Re: [PATCH v17 6/7] firmware: arm_rmm: Ensure the RMM has GPT entries for memory

[Gavin Shan <gshan@redhat.com>]

On 9/7/26 7:59 PM, Suzuki K Poulose wrote:
> From: Steven Price <steven.price@arm.com>
> 
> The RMM maintains the state of all the granules in the system to make
> sure that the host is abiding by the rules. This state can be maintained
> at different granularity, per page (TRACKING_FINE) or per region
> (TRACKING_COARSE or TRACKING_INTERMEDIATE). The region size depends on the
> underlying "RMI_GRANULE_SIZE". For a "coarse"/"intermediate" region, all pages
> in the region must be of the same state, this implies we need to have "fine"
> tracking for DRAM, so that we can delegate individual pages.
> 
> For now we only support a statically carved out memory for tracking
> granules for the "fine" regions. This can be extended in the future to
> allow modifying the tracking granularity and remove the need for a
> static allocation by the firmware.
> 
> Similarly, the firmware may create L0 GPT entries describing the total
> address space. But if we change the "PAS" (Physical Address Space) of a
> granule, then the firmware may need to create L1 tables to track the PAS
> at a finer granularity. Linux therefore checks if the platform firmware manages
> the PAR region. i.e., the firmware is in charge of managing the L1 GPTs
> (creation and the required memory for the GPT tables - via static carveouts)
> without host intervention. Support for dynamic GPT creation by the host will be
> added later.
> 
> If the firmware requires us to manage the tracking or GPT memory, Deactivate
> the RMM and reclaim any memory donated at RMM activation.
> 
> Apply the same checks when hotplugged memory is brought online.
> 
> Signed-off-by: Steven Price <steven.price@arm.com>
> [ Switch to RMI_GPT_L1_INFO for checking GPTs and deactivate RMM ]
> Co-Developed-by: Suzuki K Poulose <suzuki.poulose@arm.com>
> Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
> ---
>   Changes since v16:
>       * Check fine tracking and create L1 GPTs for hotplug-added memory.
>       * Clarify the L1 GPT setup and move the explanatory comment.
>       * Switch to using RMI_GPT_INFO command for checking the GPTs.
>       * Deactivate the RMM and reclaim the memory if we can't proceed.
>   Changes since v15:
>       * Skip firmware-reserved NOMAP memory in rmi_init_metadata()
>       * Handle negative error codes from wrappers.
>   Changes since v14:
>       * Move the implementation into drivers/firmware/arm_rmm.
>   Changes since v13:
>       * Moved out of KVM
> ---
>   drivers/firmware/arm_rmm/rmi.c | 139 +++++++++++++++++++++++++++++++++
>   include/linux/arm-rmi-cmds.h   |  75 ++++++++++++++++++
>   2 files changed, 214 insertions(+)
> 
> diff --git a/drivers/firmware/arm_rmm/rmi.c b/drivers/firmware/arm_rmm/rmi.c
> index d969c8738efde..34058e34188d3 100644
> --- a/drivers/firmware/arm_rmm/rmi.c
> +++ b/drivers/firmware/arm_rmm/rmi.c
> @@ -5,6 +5,7 @@
>   
>   #include <linux/cpufeature.h>
>   #include <linux/memblock.h>
> +#include <linux/memory.h>
>   #include <linux/arm-rmi-cmds.h>
>   #include <linux/processor.h>
>   #include <linux/slab.h>
> @@ -12,6 +13,8 @@
>   #include <asm/memory.h>
>   #include <asm/pgtable-hwdef.h>
>   
> +static bool arm64_rmi_is_available;
> +
>   /* Currently only the first 2 registers are used by Linux */
>   #define RMI_FEAT_REG_COUNT	2
>   static __ro_after_init unsigned long rmi_feat_reg_cache[RMI_FEAT_REG_COUNT];
> @@ -639,6 +642,124 @@ static int rmi_configure(void)
>   	return ret;
>   }
>   
> +/*
> + * Make sure the area is tracked by RMM at FINE granularity.
> + * We do not support changing the tracking yet.
> + */
> +static int rmi_verify_memory_tracking(phys_addr_t start, phys_addr_t end)
> +{
> +	while (start < end) {
> +		unsigned long ret, category, state, next;
> +
> +		ret = rmi_granule_tracking_get(start, end, &category, &state, &next);
> +		if (ret != RMI_SUCCESS)
> +			return -ENOMEM;
> +
> +		if (state != RMI_TRACKING_FINE ||
> +		    category != RMI_MEM_CATEGORY_CONVENTIONAL) {
> +			/* TODO: Set granule tracking in this case */
> +			pr_err("Granule tracking for region isn't fine/conventional: %llx-%lx\n",
> +			       start, next);
> +			return -ENODEV;
> +		}
> +		start = next;
> +	}
> +
> +	return 0;
> +}
> +
> +/*
> + * We do not support creating L1 GPTs yet. So, make sure that
> + * all the regions are managed by the firmware.
> + */
> +static int rmi_verify_gpt_firmware_managed(phys_addr_t start, phys_addr_t end)
> +{
> +	unsigned long l0gpt_sz;
> +	unsigned long next, par_state;
> +
> +	l0gpt_sz = 1UL << (30 + FIELD_GET(RMI_FEATURE_REGISTER_1_L0GPTSZ,
> +					  rmi_feat_reg(1)));
> +	start = ALIGN_DOWN(start, l0gpt_sz);
> +	end = ALIGN(end, l0gpt_sz);
> +
> +	while (start < end) {
> +		long ret = rmi_gpt_info(start, end, &next, &par_state);
> +
> +		if (ret != RMI_SUCCESS)
> +			return -ENOMEM;
> +
> +		if (par_state != RMI_GPT_PAR_PLAT) {
> +			pr_err("GPT for the region is not managed by firmware %llx-%lx\n",
> +				start, next);
> +			return -ENOMEM;

I guess -ENODEV is more appropriate:

			return -ENODEV

> +		}
> +		start = next;
> +	}
> +
> +	return 0;
> +}
> +
> +static int rmi_prepare_memory(phys_addr_t start, phys_addr_t end)
> +{
> +	int ret;
> +
> +	ret = rmi_verify_memory_tracking(start, end);
> +	if (ret)
> +		return ret;
> +
> +	return rmi_verify_gpt_firmware_managed(start, end);
> +}
> +
> +static int rmi_init_metadata(void)
> +{
> +	phys_addr_t start, end;
> +	struct memblock_region *r;
> +
> +	for_each_mem_region(r) {
> +		int ret;
> +
> +		/* Firmware-reserved NOMAP regions are not usable system RAM */
> +		if (memblock_is_nomap(r))
> +			continue;
> +
> +		start = memblock_region_memory_base_pfn(r) << PAGE_SHIFT;
> +		end = memblock_region_memory_end_pfn(r) << PAGE_SHIFT;
> +
> +		ret = rmi_prepare_memory(start, end);
> +		if (ret)
> +			return ret;

The local variable 'start' and 'end' can be dropped:

		ret = rmi_prepare_memory(PFN_PHYS(memblock_region_memory_base_pfn(r)),
					 PFN_PHYS(memblock_region_memory_end_pfn(r)));
		if (ret)
			return ret;

> +	}
> +
> +	return 0;
> +}
> +
> +static int rmi_memory_notifier(struct notifier_block *nb,
> +			       unsigned long action, void *data)
> +{
> +	struct memory_notify *arg = data;
> +	phys_addr_t start, end;
> +	int ret;
> +
> +	if (action != MEM_GOING_ONLINE)
> +		return NOTIFY_DONE;
> +
> +	start = PFN_PHYS(arg->start_pfn);
> +	end = PFN_PHYS(arg->start_pfn + arg->nr_pages);
> +	ret = rmi_prepare_memory(start, end);
> +
> +	return notifier_from_errno(ret);
> +}
> +
> +static struct notifier_block rmi_memory_nb = {
> +	.notifier_call = rmi_memory_notifier,
> +};
> +
> +bool is_rmi_available(void)
> +{
> +	return arm64_rmi_is_available;
> +}
> +EXPORT_SYMBOL_GPL(is_rmi_available);
> +
>   static int __init arm64_init_rmi(void)
>   {
>   	int ret = 0;
> @@ -666,8 +787,26 @@ static int __init arm64_init_rmi(void)
>   	if (ret) {
>   		pr_err("RMM activate failed\n");
>   		ret = ret < 0 ? ret : -ENXIO;
> +		goto out_free_sro;
>   	}
>   
> +	ret = rmi_init_metadata();
> +	if (ret)
> +		goto out_deactivate;
> +
> +	ret = register_memory_notifier(&rmi_memory_nb);
> +	if (ret)
> +		goto out_deactivate;
> +
> +	arm64_rmi_is_available = true;
> +	pr_info("RMI configured\n");
> +	kfree(sro);
> +
> +	return 0;
> +
> +out_deactivate:
> +	rmi_rmm_deactivate(sro);
> +out_free_sro:
>   	kfree(sro);
>   	return ret;
>   }
> diff --git a/include/linux/arm-rmi-cmds.h b/include/linux/arm-rmi-cmds.h
> index dea7c7004d35f..79e2c1f165112 100644
> --- a/include/linux/arm-rmi-cmds.h
> +++ b/include/linux/arm-rmi-cmds.h
> @@ -35,6 +35,8 @@ static inline int rmi_undelegate_page(phys_addr_t phys)
>   	return rmi_undelegate_range(phys, PAGE_SIZE);
>   }
>   
> +bool is_rmi_available(void);
> +
>   long rmi_sro_memxfer_execute(struct rmi_sro_state *sro, gfp_t gfp);
>   void rmi_sro_free(struct rmi_sro_state *sro);
>   long rmi_sro_execute(struct arm_smccc_1_2_regs *regs);
> @@ -64,6 +66,19 @@ static inline int rmi_rmm_config_set(unsigned long cfg_ptr)
>   	return res.a0;
>   }
>   
> +/**
> + * rmi_rmm_deactivate() - Deactivate the RMM and reclaim any memory donated at
> + * rmi_rmm_activate()
> + *
> + * @sro: Preallocated SRO context to be used
> + *
> + * Return: 0 on success, positive RMI result code or negative Linux error code
> + */
> +static inline long rmi_rmm_deactivate(struct rmi_sro_state *sro)
> +{
> +	return rmi_sro_memxfer_cmd(sro, GFP_KERNEL, SMC_RMI_RMM_DEACTIVATE);
> +}
> +

It seems rmi_rmm_deactivate() is used for once by rmi.c::arm64_init_rmi(). If so,
we needn't to expose this function and just combine the logics to rmi.c::arm64_init_rmi().

>   /**
>    * rmi_rmm_activate() - Activate the RMM
>    * @sro: Preallocated SRO context to be used
> @@ -75,6 +90,66 @@ static inline long rmi_rmm_activate(struct rmi_sro_state *sro)
>   	return rmi_sro_memxfer_cmd(sro, GFP_KERNEL, SMC_RMI_RMM_ACTIVATE);
>   }
>   
> +/**
> + * rmi_granule_tracking_get() - Get configuration of a Granule tracking region
> + * @start: Base PA of the tracking region
> + * @end: End of the PA region
> + * @out_category: Memory category
> + * @out_state: Tracking region state
> + * @out_top: Top of the memory region
> + *
> + * Return: RMI return code
> + */
> +static inline int rmi_granule_tracking_get(unsigned long start,
> +					   unsigned long end,
> +					   unsigned long *out_category,
> +					   unsigned long *out_state,
> +					   unsigned long *out_top)
> +{
> +	struct arm_smccc_res res;
> +
> +	arm_smccc_1_1_invoke(SMC_RMI_GRANULE_TRACKING_GET, start, end, &res);
> +
> +	if (res.a0 == RMI_SUCCESS) {
> +		if (out_category)
> +			*out_category = res.a1;
> +		if (out_state)
> +			*out_state = res.a2;
> +		if (out_top)
> +			*out_top = res.a3;
> +	}
> +
> +	return res.a0;
> +}
> +

rmi_granule_tracking_get() is used for once by rmi.c::rmi_verify_memory_tracking(). We needn't
expose rmi_granule_tracking_get() and combine its logic into rmi.c::rmi_verify_memory_tracking().

> +/*
> + * rmi_gpt_info - Query the GPT info for the given PAR.
> + * @base: Base of the physical address region
> + * @top: Top of the physical address region
> + * @out_top: Top of the phyiscal address region for which
> + *		the GPT @out_gpt_par_state is valid
> + * @out_gpt_par_state: State of the GPT covered by [base, out_top)
> + */
> +static inline long rmi_gpt_info(unsigned long base, unsigned long end,
> +			       unsigned long *out_top,
> +			       unsigned long *out_gpt_par_state)
> +{
> +	struct arm_smccc_1_2_regs regs = {
> +		SMC_RMI_GPT_INFO, base, end,
> +	};
> +
> +	long ret = rmi_sro_execute(&regs);
> +
> +	if (ret == RMI_SUCCESS) {
> +		if (out_top)
> +			*out_top = regs.a1;
> +		if (out_gpt_par_state)
> +			*out_gpt_par_state = regs.a2;
> +	}
> +
> +	return ret;
> +}
> +

Similarly, rmi_gpt_info() is used for once by rmi.c::rmi_verify_gpt_firmware_managed(). We
needn't expose this function and can combine the logic to rmi.c::rmi_verify_gpt_firmware_managed().

>   /**
>    * rmi_features() - Read feature register
>    * @index: Feature register index

Thanks,
Gavin