Re: [PATCH v2] i386/cpu: fixup number of addressable IDs for logical processors in the physical package

[Igor Mammedov]

On Wed, 18 Sep 2024 20:22:26 +0800
Chuang Xu <xuchuangxclwt@bytedance.com> wrote:

> Hi, Igor:
> 
> On 2024/9/18 下午8:02, Igor Mammedov wrote:
> > On Sat, 14 Sep 2024 19:01:27 +0800
> > Chuang Xu <xuchuangxclwt@bytedance.com> wrote:
> >  
> >> When QEMU is started with:
> >> -cpu host,migratable=on,host-cache-info=on,l3-cache=off
> >> -smp 180,sockets=2,dies=1,cores=45,threads=2
> >>
> >> Execute "cpuid -1 -l 1 -r" in guest, we'll get:
> >> eax=0x000806f8 ebx=0x465a0800 ecx=0xfffaba1f edx=0x3fa9fbff
> >> CPUID.01H.EBX[23:16] is 90, while the expected value is 128.
> >>
> >> Execute "cpuid -1 -l 4 -r" in guest, we'll get:
> >> eax=0xfc004121 ebx=0x02c0003f ecx=0x0000003f edx=0x00000000
> >> CPUID.04H.EAX[31:26] is 63, which is as expected.
> >>
> >> As (1+CPUID.04H.EAX[31:26]) round up to the nearest power-of-2 integer,
> >> we'd beter round up CPUID.01H.EBX[23:16] to the nearest power-of-2
> >> integer too. Otherwise we may encounter unexpected results in guest.
> >>
> >> For example, when QEMU is started with CLI above and xtopology is disabled,
> >> guest kernel 5.15.120 uses CPUID.01H.EBX[23:16]/(1+CPUID.04H.EAX[31:26]) to
> >> calculate threads-per-core in detect_ht(). Then guest will get 
> >> "90/(1+63)=1"
> >> as the result, even though theads-per-core should actually be 2.
> >>
> >> So let us round up CPUID.01H.EBX[23:16] to the nearest power-of-2 integer
> >> to solve the unexpected result.
> >>
> >> Signed-off-by: Guixiong Wei <weiguixiong@bytedance.com>
> >> Signed-off-by: Yipeng Yin <yinyipeng@bytedance.com>
> >> Signed-off-by: Chuang Xu <xuchuangxclwt@bytedance.com>
> >> ---
> >>   target/i386/cpu.c | 8 +++++++-
> >>   1 file changed, 7 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> >> index 4c2e6f3a71..24d60ead9e 100644
> >> --- a/target/i386/cpu.c
> >> +++ b/target/i386/cpu.c
> >> @@ -261,6 +261,12 @@ static uint32_t 
> >> max_thread_ids_for_cache(X86CPUTopoInfo *topo_info,
> >>       return num_ids - 1;
> >>   }
> >>   
> >> +static uint32_t max_thread_number_in_package(X86CPUTopoInfo *topo_info)
> >> +{
> >> +    uint32_t num_threads = 1 << apicid_pkg_offset(topo_info);
> >> +    return num_threads;
> >> +}
> >> +
> >>   static uint32_t max_core_ids_in_package(X86CPUTopoInfo *topo_info)
> >>   {
> >>       uint32_t num_cores = 1 << (apicid_pkg_offset(topo_info) -
> >> @@ -6417,7 +6423,7 @@ void cpu_x86_cpuid(CPUX86State *env, uint32_t index, 
> >> uint32_t count,
> >>           }
> >>           *edx = env->features[FEAT_1_EDX];
> >>           if (threads_per_pkg > 1) {
> >> -            *ebx |= threads_per_pkg << 16;
> >> +            *ebx |= max_thread_number_in_package(&topo_info) << 16;  
> > why not use pow2ceil(threads_per_pkg) instead?  
> 
> I saw in the latest code that calculations of cpuids involving CPU topology 
> all use topo_info,
> so in order to maintain consistency in code style, I also used topo_info for 
> calculation.

and we end up with a zoo of ways different topo stuff is calculated.

Given we already have threads_per_pkg calculated within the function,
is cleaner/more self-documenting to reuse it with pow2ceil() instead of
adding yet another helper with less than obvious '1 << 
apicid_pkg_offset(topo_info)' math.

> >  
> >>               *edx |= CPUID_HT;
> >>           }
> >>           if (!cpu->enable_pmu) 
>