Re: [PATCH] LoongArch: Correct the calculation logic of thread_count

From: Qiang Ma

Date: Wed Nov 26 2025 - 06:07:37 EST


在 2025/11/26 18:14, Qiang Ma 写道:

在 2025/11/26 18:06, Hengqi Chen 写道:
On Wed, Nov 26, 2025 at 5:47 PM Qiang Ma <maqianga@xxxxxxxxxxxxx> wrote:

On Wed, Nov 26, 2025 at 10:37 AM Qiang Ma<maqianga@xxxxxxxxxxxxx> wrote:

   For thread_count, the current calculation method has a maximum of 255,
   which may not be sufficient in the future. Therefore, we are correcting
   it now.

   Reference: SMBIOS Specification, 7.5 Processor Information (Type 4)[1]

[1]:https://www.dmtf.org/sites/default/files/standards/documents/DSP0134_3.9.0.pdf

   Signed-off-by: Qiang Ma<maqianga@xxxxxxxxxxxxx>
   ---
    arch/loongarch/kernel/setup.c | 9 ++++++++-
    1 file changed, 8 insertions(+), 1 deletion(-)

   diff --git a/arch/loongarch/kernel/setup.c b/arch/loongarch/kernel/setup.c
   index 25a87378e48e..760f5ce44384 100644
   --- a/arch/loongarch/kernel/setup.c
   +++ b/arch/loongarch/kernel/setup.c
   @@ -56,6 +56,7 @@
    #define SMBIOS_FREQLOW_MASK            0xFF
    #define SMBIOS_CORE_PACKAGE_OFFSET     0x23
    #define SMBIOS_THREAD_PACKAGE_OFFSET   0x25
   +#define SMBIOS_THREAD_PACKAGE_2_OFFSET 0x2E
    #define LOONGSON_EFI_ENABLE            (1 << 3)

    unsigned long fw_arg0, fw_arg1, fw_arg2;
   @@ -120,13 +121,19 @@ static void __init parse_cpu_table(const struct dmi_header *dm)
    {
           long freq_temp = 0;
           char *dmi_data = (char *)dm;
   +       u8 thread_count;
   +       u16 thread_count2;

           freq_temp = ((*(dmi_data + SMBIOS_FREQHIGH_OFFSET) << 8) +
                           ((*(dmi_data + SMBIOS_FREQLOW_OFFSET)) & SMBIOS_FREQLOW_MASK));
           cpu_clock_freq = freq_temp * 1000000;

           loongson_sysconf.cpuname = (void *)dmi_string_parse(dm, dmi_data[16]);
   -       loongson_sysconf.cores_per_package = *(dmi_data + SMBIOS_THREAD_PACKAGE_OFFSET);
   +       thread_count = *(dmi_data + SMBIOS_THREAD_PACKAGE_OFFSET);
   +       thread_count2 = *(u16 *)(dmi_data + SMBIOS_THREAD_PACKAGE_2_OFFSET);

You have `dm->length >= 0x30` below but the memory load already done here.
After reading thread_count2, it cannot be guaranteed that the data is valid.

'dm->length >= 0x30' can detect that thread_count2 is an available data,
right?

I mean you should perform the check before doing memory load.
It doesn't matter whether it's before or after.
My understanding was wrong. I will fix it in v2.

Anyway, here it's just reading. Before it's actually used later,
it will check this length.
   +       if (thread_count != 0)
   +               loongson_sysconf.cores_per_package = dm->length >= 0x30 && thread_count == 0xFF ?
   + thread_count2 : thread_count;

           pr_info("CpuClock = %llu\n", cpu_clock_freq);
    }
   --
   2.20.1