KVM: PPC: Book3S HV: Fix TLB management on SMT8 POWER9 and POWER10 processors
authorSuraj Jitindar Singh <sjitindarsingh@gmail.com>
Wed, 2 Jun 2021 04:04:41 +0000 (14:04 +1000)
committerMichael Ellerman <mpe@ellerman.id.au>
Sun, 20 Jun 2021 23:22:34 +0000 (09:22 +1000)
The POWER9 vCPU TLB management code assumes all threads in a core share
a TLB, and that TLBIEL execued by one thread will invalidate TLBs for
all threads. This is not the case for SMT8 capable POWER9 and POWER10
(big core) processors, where the TLB is split between groups of threads.
This results in TLB multi-hits, random data corruption, etc.

Fix this by introducing cpu_first_tlb_thread_sibling etc., to determine
which siblings share TLBs, and use that in the guest TLB flushing code.

[npiggin@gmail.com: add changelog and comment]

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Fabiano Rosas <farosas@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210602040441.3984352-1-npiggin@gmail.com
arch/powerpc/include/asm/cputhreads.h
arch/powerpc/kvm/book3s_hv.c
arch/powerpc/kvm/book3s_hv_builtin.c
arch/powerpc/kvm/book3s_hv_rm_mmu.c

index 98c8bd155bf9d28aaded8cb8310aed785181adcb..b167186aaee4af09e4743142c71cd752892d3d1e 100644 (file)
@@ -98,6 +98,36 @@ static inline int cpu_last_thread_sibling(int cpu)
        return cpu | (threads_per_core - 1);
 }
 
+/*
+ * tlb_thread_siblings are siblings which share a TLB. This is not
+ * architected, is not something a hypervisor could emulate and a future
+ * CPU may change behaviour even in compat mode, so this should only be
+ * used on PowerNV, and only with care.
+ */
+static inline int cpu_first_tlb_thread_sibling(int cpu)
+{
+       if (cpu_has_feature(CPU_FTR_ARCH_300) && (threads_per_core == 8))
+               return cpu & ~0x6;      /* Big Core */
+       else
+               return cpu_first_thread_sibling(cpu);
+}
+
+static inline int cpu_last_tlb_thread_sibling(int cpu)
+{
+       if (cpu_has_feature(CPU_FTR_ARCH_300) && (threads_per_core == 8))
+               return cpu | 0x6;       /* Big Core */
+       else
+               return cpu_last_thread_sibling(cpu);
+}
+
+static inline int cpu_tlb_thread_sibling_step(void)
+{
+       if (cpu_has_feature(CPU_FTR_ARCH_300) && (threads_per_core == 8))
+               return 2;               /* Big Core */
+       else
+               return 1;
+}
+
 static inline u32 get_tensr(void)
 {
 #ifdef CONFIG_BOOKE
index f4dc4f0c34b5ee9e94a1390983c0f412100b9924..c0f1299736b9568854a80a1b39a60693aa0c1ec5 100644 (file)
@@ -2820,7 +2820,7 @@ static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
        cpumask_t *cpu_in_guest;
        int i;
 
-       cpu = cpu_first_thread_sibling(cpu);
+       cpu = cpu_first_tlb_thread_sibling(cpu);
        if (nested) {
                cpumask_set_cpu(cpu, &nested->need_tlb_flush);
                cpu_in_guest = &nested->cpu_in_guest;
@@ -2834,9 +2834,10 @@ static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
         * the other side is the first smp_mb() in kvmppc_run_core().
         */
        smp_mb();
-       for (i = 0; i < threads_per_core; ++i)
-               if (cpumask_test_cpu(cpu + i, cpu_in_guest))
-                       smp_call_function_single(cpu + i, do_nothing, NULL, 1);
+       for (i = cpu; i <= cpu_last_tlb_thread_sibling(cpu);
+                                       i += cpu_tlb_thread_sibling_step())
+               if (cpumask_test_cpu(i, cpu_in_guest))
+                       smp_call_function_single(i, do_nothing, NULL, 1);
 }
 
 static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
@@ -2867,8 +2868,8 @@ static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
         */
        if (prev_cpu != pcpu) {
                if (prev_cpu >= 0 &&
-                   cpu_first_thread_sibling(prev_cpu) !=
-                   cpu_first_thread_sibling(pcpu))
+                   cpu_first_tlb_thread_sibling(prev_cpu) !=
+                   cpu_first_tlb_thread_sibling(pcpu))
                        radix_flush_cpu(kvm, prev_cpu, vcpu);
                if (nested)
                        nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
index 259492bb41531f53ef532a698d471c0bc770d3cc..be8ef1c5b1bfb187cf3b06ac914889c737985491 100644 (file)
@@ -721,7 +721,7 @@ void kvmppc_check_need_tlb_flush(struct kvm *kvm, int pcpu,
         * Thus we make all 4 threads use the same bit.
         */
        if (cpu_has_feature(CPU_FTR_ARCH_300))
-               pcpu = cpu_first_thread_sibling(pcpu);
+               pcpu = cpu_first_tlb_thread_sibling(pcpu);
 
        if (nested)
                need_tlb_flush = &nested->need_tlb_flush;
index f487ebb3a70aa3eff07338ea4313fcff82fe9c24..8b70de4595f013b3dd846a29e10edb7f78137e63 100644 (file)
@@ -71,7 +71,7 @@ static int global_invalidates(struct kvm *kvm)
                 * so use the bit for the first thread to represent the core.
                 */
                if (cpu_has_feature(CPU_FTR_ARCH_300))
-                       cpu = cpu_first_thread_sibling(cpu);
+                       cpu = cpu_first_tlb_thread_sibling(cpu);
                cpumask_clear_cpu(cpu, &kvm->arch.need_tlb_flush);
        }