From mboxrd@z Thu Jan 1 00:00:00 1970 From: Heiko =?ISO-8859-1?Q?St=FCbner?= Subject: Re: [PATCH v2 1/2] soc: rockchip: Register a soc_device to retrieve revision Date: Wed, 01 Apr 2020 18:34:16 +0200 Message-ID: <5143930.cPWVAAQKI9@diego> References: <20200401153513.423683-1-mylene.josserand@collabora.com> <20200401153513.423683-2-mylene.josserand@collabora.com> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Return-path: In-Reply-To: <20200401153513.423683-2-mylene.josserand@collabora.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: "linux-arm-kernel" Errors-To: linux-arm-kernel-bounces+linux-arm-kernel=m.gmane-mx.org@lists.infradead.org To: =?ISO-8859-1?Q?Myl=E8ne?= Josserand Cc: sboyd@kernel.org, mturquette@baylibre.com, linux-kernel@vger.kernel.org, kever.yang@rock-chips.com, linux-rockchip@lists.infradead.org, geert@linux-m68k.org, kernel@collabora.com, linux-clk@vger.kernel.org, linux-arm-kernel@lists.infradead.org List-Id: linux-rockchip.vger.kernel.org Hi Myl=E8ne, Am Mittwoch, 1. April 2020, 17:35:12 CEST schrieb Myl=E8ne Josserand: > Determine which revision of rk3288 by checking the HDMI version. > According to the Rockchip BSP kernel/u-boot, on rk3288w, the HDMI > revision equals 0x1A which is not the case for the rk3288. > = > As these SOC have some differences, this driver will help us > to know on which revision we are by using 'soc_device' registration > to be able to use 'soc_device_match' to detect rk3288/rk3288w. > = > Signed-off-by: Myl=E8ne Josserand I like your new approach quite a lot :-) There are some things we need to take into account though, see below. > --- > drivers/soc/rockchip/Makefile | 1 + > drivers/soc/rockchip/rk3288.c | 125 ++++++++++++++++++++++++++++++++++ > 2 files changed, 126 insertions(+) > create mode 100644 drivers/soc/rockchip/rk3288.c > = > diff --git a/drivers/soc/rockchip/Makefile b/drivers/soc/rockchip/Makefile > index afca0a4c4b72..9dbf12913512 100644 > --- a/drivers/soc/rockchip/Makefile > +++ b/drivers/soc/rockchip/Makefile > @@ -2,5 +2,6 @@ > # > # Rockchip Soc drivers > # > +obj-$(CONFIG_ARCH_ROCKCHIP) +=3D rk3288.o > obj-$(CONFIG_ROCKCHIP_GRF) +=3D grf.o > obj-$(CONFIG_ROCKCHIP_PM_DOMAINS) +=3D pm_domains.o > diff --git a/drivers/soc/rockchip/rk3288.c b/drivers/soc/rockchip/rk3288.c I'd really like this to be a soc.c instead of rk3288.c ;-) > new file mode 100644 > index 000000000000..83379ba2b31b > --- /dev/null > +++ b/drivers/soc/rockchip/rk3288.c > @@ -0,0 +1,125 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* > + * Copyright 2020 Collabora Ltd > + * Author: Mylene Josserand > + */ > + > +#include > +#include > +#include > +#include > +#include > +#include > +#include > + > +#define RK3288_HDMI_REV_REG 0x04 > +#define RK3288W_HDMI_REV 0x1A > + > +enum rk3288_soc_rev { > + RK3288_SOC_REV_NOT_DETECT, > + RK3288_SOC_REV_RK3288, > + RK3288_SOC_REV_RK3288W, > +}; > + > +static int rk3288_revision(void) > +{ > + static int revision =3D RK3288_SOC_REV_NOT_DETECT; > + struct device_node *dn; > + void __iomem *hdmi_base; > + > + if (revision !=3D RK3288_SOC_REV_NOT_DETECT) > + return revision; > + > + dn =3D of_find_compatible_node(NULL, NULL, "rockchip,rk3288-dw-hdmi"); > + if (!dn) { > + pr_err("%s: Couldn't find HDMI node\n", __func__); > + return -EINVAL; > + } > + > + hdmi_base =3D of_iomap(dn, 0); > + of_node_put(dn); > + > + if (!hdmi_base) { > + pr_err("%s: Couldn't map %pOF regs\n", __func__, > + hdmi_base); > + return -ENXIO; > + } The possible problem I see here is clocking and power-domain of the hdmi controller in corner-cases. In the past we already had a lot of fun with kexec, which also indicates that people actually use kexec productively. So while all clocks are ungated and all power-domains are powered on first boot, on a system without graphics the pclk+power-domain could be off when doing a kexec into a second kernel, which then would probably hang here. Of course with the hdmi-pclk being sourced from hclk_vio we run into a chicken-egg-problem, as we need pclk_hdmi_ctrl to register hclk_vio at all. So I guess one way out of this could be to - amend rk3288_clk_shutdown() to also ungate the hdmi-pclk on shutdown - add a shutdown mechanism to the power-domain driver so that it can enable PD_VIO on shutdown > + > + if (readl_relaxed(hdmi_base + RK3288_HDMI_REV_REG) > + =3D=3D RK3288W_HDMI_REV) nit: a nicer look would be something like val =3D readl_relaxed(hdmi_base + RK3288_HDMI_REV_REG); if (val =3D=3D RK3288W_HDMI_REV) > + revision =3D RK3288_SOC_REV_RK3288W; > + else > + revision =3D RK3288_SOC_REV_RK3288; > + > + iounmap(hdmi_base); > + > + return revision; > +} > + > +static const char *rk3288_socinfo_revision(u32 rev) > +{ > + const char *soc_rev; > + > + switch (rev) { > + case RK3288_SOC_REV_RK3288: > + soc_rev =3D "RK3288"; > + break; > + > + case RK3288_SOC_REV_RK3288W: > + soc_rev =3D "RK3288w"; can we maybe use lower-case letters for all here? > + break; > + > + case RK3288_SOC_REV_NOT_DETECT: > + soc_rev =3D ""; > + break; > + > + default: > + soc_rev =3D "unknown"; > + break; > + } > + > + return kstrdup_const(soc_rev, GFP_KERNEL); > +} > + > +static const struct of_device_id rk3288_soc_match[] =3D { > + { .compatible =3D "rockchip,rk3288", }, > + { } > +}; > + > +static int __init rk3288_soc_init(void) as noted at the top, I'd really like to see this more generalized so that other socs can just hook in there with a revision callback in a rockchip_soc_data struct. > +{ > + struct soc_device_attribute *soc_dev_attr; > + struct soc_device *soc_dev; > + struct device_node *np; > + > + np =3D of_find_matching_node(NULL, rk3288_soc_match); > + if (!np) > + return -ENODEV; > + > + soc_dev_attr =3D kzalloc(sizeof(*soc_dev_attr), GFP_KERNEL); > + if (!soc_dev_attr) > + return -ENOMEM; > + > + soc_dev_attr->family =3D "Rockchip"; > + soc_dev_attr->soc_id =3D "RK32xx"; nit: rk3288 instead of "32xx" please > + > + np =3D of_find_node_by_path("/"); > + of_property_read_string(np, "model", &soc_dev_attr->machine); > + of_node_put(np); > + > + soc_dev_attr->revision =3D rk3288_socinfo_revision(rk3288_revision()); > + > + soc_dev =3D soc_device_register(soc_dev_attr); > + if (IS_ERR(soc_dev)) { > + kfree_const(soc_dev_attr->revision); > + kfree_const(soc_dev_attr->soc_id); > + kfree(soc_dev_attr); > + return PTR_ERR(soc_dev); > + } > + > + dev_info(soc_device_to_device(soc_dev), "Rockchip %s %s detected\n", > + soc_dev_attr->soc_id, soc_dev_attr->revision); nit: dev_dbg should be enough, that information doesn't really matter for most people, as it's only relevant to clock internals. Heiko