Date:   Fri, 3 Mar 2023 23:28:21 +0530
From:   Sunil V L <sunilvl@ventanamicro.com>
To:     Andrew Jones <ajones@ventanamicro.com>
Cc:     linux-riscv@lists.infradead.org, linux-acpi@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-doc@vger.kernel.org,
        Palmer Dabbelt <palmer@dabbelt.com>,
        Albert Ou <aou@eecs.berkeley.edu>,
        "Rafael J . Wysocki" <rafael@kernel.org>,
        Len Brown <lenb@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Marc Zyngier <maz@kernel.org>,
        Daniel Lezcano <daniel.lezcano@linaro.org>,
        Jonathan Corbet <corbet@lwn.net>,
        Anup Patel <apatel@ventanamicro.com>,
        Atish Patra <atishp@rivosinc.com>,
        'Conor Dooley ' <conor.dooley@microchip.com>,
        "Rafael J . Wysocki" <rafael.j.wysocki@intel.com>
Subject: Re: [PATCH V3 11/20] RISC-V: ACPI: Cache and retrieve the RINTC
 structure
Message-ID: <ZAI1PbEfo1Gyco1n@sunil-laptop>
References: <20230303133647.845095-1-sunilvl@ventanamicro.com>
 <20230303133647.845095-12-sunilvl@ventanamicro.com>
 <20230303160556.lezccmwa7ymymxws@orel>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230303160556.lezccmwa7ymymxws@orel>
Precedence: bulk

On Fri, Mar 03, 2023 at 05:05:56PM +0100, Andrew Jones wrote:
> On Fri, Mar 03, 2023 at 07:06:38PM +0530, Sunil V L wrote:
> > RINTC structures in the MADT provide mapping between the hartid
> > and the CPU. This is required many times even at run time like
> > cpuinfo. So, instead of parsing the ACPI table every time, cache
> > the RINTC structures and provide a function to get the correct
> > RINTC structure for a given cpu.
> > 
> > Signed-off-by: Sunil V L <sunilvl@ventanamicro.com>
> > Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> > ---
> >  arch/riscv/include/asm/acpi.h |  9 ++++++
> >  arch/riscv/kernel/acpi.c      | 56 +++++++++++++++++++++++++++++++++++
> >  2 files changed, 65 insertions(+)
> > 
> > diff --git a/arch/riscv/include/asm/acpi.h b/arch/riscv/include/asm/acpi.h
> > index 111a8ed10af1..8be16c1ef7da 100644
> > --- a/arch/riscv/include/asm/acpi.h
> > +++ b/arch/riscv/include/asm/acpi.h
> > @@ -61,6 +61,10 @@ static inline void arch_fix_phys_package_id(int num, u32 slot) { }
> >  
> >  int acpi_get_riscv_isa(struct acpi_table_header *table,
> >  		       unsigned int cpu, const char **isa);
> > +
> > +struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu);
> > +
> > +u32 get_acpi_id_for_cpu(int cpu);
> >  #else
> >  static inline int acpi_get_riscv_isa(struct acpi_table_header *table,
> >  				     unsigned int cpu, const char **isa)
> > @@ -68,6 +72,11 @@ static inline int acpi_get_riscv_isa(struct acpi_table_header *table,
> >  	return -EINVAL;
> >  }
> >  
> > +static inline u32 get_acpi_id_for_cpu(int cpu)
> > +{
> > +	return -1;
> > +}
> 
> Why do we need this stub? I wouldn't expect non-ACPI code to need an ACPI
> ID.
> 
> > +
> >  #endif /* CONFIG_ACPI */
> >  
> >  #endif /*_ASM_ACPI_H*/
> > diff --git a/arch/riscv/kernel/acpi.c b/arch/riscv/kernel/acpi.c
> > index 81d448c41714..8b3d68d8225f 100644
> > --- a/arch/riscv/kernel/acpi.c
> > +++ b/arch/riscv/kernel/acpi.c
> > @@ -24,6 +24,62 @@ EXPORT_SYMBOL(acpi_disabled);
> >  int acpi_pci_disabled = 1;	/* skip ACPI PCI scan and IRQ initialization */
> >  EXPORT_SYMBOL(acpi_pci_disabled);
> >  
> > +static struct acpi_madt_rintc cpu_madt_rintc[NR_CPUS];
> > +
> > +static int acpi_parse_madt_rintc(union acpi_subtable_headers *header, const unsigned long end)
> > +{
> > +	struct acpi_madt_rintc *rintc = (struct acpi_madt_rintc *)header;
> > +	int cpuid;
> > +
> > +	if (!(rintc->flags & ACPI_MADT_ENABLED))
> > +		return 0;
> 
> Why not cache the data even when its disabled? We also cache the flags so
> we can always check later too.
> 
Okay, doesn't harm.

> > +
> > +	cpuid = riscv_hartid_to_cpuid(rintc->hart_id);
> > +	if (cpuid >= 0 && cpuid < NR_CPUS)
> 
> What does it mean for the above check to fail? Bad ACPI tables?
> 
This can happen when SMP is disabled but platform has more CPUs.

> > +		cpu_madt_rintc[cpuid] = *rintc;
> > +
> > +	return 0;
> > +}
> > +
> > +static int acpi_init_rintc_array(void)
> > +{
> > +	if (acpi_table_parse_madt(ACPI_MADT_TYPE_RINTC, acpi_parse_madt_rintc, 0) > 0)
> > +		return 0;
> > +
> > +	return -ENODEV;
> > +}
> > +
> > +/*
> > + * Instead of parsing (and freeing) the ACPI table, cache
> > + * the RINTC structures since they are frequently used
> > + * like in  cpuinfo.
> > + */
> > +struct acpi_madt_rintc *acpi_cpu_get_madt_rintc(int cpu)
> > +{
> > +	static bool rintc_init_done;
> > +
> > +	if (!rintc_init_done) {
> > +		if (acpi_init_rintc_array()) {
> > +			pr_err("No valid RINTC entries exist\n");
> > +			return NULL;
> > +		}
> > +
> > +		rintc_init_done = true;
> > +	}
> > +
> > +	return &cpu_madt_rintc[cpu];
> > +}
> > +
> > +u32 get_acpi_id_for_cpu(int cpu)
> > +{
> > +	struct acpi_madt_rintc *rintc = acpi_cpu_get_madt_rintc(cpu);
> > +
> > +	if (!rintc)
> > +		return -1;
> 
> Is -1 defined as an invalid ACPI ID? I'm wondering if we shouldn't just
> BUG in acpi_init_rintc_array() if we fail to initialize and then we can
> unconditionally return rintc->uid here.
>
Thanks!. Will update this.