Received-SPF: pass (google.com: domain of linux-kernel+bounces-134141-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Precedence: bulk
MIME-Version: 1.0
References: <cover.1709780590.git.haibo1.xu@intel.com> <7ca110c59cbb2fb358304a9ba4f9c7cbeb191345.1709780590.git.haibo1.xu@intel.com>
 <93e24c6c-65e6-4e99-ac56-aa264dbbf525@ghiti.fr>
In-Reply-To: <93e24c6c-65e6-4e99-ac56-aa264dbbf525@ghiti.fr>
From: Haibo Xu <xiaobo55x@gmail.com>
Date: Sun, 7 Apr 2024 11:09:19 +0800
Message-ID: <CAJve8ok=p5Dv_yc8Ni2eSFZjSXtZ80LVf8whaEJDZR7DjwVRPA@mail.gmail.com>
Subject: Re: [PATCH v2 2/6] ACPI: RISCV: Add NUMA support based on SRAT and SLIT
To: Alexandre Ghiti <alex@ghiti.fr>
Cc: Haibo Xu <haibo1.xu@intel.com>, Alison Schofield <alison.schofield@intel.com>, 
	Zong Li <zong.li@sifive.com>, "Rafael J. Wysocki" <rafael@kernel.org>, 
	Dave Hansen <dave.hansen@linux.intel.com>, linux-kernel@vger.kernel.org, 
	Conor Dooley <conor.dooley@microchip.com>, Guo Ren <guoren@kernel.org>, 
	Jisheng Zhang <jszhang@kernel.org>, James Morse <james.morse@arm.com>, 
	linux-riscv@lists.infradead.org, Ard Biesheuvel <ardb@kernel.org>, 
	Baoquan He <bhe@redhat.com>, Marc Zyngier <maz@kernel.org>, acpica-devel@lists.linux.dev, 
	Robert Moore <robert.moore@intel.com>, linux-acpi@vger.kernel.org, 
	Sami Tolvanen <samitolvanen@google.com>, Greentime Hu <greentime.hu@sifive.com>, 
	Len Brown <lenb@kernel.org>, Albert Ou <aou@eecs.berkeley.edu>, 
	Alexandre Ghiti <alexghiti@rivosinc.com>, Arnd Bergmann <arnd@arndb.de>, 
	Dan Williams <dan.j.williams@intel.com>, Chen Jiahao <chenjiahao16@huawei.com>, 
	Yuntao Wang <ytcoode@gmail.com>, Paul Walmsley <paul.walmsley@sifive.com>, 
	=?UTF-8?B?Q2zDqW1lbnQgTO+/vT1BIDlnZXI=?= <cleger@rivosinc.com>, 
	Anup Patel <apatel@ventanamicro.com>, Tony Luck <tony.luck@intel.com>, 
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>, 
	"Russell King (Oracle)" <rmk+kernel@armlinux.org.uk>, Samuel Holland <samuel.holland@sifive.com>, 
	Evan Green <evan@rivosinc.com>, Palmer Dabbelt <palmer@dabbelt.com>, ajones@ventanamicro.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Tue, Apr 2, 2024 at 5:31=E2=80=AFPM Alexandre Ghiti <alex@ghiti.fr> wrot=
e:
>
> Hi Haibo,
>
> On 07/03/2024 09:47, 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 | 131 +++++++++++++++++++++++++++++++++=
+
> >   arch/riscv/kernel/setup.c     |   4 +-
> >   arch/riscv/kernel/smpboot.c   |   2 -
> >   include/linux/acpi.h          |   6 ++
> >   7 files changed, 154 insertions(+), 10 deletions(-)
> >   create mode 100644 arch/riscv/kernel/acpi_numa.c
> >
> > diff --git a/arch/riscv/include/asm/acpi.h b/arch/riscv/include/asm/acp=
i.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 *cb=
om_size,
> >                            u32 *cboz_size, u32 *cbop_size);
> >   #else
> > @@ -87,4 +90,12 @@ static inline void acpi_get_cbo_block_size(struct ac=
pi_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-$(CONFIG_COMPAT)              +=3D compat_vdso/
> >
> >   obj-$(CONFIG_64BIT)         +=3D pi/
> >   obj-$(CONFIG_ACPI)          +=3D acpi.o
> > +obj-$(CONFIG_ACPI_NUMA)      +=3D 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(in=
t 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_i=
oremap()
> >    * or early_memremap() should be called here to for ACPI table mappin=
g.
> > diff --git a/arch/riscv/kernel/acpi_numa.c b/arch/riscv/kernel/acpi_num=
a.c
> > new file mode 100644
> > index 000000000000..0231482d6946
> > --- /dev/null
> > +++ b/arch/riscv/kernel/acpi_numa.c
> > @@ -0,0 +1,131 @@
> > +// 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 whic=
h 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 =3D { 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 =3D 0; cpu < nr_cpu_ids; cpu++)
> > +             if (uid =3D=3D get_acpi_id_for_cpu(cpu))
> > +                     return cpu;
> > +
> > +     return -EINVAL;
> > +}
> > +
> > +static int __init acpi_parse_rintc_pxm(union acpi_subtable_headers *he=
ader,
> > +                                    const unsigned long end)
> > +{
> > +     struct acpi_srat_rintc_affinity *pa;
> > +     int cpu, pxm, node;
> > +
> > +     if (srat_disabled())
> > +             return -EINVAL;
> > +
> > +     pa =3D (struct acpi_srat_rintc_affinity *)header;
> > +     if (!pa)
> > +             return -EINVAL;
> > +
> > +     if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
> > +             return 0;
> > +
> > +     pxm =3D pa->proximity_domain;
> > +     node =3D 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 =3D get_cpu_for_acpi_id(pa->acpi_processor_uid);
> > +     if (cpu < 0)
> > +             return 0;
> > +
> > +     acpi_early_node_map[cpu] =3D 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 throug=
h
> > +      * 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_pars=
e_rintc_pxm, 0);
> > +
> > +     for (i =3D 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_affin=
ity *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->head=
er.length);
> > +             bad_srat();
> > +             return;
> > +     }
> > +
> > +     if (!(pa->flags & ACPI_SRAT_RINTC_ENABLED))
> > +             return;
> > +
> > +     pxm =3D pa->proximity_domain;
> > +     node =3D acpi_map_pxm_to_node(pxm);
> > +
> > +     if (node =3D=3D NUMA_NO_NODE) {
> > +             pr_err("SRAT: Too many proximity domains %d\n", pxm);
> > +             bad_srat();
> > +             return;
> > +     }
> > +
> > +     node_set(node, numa_nodes_parsed);
> > +}
>
>
> What is riscv specific in the parsing of those tables? Can't we try to
> merge this into generic ACPI code? I know that's a burden to try and
> factorize code with other architectures instead of reusing it, but it
> showed numerous times that duplicating was even worse (I have the NAPOT
> code in mind).
>

Hi Alex,

The acpi_srat_rintc_affinity structure is RISC-V specific and that's
why I put the
callback in arch/riscv/kernel/acpi_numa.c.

Thanks,
Haibo


> Thanks,
>
> Alex
>
>
> > 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();
> > +     }
> >
> >       riscv_init_cbo_blocksizes();
> >       riscv_fill_hwcap();
> > diff --git a/arch/riscv/kernel/smpboot.c b/arch/riscv/kernel/smpboot.c
> > index cfbe4b840d42..81a2aa77680c 100644
> > --- a/arch/riscv/kernel/smpboot.c
> > +++ b/arch/riscv/kernel/smpboot.c
> > @@ -100,7 +100,6 @@ static int __init acpi_parse_rintc(union acpi_subta=
ble_headers *header, const un
> >       if (hart =3D=3D cpuid_to_hartid_map(0)) {
> >               BUG_ON(found_boot_cpu);
> >               found_boot_cpu =3D true;
> > -             early_map_cpu_to_node(0, acpi_numa_get_nid(cpu_count));
> >               return 0;
> >       }
> >
> > @@ -110,7 +109,6 @@ static int __init acpi_parse_rintc(union acpi_subta=
ble_headers *header, const un
> >       }
> >
> >       cpuid_to_hartid_map(cpu_count) =3D hart;
> > -     early_map_cpu_to_node(cpu_count, acpi_numa_get_nid(cpu_count));
> >       cpu_count++;
> >
> >       return 0;
> > diff --git a/include/linux/acpi.h b/include/linux/acpi.h
> > index b7165e52b3c6..f74c62956e07 100644
> > --- a/include/linux/acpi.h
> > +++ b/include/linux/acpi.h
> > @@ -269,6 +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_rint=
c_affinity *pa) { }
> > +#endif
> > +
> >   #ifndef PHYS_CPUID_INVALID
> >   typedef u32 phys_cpuid_t;
> >   #define PHYS_CPUID_INVALID (phys_cpuid_t)(-1)