On 14.07.2025 11:08, Zhao Liu wrote:
From: Chuang Xu <xuchuangxc...@bytedance.com>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 On Intel platform: CPUID.01H.EBX[23:16] is defined as "max number of addressable IDs for logical processors in the physical package". When executing "cpuid -1 -l 1 -r" in the guest, we obtain a value of 90 for CPUID.01H.EBX[23:16], whereas the expected value is 128. Additionally, executing "cpuid -1 -l 4 -r" in the guest yields a value of 63 for CPUID.04H.EAX[31:26], which matches the expected result. As (1+CPUID.04H.EAX[31:26]) rounds up to the nearest power-of-2 integer, it's necessary to round up CPUID.01H.EBX[23:16] to the nearest power-of-2 integer too. Otherwise there would be unexpected results in guest with older kernel. 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 threads-per-core should actually be 2. And on AMD platform: CPUID.01H.EBX[23:16] is defined as "Logical processor count". Current result meets our expectation. So round up CPUID.01H.EBX[23:16] to the nearest power-of-2 integer only for Intel platform to solve the unexpected result. Use the "x-vendor-cpuid-only-v2" compat option to fix this issue. Reviewed-by: Zhao Liu <zhao1....@intel.com> Signed-off-by: Guixiong Wei <weiguixi...@bytedance.com> Signed-off-by: Yipeng Yin <yinyip...@bytedance.com> Signed-off-by: Chuang Xu <xuchuangxc...@bytedance.com> Signed-off-by: Zhao Liu <zhao1....@intel.com> --- Changes Since New v1 [**]: * Drop Igor's Acked-by since this version uses the newly added x-vendor-cpuid-only-v2. * Add Zhaoxin since this is the behavior defined in SDM. Changes Since original v6 [*] : * Rebase on the b69801dd6b1e ("Merge tag 'for_upstream' of https://git.kernel.org/pub/scm/virt/kvm/mst/qemu into staging"). * Polish the comment in code. * Explain the change doesn't need extra compat property. [*] original v6: https://lore.kernel.org/qemu-devel/20241009035638.59330-1-xuchuangxc...@bytedance.com/ [**] new v1: https://lore.kernel.org/qemu-devel/20250227062523.124601-2-zhao1....@intel.com/ --- target/i386/cpu.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/target/i386/cpu.c b/target/i386/cpu.c index 9e110e49ab8a..7fcb6c144d94 100644 --- a/target/i386/cpu.c +++ b/target/i386/cpu.c @@ -7869,7 +7869,17 @@ 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; + /* + * For CPUID.01H.EBX[Bits 23-16], AMD requires logical processor + * count, but Intel needs maximum number of addressable IDs for + * logical processors per package. + */ + if (cpu->vendor_cpuid_only_v2 && + (IS_INTEL_CPU(env) || IS_ZHAOXIN_CPU(env))) {
Hi! Previous incarnation of this patch were Cc'd qemu-stable@, as it were supposed to be picked up for the stable qemu series. However, this incarnation is not Cc'd to stable, and, most importantly, it relies on a feature which was introduced after all released qemu versions. Namely, vendor_cpuid_only_v2 is past v10.0.0, which is commit 216d9bb6d7716 "i386/cpu: Add x-vendor-cpuid-only-v2 option for compatibility". Should I omit this change for stable-10.0 series, or should it be modified to work in 10.0? Thanks, /mjt
