forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2024-05-11 04dd17822334871b23ea2862f7798fb0e0007777 
 kernel/arch/powerpc/kvm/book3s_hv_builtin.c
....@@ -1,9 +1,6 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
12 /*
23  * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
3
- *
4
- * This program is free software; you can redistribute it and/or modify
5
- * it under the terms of the GNU General Public License, version 2, as
6
- * published by the Free Software Foundation.
74  */
85 
96 #include <linux/cpu.h>
....@@ -98,25 +95,17 @@
9895 void __init kvm_cma_reserve(void)
9996 {
10097 	unsigned long align_size;
101
-	struct memblock_region *reg;
102
-	phys_addr_t selected_size = 0;
98
+	phys_addr_t selected_size;
10399 
104100 	/*
105101 	 * We need CMA reservation only when we are in HV mode
106102 	 */
107103 	if (!cpu_has_feature(CPU_FTR_HVMODE))
108104 		return;
109
-	/*
110
-	 * We cannot use memblock_phys_mem_size() here, because
111
-	 * memblock_analyze() has not been called yet.
112
-	 */
113
-	for_each_memblock(memory, reg)
114
-		selected_size += memblock_region_memory_end_pfn(reg) -
115
-				 memblock_region_memory_base_pfn(reg);
116105 
117
-	selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT;
106
+	selected_size = PAGE_ALIGN(memblock_phys_mem_size() * kvm_cma_resv_ratio / 100);
118107 	if (selected_size) {
119
-		pr_debug("%s: reserving %ld MiB for global area\n", __func__,
108
+		pr_info("%s: reserving %ld MiB for global area\n", __func__,
120109 			 (unsigned long)selected_size / SZ_1M);
121110 		align_size = HPT_ALIGN_PAGES << PAGE_SHIFT;
122111 		cma_declare_contiguous(0, selected_size, 0, align_size,
....@@ -231,6 +220,15 @@
231220 	void __iomem *xics_phys;
232221 	unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER);
233222 
223
+	/* For a nested hypervisor, use the XICS via hcall */
224
+	if (kvmhv_on_pseries()) {
225
+		unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
226
+
227
+		plpar_hcall_raw(H_IPI, retbuf, get_hard_smp_processor_id(cpu),
228
+				IPI_PRIORITY);
229
+		return;
230
+	}
231
+
234232 	/* On POWER9 we can use msgsnd for any destination cpu. */
235233 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
236234 		msg |= get_hard_smp_processor_id(cpu);
....@@ -248,7 +246,7 @@
248246 	}
249247 
250248 	/* We should never reach this */
251
-	if (WARN_ON_ONCE(xive_enabled()))
249
+	if (WARN_ON_ONCE(xics_on_xive()))
252250 	    return;
253251 
254252 	/* Else poke the target with an IPI */
....@@ -460,12 +458,19 @@
460458 		return 1;
461459 
462460 	/* Now read the interrupt from the ICP */
463
-	xics_phys = local_paca->kvm_hstate.xics_phys;
464
-	rc = 0;
465
-	if (!xics_phys)
466
-		rc = opal_int_get_xirr(&xirr, false);
467
-	else
468
-		xirr = __raw_rm_readl(xics_phys + XICS_XIRR);
461
+	if (kvmhv_on_pseries()) {
462
+		unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
463
+
464
+		rc = plpar_hcall_raw(H_XIRR, retbuf, 0xFF);
465
+		xirr = cpu_to_be32(retbuf[0]);
466
+	} else {
467
+		xics_phys = local_paca->kvm_hstate.xics_phys;
468
+		rc = 0;
469
+		if (!xics_phys)
470
+			rc = opal_int_get_xirr(&xirr, false);
471
+		else
472
+			xirr = __raw_rm_readl(xics_phys + XICS_XIRR);
473
+	}
469474 	if (rc < 0)
470475 		return 1;
471476 
....@@ -494,7 +499,13 @@
494499 	 */
495500 	if (xisr == XICS_IPI) {
496501 		rc = 0;
497
-		if (xics_phys) {
502
+		if (kvmhv_on_pseries()) {
503
+			unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
504
+
505
+			plpar_hcall_raw(H_IPI, retbuf,
506
+					hard_smp_processor_id(), 0xff);
507
+			plpar_hcall_raw(H_EOI, retbuf, h_xirr);
508
+		} else if (xics_phys) {
498509 			__raw_rm_writeb(0xff, xics_phys + XICS_MFRR);
499510 			__raw_rm_writel(xirr, xics_phys + XICS_XIRR);
500511 		} else {
....@@ -520,7 +531,13 @@
520531 			/* We raced with the host,
521532 			 * we need to resend that IPI, bummer
522533 			 */
523
-			if (xics_phys)
534
+			if (kvmhv_on_pseries()) {
535
+				unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
536
+
537
+				plpar_hcall_raw(H_IPI, retbuf,
538
+						hard_smp_processor_id(),
539
+						IPI_PRIORITY);
540
+			} else if (xics_phys)
524541 				__raw_rm_writeb(IPI_PRIORITY,
525542 						xics_phys + XICS_MFRR);
526543 			else
....@@ -549,7 +566,7 @@
549566 {
550567 	if (!kvmppc_xics_enabled(vcpu))
551568 		return H_TOO_HARD;
552
-	if (xive_enabled()) {
569
+	if (xics_on_xive()) {
553570 		if (is_rm())
554571 			return xive_rm_h_xirr(vcpu);
555572 		if (unlikely(!__xive_vm_h_xirr))
....@@ -564,7 +581,7 @@
564581 	if (!kvmppc_xics_enabled(vcpu))
565582 		return H_TOO_HARD;
566583 	vcpu->arch.regs.gpr[5] = get_tb();
567
-	if (xive_enabled()) {
584
+	if (xics_on_xive()) {
568585 		if (is_rm())
569586 			return xive_rm_h_xirr(vcpu);
570587 		if (unlikely(!__xive_vm_h_xirr))
....@@ -578,7 +595,7 @@
578595 {
579596 	if (!kvmppc_xics_enabled(vcpu))
580597 		return H_TOO_HARD;
581
-	if (xive_enabled()) {
598
+	if (xics_on_xive()) {
582599 		if (is_rm())
583600 			return xive_rm_h_ipoll(vcpu, server);
584601 		if (unlikely(!__xive_vm_h_ipoll))
....@@ -593,7 +610,7 @@
593610 {
594611 	if (!kvmppc_xics_enabled(vcpu))
595612 		return H_TOO_HARD;
596
-	if (xive_enabled()) {
613
+	if (xics_on_xive()) {
597614 		if (is_rm())
598615 			return xive_rm_h_ipi(vcpu, server, mfrr);
599616 		if (unlikely(!__xive_vm_h_ipi))
....@@ -607,7 +624,7 @@
607624 {
608625 	if (!kvmppc_xics_enabled(vcpu))
609626 		return H_TOO_HARD;
610
-	if (xive_enabled()) {
627
+	if (xics_on_xive()) {
611628 		if (is_rm())
612629 			return xive_rm_h_cppr(vcpu, cppr);
613630 		if (unlikely(!__xive_vm_h_cppr))
....@@ -621,7 +638,7 @@
621638 {
622639 	if (!kvmppc_xics_enabled(vcpu))
623640 		return H_TOO_HARD;
624
-	if (xive_enabled()) {
641
+	if (xics_on_xive()) {
625642 		if (is_rm())
626643 			return xive_rm_h_eoi(vcpu, xirr);
627644 		if (unlikely(!__xive_vm_h_eoi))
....@@ -729,3 +746,168 @@
729746 	smp_mb();
730747 	local_paca->kvm_hstate.kvm_split_mode = NULL;
731748 }
749
+
750
+static void kvmppc_end_cede(struct kvm_vcpu *vcpu)
751
+{
752
+	vcpu->arch.ceded = 0;
753
+	if (vcpu->arch.timer_running) {
754
+		hrtimer_try_to_cancel(&vcpu->arch.dec_timer);
755
+		vcpu->arch.timer_running = 0;
756
+	}
757
+}
758
+
759
+void kvmppc_set_msr_hv(struct kvm_vcpu *vcpu, u64 msr)
760
+{
761
+	/*
762
+	 * Check for illegal transactional state bit combination
763
+	 * and if we find it, force the TS field to a safe state.
764
+	 */
765
+	if ((msr & MSR_TS_MASK) == MSR_TS_MASK)
766
+		msr &= ~MSR_TS_MASK;
767
+	vcpu->arch.shregs.msr = msr;
768
+	kvmppc_end_cede(vcpu);
769
+}
770
+EXPORT_SYMBOL_GPL(kvmppc_set_msr_hv);
771
+
772
+static void inject_interrupt(struct kvm_vcpu *vcpu, int vec, u64 srr1_flags)
773
+{
774
+	unsigned long msr, pc, new_msr, new_pc;
775
+
776
+	msr = kvmppc_get_msr(vcpu);
777
+	pc = kvmppc_get_pc(vcpu);
778
+	new_msr = vcpu->arch.intr_msr;
779
+	new_pc = vec;
780
+
781
+	/* If transactional, change to suspend mode on IRQ delivery */
782
+	if (MSR_TM_TRANSACTIONAL(msr))
783
+		new_msr |= MSR_TS_S;
784
+	else
785
+		new_msr |= msr & MSR_TS_MASK;
786
+
787
+	/*
788
+	 * Perform MSR and PC adjustment for LPCR[AIL]=3 if it is set and
789
+	 * applicable. AIL=2 is not supported.
790
+	 *
791
+	 * AIL does not apply to SRESET, MCE, or HMI (which is never
792
+	 * delivered to the guest), and does not apply if IR=0 or DR=0.
793
+	 */
794
+	if (vec != BOOK3S_INTERRUPT_SYSTEM_RESET &&
795
+	    vec != BOOK3S_INTERRUPT_MACHINE_CHECK &&
796
+	    (vcpu->arch.vcore->lpcr & LPCR_AIL) == LPCR_AIL_3 &&
797
+	    (msr & (MSR_IR|MSR_DR)) == (MSR_IR|MSR_DR) ) {
798
+		new_msr |= MSR_IR | MSR_DR;
799
+		new_pc += 0xC000000000004000ULL;
800
+	}
801
+
802
+	kvmppc_set_srr0(vcpu, pc);
803
+	kvmppc_set_srr1(vcpu, (msr & SRR1_MSR_BITS) | srr1_flags);
804
+	kvmppc_set_pc(vcpu, new_pc);
805
+	vcpu->arch.shregs.msr = new_msr;
806
+}
807
+
808
+void kvmppc_inject_interrupt_hv(struct kvm_vcpu *vcpu, int vec, u64 srr1_flags)
809
+{
810
+	inject_interrupt(vcpu, vec, srr1_flags);
811
+	kvmppc_end_cede(vcpu);
812
+}
813
+EXPORT_SYMBOL_GPL(kvmppc_inject_interrupt_hv);
814
+
815
+/*
816
+ * Is there a PRIV_DOORBELL pending for the guest (on POWER9)?
817
+ * Can we inject a Decrementer or a External interrupt?
818
+ */
819
+void kvmppc_guest_entry_inject_int(struct kvm_vcpu *vcpu)
820
+{
821
+	int ext;
822
+	unsigned long lpcr;
823
+
824
+	/* Insert EXTERNAL bit into LPCR at the MER bit position */
825
+	ext = (vcpu->arch.pending_exceptions >> BOOK3S_IRQPRIO_EXTERNAL) & 1;
826
+	lpcr = mfspr(SPRN_LPCR);
827
+	lpcr |= ext << LPCR_MER_SH;
828
+	mtspr(SPRN_LPCR, lpcr);
829
+	isync();
830
+
831
+	if (vcpu->arch.shregs.msr & MSR_EE) {
832
+		if (ext) {
833
+			inject_interrupt(vcpu, BOOK3S_INTERRUPT_EXTERNAL, 0);
834
+		} else {
835
+			long int dec = mfspr(SPRN_DEC);
836
+			if (!(lpcr & LPCR_LD))
837
+				dec = (int) dec;
838
+			if (dec < 0)
839
+				inject_interrupt(vcpu,
840
+					BOOK3S_INTERRUPT_DECREMENTER, 0);
841
+		}
842
+	}
843
+
844
+	if (vcpu->arch.doorbell_request) {
845
+		mtspr(SPRN_DPDES, 1);
846
+		vcpu->arch.vcore->dpdes = 1;
847
+		smp_wmb();
848
+		vcpu->arch.doorbell_request = 0;
849
+	}
850
+}
851
+
852
+static void flush_guest_tlb(struct kvm *kvm)
853
+{
854
+	unsigned long rb, set;
855
+
856
+	rb = PPC_BIT(52);	/* IS = 2 */
857
+	if (kvm_is_radix(kvm)) {
858
+		/* R=1 PRS=1 RIC=2 */
859
+		asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
860
+			     : : "r" (rb), "i" (1), "i" (1), "i" (2),
861
+			       "r" (0) : "memory");
862
+		for (set = 1; set < kvm->arch.tlb_sets; ++set) {
863
+			rb += PPC_BIT(51);	/* increment set number */
864
+			/* R=1 PRS=1 RIC=0 */
865
+			asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
866
+				     : : "r" (rb), "i" (1), "i" (1), "i" (0),
867
+				       "r" (0) : "memory");
868
+		}
869
+		asm volatile("ptesync": : :"memory");
870
+		// POWER9 congruence-class TLBIEL leaves ERAT. Flush it now.
871
+		asm volatile(PPC_RADIX_INVALIDATE_ERAT_GUEST : : :"memory");
872
+	} else {
873
+		for (set = 0; set < kvm->arch.tlb_sets; ++set) {
874
+			/* R=0 PRS=0 RIC=0 */
875
+			asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
876
+				     : : "r" (rb), "i" (0), "i" (0), "i" (0),
877
+				       "r" (0) : "memory");
878
+			rb += PPC_BIT(51);	/* increment set number */
879
+		}
880
+		asm volatile("ptesync": : :"memory");
881
+		// POWER9 congruence-class TLBIEL leaves ERAT. Flush it now.
882
+		if (cpu_has_feature(CPU_FTR_ARCH_300))
883
+			asm volatile(PPC_ISA_3_0_INVALIDATE_ERAT : : :"memory");
884
+	}
885
+}
886
+
887
+void kvmppc_check_need_tlb_flush(struct kvm *kvm, int pcpu,
888
+				 struct kvm_nested_guest *nested)
889
+{
890
+	cpumask_t *need_tlb_flush;
891
+
892
+	/*
893
+	 * On POWER9, individual threads can come in here, but the
894
+	 * TLB is shared between the 4 threads in a core, hence
895
+	 * invalidating on one thread invalidates for all.
896
+	 * Thus we make all 4 threads use the same bit.
897
+	 */
898
+	if (cpu_has_feature(CPU_FTR_ARCH_300))
899
+		pcpu = cpu_first_tlb_thread_sibling(pcpu);
900
+
901
+	if (nested)
902
+		need_tlb_flush = &nested->need_tlb_flush;
903
+	else
904
+		need_tlb_flush = &kvm->arch.need_tlb_flush;
905
+
906
+	if (cpumask_test_cpu(pcpu, need_tlb_flush)) {
907
+		flush_guest_tlb(kvm);
908
+
909
+		/* Clear the bit after the TLB flush */
910
+		cpumask_clear_cpu(pcpu, need_tlb_flush);
911
+	}
912
+}
913
+EXPORT_SYMBOL_GPL(kvmppc_check_need_tlb_flush);