Re: [PATCH 3/4] ACPI: RISCV: Add NUMA support based on SRAT and SLIT

[Haibo Xu <xiaobo55x@gmail.com>]

On Tue, Mar 5, 2024 at 1:24 PM Sunil V L <sunilvl@ventanamicro.com> wrote:
>
> On Wed, Jan 31, 2024 at 10:32:00AM +0800, Haibo Xu wrote:
> > Add acpi_numa.c file to enable parse NUMA information from
> > ACPI SRAT and SLIT tables. SRAT table provide CPUs(Hart) and
> > memory nodes to proximity domain mapping, while SLIT table
> > provide the distance metrics between proximity domains.
> >
> > Signed-off-by: Haibo Xu <haibo1.xu@intel.com>
> > ---
> >  arch/riscv/include/asm/acpi.h |  15 +++-
> >  arch/riscv/kernel/Makefile    |   1 +
> >  arch/riscv/kernel/acpi.c      |   5 --
> >  arch/riscv/kernel/acpi_numa.c | 133 ++++++++++++++++++++++++++++++++++
> >  arch/riscv/kernel/setup.c     |   4 +-
> >  arch/riscv/kernel/smpboot.c   |   2 -
> >  drivers/acpi/numa/srat.c      |   3 +-
> >  include/linux/acpi.h          |   4 +
> >  8 files changed, 156 insertions(+), 11 deletions(-)
> >  create mode 100644 arch/riscv/kernel/acpi_numa.c
> >
> > diff --git a/arch/riscv/include/asm/acpi.h b/arch/riscv/include/asm/acpi.h
> > index 7dad0cf9d701..e0a1f84404f3 100644
> > --- a/arch/riscv/include/asm/acpi.h
> > +++ b/arch/riscv/include/asm/acpi.h
> > @@ -61,11 +61,14 @@ static inline void arch_fix_phys_package_id(int num, u32 slot) { }
> >
> >  void acpi_init_rintc_map(void);
> >  struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu);
> > -u32 get_acpi_id_for_cpu(int cpu);
> > +static inline u32 get_acpi_id_for_cpu(int cpu)
> > +{
> > +     return acpi_cpu_get_madt_rintc(cpu)->uid;
> > +}
> > +
> >  int acpi_get_riscv_isa(struct acpi_table_header *table,
> >                      unsigned int cpu, const char **isa);
> >
> > -static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
> >  void acpi_get_cbo_block_size(struct acpi_table_header *table, u32 *cbom_size,
> >                            u32 *cboz_size, u32 *cbop_size);
> >  #else
> > @@ -87,4 +90,12 @@ static inline void acpi_get_cbo_block_size(struct acpi_table_header *table,
> >
> >  #endif /* CONFIG_ACPI */
> >
> > +#ifdef CONFIG_ACPI_NUMA
> > +int acpi_numa_get_nid(unsigned int cpu);
> > +void acpi_map_cpus_to_nodes(void);
> > +#else
> > +static inline int acpi_numa_get_nid(unsigned int cpu) { return NUMA_NO_NODE; }
> > +static inline void acpi_map_cpus_to_nodes(void) { }
> > +#endif /* CONFIG_ACPI_NUMA */
> > +
> >  #endif /*_ASM_ACPI_H*/
> > diff --git a/arch/riscv/kernel/Makefile b/arch/riscv/kernel/Makefile
> > index f71910718053..5d3e9cf89b76 100644
> > --- a/arch/riscv/kernel/Makefile
> > +++ b/arch/riscv/kernel/Makefile
> > @@ -105,3 +105,4 @@ obj-$(CONFIG_COMPAT)              += compat_vdso/
> >
> >  obj-$(CONFIG_64BIT)          += pi/
> >  obj-$(CONFIG_ACPI)           += acpi.o
> > +obj-$(CONFIG_ACPI_NUMA)      += acpi_numa.o
> > diff --git a/arch/riscv/kernel/acpi.c b/arch/riscv/kernel/acpi.c
> > index e619edc8b0cc..040bdbfea2b4 100644
> > --- a/arch/riscv/kernel/acpi.c
> > +++ b/arch/riscv/kernel/acpi.c
> > @@ -191,11 +191,6 @@ struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu)
> >       return &cpu_madt_rintc[cpu];
> >  }
> >
> > -u32 get_acpi_id_for_cpu(int cpu)
> > -{
> > -     return acpi_cpu_get_madt_rintc(cpu)->uid;
> > -}
> > -
> >  /*
> >   * __acpi_map_table() will be called before paging_init(), so early_ioremap()
> >   * or early_memremap() should be called here to for ACPI table mapping.
> > diff --git a/arch/riscv/kernel/acpi_numa.c b/arch/riscv/kernel/acpi_numa.c
> > new file mode 100644
> > index 000000000000..493642a61457
> > --- /dev/null
> > +++ b/arch/riscv/kernel/acpi_numa.c
> > @@ -0,0 +1,133 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * ACPI 6.6 based NUMA setup for RISCV
> > + * Lots of code was borrowed from arch/arm64/kernel/acpi_numa.c
> > + *
> > + * Copyright 2004 Andi Kleen, SuSE Labs.
> > + * Copyright (C) 2013-2016, Linaro Ltd.
> > + *           Author: Hanjun Guo <hanjun.guo@linaro.org>
> > + * Copyright (C) 2024 Intel Corporation.
> > + *
> > + * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs.
> > + *
> > + * Called from acpi_numa_init while reading the SRAT and SLIT tables.
> > + * Assumes all memory regions belonging to a single proximity domain
> > + * are in one chunk. Holes between them will be included in the node.
> > + */
> > +
> > +#define pr_fmt(fmt) "ACPI: NUMA: " fmt
> > +
> > +#include <linux/acpi.h>
> > +#include <linux/bitmap.h>
> > +#include <linux/kernel.h>
> > +#include <linux/mm.h>
> > +#include <linux/memblock.h>
> > +#include <linux/mmzone.h>
> > +#include <linux/module.h>
> > +#include <linux/topology.h>
> > +
> > +#include <asm/numa.h>
> > +
> > +static int acpi_early_node_map[NR_CPUS] __initdata = { NUMA_NO_NODE };
> > +
> > +int __init acpi_numa_get_nid(unsigned int cpu)
> > +{
> > +     return acpi_early_node_map[cpu];
> > +}
> > +
> > +static inline int get_cpu_for_acpi_id(u32 uid)
> > +{
> > +     int cpu;
> > +
> > +     for (cpu = 0; cpu < nr_cpu_ids; cpu++)
> > +             if (uid == get_acpi_id_for_cpu(cpu))
> > +                     return cpu;
> > +
> > +     return -EINVAL;
> > +}
> > +
> > +static int __init acpi_parse_rintc_pxm(union acpi_subtable_headers *header,
> > +                                   const unsigned long end)
>
> Please check alignment.
>

Sure.

> > +{
> > +     struct acpi_srat_rintc_affinity *pa;
> > +     int cpu, pxm, node;
> > +
> > +     if (srat_disabled())
> > +             return -EINVAL;
> > +
> > +     pa = (struct acpi_srat_rintc_affinity *)header;
> > +     if (!pa)
> > +             return -EINVAL;
> > +
> > +     if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
> > +             return 0;
> > +
> > +     pxm = pa->proximity_domain;
> > +     node = pxm_to_node(pxm);
> > +
> > +     /*
> > +      * If we can't map the UID to a logical cpu this
> > +      * means that the UID is not part of possible cpus
> > +      * so we do not need a NUMA mapping for it, skip
> > +      * the SRAT entry and keep parsing.
> > +      */
> > +     cpu = get_cpu_for_acpi_id(pa->acpi_processor_uid);
> > +     if (cpu < 0)
> > +             return 0;
> > +
> > +     acpi_early_node_map[cpu] = node;
> > +     pr_info("SRAT: PXM %d -> HARTID 0x%lx -> Node %d\n", pxm,
> > +             cpuid_to_hartid_map(cpu), node);
> > +
> > +     return 0;
> > +}
> > +
> > +void __init acpi_map_cpus_to_nodes(void)
> > +{
> > +     int i;
> > +
> > +     /*
> > +      * In ACPI, SMP and CPU NUMA information is provided in separate
> > +      * static tables, namely the MADT and the SRAT.
> > +      *
> > +      * Thus, it is simpler to first create the cpu logical map through
> > +      * an MADT walk and then map the logical cpus to their node ids
> > +      * as separate steps.
> > +      */
> > +     acpi_table_parse_entries(ACPI_SIG_SRAT, sizeof(struct acpi_table_srat),
> > +                                         ACPI_SRAT_TYPE_RINTC_AFFINITY,
> > +                                         acpi_parse_rintc_pxm, 0);
> > +
> Alignment here as well.
>

Sure.

> > +     for (i = 0; i < nr_cpu_ids; i++)
> > +             early_map_cpu_to_node(i, acpi_numa_get_nid(i));
> > +}
> > +
> > +/* Callback for Proximity Domain -> logical node ID mapping */
> > +void __init acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa)
> > +{
> > +     int pxm, node;
> > +
> > +     if (srat_disabled())
> > +             return;
> > +
> > +     if (pa->header.length < sizeof(struct acpi_srat_rintc_affinity)) {
> > +             pr_err("SRAT: Invalid SRAT header length: %d\n",
> > +                     pa->header.length);
> Can we merge these into single line?
>
> > +             bad_srat();
> > +             return;
> > +     }
> > +
> > +     if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
> > +             return;
> > +
> > +     pxm = pa->proximity_domain;
> > +     node = acpi_map_pxm_to_node(pxm);
> > +
> > +     if (node == NUMA_NO_NODE) {
> > +             pr_err("SRAT: Too many proximity domains %d\n", pxm);
> > +             bad_srat();
> > +             return;
> > +     }
> > +
> > +     node_set(node, numa_nodes_parsed);
> > +}
> > diff --git a/arch/riscv/kernel/setup.c b/arch/riscv/kernel/setup.c
> > index 4f73c0ae44b2..a2cde65b69e9 100644
> > --- a/arch/riscv/kernel/setup.c
> > +++ b/arch/riscv/kernel/setup.c
> > @@ -281,8 +281,10 @@ void __init setup_arch(char **cmdline_p)
> >       setup_smp();
> >  #endif
> >
> > -     if (!acpi_disabled)
> > +     if (!acpi_disabled) {
> >               acpi_init_rintc_map();
> > +             acpi_map_cpus_to_nodes();
> Is it not possible to fill up both in single parsing of MADT?
>

I think it's not possible to fill both in a single MADT parse since
the NUMA info is provided in a separate SRAT table.

For RISC-V, currently we parsed 2 times of the MADT.
FIrst one in setup_smp()->acpi_parse_and_init_cpus() call to build up
cpuid_to_hartid_map.
Second one in acpi_init_rintc_map call to build the cpu_madt_rintc[]
cache structure.

Since the first one depends on the CONFIG_SMP, I am not sure whether
it's possible to combine
these two parts into one.

> > +     }
> >
> >       riscv_init_cbo_blocksizes();
> >       riscv_fill_hwcap();
> > diff --git a/arch/riscv/kernel/smpboot.c b/arch/riscv/kernel/smpboot.c
> > index 519b6bd946e5..b188d83d1ec4 100644
> > --- a/arch/riscv/kernel/smpboot.c
> > +++ b/arch/riscv/kernel/smpboot.c
> > @@ -101,7 +101,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un
> >       if (hart == cpuid_to_hartid_map(0)) {
> >               BUG_ON(found_boot_cpu);
> >               found_boot_cpu = true;
> > -             early_map_cpu_to_node(0, acpi_numa_get_nid(cpu_count));
> >               return 0;
> >       }
> >
> > @@ -111,7 +110,6 @@ static int __init acpi_parse_rintc(union acpi_subtable_headers *header, const un
> >       }
> >
> >       cpuid_to_hartid_map(cpu_count) = hart;
> > -     early_map_cpu_to_node(cpu_count, acpi_numa_get_nid(cpu_count));
> >       cpu_count++;
> >
> >       return 0;
> > diff --git a/drivers/acpi/numa/srat.c b/drivers/acpi/numa/srat.c
> > index 503abcf6125d..1f0462cef47c 100644
> > --- a/drivers/acpi/numa/srat.c
> > +++ b/drivers/acpi/numa/srat.c
> > @@ -219,7 +219,8 @@ int __init srat_disabled(void)
> >       return acpi_numa < 0;
> >  }
> >
> > -#if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH)
> > +#if defined(CONFIG_X86) || defined(CONFIG_ARM64) || defined(CONFIG_LOONGARCH) || \
> > +     defined(CONFIG_RISCV)
> Please check alignment. Or make it single line if fits in 100 chars.
> Also, it looks it covers most of the architectures now. Is it possible
> to simplify / remove the condition? I hope IA64 is removed now?
>

Good catch!

Since IA64 support was removed in commit
cf8e8658100d4(https://lwn.net/Articles/923376/).
I think it's possible to remove the condition. Will fix it in v2.

Thanks,
Haibo

> May be you need to update the comment at #endif too.
>
> Thanks
> Sunil
>
> >  /*
> >   * Callback for SLIT parsing.  pxm_to_node() returns NUMA_NO_NODE for
> >   * I/O localities since SRAT does not list them.  I/O localities are
> > diff --git a/include/linux/acpi.h b/include/linux/acpi.h
> > index a65273db55c6..be78a9d28927 100644
> > --- a/include/linux/acpi.h
> > +++ b/include/linux/acpi.h
> > @@ -269,8 +269,12 @@ acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }
> >
> >  int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);
> >
> > +#ifdef CONFIG_RISCV
> > +void acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa);
> > +#else
> >  static inline void
> >  acpi_numa_rintc_affinity_init(struct acpi_srat_rintc_affinity *pa) { }
> > +#endif
> >
> >  #ifndef PHYS_CPUID_INVALID
> >  typedef u32 phys_cpuid_t;
> > --
> > 2.34.1
> >

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv