Message-ID: <mhng-24b22694-82f4-467b-b6a9-0fb2e186d3f2@palmerdabbelt-glaptop>
Date:   Thu, 16 Jan 2020 11:51:41 -0800 (PST)
From:   Palmer Dabbelt <palmerdabbelt@...gle.com>
To:     Anup Patel <Anup.Patel@....com>
CC:     Paul Walmsley <paul.walmsley@...ive.com>, aou@...s.berkeley.edu,
        pbonzini@...hat.com, rkrcmar@...hat.com, graf@...zon.com,
        Atish Patra <Atish.Patra@....com>,
        Alistair Francis <Alistair.Francis@....com>,
        Damien Le Moal <Damien.LeMoal@....com>,
        Christoph Hellwig <hch@....de>, anup@...infault.org,
        kvm@...r.kernel.org, kvm-riscv@...ts.infradead.org,
        linux-riscv@...ts.infradead.org, linux-kernel@...r.kernel.org,
        Anup Patel <Anup.Patel@....com>
Subject:     Re: [PATCH v10 02/19] RISC-V: Add bitmap reprensenting ISA features common across CPUs

On Mon, 23 Dec 2019 03:35:26 PST (-0800), Anup Patel wrote:
> This patch adds riscv_isa bitmap which represents Host ISA features
> common across all Host CPUs. The riscv_isa is not same as elf_hwcap
> because elf_hwcap will only have ISA features relevant for user-space
> apps whereas riscv_isa will have ISA features relevant to both kernel
> and user-space apps.
>
> One of the use-case for riscv_isa bitmap is in KVM hypervisor where
> we will use it to do following operations:
>
> 1. Check whether hypervisor extension is available
> 2. Find ISA features that need to be virtualized (e.g. floating
>    point support, vector extension, etc.)
>
> Signed-off-by: Anup Patel <anup.patel@....com>
> Signed-off-by: Atish Patra <atish.patra@....com>
> Reviewed-by: Alexander Graf <graf@...zon.com>
> ---
>  arch/riscv/include/asm/hwcap.h | 22 +++++++++
>  arch/riscv/kernel/cpufeature.c | 83 ++++++++++++++++++++++++++++++++--
>  2 files changed, 102 insertions(+), 3 deletions(-)
>
> diff --git a/arch/riscv/include/asm/hwcap.h b/arch/riscv/include/asm/hwcap.h
> index 1bb0cd04aec3..5589c012e004 100644
> --- a/arch/riscv/include/asm/hwcap.h
> +++ b/arch/riscv/include/asm/hwcap.h
> @@ -8,6 +8,7 @@
>  #ifndef _ASM_RISCV_HWCAP_H
>  #define _ASM_RISCV_HWCAP_H
>  
> +#include <linux/bits.h>
>  #include <uapi/asm/hwcap.h>
>  
>  #ifndef __ASSEMBLY__
> @@ -22,6 +23,27 @@ enum {
>  };
>  
>  extern unsigned long elf_hwcap;
> +
> +#define RISCV_ISA_EXT_a		('a' - 'a')
> +#define RISCV_ISA_EXT_c		('c' - 'a')
> +#define RISCV_ISA_EXT_d		('d' - 'a')
> +#define RISCV_ISA_EXT_f		('f' - 'a')
> +#define RISCV_ISA_EXT_h		('h' - 'a')
> +#define RISCV_ISA_EXT_i		('i' - 'a')
> +#define RISCV_ISA_EXT_m		('m' - 'a')
> +#define RISCV_ISA_EXT_s		('s' - 'a')
> +#define RISCV_ISA_EXT_u		('u' - 'a')

Unfortunately the ISA doesn't really work this way any more: the single-letter
extensions are just aliases for longer extension strings, each of which
represents a single instruction.  I know we're saddled with some ABI that looks
this way, but I really don't want to add new code that depends on these defunct
assumptions -- there isn't that much in Linux right now, but there's a lot in
the FSF toolchain and getting that all out is going to be a long project.

> +
> +#define RISCV_ISA_EXT_MAX	256

Why so big?  It looks like the rest of the code just touches the first word,
and most of that is explicit.

> +
> +unsigned long riscv_isa_extension_base(const unsigned long *isa_bitmap);
> +
> +#define riscv_isa_extension_mask(ext) BIT_MASK(RISCV_ISA_EXT_##ext)
> +
> +bool __riscv_isa_extension_available(const unsigned long *isa_bitmap, int bit);
> +#define riscv_isa_extension_available(isa_bitmap, ext)	\
> +	__riscv_isa_extension_available(isa_bitmap, RISCV_ISA_EXT_##ext)
> +
>  #endif
>  
>  #endif /* _ASM_RISCV_HWCAP_H */
> diff --git a/arch/riscv/kernel/cpufeature.c b/arch/riscv/kernel/cpufeature.c
> index 0b40705567b7..e172a2322b34 100644
> --- a/arch/riscv/kernel/cpufeature.c
> +++ b/arch/riscv/kernel/cpufeature.c
> @@ -6,6 +6,7 @@
>   * Copyright (C) 2017 SiFive
>   */
>  
> +#include <linux/bitmap.h>
>  #include <linux/of.h>
>  #include <asm/processor.h>
>  #include <asm/hwcap.h>
> @@ -13,15 +14,57 @@
>  #include <asm/switch_to.h>
>  
>  unsigned long elf_hwcap __read_mostly;
> +
> +/* Host ISA bitmap */
> +static DECLARE_BITMAP(riscv_isa, RISCV_ISA_EXT_MAX) __read_mostly;
> +
>  #ifdef CONFIG_FPU
>  bool has_fpu __read_mostly;
>  #endif
>  
> +/**
> + * riscv_isa_extension_base() - Get base extension word
> + *
> + * @isa_bitmap: ISA bitmap to use
> + * Return: base extension word as unsigned long value
> + *
> + * NOTE: If isa_bitmap is NULL then Host ISA bitmap will be used.
> + */
> +unsigned long riscv_isa_extension_base(const unsigned long *isa_bitmap)
> +{
> +	if (!isa_bitmap)
> +		return riscv_isa[0];
> +	return isa_bitmap[0];
> +}
> +EXPORT_SYMBOL_GPL(riscv_isa_extension_base);

This isn't used, which makes it hard to review.  Can you please split out the
changes that don't depend on the V extension to come out of draft?  That would
make it easier to take some of the code early, which lets us keep around less
diff.

> +
> +/**
> + * __riscv_isa_extension_available() - Check whether given extension
> + * is available or not
> + *
> + * @isa_bitmap: ISA bitmap to use
> + * @bit: bit position of the desired extension
> + * Return: true or false
> + *
> + * NOTE: If isa_bitmap is NULL then Host ISA bitmap will be used.
> + */
> +bool __riscv_isa_extension_available(const unsigned long *isa_bitmap, int bit)
> +{
> +	const unsigned long *bmap = (isa_bitmap) ? isa_bitmap : riscv_isa;
> +
> +	if (bit >= RISCV_ISA_EXT_MAX)
> +		return false;
> +
> +	return test_bit(bit, bmap) ? true : false;
> +}
> +EXPORT_SYMBOL_GPL(__riscv_isa_extension_available);
> +
>  void riscv_fill_hwcap(void)
>  {
>  	struct device_node *node;
>  	const char *isa;
> -	size_t i;
> +	char print_str[BITS_PER_LONG + 1];
> +	size_t i, j, isa_len;
>  	static unsigned long isa2hwcap[256] = {0};
>  
>  	isa2hwcap['i'] = isa2hwcap['I'] = COMPAT_HWCAP_ISA_I;
> @@ -33,8 +76,11 @@ void riscv_fill_hwcap(void)
>  
>  	elf_hwcap = 0;
>  
> +	bitmap_zero(riscv_isa, RISCV_ISA_EXT_MAX);
> +
>  	for_each_of_cpu_node(node) {
>  		unsigned long this_hwcap = 0;
> +		unsigned long this_isa = 0;
>  
>  		if (riscv_of_processor_hartid(node) < 0)
>  			continue;
> @@ -42,8 +88,24 @@ void riscv_fill_hwcap(void)
>  		if (riscv_read_check_isa(node, &isa) < 0)
>  			continue;
>  
> -		for (i = 0; i < strlen(isa); ++i)
> +		i = 0;
> +		isa_len = strlen(isa);
> +#if IS_ENABLED(CONFIG_32BIT)
> +		if (!strncmp(isa, "rv32", 4))
> +			i += 4;
> +#elif IS_ENABLED(CONFIG_64BIT)
> +		if (!strncmp(isa, "rv64", 4))
> +			i += 4;

We shouldn't be accepting arbitrary inputs and attempting to correct them, just
enforce that an actual ISA string is provided and check it against what the
kernel can support.

> +#endif
> +		for (; i < isa_len; ++i) {
>  			this_hwcap |= isa2hwcap[(unsigned char)(isa[i])];
> +			/*
> +			 * TODO: X, Y and Z extension parsing for Host ISA
> +			 * bitmap will be added in-future.
> +			 */
> +			if ('a' <= isa[i] && isa[i] < 'x')
> +				this_isa |= (1UL << (isa[i] - 'a'));
> +		}
>  
>  		/*
>  		 * All "okay" hart should have same isa. Set HWCAP based on
> @@ -54,6 +116,11 @@ void riscv_fill_hwcap(void)
>  			elf_hwcap &= this_hwcap;
>  		else
>  			elf_hwcap = this_hwcap;
> +
> +		if (riscv_isa[0])
> +			riscv_isa[0] &= this_isa;
> +		else
> +			riscv_isa[0] = this_isa;
>  	}
>  
>  	/* We don't support systems with F but without D, so mask those out
> @@ -63,7 +130,17 @@ void riscv_fill_hwcap(void)
>  		elf_hwcap &= ~COMPAT_HWCAP_ISA_F;
>  	}
>  
> -	pr_info("elf_hwcap is 0x%lx\n", elf_hwcap);
> +	memset(print_str, 0, sizeof(print_str));
> +	for (i = 0, j = 0; i < BITS_PER_LONG; i++)
> +		if (riscv_isa[0] & BIT_MASK(i))
> +			print_str[j++] = (char)('a' + i);
> +	pr_info("riscv: ISA extensions %s\n", print_str);
> +
> +	memset(print_str, 0, sizeof(print_str));
> +	for (i = 0, j = 0; i < BITS_PER_LONG; i++)
> +		if (elf_hwcap & BIT_MASK(i))
> +			print_str[j++] = (char)('a' + i);
> +	pr_info("riscv: ELF capabilities %s\n", print_str);
>  
>  #ifdef CONFIG_FPU
>  	if (elf_hwcap & (COMPAT_HWCAP_ISA_F | COMPAT_HWCAP_ISA_D))
> -- 
> 2.17.1

Powered by blists - more mailing lists