[RISC-V] Read isa line in /proc/cpuinfo when hwprobe not available

[fuad1502]

hwprobe was introduced in Linux kernel version 6.4 https://www.kernel.org/doc/html/v6.4/riscv/hwprobe.html (although on version 6.5, it only started supporting Zba, Zbb, and Zbs detection)

isa line in /proc/cpuinfo is available from Linux kernel version 6.3 upwards https://www.kernel.org/doc/html/v6.3/riscv/uabi.html (CMIIW)

hwprobe also exports less extensions than /proc/cpuinfo when first introduced. Should we prioritize /proc/cpuinfo on kernel version 6.11 and lower (on version 6.11, the extension supported in hwprobe looks complete https://www.kernel.org/doc/html/v6.11/arch/riscv/hwprobe.html)?

TODO:

• Study further on hwprobe and /proc/cpuinfo isa line support on Linux.

Part of #84834, cc @dotnet/samsung

[risc-vv]

RISC-V Release-CLR-VF2: 9702 / 9750 (99.51%)

=======================
      passed: 9702
      failed: 28
     skipped: 70
      killed: 20
------------------------
  TOTAL libs: 9820
 TOTAL tests: 9820
   REAL time: 1h 59min 38s 839ms
=======================

Release-CLR-VF2.md, Release-CLR-VF2.xml, testclr_output.tar.gz

Build information and commands

GIT: 5c23311a4c98fb62130781025f5aafc02c34fe41
CI: 2d916d20de463f9bba05ae71b3d1f37d439a8cb1
REPO: fuad1502/runtime
BRANCH: riscv-ext-cpuinfo
CONFIG: Release
LIB_CONFIG: Release

RISC-V Release-CLR-QEMU: 9701 / 9750 (99.50%)

=======================
      passed: 9701
      failed: 29
     skipped: 70
      killed: 20
------------------------
  TOTAL libs: 9820
 TOTAL tests: 9820
   REAL time: 3h 28min 15s 56ms
=======================

Release-CLR-QEMU.md, Release-CLR-QEMU.xml, testclr_output.tar.gz

Build information and commands

GIT: 5c23311a4c98fb62130781025f5aafc02c34fe41
CI: 2d916d20de463f9bba05ae71b3d1f37d439a8cb1
REPO: fuad1502/runtime
BRANCH: riscv-ext-cpuinfo
CONFIG: Release
LIB_CONFIG: Release

RISC-V Release-FX-QEMU: 706582 / 721531 (97.93%)

=======================
      passed: 706582
      failed: 1623
     skipped: 1537
      killed: 13326
------------------------
  TOTAL libs: 259
 TOTAL tests: 723068
   REAL time: 2h 28min 12s 448ms
=======================

Release-FX-QEMU.md, Release-FX-QEMU.xml, testfx_output.tar.gz

Build information and commands

GIT: 5c23311a4c98fb62130781025f5aafc02c34fe41
CI: 2d916d20de463f9bba05ae71b3d1f37d439a8cb1
REPO: fuad1502/runtime
BRANCH: riscv-ext-cpuinfo
CONFIG: Release
LIB_CONFIG: Release

[am11]

Linux procfs is considered optional for number of reasons.

Suggested change

	assert(cpuInfo != NULL);
	if (cpuInfo == NULL) return result;

(or wrap the handling under != NULL)

[am11]

hwprobe was introduced in Linux kernel version 6.4 kernel.org/doc/html/v6.4/riscv/hwprobe.html (although on version 6.5, it only started supporting Zba, Zbb, and Zbs detection)

Have we decided what is the .NET baseline kernel version for riscv64? If it is higher than 6.5 for other reasons, we can ignore the #else branch. https://lf-rise.atlassian.net/wiki/spaces/HOME/pages/8587962/RVA23+Profile

[fuad1502]

Have we decided what is the .NET baseline kernel version for riscv64? If it is higher than 6.5 for other reasons, we can ignore the #else branch. https://lf-rise.atlassian.net/wiki/spaces/HOME/pages/8587962/RVA23+Profile

Hmm not sure, but you're right. And thanks for the reference BTW 👍

So actually, I added this detection through /proc/cpuinfo since the extensions are not detected on a platform I am testing on, despite the extensions are clearly available (exists in /proc/cpuinfo and run fine if I force the flag). But it turns out it's running kernel 6.6, so it should be fine. So, there might be some other issue that I'm not aware of.

I'll investigate further and closing this PR for now. I'll reopen it if it's clearly needed.

Thank you for your time reviewing it! 🙏 @am11

[am11]

@fuad1502, I am also on the same kernel and getting:

$ neofetch
        #####           am11@k1 
       #######          ------- 
       ##O#O##          OS: Bianbu 2.1 riscv64 
       #######          Host: M1-MUSE-BOOK 
     ###########        Kernel: 6.6.63 
    #############       Uptime: 4 days, 1 hour, 13 mins 
   ###############      Packages: 2325 (dpkg) 
   ################     Shell: bash 5.2.21 
  #################     Resolution: 1920x1080 
#####################   Terminal: /dev/pts/1 
#####################   CPU: Spacemit X60 (8) @ 1.600GHz 
  #################     Memory: 723MiB / 15904MiB 

$ cc -xc - <<EOF
#include <stdio.h>
#include <assert.h>
#include <asm/hwprobe.h>
#include <asm/unistd.h>
#include <unistd.h>

enum RiscV64IntrinsicConstants
{
    RiscV64IntrinsicConstants_Zba = 0x0001,
    RiscV64IntrinsicConstants_Zbb = 0x0002,
};

int main(void){
int result;
    struct riscv_hwprobe pairs[1] = {{RISCV_HWPROBE_KEY_IMA_EXT_0, 0}};

    if (syscall(__NR_riscv_hwprobe, pairs, 1, 0, NULL, 0) == 0)
    {
        // Our baseline support is for RV64GC (see #73437)
        assert(pairs[0].value & RISCV_HWPROBE_IMA_FD);
        assert(pairs[0].value & RISCV_HWPROBE_IMA_C);

        if (pairs[0].value & RISCV_HWPROBE_EXT_ZBA)
        {
            result |= RiscV64IntrinsicConstants_Zba;
        }

        if (pairs[0].value & RISCV_HWPROBE_EXT_ZBB)
        {
            result |= RiscV64IntrinsicConstants_Zbb;
        }
    }

    printf("result: %d\n", result);
    return 0;
}
EOF


$ ./a.out
result: 3

Please double check the build logs; cmake logs should be showing lines like:

  -- Looking for include files windows.h, bcrypt.h - not found
  -- Looking for include file asm/hwprobe.h
  -- Looking for include file asm/hwprobe.h - found
  -- Looking for arc4random_buf

https://github.com/am11/CrossRepoCITesting/actions/runs/14564508979/job/40851797145

[fuad1502]

Thanks for looking into it! I found that the issue was indeed with our internal build, it turns out we're compiling with Linux 5.4 kernel headers 😅

I think we can safely assume hwprobe is available here for now 👍