KVM: PPC: Book3S HV: Use bitmap of active threads rather than count

Currently, the entry_exit_count field in the kvmppc_vcore struct
contains two 8-bit counts, one of the threads that have started entering
the guest, and one of the threads that have started exiting the guest.
This changes it to an entry_exit_map field which contains two bitmaps
of 8 bits each. The advantage of doing this is that it gives us a
bitmap of which threads need to be signalled when exiting the guest.
That means that we no longer need to use the trick of setting the
HDEC to 0 to pull the other threads out of the guest, which led in
some cases to a spurious HDEC interrupt on the next guest entry.

Signed-off-by: Paul Mackerras <[e-mail]>
Signed-off-by: Alexander Graf <[e-mail]>

arch/powerpc/include/asm/kvm_host.h

···
 
 /*
  * Struct for a virtual core.
- * Note: entry_exit_count combines an entry count in the bottom 8 bits
- * and an exit count in the next 8 bits.  This is so that we can
- * atomically increment the entry count iff the exit count is 0
- * without taking the lock.
+ * Note: entry_exit_map combines a bitmap of threads that have entered
+ * in the bottom 8 bits and a bitmap of threads that have exited in the
+ * next 8 bits.  This is so that we can atomically set the entry bit
+ * iff the exit map is 0 without taking a lock.
  */
 struct kvmppc_vcore {
 	int n_runnable;
 	int num_threads;
-	int entry_exit_count;
+	int entry_exit_map;
 	int napping_threads;
 	int first_vcpuid;
 	u16 pcpu;
···
 	ulong conferring_threads;
 };
 
-#define VCORE_ENTRY_COUNT(vc)	((vc)->entry_exit_count & 0xff)
-#define VCORE_EXIT_COUNT(vc)	((vc)->entry_exit_count >> 8)
+#define VCORE_ENTRY_MAP(vc)	((vc)->entry_exit_map & 0xff)
+#define VCORE_EXIT_MAP(vc)	((vc)->entry_exit_map >> 8)
+#define VCORE_IS_EXITING(vc)	(VCORE_EXIT_MAP(vc) != 0)
 
 /* Values for vcore_state */
 #define VCORE_INACTIVE	0

arch/powerpc/kernel/asm-offsets.c

···
 	DEFINE(VCPU_ACOP, offsetof(struct kvm_vcpu, arch.acop));
 	DEFINE(VCPU_WORT, offsetof(struct kvm_vcpu, arch.wort));
 	DEFINE(VCPU_SHADOW_SRR1, offsetof(struct kvm_vcpu, arch.shadow_srr1));
-	DEFINE(VCORE_ENTRY_EXIT, offsetof(struct kvmppc_vcore, entry_exit_count));
+	DEFINE(VCORE_ENTRY_EXIT, offsetof(struct kvmppc_vcore, entry_exit_map));
 	DEFINE(VCORE_IN_GUEST, offsetof(struct kvmppc_vcore, in_guest));
 	DEFINE(VCORE_NAPPING_THREADS, offsetof(struct kvmppc_vcore, napping_threads));
 	DEFINE(VCORE_KVM, offsetof(struct kvmppc_vcore, kvm));

arch/powerpc/kvm/book3s_hv.c

···
 	/*
 	 * Initialize *vc.
 	 */
-	vc->entry_exit_count = 0;
+	vc->entry_exit_map = 0;
 	vc->preempt_tb = TB_NIL;
 	vc->in_guest = 0;
 	vc->napping_threads = 0;
···
 	 * this thread straight away and have it join in.
 	 */
 	if (!signal_pending(current)) {
-		if (vc->vcore_state == VCORE_RUNNING &&
-		    VCORE_EXIT_COUNT(vc) == 0) {
+		if (vc->vcore_state == VCORE_RUNNING && !VCORE_IS_EXITING(vc)) {
 			kvmppc_create_dtl_entry(vcpu, vc);
 			kvmppc_start_thread(vcpu);
 			trace_kvm_guest_enter(vcpu);

arch/powerpc/kvm/book3s_hv_builtin.c

···
 	int rv = H_SUCCESS; /* => don't yield */
 
 	set_bit(vcpu->arch.ptid, &vc->conferring_threads);
-	while ((get_tb() < stop) && (VCORE_EXIT_COUNT(vc) == 0)) {
-		threads_running = VCORE_ENTRY_COUNT(vc);
-		threads_ceded = hweight32(vc->napping_threads);
-		threads_conferring = hweight32(vc->conferring_threads);
-		if (threads_ceded + threads_conferring >= threads_running) {
+	while ((get_tb() < stop) && !VCORE_IS_EXITING(vc)) {
+		threads_running = VCORE_ENTRY_MAP(vc);
+		threads_ceded = vc->napping_threads;
+		threads_conferring = vc->conferring_threads;
+		if ((threads_ceded | threads_conferring) == threads_running) {
 			rv = H_TOO_HARD; /* => do yield */
 			break;
 		}

arch/powerpc/kvm/book3s_hv_rmhandlers.S

···
 	or	r3, r3, r0
 	stwcx.	r3, 0, r6
 	bne	1b
-	/* order napping_threads update vs testing entry_exit_count */
+	/* order napping_threads update vs testing entry_exit_map */
 	isync
 	li	r12, 0
 	lwz	r7, VCORE_ENTRY_EXIT(r5)
···
 	 * We don't have to lock against concurrent tlbies,
 	 * but we do have to coordinate across hardware threads.
 	 */
-	/* Increment entry count iff exit count is zero. */
-	ld	r5,HSTATE_KVM_VCORE(r13)
-	addi	r9,r5,VCORE_ENTRY_EXIT
-21:	lwarx	r3,0,r9
-	cmpwi	r3,0x100		/* any threads starting to exit? */
+	/* Set bit in entry map iff exit map is zero. */
+	ld	r5, HSTATE_KVM_VCORE(r13)
+	li	r7, 1
+	lbz	r6, HSTATE_PTID(r13)
+	sld	r7, r7, r6
+	addi	r9, r5, VCORE_ENTRY_EXIT
+21:	lwarx	r3, 0, r9
+	cmpwi	r3, 0x100		/* any threads starting to exit? */
 	bge	secondary_too_late	/* if so we're too late to the party */
-	addi	r3,r3,1
-	stwcx.	r3,0,r9
+	or	r3, r3, r7
+	stwcx.	r3, 0, r9
 	bne	21b
 
 	/* Primary thread switches to guest partition. */
 	ld	r9,VCORE_KVM(r5)	/* pointer to struct kvm */
-	lbz	r6,HSTATE_PTID(r13)
 	cmpwi	r6,0
 	bne	20f
 	ld	r6,KVM_SDR1(r9)
···
 	 * We don't have to lock against tlbies but we do
 	 * have to coordinate the hardware threads.
 	 */
-	/* Increment the threads-exiting-guest count in the 0xff00
-	   bits of vcore->entry_exit_count */
-	ld	r5,HSTATE_KVM_VCORE(r13)
-	addi	r6,r5,VCORE_ENTRY_EXIT
-41:	lwarx	r3,0,r6
-	addi	r0,r3,0x100
-	stwcx.	r0,0,r6
+	/* Set our bit in the threads-exiting-guest map in the 0xff00
+	   bits of vcore->entry_exit_map */
+	ld	r5, HSTATE_KVM_VCORE(r13)
+	lbz	r4, HSTATE_PTID(r13)
+	li	r7, 0x100
+	sld	r7, r7, r4
+	addi	r6, r5, VCORE_ENTRY_EXIT
+41:	lwarx	r3, 0, r6
+	or	r0, r3, r7
+	stwcx.	r0, 0, r6
 	bne	41b
 	isync		/* order stwcx. vs. reading napping_threads */
 
···
 	 * up to the kernel or qemu; we can't handle it in real mode.
 	 * Thus we have to do a partition switch, so we have to
 	 * collect the other threads, if we are the first thread
-	 * to take an interrupt.  To do this, we set the HDEC to 0,
-	 * which causes an HDEC interrupt in all threads within 2ns
-	 * because the HDEC register is shared between all 4 threads.
+	 * to take an interrupt.  To do this, we send a message or
+	 * IPI to all the threads that have their bit set in the entry
+	 * map in vcore->entry_exit_map (other than ourselves).
 	 * However, we don't need to bother if this is an HDEC
 	 * interrupt, since the other threads will already be on their
 	 * way here in that case.
···
 	bge	43f
 	cmpwi	r12,BOOK3S_INTERRUPT_HV_DECREMENTER
 	beq	43f
-	li	r0,0
-	mtspr	SPRN_HDEC,r0
 
-	/*
-	 * Send an IPI to any napping threads, since an HDEC interrupt
-	 * doesn't wake CPUs up from nap.
-	 */
-	lwz	r3,VCORE_NAPPING_THREADS(r5)
-	lbz	r4,HSTATE_PTID(r13)
-	li	r0,1
-	sld	r0,r0,r4
+	srwi	r0,r7,8
 	andc.	r3,r3,r0		/* no sense IPI'ing ourselves */
 	beq	43f
 	/* Order entry/exit update vs. IPIs */
···
 	addi	r6,r5,VCORE_NAPPING_THREADS
 31:	lwarx	r4,0,r6
 	or	r4,r4,r0
-	PPC_POPCNTW(R7,R4)
-	cmpw	r7,r8
-	bge	kvm_cede_exit
+	cmpw	r4,r8
+	beq	kvm_cede_exit
 	stwcx.	r4,0,r6
 	bne	31b
-	/* order napping_threads update vs testing entry_exit_count */
+	/* order napping_threads update vs testing entry_exit_map */
 	isync
 	li	r0,NAPPING_CEDE
 	stb	r0,HSTATE_NAPPING(r13)