Re: [PATCH v5 03/17] riscv: vector: Use vlenb from DT

[Charlie Jenkins <charlie@rivosinc.com>]

On Fri, May 03, 2024 at 05:59:33PM +0100, Conor Dooley wrote:
> On Thu, May 02, 2024 at 09:46:38PM -0700, Charlie Jenkins wrote:
> > If vlenb is provided in the device tree, prefer that over reading the
> > vlenb csr.
> > 
> > Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
> > ---
> >  arch/riscv/include/asm/cpufeature.h |  2 ++
> >  arch/riscv/kernel/cpufeature.c      | 43 +++++++++++++++++++++++++++++++++++++
> >  arch/riscv/kernel/vector.c          | 12 ++++++++++-
> >  3 files changed, 56 insertions(+), 1 deletion(-)
> > 
> > diff --git a/arch/riscv/include/asm/cpufeature.h b/arch/riscv/include/asm/cpufeature.h
> > index 347805446151..0c4f08577015 100644
> > --- a/arch/riscv/include/asm/cpufeature.h
> > +++ b/arch/riscv/include/asm/cpufeature.h
> > @@ -31,6 +31,8 @@ DECLARE_PER_CPU(struct riscv_cpuinfo, riscv_cpuinfo);
> >  /* Per-cpu ISA extensions. */
> >  extern struct riscv_isainfo hart_isa[NR_CPUS];
> >  
> > +extern u32 riscv_vlenb_of;
> > +
> >  void riscv_user_isa_enable(void);
> >  
> >  #if defined(CONFIG_RISCV_MISALIGNED)
> > diff --git a/arch/riscv/kernel/cpufeature.c b/arch/riscv/kernel/cpufeature.c
> > index 3ed2359eae35..12c79db0b0bb 100644
> > --- a/arch/riscv/kernel/cpufeature.c
> > +++ b/arch/riscv/kernel/cpufeature.c
> > @@ -35,6 +35,8 @@ static DECLARE_BITMAP(riscv_isa, RISCV_ISA_EXT_MAX) __read_mostly;
> >  /* Per-cpu ISA extensions. */
> >  struct riscv_isainfo hart_isa[NR_CPUS];
> >  
> > +u32 riscv_vlenb_of;
> > +
> >  /**
> >   * riscv_isa_extension_base() - Get base extension word
> >   *
> > @@ -648,6 +650,42 @@ static int __init riscv_isa_fallback_setup(char *__unused)
> >  early_param("riscv_isa_fallback", riscv_isa_fallback_setup);
> >  #endif
> >  
> > +static int has_riscv_homogeneous_vlenb(void)
> > +{
> > +	int cpu;
> > +	u32 prev_vlenb = 0;
> > +	u32 vlenb;
> > +
> > +	for_each_possible_cpu(cpu) {
> > +		struct device_node *cpu_node;
> > +
> > +		cpu_node = of_cpu_device_node_get(cpu);
> > +		if (!cpu_node) {
> > +			pr_warn("Unable to find cpu node\n");
> > +			return -ENOENT;
> > +		}
> > +
> > +		if (of_property_read_u32(cpu_node, "riscv,vlenb", &vlenb)) {
> > +			of_node_put(cpu_node);
> > +
> > +			if (prev_vlenb)
> > +				return -ENOENT;
> > +			continue;
> > +		}
> > +
> > +		if (prev_vlenb && vlenb != prev_vlenb) {
> > +			of_node_put(cpu_node);
> > +			return -ENOENT;
> > +		}
> > +
> > +		prev_vlenb = vlenb;
> > +		of_node_put(cpu_node);
> > +	}
> > +
> > +	riscv_vlenb_of = vlenb;
> > +	return 0;
> > +}
> > +
> >  void __init riscv_fill_hwcap(void)
> >  {
> >  	char print_str[NUM_ALPHA_EXTS + 1];
> > @@ -671,6 +709,11 @@ void __init riscv_fill_hwcap(void)
> >  			pr_info("Falling back to deprecated \"riscv,isa\"\n");
> >  			riscv_fill_hwcap_from_isa_string(isa2hwcap);
> >  		}
> > +
> > +		if (elf_hwcap & COMPAT_HWCAP_ISA_V && has_riscv_homogeneous_vlenb() < 0) {
> 
> I still think this isn't quite right, as it will emit a warning when
> RISCV_ISA_V is disabled. The simplest thing to do probably is just
> add an `if (IS_ENABLED(CONFIG_RISCV_ISA_V) return 0` shortcut the to
> function? It'll get disabled a few lines later so I think a zero is
> safe.

That seems like a good idea. It is weird to throw a warning about this
even when they have V disabled in the kernel. The DT is improperly
formatted since it has heterogeneous vlenb entries and has V enabled,
but since the user disabled V in the kernel skipping the warning is
reasonable.

- Charlie

> 
> > +			pr_warn("Unsupported heterogeneous vlen detected, vector extension disabled.\n");
> > +			elf_hwcap &= ~COMPAT_HWCAP_ISA_V;
> > +		}
> >  	}
> >  
> >  	/*
> > diff --git a/arch/riscv/kernel/vector.c b/arch/riscv/kernel/vector.c
> > index 6727d1d3b8f2..e04586cdb7f0 100644
> > --- a/arch/riscv/kernel/vector.c
> > +++ b/arch/riscv/kernel/vector.c
> > @@ -33,7 +33,17 @@ int riscv_v_setup_vsize(void)
> >  {
> >  	unsigned long this_vsize;
> >  
> > -	/* There are 32 vector registers with vlenb length. */
> > +	/*
> > +	 * There are 32 vector registers with vlenb length.
> > +	 *
> > +	 * If the riscv,vlenb property was provided by the firmware, use that
> > +	 * instead of probing the CSRs.
> > +	 */
> > +	if (riscv_vlenb_of) {
> > +		this_vsize = riscv_vlenb_of * 32;
> > +		return 0;
> > +	}
> > +
> >  	riscv_v_enable();
> >  	this_vsize = csr_read(CSR_VLENB) * 32;
> >  	riscv_v_disable();
> > 
> > -- 
> > 2.44.0
> >