From 1e175d2e07c71d9574f5b1c74523abca54e2654f Mon Sep 17 00:00:00 2001
From: Sam Bobroff <sam.bobroff@au1.ibm.com>
Date: Wed, 25 Jul 2018 16:12:02 +1000
Subject: [PATCH] KVM: PPC: Book3S HV: Pack VCORE IDs to access full VCPU ID
space
References: bsc#1061840
Patch-mainline: v4.19-rc1
Git-commit: 1e175d2e07c71d9574f5b1c74523abca54e2654f
It is not currently possible to create the full number of possible
VCPUs (KVM_MAX_VCPUS) on Power9 with KVM-HV when the guest uses fewer
threads per core than its core stride (or "VSMT mode"). This is
because the VCORE ID and XIVE offsets grow beyond KVM_MAX_VCPUS
even though the VCPU ID is less than KVM_MAX_VCPU_ID.
To address this, "pack" the VCORE ID and XIVE offsets by using
knowledge of the way the VCPU IDs will be used when there are fewer
guest threads per core than the core stride. The primary thread of
each core will always be used first. Then, if the guest uses more than
one thread per core, these secondary threads will sequentially follow
the primary in each core.
So, the only way an ID above KVM_MAX_VCPUS can be seen, is if the
VCPUs are being spaced apart, so at least half of each core is empty,
and IDs between KVM_MAX_VCPUS and (KVM_MAX_VCPUS * 2) can be mapped
into the second half of each core (4..7, in an 8-thread core).
Similarly, if IDs above KVM_MAX_VCPUS * 2 are seen, at least 3/4 of
each core is being left empty, and we can map down into the second and
third quarters of each core (2, 3 and 5, 6 in an 8-thread core).
Lastly, if IDs above KVM_MAX_VCPUS * 4 are seen, only the primary
threads are being used and 7/8 of the core is empty, allowing use of
the 1, 5, 3 and 7 thread slots.
(Strides less than 8 are handled similarly.)
This allows the VCORE ID or offset to be calculated quickly from the
VCPU ID or XIVE server numbers, without access to the VCPU structure.
[paulus@ozlabs.org - tidied up comment a little, changed some WARN_ONCE
to pr_devel, wrapped line, fixed id check.]
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Acked-by: Michal Suchanek <msuchanek@suse.de>
---
arch/powerpc/include/asm/kvm_book3s.h | 47 +++++++++++++++++++++++++++++++++++
arch/powerpc/kvm/book3s_hv.c | 27 +++++++++++++++-----
arch/powerpc/kvm/book3s_xive.c | 19 ++++++++------
3 files changed, 80 insertions(+), 13 deletions(-)
diff --git a/arch/powerpc/include/asm/kvm_book3s.h b/arch/powerpc/include/asm/kvm_book3s.h
index 1f345a0b6ba2..83a9aa3cf689 100644
--- a/arch/powerpc/include/asm/kvm_book3s.h
+++ b/arch/powerpc/include/asm/kvm_book3s.h
@@ -390,4 +390,51 @@ extern int kvmppc_h_logical_ci_store(struct kvm_vcpu *vcpu);
#define SPLIT_HACK_MASK 0xff000000
#define SPLIT_HACK_OFFS 0xfb000000
+/*
+ * This packs a VCPU ID from the [0..KVM_MAX_VCPU_ID) space down to the
+ * [0..KVM_MAX_VCPUS) space, using knowledge of the guest's core stride
+ * (but not its actual threading mode, which is not available) to avoid
+ * collisions.
+ *
+ * The implementation leaves VCPU IDs from the range [0..KVM_MAX_VCPUS) (block
+ * 0) unchanged: if the guest is filling each VCORE completely then it will be
+ * using consecutive IDs and it will fill the space without any packing.
+ *
+ * For higher VCPU IDs, the packed ID is based on the VCPU ID modulo
+ * KVM_MAX_VCPUS (effectively masking off the top bits) and then an offset is
+ * added to avoid collisions.
+ *
+ * VCPU IDs in the range [KVM_MAX_VCPUS..(KVM_MAX_VCPUS*2)) (block 1) are only
+ * possible if the guest is leaving at least 1/2 of each VCORE empty, so IDs
+ * can be safely packed into the second half of each VCORE by adding an offset
+ * of (stride / 2).
+ *
+ * Similarly, if VCPU IDs in the range [(KVM_MAX_VCPUS*2)..(KVM_MAX_VCPUS*4))
+ * (blocks 2 and 3) are seen, the guest must be leaving at least 3/4 of each
+ * VCORE empty so packed IDs can be offset by (stride / 4) and (stride * 3 / 4).
+ *
+ * Finally, VCPU IDs from blocks 5..7 will only be seen if the guest is using a
+ * stride of 8 and 1 thread per core so the remaining offsets of 1, 5, 3 and 7
+ * must be free to use.
+ *
+ * (The offsets for each block are stored in block_offsets[], indexed by the
+ * block number if the stride is 8. For cases where the guest's stride is less
+ * than 8, we can re-use the block_offsets array by multiplying the block
+ * number by (MAX_SMT_THREADS / stride) to reach the correct entry.)
+ */
+static inline u32 kvmppc_pack_vcpu_id(struct kvm *kvm, u32 id)
+{
+ const int block_offsets[MAX_SMT_THREADS] = {0, 4, 2, 6, 1, 5, 3, 7};
+ int stride = kvm->arch.emul_smt_mode;
+ int block = (id / KVM_MAX_VCPUS) * (MAX_SMT_THREADS / stride);
+ u32 packed_id;
+
+ if (WARN_ONCE(block >= MAX_SMT_THREADS, "VCPU ID too large to pack"))
+ return 0;
+ packed_id = (id % KVM_MAX_VCPUS) + block_offsets[block];
+ if (WARN_ONCE(packed_id >= KVM_MAX_VCPUS, "VCPU ID packing failed"))
+ return 0;
+ return packed_id;
+}
+
#endif /* __ASM_KVM_BOOK3S_H__ */
diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index d73b29b6aaa1..785245e09f32 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -1816,7 +1816,7 @@ static int threads_per_vcore(struct kvm *kvm)
return threads_per_subcore;
}
-static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
+static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int id)
{
struct kvmppc_vcore *vcore;
@@ -1830,7 +1830,7 @@ static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, int core)
init_swait_queue_head(&vcore->wq);
vcore->preempt_tb = TB_NIL;
vcore->lpcr = kvm->arch.lpcr;
- vcore->first_vcpuid = core * kvm->arch.smt_mode;
+ vcore->first_vcpuid = id;
vcore->kvm = kvm;
INIT_LIST_HEAD(&vcore->preempt_list);
@@ -1989,10 +1989,16 @@ static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
unsigned int id)
{
struct kvm_vcpu *vcpu;
- int err;
+ int err = -EINVAL;
int core;
struct kvmppc_vcore *vcore;
+ if (id >= (KVM_MAX_VCPUS * kvm->arch.emul_smt_mode) &&
+ cpu_has_feature(CPU_FTR_ARCH_300)) {
+ pr_devel("DNCI: VCPU ID too high\n");
+ goto out;
+ }
+
err = -ENOMEM;
vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
if (!vcpu)
@@ -2048,12 +2054,21 @@ static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
mutex_lock(&kvm->lock);
vcore = NULL;
err = -EINVAL;
- core = id / kvm->arch.smt_mode;
+ if (cpu_has_feature(CPU_FTR_ARCH_300)) {
+ BUG_ON(kvm->arch.smt_mode != 1);
+ core = kvmppc_pack_vcpu_id(kvm, id);
+ } else {
+ core = id / kvm->arch.smt_mode;
+ }
if (core < KVM_MAX_VCORES) {
vcore = kvm->arch.vcores[core];
- if (!vcore) {
+ if (vcore && cpu_has_feature(CPU_FTR_ARCH_300)) {
+ pr_devel("KVM: collision on id %u", id);
+ vcore = NULL;
+ } else if (!vcore) {
err = -ENOMEM;
- vcore = kvmppc_vcore_create(kvm, core);
+ vcore = kvmppc_vcore_create(kvm,
+ id & ~(kvm->arch.smt_mode - 1));
kvm->arch.vcores[core] = vcore;
kvm->arch.online_vcores++;
}
diff --git a/arch/powerpc/kvm/book3s_xive.c b/arch/powerpc/kvm/book3s_xive.c
index f9818d7d3381..126f02b3ffb8 100644
--- a/arch/powerpc/kvm/book3s_xive.c
+++ b/arch/powerpc/kvm/book3s_xive.c
@@ -317,6 +317,11 @@ static int xive_select_target(struct kvm *kvm, u32 *server, u8 prio)
return -EBUSY;
}
+static u32 xive_vp(struct kvmppc_xive *xive, u32 server)
+{
+ return xive->vp_base + kvmppc_pack_vcpu_id(xive->kvm, server);
+}
+
static u8 xive_lock_and_mask(struct kvmppc_xive *xive,
struct kvmppc_xive_src_block *sb,
struct kvmppc_xive_irq_state *state)
@@ -362,7 +367,7 @@ static u8 xive_lock_and_mask(struct kvmppc_xive *xive,
*/
if (xd->flags & OPAL_XIVE_IRQ_MASK_VIA_FW) {
xive_native_configure_irq(hw_num,
- xive->vp_base + state->act_server,
+ xive_vp(xive, state->act_server),
MASKED, state->number);
/* set old_p so we can track if an H_EOI was done */
state->old_p = true;
@@ -418,7 +423,7 @@ static void xive_finish_unmask(struct kvmppc_xive *xive,
*/
if (xd->flags & OPAL_XIVE_IRQ_MASK_VIA_FW) {
xive_native_configure_irq(hw_num,
- xive->vp_base + state->act_server,
+ xive_vp(xive, state->act_server),
state->act_priority, state->number);
/* If an EOI is needed, do it here */
if (!state->old_p)
@@ -495,7 +500,7 @@ static int xive_target_interrupt(struct kvm *kvm,
kvmppc_xive_select_irq(state, &hw_num, NULL);
return xive_native_configure_irq(hw_num,
- xive->vp_base + server,
+ xive_vp(xive, server),
prio, state->number);
}
@@ -883,7 +888,7 @@ int kvmppc_xive_set_mapped(struct kvm *kvm, unsigned long guest_irq,
* which is fine for a never started interrupt.
*/
xive_native_configure_irq(hw_irq,
- xive->vp_base + state->act_server,
+ xive_vp(xive, state->act_server),
state->act_priority, state->number);
/*
@@ -959,7 +964,7 @@ int kvmppc_xive_clr_mapped(struct kvm *kvm, unsigned long guest_irq,
/* Reconfigure the IPI */
xive_native_configure_irq(state->ipi_number,
- xive->vp_base + state->act_server,
+ xive_vp(xive, state->act_server),
state->act_priority, state->number);
/*
@@ -1084,7 +1089,7 @@ int kvmppc_xive_connect_vcpu(struct kvm_device *dev,
pr_devel("Duplicate !\n");
return -EEXIST;
}
- if (cpu >= KVM_MAX_VCPUS) {
+ if (cpu >= (KVM_MAX_VCPUS * vcpu->kvm->arch.emul_smt_mode)) {
pr_devel("Out of bounds !\n");
return -EINVAL;
}
@@ -1098,7 +1103,7 @@ int kvmppc_xive_connect_vcpu(struct kvm_device *dev,
xc->xive = xive;
xc->vcpu = vcpu;
xc->server_num = cpu;
- xc->vp_id = xive->vp_base + cpu;
+ xc->vp_id = xive_vp(xive, cpu);
xc->mfrr = 0xff;
xc->valid = true;
--
2.13.7