1aa04b4ccSPaul Mackerras /* 2aa04b4ccSPaul Mackerras * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> 3aa04b4ccSPaul Mackerras * 4aa04b4ccSPaul Mackerras * This program is free software; you can redistribute it and/or modify 5aa04b4ccSPaul Mackerras * it under the terms of the GNU General Public License, version 2, as 6aa04b4ccSPaul Mackerras * published by the Free Software Foundation. 7aa04b4ccSPaul Mackerras */ 8aa04b4ccSPaul Mackerras 9441c19c8SMichael Ellerman #include <linux/cpu.h> 10aa04b4ccSPaul Mackerras #include <linux/kvm_host.h> 11aa04b4ccSPaul Mackerras #include <linux/preempt.h> 1266b15db6SPaul Gortmaker #include <linux/export.h> 13aa04b4ccSPaul Mackerras #include <linux/sched.h> 14aa04b4ccSPaul Mackerras #include <linux/spinlock.h> 15aa04b4ccSPaul Mackerras #include <linux/init.h> 16fa61a4e3SAneesh Kumar K.V #include <linux/memblock.h> 17fa61a4e3SAneesh Kumar K.V #include <linux/sizes.h> 18fc95ca72SJoonsoo Kim #include <linux/cma.h> 1990fd09f8SSam Bobroff #include <linux/bitops.h> 20aa04b4ccSPaul Mackerras 21aa04b4ccSPaul Mackerras #include <asm/cputable.h> 22aa04b4ccSPaul Mackerras #include <asm/kvm_ppc.h> 23aa04b4ccSPaul Mackerras #include <asm/kvm_book3s.h> 24e928e9cbSMichael Ellerman #include <asm/archrandom.h> 25eddb60fbSPaul Mackerras #include <asm/xics.h> 2666feed61SPaul Mackerras #include <asm/dbell.h> 2766feed61SPaul Mackerras #include <asm/cputhreads.h> 28aa04b4ccSPaul Mackerras 29fc95ca72SJoonsoo Kim #define KVM_CMA_CHUNK_ORDER 18 30fc95ca72SJoonsoo Kim 31fa61a4e3SAneesh Kumar K.V /* 32fa61a4e3SAneesh Kumar K.V * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206) 33fa61a4e3SAneesh Kumar K.V * should be power of 2. 34fa61a4e3SAneesh Kumar K.V */ 35fa61a4e3SAneesh Kumar K.V #define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */ 36fa61a4e3SAneesh Kumar K.V /* 37fa61a4e3SAneesh Kumar K.V * By default we reserve 5% of memory for hash pagetable allocation. 38fa61a4e3SAneesh Kumar K.V */ 39fa61a4e3SAneesh Kumar K.V static unsigned long kvm_cma_resv_ratio = 5; 40aa04b4ccSPaul Mackerras 41fc95ca72SJoonsoo Kim static struct cma *kvm_cma; 42fc95ca72SJoonsoo Kim 43fa61a4e3SAneesh Kumar K.V static int __init early_parse_kvm_cma_resv(char *p) 44d2a1b483SAlexander Graf { 45fa61a4e3SAneesh Kumar K.V pr_debug("%s(%s)\n", __func__, p); 46d2a1b483SAlexander Graf if (!p) 47fa61a4e3SAneesh Kumar K.V return -EINVAL; 48fa61a4e3SAneesh Kumar K.V return kstrtoul(p, 0, &kvm_cma_resv_ratio); 49d2a1b483SAlexander Graf } 50fa61a4e3SAneesh Kumar K.V early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv); 51d2a1b483SAlexander Graf 52fa61a4e3SAneesh Kumar K.V struct page *kvm_alloc_hpt(unsigned long nr_pages) 53d2a1b483SAlexander Graf { 54c04fa583SAlexey Kardashevskiy VM_BUG_ON(order_base_2(nr_pages) < KVM_CMA_CHUNK_ORDER - PAGE_SHIFT); 55fc95ca72SJoonsoo Kim 56c17b98cfSPaul Mackerras return cma_alloc(kvm_cma, nr_pages, order_base_2(HPT_ALIGN_PAGES)); 57d2a1b483SAlexander Graf } 58d2a1b483SAlexander Graf EXPORT_SYMBOL_GPL(kvm_alloc_hpt); 59d2a1b483SAlexander Graf 60fa61a4e3SAneesh Kumar K.V void kvm_release_hpt(struct page *page, unsigned long nr_pages) 61d2a1b483SAlexander Graf { 62fc95ca72SJoonsoo Kim cma_release(kvm_cma, page, nr_pages); 63d2a1b483SAlexander Graf } 64d2a1b483SAlexander Graf EXPORT_SYMBOL_GPL(kvm_release_hpt); 65d2a1b483SAlexander Graf 66fa61a4e3SAneesh Kumar K.V /** 67fa61a4e3SAneesh Kumar K.V * kvm_cma_reserve() - reserve area for kvm hash pagetable 68fa61a4e3SAneesh Kumar K.V * 69fa61a4e3SAneesh Kumar K.V * This function reserves memory from early allocator. It should be 7014ed7409SAnton Blanchard * called by arch specific code once the memblock allocator 71fa61a4e3SAneesh Kumar K.V * has been activated and all other subsystems have already allocated/reserved 72fa61a4e3SAneesh Kumar K.V * memory. 73fa61a4e3SAneesh Kumar K.V */ 74fa61a4e3SAneesh Kumar K.V void __init kvm_cma_reserve(void) 75fa61a4e3SAneesh Kumar K.V { 76fa61a4e3SAneesh Kumar K.V unsigned long align_size; 77fa61a4e3SAneesh Kumar K.V struct memblock_region *reg; 78fa61a4e3SAneesh Kumar K.V phys_addr_t selected_size = 0; 79cec26bc3SAneesh Kumar K.V 80cec26bc3SAneesh Kumar K.V /* 81cec26bc3SAneesh Kumar K.V * We need CMA reservation only when we are in HV mode 82cec26bc3SAneesh Kumar K.V */ 83cec26bc3SAneesh Kumar K.V if (!cpu_has_feature(CPU_FTR_HVMODE)) 84cec26bc3SAneesh Kumar K.V return; 85fa61a4e3SAneesh Kumar K.V /* 86fa61a4e3SAneesh Kumar K.V * We cannot use memblock_phys_mem_size() here, because 87fa61a4e3SAneesh Kumar K.V * memblock_analyze() has not been called yet. 88fa61a4e3SAneesh Kumar K.V */ 89fa61a4e3SAneesh Kumar K.V for_each_memblock(memory, reg) 90fa61a4e3SAneesh Kumar K.V selected_size += memblock_region_memory_end_pfn(reg) - 91fa61a4e3SAneesh Kumar K.V memblock_region_memory_base_pfn(reg); 92fa61a4e3SAneesh Kumar K.V 93fa61a4e3SAneesh Kumar K.V selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT; 94fa61a4e3SAneesh Kumar K.V if (selected_size) { 95fa61a4e3SAneesh Kumar K.V pr_debug("%s: reserving %ld MiB for global area\n", __func__, 96fa61a4e3SAneesh Kumar K.V (unsigned long)selected_size / SZ_1M); 97fa61a4e3SAneesh Kumar K.V align_size = HPT_ALIGN_PAGES << PAGE_SHIFT; 98c1f733aaSJoonsoo Kim cma_declare_contiguous(0, selected_size, 0, align_size, 99c1f733aaSJoonsoo Kim KVM_CMA_CHUNK_ORDER - PAGE_SHIFT, false, &kvm_cma); 100fa61a4e3SAneesh Kumar K.V } 101fa61a4e3SAneesh Kumar K.V } 102441c19c8SMichael Ellerman 103441c19c8SMichael Ellerman /* 10490fd09f8SSam Bobroff * Real-mode H_CONFER implementation. 10590fd09f8SSam Bobroff * We check if we are the only vcpu out of this virtual core 10690fd09f8SSam Bobroff * still running in the guest and not ceded. If so, we pop up 10790fd09f8SSam Bobroff * to the virtual-mode implementation; if not, just return to 10890fd09f8SSam Bobroff * the guest. 10990fd09f8SSam Bobroff */ 11090fd09f8SSam Bobroff long int kvmppc_rm_h_confer(struct kvm_vcpu *vcpu, int target, 11190fd09f8SSam Bobroff unsigned int yield_count) 11290fd09f8SSam Bobroff { 113ec257165SPaul Mackerras struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore; 114ec257165SPaul Mackerras int ptid = local_paca->kvm_hstate.ptid; 11590fd09f8SSam Bobroff int threads_running; 11690fd09f8SSam Bobroff int threads_ceded; 11790fd09f8SSam Bobroff int threads_conferring; 11890fd09f8SSam Bobroff u64 stop = get_tb() + 10 * tb_ticks_per_usec; 11990fd09f8SSam Bobroff int rv = H_SUCCESS; /* => don't yield */ 12090fd09f8SSam Bobroff 121ec257165SPaul Mackerras set_bit(ptid, &vc->conferring_threads); 1227d6c40daSPaul Mackerras while ((get_tb() < stop) && !VCORE_IS_EXITING(vc)) { 1237d6c40daSPaul Mackerras threads_running = VCORE_ENTRY_MAP(vc); 1247d6c40daSPaul Mackerras threads_ceded = vc->napping_threads; 1257d6c40daSPaul Mackerras threads_conferring = vc->conferring_threads; 1267d6c40daSPaul Mackerras if ((threads_ceded | threads_conferring) == threads_running) { 12790fd09f8SSam Bobroff rv = H_TOO_HARD; /* => do yield */ 12890fd09f8SSam Bobroff break; 12990fd09f8SSam Bobroff } 13090fd09f8SSam Bobroff } 131ec257165SPaul Mackerras clear_bit(ptid, &vc->conferring_threads); 13290fd09f8SSam Bobroff return rv; 13390fd09f8SSam Bobroff } 13490fd09f8SSam Bobroff 13590fd09f8SSam Bobroff /* 136441c19c8SMichael Ellerman * When running HV mode KVM we need to block certain operations while KVM VMs 137441c19c8SMichael Ellerman * exist in the system. We use a counter of VMs to track this. 138441c19c8SMichael Ellerman * 139441c19c8SMichael Ellerman * One of the operations we need to block is onlining of secondaries, so we 140441c19c8SMichael Ellerman * protect hv_vm_count with get/put_online_cpus(). 141441c19c8SMichael Ellerman */ 142441c19c8SMichael Ellerman static atomic_t hv_vm_count; 143441c19c8SMichael Ellerman 144441c19c8SMichael Ellerman void kvm_hv_vm_activated(void) 145441c19c8SMichael Ellerman { 146441c19c8SMichael Ellerman get_online_cpus(); 147441c19c8SMichael Ellerman atomic_inc(&hv_vm_count); 148441c19c8SMichael Ellerman put_online_cpus(); 149441c19c8SMichael Ellerman } 150441c19c8SMichael Ellerman EXPORT_SYMBOL_GPL(kvm_hv_vm_activated); 151441c19c8SMichael Ellerman 152441c19c8SMichael Ellerman void kvm_hv_vm_deactivated(void) 153441c19c8SMichael Ellerman { 154441c19c8SMichael Ellerman get_online_cpus(); 155441c19c8SMichael Ellerman atomic_dec(&hv_vm_count); 156441c19c8SMichael Ellerman put_online_cpus(); 157441c19c8SMichael Ellerman } 158441c19c8SMichael Ellerman EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated); 159441c19c8SMichael Ellerman 160441c19c8SMichael Ellerman bool kvm_hv_mode_active(void) 161441c19c8SMichael Ellerman { 162441c19c8SMichael Ellerman return atomic_read(&hv_vm_count) != 0; 163441c19c8SMichael Ellerman } 164ae2113a4SPaul Mackerras 165ae2113a4SPaul Mackerras extern int hcall_real_table[], hcall_real_table_end[]; 166ae2113a4SPaul Mackerras 167ae2113a4SPaul Mackerras int kvmppc_hcall_impl_hv_realmode(unsigned long cmd) 168ae2113a4SPaul Mackerras { 169ae2113a4SPaul Mackerras cmd /= 4; 170ae2113a4SPaul Mackerras if (cmd < hcall_real_table_end - hcall_real_table && 171ae2113a4SPaul Mackerras hcall_real_table[cmd]) 172ae2113a4SPaul Mackerras return 1; 173ae2113a4SPaul Mackerras 174ae2113a4SPaul Mackerras return 0; 175ae2113a4SPaul Mackerras } 176ae2113a4SPaul Mackerras EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode); 177e928e9cbSMichael Ellerman 178e928e9cbSMichael Ellerman int kvmppc_hwrng_present(void) 179e928e9cbSMichael Ellerman { 180e928e9cbSMichael Ellerman return powernv_hwrng_present(); 181e928e9cbSMichael Ellerman } 182e928e9cbSMichael Ellerman EXPORT_SYMBOL_GPL(kvmppc_hwrng_present); 183e928e9cbSMichael Ellerman 184e928e9cbSMichael Ellerman long kvmppc_h_random(struct kvm_vcpu *vcpu) 185e928e9cbSMichael Ellerman { 186e928e9cbSMichael Ellerman if (powernv_get_random_real_mode(&vcpu->arch.gpr[4])) 187e928e9cbSMichael Ellerman return H_SUCCESS; 188e928e9cbSMichael Ellerman 189e928e9cbSMichael Ellerman return H_HARDWARE; 190e928e9cbSMichael Ellerman } 191eddb60fbSPaul Mackerras 192eddb60fbSPaul Mackerras static inline void rm_writeb(unsigned long paddr, u8 val) 193eddb60fbSPaul Mackerras { 194eddb60fbSPaul Mackerras __asm__ __volatile__("stbcix %0,0,%1" 195eddb60fbSPaul Mackerras : : "r" (val), "r" (paddr) : "memory"); 196eddb60fbSPaul Mackerras } 197eddb60fbSPaul Mackerras 198eddb60fbSPaul Mackerras /* 19966feed61SPaul Mackerras * Send an interrupt or message to another CPU. 200eddb60fbSPaul Mackerras * This can only be called in real mode. 201eddb60fbSPaul Mackerras * The caller needs to include any barrier needed to order writes 202eddb60fbSPaul Mackerras * to memory vs. the IPI/message. 203eddb60fbSPaul Mackerras */ 204eddb60fbSPaul Mackerras void kvmhv_rm_send_ipi(int cpu) 205eddb60fbSPaul Mackerras { 206eddb60fbSPaul Mackerras unsigned long xics_phys; 207eddb60fbSPaul Mackerras 20866feed61SPaul Mackerras /* On POWER8 for IPIs to threads in the same core, use msgsnd */ 20966feed61SPaul Mackerras if (cpu_has_feature(CPU_FTR_ARCH_207S) && 21066feed61SPaul Mackerras cpu_first_thread_sibling(cpu) == 21166feed61SPaul Mackerras cpu_first_thread_sibling(raw_smp_processor_id())) { 21266feed61SPaul Mackerras unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER); 21366feed61SPaul Mackerras msg |= cpu_thread_in_core(cpu); 21466feed61SPaul Mackerras __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); 21566feed61SPaul Mackerras return; 21666feed61SPaul Mackerras } 21766feed61SPaul Mackerras 21866feed61SPaul Mackerras /* Else poke the target with an IPI */ 219eddb60fbSPaul Mackerras xics_phys = paca[cpu].kvm_hstate.xics_phys; 220eddb60fbSPaul Mackerras rm_writeb(xics_phys + XICS_MFRR, IPI_PRIORITY); 221eddb60fbSPaul Mackerras } 222eddb60fbSPaul Mackerras 223eddb60fbSPaul Mackerras /* 224eddb60fbSPaul Mackerras * The following functions are called from the assembly code 225eddb60fbSPaul Mackerras * in book3s_hv_rmhandlers.S. 226eddb60fbSPaul Mackerras */ 227eddb60fbSPaul Mackerras static void kvmhv_interrupt_vcore(struct kvmppc_vcore *vc, int active) 228eddb60fbSPaul Mackerras { 229eddb60fbSPaul Mackerras int cpu = vc->pcpu; 230eddb60fbSPaul Mackerras 231eddb60fbSPaul Mackerras /* Order setting of exit map vs. msgsnd/IPI */ 232eddb60fbSPaul Mackerras smp_mb(); 233eddb60fbSPaul Mackerras for (; active; active >>= 1, ++cpu) 234eddb60fbSPaul Mackerras if (active & 1) 235eddb60fbSPaul Mackerras kvmhv_rm_send_ipi(cpu); 236eddb60fbSPaul Mackerras } 237eddb60fbSPaul Mackerras 238eddb60fbSPaul Mackerras void kvmhv_commence_exit(int trap) 239eddb60fbSPaul Mackerras { 240eddb60fbSPaul Mackerras struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore; 241eddb60fbSPaul Mackerras int ptid = local_paca->kvm_hstate.ptid; 242eddb60fbSPaul Mackerras int me, ee; 243eddb60fbSPaul Mackerras 244eddb60fbSPaul Mackerras /* Set our bit in the threads-exiting-guest map in the 0xff00 245eddb60fbSPaul Mackerras bits of vcore->entry_exit_map */ 246eddb60fbSPaul Mackerras me = 0x100 << ptid; 247eddb60fbSPaul Mackerras do { 248eddb60fbSPaul Mackerras ee = vc->entry_exit_map; 249eddb60fbSPaul Mackerras } while (cmpxchg(&vc->entry_exit_map, ee, ee | me) != ee); 250eddb60fbSPaul Mackerras 251eddb60fbSPaul Mackerras /* Are we the first here? */ 252eddb60fbSPaul Mackerras if ((ee >> 8) != 0) 253eddb60fbSPaul Mackerras return; 254eddb60fbSPaul Mackerras 255eddb60fbSPaul Mackerras /* 256eddb60fbSPaul Mackerras * Trigger the other threads in this vcore to exit the guest. 257eddb60fbSPaul Mackerras * If this is a hypervisor decrementer interrupt then they 258eddb60fbSPaul Mackerras * will be already on their way out of the guest. 259eddb60fbSPaul Mackerras */ 260eddb60fbSPaul Mackerras if (trap != BOOK3S_INTERRUPT_HV_DECREMENTER) 261eddb60fbSPaul Mackerras kvmhv_interrupt_vcore(vc, ee & ~(1 << ptid)); 262eddb60fbSPaul Mackerras } 263