| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CADrjBPqT=F+MALssnWsMa1E1RcPN7Sz8s+HfEZRU3pSgPO+z6w@mail.gmail.com>
Date: Tue, 6 Feb 2024 15:56:39 +0000
From: Peter Griffin <peter.griffin@...aro.org>
To: Sam Protsenko <semen.protsenko@...aro.org>
Cc: arnd@...db.de, krzysztof.kozlowski@...aro.org, linux@...ck-us.net,
wim@...ux-watchdog.org, alim.akhtar@...sung.com, jaewon02.kim@...sung.com,
kernel-team@...roid.com, alexey.klimov@...aro.org, tudor.ambarus@...aro.org,
andre.draszik@...aro.org, saravanak@...gle.com, willmcvicker@...gle.com,
linux-fsd@...la.com, linux-watchdog@...r.kernel.org,
devicetree@...r.kernel.org, linux-kernel@...r.kernel.org,
linux-arm-kernel@...ts.infradead.org, linux-samsung-soc@...r.kernel.org
Subject: Re: [PATCH v3 1/2] soc: samsung: exynos-pmu: Add regmap support for
SoCs that protect PMU regs
Hi Sam,
Thanks for the great review feedback.
On Fri, 2 Feb 2024 at 17:35, Sam Protsenko <semen.protsenko@...aro.org> wrote:
>
> On Fri, Feb 2, 2024 at 8:57 AM Peter Griffin <peter.griffin@...aro.org> wrote:
> >
> > Some Exynos based SoCs like Tensor gs101 protect the PMU registers for
> > security hardening reasons so that they are only write accessible in el3
> > via an SMC call.
> >
> > As most Exynos drivers that need to write PMU registers currently obtain a
> > regmap via syscon (phys, pinctrl, watchdog). Support for the above usecase
> > is implemented in this driver using a custom regmap similar to syscon to
> > handle the SMC call. Platforms that don't secure PMU registers, get a mmio
> > regmap like before. As regmaps abstract out the underlying register access
> > changes to the leaf drivers are minimal.
> >
> > A new API exynos_get_pmu_regmap_by_phandle() is provided for leaf drivers
> > that currently use syscon_regmap_lookup_by_phandle(). This also handles
> > deferred probing.
> >
> > Signed-off-by: Peter Griffin <peter.griffin@...aro.org>
> > ---
> > Changes since v2
> > - Add select REGMAP to Kconfig
> > - Add constant for SET/CLEAR bits
> > - Replace kerneldoc with one line comment
> > - Fix kerneldoc for EXPORT_SYMBOL_GPL funcs
> > - remove superflous extern keyword
> > - dev_err_probe() on probe error
> > - shorten regmcfg name
> > - no compatibles inside probe, use match data
> > - don't mix declarations with/without initializations
> > - tensor_sec_reg_read() use mmio to avoid access restrictions
> > - Collect up Reviewed-by
> > - const for regmap_config structs
> > ---
> > drivers/soc/samsung/Kconfig | 1 +
> > drivers/soc/samsung/exynos-pmu.c | 233 ++++++++++++++++++++++++-
> > drivers/soc/samsung/exynos-pmu.h | 1 +
> > include/linux/soc/samsung/exynos-pmu.h | 11 +-
> > 4 files changed, 241 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/soc/samsung/Kconfig b/drivers/soc/samsung/Kconfig
> > index 27ec99af77e3..1a5dfdc978dc 100644
> > --- a/drivers/soc/samsung/Kconfig
> > +++ b/drivers/soc/samsung/Kconfig
> > @@ -42,6 +42,7 @@ config EXYNOS_PMU
> > depends on ARCH_EXYNOS || ((ARM || ARM64) && COMPILE_TEST)
> > select EXYNOS_PMU_ARM_DRIVERS if ARM && ARCH_EXYNOS
> > select MFD_CORE
> > + select REGMAP_MMIO
> >
> > # There is no need to enable these drivers for ARMv8
> > config EXYNOS_PMU_ARM_DRIVERS
> > diff --git a/drivers/soc/samsung/exynos-pmu.c b/drivers/soc/samsung/exynos-pmu.c
> > index 250537d7cfd6..adf3549370d6 100644
> > --- a/drivers/soc/samsung/exynos-pmu.c
> > +++ b/drivers/soc/samsung/exynos-pmu.c
> > @@ -5,6 +5,7 @@
> > //
> > // Exynos - CPU PMU(Power Management Unit) support
> >
> > +#include <linux/arm-smccc.h>
> > #include <linux/of.h>
> > #include <linux/of_address.h>
> > #include <linux/mfd/core.h>
> > @@ -12,19 +13,130 @@
> > #include <linux/of_platform.h>
> > #include <linux/platform_device.h>
> > #include <linux/delay.h>
> > +#include <linux/regmap.h>
> >
> > #include <linux/soc/samsung/exynos-regs-pmu.h>
> > #include <linux/soc/samsung/exynos-pmu.h>
> >
> > #include "exynos-pmu.h"
> >
> > +#define PMUALIVE_MASK GENMASK(14, 0)
>
> Are you sure it's not GENMASK(13, 0)?
Yes, you're right, mask should be 0x3fff.
> Because SET_BITS has bit #14
> set, which overlaps with bit #14 from PMUALIVE_MASK, when being added
> in tensor_set_bit_atomic().
>
> This also can be aligned with below definitions.
Will update alignment
>
> > +#define SET_BITS 0xc000
> > +#define CLEAR_BITS 0x8000
>
> All 3 above values seem to be gs101 specific. At least I can't find
> any similar atomic registers in Exynos850 TRM, in PMU block. So I'd
> suggest also adding TENSOR_ prefix to those to make it clear and to
> prevent possible naming conflicts in future.
Sure, I can add a Tensor prefix. Unfortunately these atomic registers
aren't mentioned anywhere in gs101 TRM's either, so it is a bit of a
nature study :( It was implemented and used in the downstream drivers
though (and doesn't look to be implemented downstream for exynos850,
so you're likely correct it doesn't have it.). I tried to check some
other downstream product tree's, I found the hardware looks to exist
on Exynos 9820 [1]. But I think we can update the name if somebody
tries to upstream that platform in the future.
[1] https://github.com/PixelOS-Devices/kernel_samsung_exynos9820/blob/thirteen/drivers/soc/samsung/exynos-pmu.c#L32
>
> Also, not sure if it makes things better, but FWIW:
>
> #define CLEAR_BITS BIT(15)
> #define SET_BITS (BIT(15) | BIT(14))
I agree that would make it clearer.
>
> Just to show that those two bits don't overlap with PMUALIVE_MASK (if
> it can be fixed to 13:0), and show their relation. If I understand
> correctly what's going on anyway.
I think you've understood what's happening here perfectly
>
> > +
> > +#define TENSOR_SMC_PMU_SEC_REG 0x82000504
> > +#define TENSOR_PMUREG_READ 0
> > +#define TENSOR_PMUREG_WRITE 1
> > +#define TENSOR_PMUREG_RMW 2
> > +
> > struct exynos_pmu_context {
> > struct device *dev;
> > const struct exynos_pmu_data *pmu_data;
> > + struct regmap *pmureg;
> > };
> >
> > void __iomem *pmu_base_addr;
> > static struct exynos_pmu_context *pmu_context;
> > +static struct platform_driver exynos_pmu_driver;
>
> Just an idea: maybe add a comment saying it's a forward declaration,
> and the variable is assigned below, as it might be confusing. Not sure
> if it's worth it though.
>
> > +
> > +/*
> > + * Tensor SoCs are configured so that PMU_ALIVE registers can only be written
> > + * from EL3, but are still read accessible. As Linux needs to write some of
> > + * these registers, the following functions are provided and exposed via
> > + * regmap.
> > + *
> > + * Note: This SMC interface is known to be implemented on gs101 and derivative
> > + * SoCs.
> > + */
> > +
> > +/* Write to a protected PMU register. */
> > +static int tensor_sec_reg_write(void *base, unsigned int reg, unsigned int val)
> > +{
> > + struct arm_smccc_res res;
> > + unsigned long pmu_base = (unsigned long)base;
> > +
> > + arm_smccc_smc(TENSOR_SMC_PMU_SEC_REG, pmu_base + reg,
> > + TENSOR_PMUREG_WRITE, val, 0, 0, 0, 0, &res);
> > +
> > + /* returns -EINVAL if access isn't allowed or 0 */
> > + if (res.a0)
> > + pr_warn("%s(): SMC failed: %d\n", __func__, (int)res.a0);
> > +
> > + return (int)res.a0;
> > +}
> > +
> > +/* Read/Modify/Write a protected PMU register. */
> > +static int tensor_sec_reg_rmw(void *base, unsigned int reg,
> > + unsigned int mask, unsigned int val)
> > +{
> > + struct arm_smccc_res res;
> > + unsigned long pmu_base = (unsigned long)base;
> > +
> > + arm_smccc_smc(TENSOR_SMC_PMU_SEC_REG, pmu_base + reg,
> > + TENSOR_PMUREG_RMW, mask, val, 0, 0, 0, &res);
> > +
> > + /* returns -EINVAL if access isn't allowed or 0*/
> > + if (res.a0)
> > + pr_warn("%s(): SMC failed: %d\n", __func__, (int)res.a0);
> > +
> > + return (int)res.a0;
> > +}
> > +
> > +/*
> > + * Read a protected PMU register. All PMU registers can be read by Linux.
> > + * Note: The SMC read register is not used, as only registers that can be
> > + * written are readable via SMC.
> > + */
> > +static int tensor_sec_reg_read(void *base, unsigned int reg, unsigned int *val)
> > +{
> > + *val = pmu_raw_readl(reg);
> > + return 0;
> > +}
> > +
> > +/*
> > + * For SoCs that have set/clear bit hardware this function can be used when
> > + * the PMU register will be accessed by multiple masters.
> > + *
> > + * For example, to set bits 13:8 in PMU reg offset 0x3e80
> > + * tensor_set_bit_atomic(0x3e80, 0x3f00, 0x3f00);
> > + *
> > + * To clear bits 13:8 in PMU offset 0x3e80
> > + * tensor_set_bit_atomic(0x3e80, 0x0, 0x3f00);
> > + */
> > +static inline int tensor_set_bit_atomic(void *ctx, unsigned int offset,
>
> set_bit -> set_bits?
Will change
>
> > + u32 val, u32 mask)
> > +{
> > + int ret;
> > + unsigned int i;
> > +
> > + for (i = 0; i < 32; i++) {
> > + if (mask & BIT(i)) {
>
> Maybe replace it with:
>
> if (!(mask & BIT(i)))
> continue;
>
> to reduce the indentation level?
Will do
>
> > + if (val & BIT(i))
> > + offset |= SET_BITS;
> > + else
> > + offset |= CLEAR_BITS;
>
> What if someone calls this functions like this:
>
> tensor_set_bit_atomic(0x3e80, 0x100, 0x3f00);
>
> which means "set bit #8, and clear bits 13:9). But because the offset
> variable will hold SET_BITS set during bit #8 handling, bits 13:9 are
> also going to be set, effectively making that call act like
> tensor_set_bit_atomic(0x3e80, 0x3f00, 0x3f00) instead. So I'd add
> something like:
>
> offset &= ~SET_BITS;
>
> before doing |= operations.
Good catch! Will fix in v4
>
> > +
> > + ret = tensor_sec_reg_write(ctx, offset, i);
> > + if (ret)
> > + goto out;
>
> Maybe remove "out" and just do return ret here?
Will update
>
> > + }
> > + }
> > +out:
> > + return ret;
> > +}
> > +
> > +static int tensor_sec_update_bits(void *ctx, unsigned int reg,
> > + unsigned int mask, unsigned int val)
> > +{
> > + /*
> > + * Use atomic operations for PMU_ALIVE registers (offset 0~0x3FFF)
> > + * as the target registers can be accessed by multiple masters.
> > + */
> > + if (reg > PMUALIVE_MASK)
> > + return tensor_sec_reg_rmw(ctx, reg, mask, val);
> > +
> > + return tensor_set_bit_atomic(ctx, reg, val, mask);
> > +}
> >
> > void pmu_raw_writel(u32 val, u32 offset)
> > {
> > @@ -75,11 +187,41 @@ void exynos_sys_powerdown_conf(enum sys_powerdown mode)
> > #define exynos_pmu_data_arm_ptr(data) NULL
> > #endif
> >
> > +static const struct regmap_config regmap_smccfg = {
> > + .name = "pmu_regs",
> > + .reg_bits = 32,
> > + .reg_stride = 4,
> > + .val_bits = 32,
> > + .fast_io = true,
> > + .use_single_read = true,
> > + .use_single_write = true,
> > + .reg_read = tensor_sec_reg_read,
> > + .reg_write = tensor_sec_reg_write,
> > + .reg_update_bits = tensor_sec_update_bits,
> > +};
> > +
> > +static const struct regmap_config regmap_mmiocfg = {
> > + .name = "pmu_regs",
> > + .reg_bits = 32,
> > + .reg_stride = 4,
> > + .val_bits = 32,
> > + .fast_io = true,
> > + .use_single_read = true,
> > + .use_single_write = true,
> > +};
> > +
> > +static const struct exynos_pmu_data gs101_pmu_data = {
> > + .pmu_secure = true
> > +};
> > +
> > /*
> > * PMU platform driver and devicetree bindings.
> > */
> > static const struct of_device_id exynos_pmu_of_device_ids[] = {
> > {
> > + .compatible = "google,gs101-pmu",
> > + .data = &gs101_pmu_data,
> > + }, {
> > .compatible = "samsung,exynos3250-pmu",
> > .data = exynos_pmu_data_arm_ptr(exynos3250_pmu_data),
> > }, {
> > @@ -113,19 +255,73 @@ static const struct mfd_cell exynos_pmu_devs[] = {
> > { .name = "exynos-clkout", },
> > };
> >
> > +/**
> > + * exynos_get_pmu_regmap() - Obtain pmureg regmap
> > + *
> > + * Find the pmureg regmap previously configured in probe() and return regmap
> > + * pointer.
> > + *
> > + * Return: A pointer to regmap if found or ERR_PTR error value.
> > + */
> > struct regmap *exynos_get_pmu_regmap(void)
> > {
> > struct device_node *np = of_find_matching_node(NULL,
> > exynos_pmu_of_device_ids);
> > if (np)
> > - return syscon_node_to_regmap(np);
> > + return exynos_get_pmu_regmap_by_phandle(np, NULL);
> > return ERR_PTR(-ENODEV);
> > }
> > EXPORT_SYMBOL_GPL(exynos_get_pmu_regmap);
> >
> > +/**
> > + * exynos_get_pmu_regmap_by_phandle() - Obtain pmureg regmap via phandle
> > + * @np: Pointer to device's Device Tree node
> > + * @property: Device Tree property name which references the pmu
> > + *
> > + * Find the pmureg regmap previously configured in probe() and return regmap
> > + * pointer.
> > + *
> > + * Return: A pointer to regmap if found or ERR_PTR error value.
> > + */
> > +struct regmap *exynos_get_pmu_regmap_by_phandle(struct device_node *np,
> > + const char *property)
> > +{
> > + struct device *dev;
> > + struct exynos_pmu_context *ctx;
> > + struct device_node *pmu_np;
> > +
> > + if (property)
> > + pmu_np = of_parse_phandle(np, property, 0);
> > + else
> > + pmu_np = np;
> > +
> > + if (!pmu_np)
> > + return ERR_PTR(-ENODEV);
> > +
> > + /*
> > + * Determine if exynos-pmu device has probed and therefore regmap
> > + * has been created and can be returned to the caller. Otherwise we
> > + * return -EPROBE_DEFER.
> > + */
> > + dev = driver_find_device_by_of_node(&exynos_pmu_driver.driver,
> > + (void *)pmu_np);
> > +
> > + of_node_put(pmu_np);
> > + if (!dev)
> > + return ERR_PTR(-EPROBE_DEFER);
> > +
> > + ctx = dev_get_drvdata(dev);
> > +
> > + return ctx->pmureg;
> > +}
> > +EXPORT_SYMBOL_GPL(exynos_get_pmu_regmap_by_phandle);
> > +
> > static int exynos_pmu_probe(struct platform_device *pdev)
> > {
> > struct device *dev = &pdev->dev;
> > + struct regmap_config pmu_regmcfg;
> > + struct regmap *regmap;
> > + struct resource *res;
> > int ret;
> >
> > pmu_base_addr = devm_platform_ioremap_resource(pdev, 0);
> > @@ -133,13 +329,42 @@ static int exynos_pmu_probe(struct platform_device *pdev)
> > return PTR_ERR(pmu_base_addr);
> >
> > pmu_context = devm_kzalloc(&pdev->dev,
> > - sizeof(struct exynos_pmu_context),
> > - GFP_KERNEL);
> > + sizeof(struct exynos_pmu_context),
> > + GFP_KERNEL);
> > if (!pmu_context)
> > return -ENOMEM;
> > - pmu_context->dev = dev;
> > +
> > + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> > + if (!res)
> > + return -ENODEV;
> > +
> > pmu_context->pmu_data = of_device_get_match_data(dev);
> >
> > + /* For SoCs that secure PMU register writes use custom regmap */
> > + if (pmu_context->pmu_data && pmu_context->pmu_data->pmu_secure) {
> > + pmu_regmcfg = regmap_smccfg;
> > + pmu_regmcfg.max_register = resource_size(res) -
> > + pmu_regmcfg.reg_stride;
> > + /* Need physical address for SMC call */
> > + regmap = devm_regmap_init(dev, NULL,
> > + (void *)(uintptr_t)res->start,
> > + &pmu_regmcfg);
> > + } else {
> > + /* all other SoCs use a MMIO regmap */
>
> Suggest starting with a capital letter, for consistency with previous comments.
Will update
>
> > + pmu_regmcfg = regmap_mmiocfg;
> > + pmu_regmcfg.max_register = resource_size(res) -
> > + pmu_regmcfg.reg_stride;
> > + regmap = devm_regmap_init_mmio(dev, pmu_base_addr,
> > + &pmu_regmcfg);
> > + }
> > +
> > + if (IS_ERR(regmap))
> > + dev_err_probe(&pdev->dev, PTR_ERR(regmap),
> > + "regmap init failed\n");
>
> Why not "return dev_err_probe()"? Is it ok to continue with no regmap created?
That should have been return dev_err_probe. Will fix
Peter
>
> > +
> > + pmu_context->pmureg = regmap;
> > + pmu_context->dev = dev;
> > +
> > if (pmu_context->pmu_data && pmu_context->pmu_data->pmu_init)
> > pmu_context->pmu_data->pmu_init();
> >
> > diff --git a/drivers/soc/samsung/exynos-pmu.h b/drivers/soc/samsung/exynos-pmu.h
> > index 1c652ffd79b4..0a49a2c9a08e 100644
> > --- a/drivers/soc/samsung/exynos-pmu.h
> > +++ b/drivers/soc/samsung/exynos-pmu.h
> > @@ -21,6 +21,7 @@ struct exynos_pmu_conf {
> > struct exynos_pmu_data {
> > const struct exynos_pmu_conf *pmu_config;
> > const struct exynos_pmu_conf *pmu_config_extra;
> > + bool pmu_secure;
> >
> > void (*pmu_init)(void);
> > void (*powerdown_conf)(enum sys_powerdown);
> > diff --git a/include/linux/soc/samsung/exynos-pmu.h b/include/linux/soc/samsung/exynos-pmu.h
> > index a4f5516cc956..406ed73614fd 100644
> > --- a/include/linux/soc/samsung/exynos-pmu.h
> > +++ b/include/linux/soc/samsung/exynos-pmu.h
> > @@ -20,12 +20,21 @@ enum sys_powerdown {
> >
> > extern void exynos_sys_powerdown_conf(enum sys_powerdown mode);
> > #ifdef CONFIG_EXYNOS_PMU
> > -extern struct regmap *exynos_get_pmu_regmap(void);
> > +struct regmap *exynos_get_pmu_regmap(void);
> > +
>
> Usually empty line delimeter is not needed in cases like that.
>
> > +struct regmap *exynos_get_pmu_regmap_by_phandle(struct device_node *np,
> > + const char *property);
> > #else
> > static inline struct regmap *exynos_get_pmu_regmap(void)
> > {
> > return ERR_PTR(-ENODEV);
> > }
> > +
> > +static inline struct regmap *exynos_get_pmu_regmap_by_phandle(struct device_node *np,
> > + const char *property)
> > +{
> > + return ERR_PTR(-ENODEV);
> > +}
> > #endif
> >
> > #endif /* __LINUX_SOC_EXYNOS_PMU_H */
> > --
> > 2.43.0.594.gd9cf4e227d-goog
> >
Powered by blists - more mailing lists