> >> >> >
> >> >> >> > diff --git a/MdeModulePkg/Bus/Pci/SdMmcPciHcDxe/SdMmcPciHci.h b/MdeModulePkg/Bus/Pci/SdMmcPciHcDxe/SdMmcPciHci.h
> >> >> >> > index e389d52..a03160d 100644
> >> >> >> > --- a/MdeModulePkg/Bus/Pci/SdMmcPciHcDxe/SdMmcPciHci.h
> >> >> >> > +++ b/MdeModulePkg/Bus/Pci/SdMmcPciHcDxe/SdMmcPciHci.h
> >> >> >> > @@ -63,6 +63,39 @@ WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
> >> >> >> >  #define SD_MMC_HC_CTRL_VER            0xFE
> >> >> >> >
> >> >> >> >  //
> >> >> >> > +// SD Host Controler bits to HOST_CTRL2 register
> >> >> >> > +//
> >> >> >> > +#define SD_MMC_HC_CTRL_UHS_MASK       0x0007
> >> >> >> > +#define SD_MMC_HC_CTRL_UHS_SDR12      0x0000
> >> >> >> > +#define SD_MMC_HC_CTRL_UHS_SDR25      0x0001
> >> >> >> > +#define SD_MMC_HC_CTRL_UHS_SDR50      0x0002
> >> >> >> > +#define SD_MMC_HC_CTRL_UHS_SDR104     0x0003
> >> >> >> > +#define SD_MMC_HC_CTRL_UHS_DDR50      0x0004
> >> >> >> > +#define SD_MMC_HC_CTRL_MMC_DDR52      0x0004
> >> >> >> > +#define SD_MMC_HC_CTRL_MMC_SDR50      0x0002
> >> >> >> > +#define SD_MMC_HC_CTRL_MMC_SDR25      0x0001
> >> >> >> > +#define SD_MMC_HC_CTRL_MMC_SDR12      0x0000
> >> >> >> > +#define SD_MMC_HC_CTRL_HS200          0x0003
> >> >> >> > +#define SD_MMC_HC_CTRL_HS400          0x0005
> >> >> >> > +
> >
> > In case we move enums to SdMmcOverride.h, would it be desired, to move
> > there register fields values as well? Or should I rather use Xenon
> > macros for all of above locally?
> >
>
> No, I think the macros should be kept locally.
>
> >> >> >> > +//
> >> >> >> > +// Timing modes for uhs
> >> >> >> > +//
> >> >> >> > +typedef enum {
> >> >> >> > +  SdMmcUhsSdr12,
> >> >> >> > +  SdMmcUhsSdr25,
> >> >> >> > +  SdMmcUhsSdr50,
> >> >> >> > +  SdMmcUhsSdr104,
> >> >> >> > +  SdMmcUhsDdr50,
> >> >> >> > +  SdMmcMmcDdr52,
> >> >> >> > +  SdMmcMmcSdr50,
> >> >> >> > +  SdMmcMmcSdr25,
> >> >> >> > +  SdMmcMmcSdr12,
> >> >> >> > +  SdMmcMmcHs200,
> >> >> >> > +  SdMmcMmcHs400,
> >> >> >> > +} SD_MMC_UHS_TIMING;
> >> >> >> > +
> >> >> >>
> >> >> >> Here, we end up with two sets of symbolic constants for the same
> >> >> >> thing, and I suppose this enum will be duplicated in your
> >> >> >> SdMmcOverride implementation?
> >> >> >>
> >> >> >
> >> >> > Why duplicated? Macros are for generic UHS_MODE_SEL field values for
> >> >> > SD and MMC in HostControl2Register.
> >> >> >
> >> >> > SD_MMC_UHS_TIMING is just a timing mode indicator, it can be used not
> >> >> > only in UhsSignaling routine (actually the next patch, with
> >> >> > SwitchClockFreqPost, use it...).
> >> >> >
> >> >> > In my SdMmcOverride implementation this enum is not duplicated,
> >> >> > because this file (SdMmcPciHci.h) is included via
> >> >> > Protocol/SdMmcOverride.h.
> >> >> >
> >> >>
> >> >> Ah ok. Please don't expose internal headers of the SD/MMC driver via
> >> >> Protocol/SdMmcOverride.h
> >> >>
> >> >
> >> > OK.
> >> >
> >> >> I think it should be fine to add the enum definition to
> >> >> Protocol/SdMmcOverride.h instead.
> >> >>
> >> >
> >> > OK.
> >> >
> >> >> But wouldn't it be much easier to have a hook for setting
> >> >> HostControl2Register that decodes the value and modifies it according
> >> >> to what the platform requires?
> >> >>
> >> >
> >> > Can you please explain, how it will be different from UhsSignaling in
> >> > current shape (read required timing value and update UHS_MODE_SEL
> >> > field)?
> >> >
> >>
> >> Well, you decode the value, and if, e.g., the SD_MMC_HC_CTRL_HS200
> >> bits are set, you substitute them with the appropriate xenon values.
> >
> > Because values can be same for SD and MMC (e.g. UHS_104 and HS200),
> > from the controller driver perspective, how would I know, which mode
> > is requested?
> >
>
> Good point.
>
> >>
> >> Also, how important is it to drive the SD/MMC at its max rated speed
> >> at boot time? On Synquacer, I just disable HS200 in the capability
> >> struct so I can forget about all this stuff
> >
> > Some customers want it - a real life scenario from one of them:
> > applications, Linux binaries and rootfs stores in the MMC. Each boot a
> > couple of hundreds of MB to be loaded. Thanks to HS200 we have huge
> > time saving.
> >
>
> Do you mean in the initrd? Because otherwise, Linux will use its own
> driver and select its own mode.

No, I mean loading >300MB images from eMMC to memory in the DXE phase,
before booting anything.

>
> And btw, does the spec permit using different HC2 values for HS200 / HS400 ?

According to SD Host Controller Specification v4.20, UHS_MODE_SEL
values 0x5 and 0x6 are "reserved". According to Linux code, HS400
value is treated as "non-standard" and HS200 should be same as for
SDR104 (0x3). Nothing is written about permiting different values, but
given Linux 'sdhci_set_uhs_signaling' and whole bunch of other quirks
used under drivers/mmc overriding standard behavior is very common.
Please let know your desired way of handling custom UhsSignaling.

Best regards,
MArcin