On 14:26 Tue 21 May , Stefan Wahren wrote:
> Hi Andrea,
>
> Am 21.05.24 um 10:35 schrieb Andrea della Porta:
> > Broadcom BCM2712 SoC has an SDHCI card controller using the SDIO CFG
> > register block present on other STB chips. Add support for BCM2712
> > SD capabilities of this chipset.
> > The silicon is SD Express capable but this driver port does not currently
> > include that feature yet.
> > Based on downstream driver by raspberry foundation maintained kernel.
> >
> > Signed-off-by: Andrea della Porta <andrea.porta@xxxxxxxx>
> > ---
> > drivers/mmc/host/sdhci-brcmstb.c | 65 ++++++++++++++++++++++++++++++++
> > 1 file changed, 65 insertions(+)
> >
> > diff --git a/drivers/mmc/host/sdhci-brcmstb.c b/drivers/mmc/host/sdhci-brcmstb.c
> > index 9053526fa212..b349262da36e 100644
> > --- a/drivers/mmc/host/sdhci-brcmstb.c
> > +++ b/drivers/mmc/host/sdhci-brcmstb.c
> > @@ -30,6 +30,21 @@
> >
> > #define SDHCI_ARASAN_CQE_BASE_ADDR 0x200
> >
> > +#define SDIO_CFG_CQ_CAPABILITY 0x4c
> > +#define SDIO_CFG_CQ_CAPABILITY_FMUL GENMASK(13, 12)
> > +
> > +#define SDIO_CFG_CTRL 0x0
> > +#define SDIO_CFG_CTRL_SDCD_N_TEST_EN BIT(31)
> > +#define SDIO_CFG_CTRL_SDCD_N_TEST_LEV BIT(30)
> > +
> > +#define SDIO_CFG_MAX_50MHZ_MODE 0x1ac
> > +#define SDIO_CFG_MAX_50MHZ_MODE_STRAP_OVERRIDE BIT(31)
> > +#define SDIO_CFG_MAX_50MHZ_MODE_ENABLE BIT(0)
> > +
> > +#define MMC_CAP_HSE_MASK (MMC_CAP2_HSX00_1_2V | MMC_CAP2_HSX00_1_8V)
> > +/* Select all SD UHS type I SDR speed above 50MB/s */
> > +#define MMC_CAP_UHS_I_SDR_MASK (MMC_CAP_UHS_SDR50 | MMC_CAP_UHS_SDR104)
> > +
> > struct sdhci_brcmstb_priv {
> > void __iomem *cfg_regs;
> > unsigned int flags;
> > @@ -38,6 +53,7 @@ struct sdhci_brcmstb_priv {
> > };
> >
> > struct brcmstb_match_priv {
> > + void (*cfginit)(struct sdhci_host *host);
> > void (*hs400es)(struct mmc_host *mmc, struct mmc_ios *ios);
> > struct sdhci_ops *ops;
> > const unsigned int flags;
> > @@ -168,6 +184,38 @@ static void sdhci_brcmstb_set_uhs_signaling(struct sdhci_host *host,
> > sdhci_writew(host, ctrl_2, SDHCI_HOST_CONTROL2);
> > }
> >
> > +static void sdhci_brcmstb_cfginit_2712(struct sdhci_host *host)
> > +{
> > + struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
> > + struct sdhci_brcmstb_priv *brcmstb_priv = sdhci_pltfm_priv(pltfm_host);
> > + u32 reg, base_clk_mhz;
> > +
> > + /*
> > + * If we support a speed that requires tuning,
> > + * then select the delay line PHY as the clock source.
> > + */
> > + if ((host->mmc->caps & MMC_CAP_UHS_I_SDR_MASK) || (host->mmc->caps2 & MMC_CAP_HSE_MASK)) {
> > + reg = readl(brcmstb_priv->cfg_regs + SDIO_CFG_MAX_50MHZ_MODE);
> > + reg &= ~SDIO_CFG_MAX_50MHZ_MODE_ENABLE;
> > + reg |= SDIO_CFG_MAX_50MHZ_MODE_STRAP_OVERRIDE;
> > + writel(reg, brcmstb_priv->cfg_regs + SDIO_CFG_MAX_50MHZ_MODE);
> > + }
> > +
> > + if ((host->mmc->caps & MMC_CAP_NONREMOVABLE) ||
> > + (host->mmc->caps & MMC_CAP_NEEDS_POLL)) {
> > + /* Force presence */
> > + reg = readl(brcmstb_priv->cfg_regs + SDIO_CFG_CTRL);
> > + reg &= ~SDIO_CFG_CTRL_SDCD_N_TEST_LEV;
> > + reg |= SDIO_CFG_CTRL_SDCD_N_TEST_EN;
> > + writel(reg, brcmstb_priv->cfg_regs + SDIO_CFG_CTRL);
> > + }
> > +
> > + /* Guesstimate the timer frequency (controller base clock) */
> > + base_clk_mhz = max_t(u32, clk_get_rate(pltfm_host->clk) / (1000 * 1000), 1);
> > + reg = SDIO_CFG_CQ_CAPABILITY_FMUL | base_clk_mhz;
> > + writel(reg, brcmstb_priv->cfg_regs + SDIO_CFG_CQ_CAPABILITY);
> This part assumes the clock isn't changed afterwards, see below ...
> > +}
> > +
> > static void sdhci_brcmstb_dumpregs(struct mmc_host *mmc)
> > {
> > sdhci_dumpregs(mmc_priv(mmc));
> > @@ -200,6 +248,14 @@ static struct sdhci_ops sdhci_brcmstb_ops = {
> > .set_uhs_signaling = sdhci_set_uhs_signaling,
> > };
> >
> > +static struct sdhci_ops sdhci_brcmstb_ops_2712 = {
> > + .set_clock = sdhci_set_clock,
> > + .set_power = sdhci_set_power_and_bus_voltage,
> > + .set_bus_width = sdhci_set_bus_width,
> > + .reset = sdhci_reset,
> > + .set_uhs_signaling = sdhci_set_uhs_signaling,
> > +};
> > +
> > static struct sdhci_ops sdhci_brcmstb_ops_7216 = {
> > .set_clock = sdhci_brcmstb_set_clock,
> > .set_bus_width = sdhci_set_bus_width,
> > @@ -214,6 +270,11 @@ static struct sdhci_ops sdhci_brcmstb_ops_74165b0 = {
> > .set_uhs_signaling = sdhci_brcmstb_set_uhs_signaling,
> > };
> >
> > +static const struct brcmstb_match_priv match_priv_2712 = {
> > + .cfginit = sdhci_brcmstb_cfginit_2712,
> > + .ops = &sdhci_brcmstb_ops_2712,
> > +};
> > +
> > static struct brcmstb_match_priv match_priv_7425 = {
> > .flags = BRCMSTB_MATCH_FLAGS_NO_64BIT |
> > BRCMSTB_MATCH_FLAGS_BROKEN_TIMEOUT,
> > @@ -238,6 +299,7 @@ static struct brcmstb_match_priv match_priv_74165b0 = {
> > };
> >
> > static const struct of_device_id __maybe_unused sdhci_brcm_of_match[] = {
> > + { .compatible = "brcm,bcm2712-sdhci", .data = &match_priv_2712 },
> > { .compatible = "brcm,bcm7425-sdhci", .data = &match_priv_7425 },
> > { .compatible = "brcm,bcm7445-sdhci", .data = &match_priv_7445 },
> > { .compatible = "brcm,bcm7216-sdhci", .data = &match_priv_7216 },
> > @@ -370,6 +432,9 @@ static int sdhci_brcmstb_probe(struct platform_device *pdev)
> > (host->mmc->caps2 & MMC_CAP2_HS400_ES))
> > host->mmc_host_ops.hs400_enhanced_strobe = match_priv->hs400es;
> >
> > + if (match_priv->cfginit)
> > + match_priv->cfginit(host);
> > +
> I'm not sure this is the right place to call cfginit.
> sdhci_brcmstb_cfginit_2712 retrives the current controller base clock,
> but at the end of sdhci_brcmstb_probe this clock frequency could be
> adjusted by the device property "clock-frequency". So i'm not sure this
> is intended.
I've tried to interpret the meaning of those two clocks since unfortunately I don't
own the datasheet for any of the platforms involved, so please take the following
as the result of my own (possibly wrong) intuition and (mostly wild) guessing.
The main clock is 'sw_sdio' while 'sdio_freq' is optional and the latter seems to be
orthogonal to the former.
While sw_sdio is mostly used for SD storage card, sdio_freq seems more related to
SDIO family of cards (wifi, gps, camera, etc) for which you could specify a particular
(and higher) base frequency.
Unfortunately I wasn't able to find any reference to sdio_freq in current devicetree
so it's probably only specific to custom appliances: to be honest I'm not even sure
that BCM2712 is supporting that improved clock source. Also, from the following lines
at the end of cfginit function:
/* Guesstimate the timer frequency (controller base clock) */
base_clk_mhz = max_t(u32, clk_get_rate(pltfm_host->clk) / (1000 * 1000), 1);
reg = SDIO_CFG_CQ_CAPABILITY_FMUL | base_clk_mhz;
writel(reg, brcmstb_priv->cfg_regs + SDIO_CFG_CQ_CAPABILITY);
judging from the name of SDIO_CFG_CQ_CAPABILITY register, I'd guess that it relates
to some Command Queue (timeout?) setting so it's probably only important if CQE is
enabled specifying 'supports-cqe' property, which is not in current devicetree (nor
in downstream one). If this is the case it's mostly a performance improvement, and
as such something that we are not necessarily interested in right now since this
patchset adds just minimal boot support. I would then drop those lines, as we could
just reintroduce them if they need be once we have a better understanding of that
specific register and/if the cqe support will be enabled. As a matter of fact those
lines are not working as expected in any case since pltfm_host->clk is set at the
very end of sdhci_brcmstb_probe() while the cfginit function is invoked much earlier.
The result is that right now the value set ito SDIO_CFG_CQ_CAPABILITY register is always
equal to 1MHz. Further testing reveals that it is indeed working fine even with those
lines dropped, so I would deem that code as unnecessary for this early patchset.
Is it a viable solution?
Many thanks,
Andrea
> > /*
> > * Supply the existing CAPS, but clear the UHS modes. This
> > * will allow these modes to be specified by device tree
>
