KVM: MIPS/VZ: VZ hardware setup for Octeon III

Set up hardware virtualisation on Octeon III cores, configuring guest
interrupt routing and carving out half of the root TLB for guest use,
restoring it back again afterwards.

We need to be careful to inhibit TLB shutdown machine check exceptions
while invalidating guest TLB entries, since TLB invalidation is not
available so guest entries must be invalidated by setting them to unique
unmapped addresses, which could conflict with mappings set by the guest
or root if recently repartitioned.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: David Daney <david.daney@cavium.com>
Cc: Andreas Herrmann <andreas.herrmann@caviumnetworks.com>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org

arch/mips/kvm/tlb.c

@@ -447,6 +447,7 @@ void kvm_vz_local_flush_guesttlb_all(void)
 	unsigned long old_entrylo[2];
 	unsigned long old_pagemask;
 	int entry;
+	u64 cvmmemctl2 = 0;
 
 	local_irq_save(flags);
 
@@ -457,6 +458,15 @@ void kvm_vz_local_flush_guesttlb_all(void)
 	old_entrylo[1] = read_gc0_entrylo1();
 	old_pagemask = read_gc0_pagemask();
 
+	switch (current_cpu_type()) {
+	case CPU_CAVIUM_OCTEON3:
+		/* Inhibit machine check due to multiple matching TLB entries */
+		cvmmemctl2 = read_c0_cvmmemctl2();
+		cvmmemctl2 |= CVMMEMCTL2_INHIBITTS;
+		write_c0_cvmmemctl2(cvmmemctl2);
+		break;
+	};
+
 	/* Invalidate guest entries in guest TLB */
 	write_gc0_entrylo0(0);
 	write_gc0_entrylo1(0);
@@ -468,6 +478,12 @@ void kvm_vz_local_flush_guesttlb_all(void)
 		mtc0_tlbw_hazard();
 		guest_tlb_write_indexed();
 	}
+
+	if (cvmmemctl2) {
+		cvmmemctl2 &= ~CVMMEMCTL2_INHIBITTS;
+		write_c0_cvmmemctl2(cvmmemctl2);
+	};
+
 	write_gc0_index(old_index);
 	write_gc0_entryhi(old_entryhi);
 	write_gc0_entrylo0(old_entrylo[0]);

arch/mips/kvm/vz.c

@@ -2747,37 +2747,73 @@ static unsigned int kvm_vz_resize_guest_vtlb(unsigned int size)
 static int kvm_vz_hardware_enable(void)
 {
 	unsigned int mmu_size, guest_mmu_size, ftlb_size;
+	u64 guest_cvmctl, cvmvmconfig;
+
+	switch (current_cpu_type()) {
+	case CPU_CAVIUM_OCTEON3:
+		/* Set up guest timer/perfcount IRQ lines */
+		guest_cvmctl = read_gc0_cvmctl();
+		guest_cvmctl &= ~CVMCTL_IPTI;
+		guest_cvmctl |= 7ull << CVMCTL_IPTI_SHIFT;
+		guest_cvmctl &= ~CVMCTL_IPPCI;
+		guest_cvmctl |= 6ull << CVMCTL_IPPCI_SHIFT;
+		write_gc0_cvmctl(guest_cvmctl);
+
+		cvmvmconfig = read_c0_cvmvmconfig();
+		/* No I/O hole translation. */
+		cvmvmconfig |= CVMVMCONF_DGHT;
+		/* Halve the root MMU size */
+		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
+			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
+		guest_mmu_size = mmu_size / 2;
+		mmu_size -= guest_mmu_size;
+		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
+		cvmvmconfig |= mmu_size - 1;
+		write_c0_cvmvmconfig(cvmvmconfig);
+
+		/* Update our records */
+		current_cpu_data.tlbsize = mmu_size;
+		current_cpu_data.tlbsizevtlb = mmu_size;
+		current_cpu_data.guest.tlbsize = guest_mmu_size;
+
+		/* Flush moved entries in new (guest) context */
+		kvm_vz_local_flush_guesttlb_all();
+		break;
+	default:
+		/*
+		 * ImgTec cores tend to use a shared root/guest TLB. To avoid
+		 * overlap of root wired and guest entries, the guest TLB may
+		 * need resizing.
+		 */
+		mmu_size = current_cpu_data.tlbsizevtlb;
+		ftlb_size = current_cpu_data.tlbsize - mmu_size;
 
-	/*
-	 * ImgTec cores tend to use a shared root/guest TLB. To avoid overlap of
-	 * root wired and guest entries, the guest TLB may need resizing.
-	 */
-	mmu_size = current_cpu_data.tlbsizevtlb;
-	ftlb_size = current_cpu_data.tlbsize - mmu_size;
-
-	/* Try switching to maximum guest VTLB size for flush */
-	guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
-	current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
-	kvm_vz_local_flush_guesttlb_all();
+		/* Try switching to maximum guest VTLB size for flush */
+		guest_mmu_size = kvm_vz_resize_guest_vtlb(mmu_size);
+		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
+		kvm_vz_local_flush_guesttlb_all();
 
-	/*
-	 * Reduce to make space for root wired entries and at least 2 root
-	 * non-wired entries. This does assume that long-term wired entries
-	 * won't be added later.
-	 */
-	guest_mmu_size = mmu_size - num_wired_entries() - 2;
-	guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
-	current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
+		/*
+		 * Reduce to make space for root wired entries and at least 2
+		 * root non-wired entries. This does assume that long-term wired
+		 * entries won't be added later.
+		 */
+		guest_mmu_size = mmu_size - num_wired_entries() - 2;
+		guest_mmu_size = kvm_vz_resize_guest_vtlb(guest_mmu_size);
+		current_cpu_data.guest.tlbsize = guest_mmu_size + ftlb_size;
 
-	/*
-	 * Write the VTLB size, but if another CPU has already written, check it
-	 * matches or we won't provide a consistent view to the guest. If this
-	 * ever happens it suggests an asymmetric number of wired entries.
-	 */
-	if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
-	    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
-		 "Available guest VTLB size mismatch"))
-		return -EINVAL;
+		/*
+		 * Write the VTLB size, but if another CPU has already written,
+		 * check it matches or we won't provide a consistent view to the
+		 * guest. If this ever happens it suggests an asymmetric number
+		 * of wired entries.
+		 */
+		if (cmpxchg(&kvm_vz_guest_vtlb_size, 0, guest_mmu_size) &&
+		    WARN(guest_mmu_size != kvm_vz_guest_vtlb_size,
+			 "Available guest VTLB size mismatch"))
+			return -EINVAL;
+		break;
+	}
 
 	/*
 	 * Enable virtualization features granting guest direct control of
@@ -2814,8 +2850,36 @@ static int kvm_vz_hardware_enable(void)
 
 static void kvm_vz_hardware_disable(void)
 {
+	u64 cvmvmconfig;
+	unsigned int mmu_size;
+
+	/* Flush any remaining guest TLB entries */
 	kvm_vz_local_flush_guesttlb_all();
 
+	switch (current_cpu_type()) {
+	case CPU_CAVIUM_OCTEON3:
+		/*
+		 * Allocate whole TLB for root. Existing guest TLB entries will
+		 * change ownership to the root TLB. We should be safe though as
+		 * they've already been flushed above while in guest TLB.
+		 */
+		cvmvmconfig = read_c0_cvmvmconfig();
+		mmu_size = ((cvmvmconfig & CVMVMCONF_MMUSIZEM1)
+			    >> CVMVMCONF_MMUSIZEM1_S) + 1;
+		cvmvmconfig &= ~CVMVMCONF_RMMUSIZEM1;
+		cvmvmconfig |= mmu_size - 1;
+		write_c0_cvmvmconfig(cvmvmconfig);
+
+		/* Update our records */
+		current_cpu_data.tlbsize = mmu_size;
+		current_cpu_data.tlbsizevtlb = mmu_size;
+		current_cpu_data.guest.tlbsize = 0;
+
+		/* Flush moved entries in new (root) context */
+		local_flush_tlb_all();
+		break;
+	}
+
 	if (cpu_has_guestid) {
 		write_c0_guestctl1(0);
 		kvm_vz_local_flush_roottlb_all_guests();