Re: [PATCH v2 2/2] cache: add SMCCC-backed cache invalidate provider

["Bence Csókás" <bence.csokas@arm.com>]

Hi,

I'm reading the 1.7 H BET0 [1] version of the spec (released 
2026-04-20), and I see some discrepancies. Not sure if I'm the one 
reading it wrong, or if you had a different version than me.

[1] https://developer.arm.com/documentation/den0028/h/?lang=en

Either way, thanks for posting this patch, I'm also looking forward to 
seeing it get merged!

On 2026. 06. 09. 0:07, Srirangan Madhavan wrote:
> Add a cache maintenance provider for the Arm SMCCC cache clean+invalidate
> interface.
> 
> The provider discovers SMCCC support and attributes at init time,
> serializes firmware calls, handles transient BUSY and RATE_LIMITED
> responses with bounded retries, and registers with the generic cache
> coherency framework used by memregion callers.
> 
> Signed-off-by: Srirangan Madhavan <smadhavan@nvidia.com>
> Reviewed-by: Jonathan Cameron <jic23@kernel.org>
> ---
>   drivers/cache/Kconfig           |  11 +++
>   drivers/cache/Makefile          |   2 +
>   drivers/cache/arm_smccc_cache.c | 157 ++++++++++++++++++++++++++++++++
>   3 files changed, 170 insertions(+)
>   create mode 100644 drivers/cache/arm_smccc_cache.c
> 
> diff --git a/drivers/cache/Kconfig b/drivers/cache/Kconfig
> index 1518449d47b5..57fd1823dec5 100644
> --- a/drivers/cache/Kconfig
> +++ b/drivers/cache/Kconfig
> @@ -42,6 +42,17 @@ menuconfig CACHEMAINT_FOR_HOTPLUG
>   
>   if CACHEMAINT_FOR_HOTPLUG
>   
> +config ARM_SMCCC_CACHE
> +	bool "Arm SMCCC cache maintenance provider"
> +	depends on ARM64 && HAVE_ARM_SMCCC_DISCOVERY
> +	help
> +	  Enable support for the Arm SMCCC cache clean+invalidate
> +	  interface as a provider for memory hotplug-like cache
> +	  maintenance operations.
> +	  The provider registers only when firmware advertises the
> +	  SMCCC calls and attributes, so systems without firmware support
> +	  continue without this registered provider.
> +
>   config HISI_SOC_HHA
>   	tristate "HiSilicon Hydra Home Agent (HHA) device driver"
>   	depends on (ARM64 && ACPI) || COMPILE_TEST
> diff --git a/drivers/cache/Makefile b/drivers/cache/Makefile
> index b3362b15d6c1..55736a032d6f 100644
> --- a/drivers/cache/Makefile
> +++ b/drivers/cache/Makefile
> @@ -4,4 +4,6 @@ obj-$(CONFIG_AX45MP_L2_CACHE)		+= ax45mp_cache.o
>   obj-$(CONFIG_SIFIVE_CCACHE)		+= sifive_ccache.o
>   obj-$(CONFIG_STARFIVE_STARLINK_CACHE)	+= starfive_starlink_cache.o
>   
> +# Providers below depend on CACHEMAINT_FOR_HOTPLUG.
> +obj-$(CONFIG_ARM_SMCCC_CACHE)		+= arm_smccc_cache.o
>   obj-$(CONFIG_HISI_SOC_HHA)		+= hisi_soc_hha.o
> diff --git a/drivers/cache/arm_smccc_cache.c b/drivers/cache/arm_smccc_cache.c
> new file mode 100644
> index 000000000000..82b9efdb190b
> --- /dev/null
> +++ b/drivers/cache/arm_smccc_cache.c
> @@ -0,0 +1,157 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright (C) 2026 NVIDIA Corporation
> + *
> + * Arm SMCCC cache maintenance provider using cache clean+invalidate calls.
> + */
> +
> +#include <linux/arm-smccc.h>
> +#include <linux/cache_coherency.h>
> +#include <linux/cleanup.h>
> +#include <linux/delay.h>
> +#include <linux/errno.h>
> +#include <linux/init.h>
> +#include <linux/kernel.h>
> +#include <linux/mutex.h>
> +#include <linux/nmi.h>
> +
> +#define SMCCC_CACHE_MAX_RETRIES		5

Indentation seems off here.

> +#define SMCCC_CACHE_DEFAULT_DELAY_US	1000UL
> +#define SMCCC_CACHE_MAX_DELAY_US	20000UL
> +
> +struct smccc_cache {
> +	/* Must be first member */
> +	struct cache_coherency_ops_inst cci;
> +	struct mutex lock; /* Serializes SMCCC cache maintenance calls. */
> +	u32 latency_us;
> +	u32 rate_limit;
> +};
> +
> +static int smccc_cache_status_to_errno(s32 status)

I could see this being useful in the common smccc.c, not just here.

> +{
> +	switch (status) {
> +	case SMCCC_RET_SUCCESS:
> +		return 0;
> +	case SMCCC_RET_NOT_SUPPORTED:
> +		return -EOPNOTSUPP;
> +	case SMCCC_RET_INVALID_PARAMETER:
> +		return -EINVAL;
> +	case SMCCC_RET_RATE_LIMITED:
> +		return -EAGAIN;
> +	case SMCCC_RET_BUSY:
> +		return -EBUSY;
> +	default:
> +		return -EIO;
> +	}
> +}
> +
> +static unsigned long smccc_cache_delay_us(const struct smccc_cache *cache)
> +{
> +	unsigned long delay_us = 0;
> +
> +	if (cache->rate_limit)
> +		delay_us = DIV_ROUND_UP_ULL(USEC_PER_SEC, cache->rate_limit);
> +
> +	if (cache->latency_us)
> +		delay_us = max(delay_us, (unsigned long)cache->latency_us);
> +
> +	/*
> +	 * Firmware may advertise neither a rate limit nor a latency hint; use
> +	 * a small bounded backoff instead of retrying in a tight loop.
> +	 */
> +	if (!delay_us)
> +		delay_us = SMCCC_CACHE_DEFAULT_DELAY_US;
> +
> +	return min(delay_us, SMCCC_CACHE_MAX_DELAY_US);
> +}
> +
> +static int smccc_cache_wbinv(struct cache_coherency_ops_inst *cci,
> +			     struct cc_inval_params *invp)
> +{
> +	struct smccc_cache *cache = container_of(cci, struct smccc_cache, cci);
> +	struct arm_smccc_res res = {};

I would move this down one, for aesthetics (reverse fir tree).

> +	unsigned long delay_us = smccc_cache_delay_us(cache);
> +	int ret;
> +
> +	if (!invp->size)
> +		return -EINVAL;
> +
> +	/*
> +	 * Serialize the full retry sequence. With the default bounds, a caller
> +	 * may hold the mutex across up to five 20ms backoff sleeps.
> +	 */
> +	guard(mutex)(&cache->lock);
> +
> +	for (unsigned int i = 0; i < SMCCC_CACHE_MAX_RETRIES; i++) {
> +		/* Long firmware operations can trigger watchdog checks. */
> +		touch_nmi_watchdog();
> +
> +		arm_smccc_1_1_invoke(ARM_SMCCC_ARCH_CLEAN_INV_MEMREGION,
> +				     invp->addr, invp->size, 0UL, &res);
> +
> +		ret = smccc_cache_status_to_errno((s32)res.a0);
> +		if (!ret)
> +			return 0;
> +
> +		if (ret != -EBUSY && ret != -EAGAIN)
> +			return ret;
> +
> +		fsleep(delay_us);
> +	}
> +
> +	return -EBUSY;

Minor: I would do `return ret;` so that we get the last error message, 
which could be either EBUSY or EAGAIN (depending on if FW responded BUSY 
or RATE_LIMITED).

> +}
> +
> +static const struct cache_coherency_ops smccc_cache_ops = {
> +	.wbinv = smccc_cache_wbinv,
> +};
> +
> +static int __init smccc_cache_init(void)
> +{
> +	struct smccc_cache *cache;

Again, I would move this line down one.

> +	struct arm_smccc_res res = {};
> +	int ret;
> +
> +	if (arm_smccc_get_version() < ARM_SMCCC_VERSION_1_1)
> +		return -ENODEV;
> +
> +	if (arm_smccc_1_1_get_conduit() == SMCCC_CONDUIT_NONE)
> +		return -ENODEV;
> +
> +	arm_smccc_1_1_invoke(ARM_SMCCC_ARCH_FEATURES_FUNC_ID,
> +			     ARM_SMCCC_ARCH_CLEAN_INV_MEMREGION, &res);
> +	if ((s32)res.a0 < 0)
> +		return -ENODEV;
> +
> +	arm_smccc_1_1_invoke(ARM_SMCCC_ARCH_FEATURES_FUNC_ID,
> +			     ARM_SMCCC_ARCH_CLEAN_INV_MEMREGION_ATTRIBUTES,
> +			     &res);
> +	if ((s32)res.a0 < 0)
> +		return -ENODEV;

7.11.2 states (top of page 47):
   This function must be implemented if SMCCC_ARCH_CLEAN_INV_MEMREGION 
is implemented.

Therefore, this check can be dropped, just checking for INV_MEMREGION's 
existence should be enough.

> +	arm_smccc_1_1_invoke(ARM_SMCCC_ARCH_CLEAN_INV_MEMREGION_ATTRIBUTES,
> +			     &res);
> +	if ((s32)res.a0)
> +		return -ENODEV;
> +
> +	cache = cache_coherency_ops_instance_alloc(&smccc_cache_ops,
> +						   struct smccc_cache, cci);
> +	if (!cache)
> +		return -ENOMEM;
> +
> +	mutex_init(&cache->lock);
> +	cache->latency_us = lower_32_bits(res.a2);
> +	cache->rate_limit = lower_32_bits(res.a3);

The spec only says to truncate X2 to 32 bits, not X3 (chapter 7.11, page 
46).

> +	ret = cache_coherency_ops_instance_register(&cache->cci);
> +	if (ret) {
> +		mutex_destroy(&cache->lock);
> +		cache_coherency_ops_instance_put(&cache->cci);
> +		return ret;
> +	}
> +
> +	pr_info("SMCCC cache clean+invalidate provider registered\n");

Is this log line useful for _production_ environments?

> +	return 0;
> +}
> +arch_initcall(smccc_cache_init);

Bence