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> 2837f55d30SSuresh Warrier #include <asm/io.h> 29f725758bSPaul Mackerras #include <asm/opal.h> 30e2702871SPaul Mackerras #include <asm/smp.h> 31aa04b4ccSPaul Mackerras 32fc95ca72SJoonsoo Kim #define KVM_CMA_CHUNK_ORDER 18 33fc95ca72SJoonsoo Kim 34fa61a4e3SAneesh Kumar K.V /* 35fa61a4e3SAneesh Kumar K.V * Hash page table alignment on newer cpus(CPU_FTR_ARCH_206) 36fa61a4e3SAneesh Kumar K.V * should be power of 2. 37fa61a4e3SAneesh Kumar K.V */ 38fa61a4e3SAneesh Kumar K.V #define HPT_ALIGN_PAGES ((1 << 18) >> PAGE_SHIFT) /* 256k */ 39fa61a4e3SAneesh Kumar K.V /* 40fa61a4e3SAneesh Kumar K.V * By default we reserve 5% of memory for hash pagetable allocation. 41fa61a4e3SAneesh Kumar K.V */ 42fa61a4e3SAneesh Kumar K.V static unsigned long kvm_cma_resv_ratio = 5; 43aa04b4ccSPaul Mackerras 44fc95ca72SJoonsoo Kim static struct cma *kvm_cma; 45fc95ca72SJoonsoo Kim 46fa61a4e3SAneesh Kumar K.V static int __init early_parse_kvm_cma_resv(char *p) 47d2a1b483SAlexander Graf { 48fa61a4e3SAneesh Kumar K.V pr_debug("%s(%s)\n", __func__, p); 49d2a1b483SAlexander Graf if (!p) 50fa61a4e3SAneesh Kumar K.V return -EINVAL; 51fa61a4e3SAneesh Kumar K.V return kstrtoul(p, 0, &kvm_cma_resv_ratio); 52d2a1b483SAlexander Graf } 53fa61a4e3SAneesh Kumar K.V early_param("kvm_cma_resv_ratio", early_parse_kvm_cma_resv); 54d2a1b483SAlexander Graf 55db9a290dSDavid Gibson struct page *kvm_alloc_hpt_cma(unsigned long nr_pages) 56d2a1b483SAlexander Graf { 57c04fa583SAlexey Kardashevskiy VM_BUG_ON(order_base_2(nr_pages) < KVM_CMA_CHUNK_ORDER - PAGE_SHIFT); 58fc95ca72SJoonsoo Kim 59e2f466e3SLucas Stach return cma_alloc(kvm_cma, nr_pages, order_base_2(HPT_ALIGN_PAGES), 60e2f466e3SLucas Stach GFP_KERNEL); 61d2a1b483SAlexander Graf } 62db9a290dSDavid Gibson EXPORT_SYMBOL_GPL(kvm_alloc_hpt_cma); 63d2a1b483SAlexander Graf 64db9a290dSDavid Gibson void kvm_free_hpt_cma(struct page *page, unsigned long nr_pages) 65d2a1b483SAlexander Graf { 66fc95ca72SJoonsoo Kim cma_release(kvm_cma, page, nr_pages); 67d2a1b483SAlexander Graf } 68db9a290dSDavid Gibson EXPORT_SYMBOL_GPL(kvm_free_hpt_cma); 69d2a1b483SAlexander Graf 70fa61a4e3SAneesh Kumar K.V /** 71fa61a4e3SAneesh Kumar K.V * kvm_cma_reserve() - reserve area for kvm hash pagetable 72fa61a4e3SAneesh Kumar K.V * 73fa61a4e3SAneesh Kumar K.V * This function reserves memory from early allocator. It should be 7414ed7409SAnton Blanchard * called by arch specific code once the memblock allocator 75fa61a4e3SAneesh Kumar K.V * has been activated and all other subsystems have already allocated/reserved 76fa61a4e3SAneesh Kumar K.V * memory. 77fa61a4e3SAneesh Kumar K.V */ 78fa61a4e3SAneesh Kumar K.V void __init kvm_cma_reserve(void) 79fa61a4e3SAneesh Kumar K.V { 80fa61a4e3SAneesh Kumar K.V unsigned long align_size; 81fa61a4e3SAneesh Kumar K.V struct memblock_region *reg; 82fa61a4e3SAneesh Kumar K.V phys_addr_t selected_size = 0; 83cec26bc3SAneesh Kumar K.V 84cec26bc3SAneesh Kumar K.V /* 85cec26bc3SAneesh Kumar K.V * We need CMA reservation only when we are in HV mode 86cec26bc3SAneesh Kumar K.V */ 87cec26bc3SAneesh Kumar K.V if (!cpu_has_feature(CPU_FTR_HVMODE)) 88cec26bc3SAneesh Kumar K.V return; 89fa61a4e3SAneesh Kumar K.V /* 90fa61a4e3SAneesh Kumar K.V * We cannot use memblock_phys_mem_size() here, because 91fa61a4e3SAneesh Kumar K.V * memblock_analyze() has not been called yet. 92fa61a4e3SAneesh Kumar K.V */ 93fa61a4e3SAneesh Kumar K.V for_each_memblock(memory, reg) 94fa61a4e3SAneesh Kumar K.V selected_size += memblock_region_memory_end_pfn(reg) - 95fa61a4e3SAneesh Kumar K.V memblock_region_memory_base_pfn(reg); 96fa61a4e3SAneesh Kumar K.V 97fa61a4e3SAneesh Kumar K.V selected_size = (selected_size * kvm_cma_resv_ratio / 100) << PAGE_SHIFT; 98fa61a4e3SAneesh Kumar K.V if (selected_size) { 99fa61a4e3SAneesh Kumar K.V pr_debug("%s: reserving %ld MiB for global area\n", __func__, 100fa61a4e3SAneesh Kumar K.V (unsigned long)selected_size / SZ_1M); 101fa61a4e3SAneesh Kumar K.V align_size = HPT_ALIGN_PAGES << PAGE_SHIFT; 102c1f733aaSJoonsoo Kim cma_declare_contiguous(0, selected_size, 0, align_size, 103f318dd08SLaura Abbott KVM_CMA_CHUNK_ORDER - PAGE_SHIFT, false, "kvm_cma", 104f318dd08SLaura Abbott &kvm_cma); 105fa61a4e3SAneesh Kumar K.V } 106fa61a4e3SAneesh Kumar K.V } 107441c19c8SMichael Ellerman 108441c19c8SMichael Ellerman /* 10990fd09f8SSam Bobroff * Real-mode H_CONFER implementation. 11090fd09f8SSam Bobroff * We check if we are the only vcpu out of this virtual core 11190fd09f8SSam Bobroff * still running in the guest and not ceded. If so, we pop up 11290fd09f8SSam Bobroff * to the virtual-mode implementation; if not, just return to 11390fd09f8SSam Bobroff * the guest. 11490fd09f8SSam Bobroff */ 11590fd09f8SSam Bobroff long int kvmppc_rm_h_confer(struct kvm_vcpu *vcpu, int target, 11690fd09f8SSam Bobroff unsigned int yield_count) 11790fd09f8SSam Bobroff { 118ec257165SPaul Mackerras struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore; 119ec257165SPaul Mackerras int ptid = local_paca->kvm_hstate.ptid; 12090fd09f8SSam Bobroff int threads_running; 12190fd09f8SSam Bobroff int threads_ceded; 12290fd09f8SSam Bobroff int threads_conferring; 12390fd09f8SSam Bobroff u64 stop = get_tb() + 10 * tb_ticks_per_usec; 12490fd09f8SSam Bobroff int rv = H_SUCCESS; /* => don't yield */ 12590fd09f8SSam Bobroff 126ec257165SPaul Mackerras set_bit(ptid, &vc->conferring_threads); 1277d6c40daSPaul Mackerras while ((get_tb() < stop) && !VCORE_IS_EXITING(vc)) { 1287d6c40daSPaul Mackerras threads_running = VCORE_ENTRY_MAP(vc); 1297d6c40daSPaul Mackerras threads_ceded = vc->napping_threads; 1307d6c40daSPaul Mackerras threads_conferring = vc->conferring_threads; 1317d6c40daSPaul Mackerras if ((threads_ceded | threads_conferring) == threads_running) { 13290fd09f8SSam Bobroff rv = H_TOO_HARD; /* => do yield */ 13390fd09f8SSam Bobroff break; 13490fd09f8SSam Bobroff } 13590fd09f8SSam Bobroff } 136ec257165SPaul Mackerras clear_bit(ptid, &vc->conferring_threads); 13790fd09f8SSam Bobroff return rv; 13890fd09f8SSam Bobroff } 13990fd09f8SSam Bobroff 14090fd09f8SSam Bobroff /* 141441c19c8SMichael Ellerman * When running HV mode KVM we need to block certain operations while KVM VMs 142441c19c8SMichael Ellerman * exist in the system. We use a counter of VMs to track this. 143441c19c8SMichael Ellerman * 144441c19c8SMichael Ellerman * One of the operations we need to block is onlining of secondaries, so we 145441c19c8SMichael Ellerman * protect hv_vm_count with get/put_online_cpus(). 146441c19c8SMichael Ellerman */ 147441c19c8SMichael Ellerman static atomic_t hv_vm_count; 148441c19c8SMichael Ellerman 149441c19c8SMichael Ellerman void kvm_hv_vm_activated(void) 150441c19c8SMichael Ellerman { 151441c19c8SMichael Ellerman get_online_cpus(); 152441c19c8SMichael Ellerman atomic_inc(&hv_vm_count); 153441c19c8SMichael Ellerman put_online_cpus(); 154441c19c8SMichael Ellerman } 155441c19c8SMichael Ellerman EXPORT_SYMBOL_GPL(kvm_hv_vm_activated); 156441c19c8SMichael Ellerman 157441c19c8SMichael Ellerman void kvm_hv_vm_deactivated(void) 158441c19c8SMichael Ellerman { 159441c19c8SMichael Ellerman get_online_cpus(); 160441c19c8SMichael Ellerman atomic_dec(&hv_vm_count); 161441c19c8SMichael Ellerman put_online_cpus(); 162441c19c8SMichael Ellerman } 163441c19c8SMichael Ellerman EXPORT_SYMBOL_GPL(kvm_hv_vm_deactivated); 164441c19c8SMichael Ellerman 165441c19c8SMichael Ellerman bool kvm_hv_mode_active(void) 166441c19c8SMichael Ellerman { 167441c19c8SMichael Ellerman return atomic_read(&hv_vm_count) != 0; 168441c19c8SMichael Ellerman } 169ae2113a4SPaul Mackerras 170ae2113a4SPaul Mackerras extern int hcall_real_table[], hcall_real_table_end[]; 171ae2113a4SPaul Mackerras 172ae2113a4SPaul Mackerras int kvmppc_hcall_impl_hv_realmode(unsigned long cmd) 173ae2113a4SPaul Mackerras { 174ae2113a4SPaul Mackerras cmd /= 4; 175ae2113a4SPaul Mackerras if (cmd < hcall_real_table_end - hcall_real_table && 176ae2113a4SPaul Mackerras hcall_real_table[cmd]) 177ae2113a4SPaul Mackerras return 1; 178ae2113a4SPaul Mackerras 179ae2113a4SPaul Mackerras return 0; 180ae2113a4SPaul Mackerras } 181ae2113a4SPaul Mackerras EXPORT_SYMBOL_GPL(kvmppc_hcall_impl_hv_realmode); 182e928e9cbSMichael Ellerman 183e928e9cbSMichael Ellerman int kvmppc_hwrng_present(void) 184e928e9cbSMichael Ellerman { 185e928e9cbSMichael Ellerman return powernv_hwrng_present(); 186e928e9cbSMichael Ellerman } 187e928e9cbSMichael Ellerman EXPORT_SYMBOL_GPL(kvmppc_hwrng_present); 188e928e9cbSMichael Ellerman 189e928e9cbSMichael Ellerman long kvmppc_h_random(struct kvm_vcpu *vcpu) 190e928e9cbSMichael Ellerman { 191e928e9cbSMichael Ellerman if (powernv_get_random_real_mode(&vcpu->arch.gpr[4])) 192e928e9cbSMichael Ellerman return H_SUCCESS; 193e928e9cbSMichael Ellerman 194e928e9cbSMichael Ellerman return H_HARDWARE; 195e928e9cbSMichael Ellerman } 196eddb60fbSPaul Mackerras 197eddb60fbSPaul Mackerras static inline void rm_writeb(unsigned long paddr, u8 val) 198eddb60fbSPaul Mackerras { 199eddb60fbSPaul Mackerras __asm__ __volatile__("stbcix %0,0,%1" 200eddb60fbSPaul Mackerras : : "r" (val), "r" (paddr) : "memory"); 201eddb60fbSPaul Mackerras } 202eddb60fbSPaul Mackerras 203eddb60fbSPaul Mackerras /* 20466feed61SPaul Mackerras * Send an interrupt or message to another CPU. 205eddb60fbSPaul Mackerras * The caller needs to include any barrier needed to order writes 206eddb60fbSPaul Mackerras * to memory vs. the IPI/message. 207eddb60fbSPaul Mackerras */ 208eddb60fbSPaul Mackerras void kvmhv_rm_send_ipi(int cpu) 209eddb60fbSPaul Mackerras { 210eddb60fbSPaul Mackerras unsigned long xics_phys; 2111704a81cSPaul Mackerras unsigned long msg = PPC_DBELL_TYPE(PPC_DBELL_SERVER); 212eddb60fbSPaul Mackerras 2131704a81cSPaul Mackerras /* On POWER9 we can use msgsnd for any destination cpu. */ 2141704a81cSPaul Mackerras if (cpu_has_feature(CPU_FTR_ARCH_300)) { 2151704a81cSPaul Mackerras msg |= get_hard_smp_processor_id(cpu); 2161704a81cSPaul Mackerras __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); 2171704a81cSPaul Mackerras return; 2181704a81cSPaul Mackerras } 2191704a81cSPaul Mackerras /* On POWER8 for IPIs to threads in the same core, use msgsnd. */ 22066feed61SPaul Mackerras if (cpu_has_feature(CPU_FTR_ARCH_207S) && 22166feed61SPaul Mackerras cpu_first_thread_sibling(cpu) == 22266feed61SPaul Mackerras cpu_first_thread_sibling(raw_smp_processor_id())) { 22366feed61SPaul Mackerras msg |= cpu_thread_in_core(cpu); 22466feed61SPaul Mackerras __asm__ __volatile__ (PPC_MSGSND(%0) : : "r" (msg)); 22566feed61SPaul Mackerras return; 22666feed61SPaul Mackerras } 22766feed61SPaul Mackerras 22866feed61SPaul Mackerras /* Else poke the target with an IPI */ 229eddb60fbSPaul Mackerras xics_phys = paca[cpu].kvm_hstate.xics_phys; 230ab9bad0eSBenjamin Herrenschmidt if (xics_phys) 231eddb60fbSPaul Mackerras rm_writeb(xics_phys + XICS_MFRR, IPI_PRIORITY); 232f725758bSPaul Mackerras else 233ab9bad0eSBenjamin Herrenschmidt opal_int_set_mfrr(get_hard_smp_processor_id(cpu), IPI_PRIORITY); 234eddb60fbSPaul Mackerras } 235eddb60fbSPaul Mackerras 236eddb60fbSPaul Mackerras /* 237eddb60fbSPaul Mackerras * The following functions are called from the assembly code 238eddb60fbSPaul Mackerras * in book3s_hv_rmhandlers.S. 239eddb60fbSPaul Mackerras */ 240eddb60fbSPaul Mackerras static void kvmhv_interrupt_vcore(struct kvmppc_vcore *vc, int active) 241eddb60fbSPaul Mackerras { 242eddb60fbSPaul Mackerras int cpu = vc->pcpu; 243eddb60fbSPaul Mackerras 244eddb60fbSPaul Mackerras /* Order setting of exit map vs. msgsnd/IPI */ 245eddb60fbSPaul Mackerras smp_mb(); 246eddb60fbSPaul Mackerras for (; active; active >>= 1, ++cpu) 247eddb60fbSPaul Mackerras if (active & 1) 248eddb60fbSPaul Mackerras kvmhv_rm_send_ipi(cpu); 249eddb60fbSPaul Mackerras } 250eddb60fbSPaul Mackerras 251eddb60fbSPaul Mackerras void kvmhv_commence_exit(int trap) 252eddb60fbSPaul Mackerras { 253eddb60fbSPaul Mackerras struct kvmppc_vcore *vc = local_paca->kvm_hstate.kvm_vcore; 254eddb60fbSPaul Mackerras int ptid = local_paca->kvm_hstate.ptid; 255b4deba5cSPaul Mackerras struct kvm_split_mode *sip = local_paca->kvm_hstate.kvm_split_mode; 256b4deba5cSPaul Mackerras int me, ee, i; 257eddb60fbSPaul Mackerras 258eddb60fbSPaul Mackerras /* Set our bit in the threads-exiting-guest map in the 0xff00 259eddb60fbSPaul Mackerras bits of vcore->entry_exit_map */ 260eddb60fbSPaul Mackerras me = 0x100 << ptid; 261eddb60fbSPaul Mackerras do { 262eddb60fbSPaul Mackerras ee = vc->entry_exit_map; 263eddb60fbSPaul Mackerras } while (cmpxchg(&vc->entry_exit_map, ee, ee | me) != ee); 264eddb60fbSPaul Mackerras 265eddb60fbSPaul Mackerras /* Are we the first here? */ 266eddb60fbSPaul Mackerras if ((ee >> 8) != 0) 267eddb60fbSPaul Mackerras return; 268eddb60fbSPaul Mackerras 269eddb60fbSPaul Mackerras /* 270eddb60fbSPaul Mackerras * Trigger the other threads in this vcore to exit the guest. 271eddb60fbSPaul Mackerras * If this is a hypervisor decrementer interrupt then they 272eddb60fbSPaul Mackerras * will be already on their way out of the guest. 273eddb60fbSPaul Mackerras */ 274eddb60fbSPaul Mackerras if (trap != BOOK3S_INTERRUPT_HV_DECREMENTER) 275eddb60fbSPaul Mackerras kvmhv_interrupt_vcore(vc, ee & ~(1 << ptid)); 276b4deba5cSPaul Mackerras 277b4deba5cSPaul Mackerras /* 278b4deba5cSPaul Mackerras * If we are doing dynamic micro-threading, interrupt the other 279b4deba5cSPaul Mackerras * subcores to pull them out of their guests too. 280b4deba5cSPaul Mackerras */ 281b4deba5cSPaul Mackerras if (!sip) 282b4deba5cSPaul Mackerras return; 283b4deba5cSPaul Mackerras 284b4deba5cSPaul Mackerras for (i = 0; i < MAX_SUBCORES; ++i) { 285b4deba5cSPaul Mackerras vc = sip->master_vcs[i]; 286b4deba5cSPaul Mackerras if (!vc) 287b4deba5cSPaul Mackerras break; 288b4deba5cSPaul Mackerras do { 289b4deba5cSPaul Mackerras ee = vc->entry_exit_map; 290b4deba5cSPaul Mackerras /* Already asked to exit? */ 291b4deba5cSPaul Mackerras if ((ee >> 8) != 0) 292b4deba5cSPaul Mackerras break; 293b4deba5cSPaul Mackerras } while (cmpxchg(&vc->entry_exit_map, ee, 294b4deba5cSPaul Mackerras ee | VCORE_EXIT_REQ) != ee); 295b4deba5cSPaul Mackerras if ((ee >> 8) == 0) 296b4deba5cSPaul Mackerras kvmhv_interrupt_vcore(vc, ee); 297b4deba5cSPaul Mackerras } 298eddb60fbSPaul Mackerras } 29979b6c247SSuresh Warrier 30079b6c247SSuresh Warrier struct kvmppc_host_rm_ops *kvmppc_host_rm_ops_hv; 30179b6c247SSuresh Warrier EXPORT_SYMBOL_GPL(kvmppc_host_rm_ops_hv); 30237f55d30SSuresh Warrier 303e3c13e56SSuresh Warrier #ifdef CONFIG_KVM_XICS 304e3c13e56SSuresh Warrier static struct kvmppc_irq_map *get_irqmap(struct kvmppc_passthru_irqmap *pimap, 305e3c13e56SSuresh Warrier u32 xisr) 306e3c13e56SSuresh Warrier { 307e3c13e56SSuresh Warrier int i; 308e3c13e56SSuresh Warrier 309e3c13e56SSuresh Warrier /* 310e3c13e56SSuresh Warrier * We access the mapped array here without a lock. That 311e3c13e56SSuresh Warrier * is safe because we never reduce the number of entries 312e3c13e56SSuresh Warrier * in the array and we never change the v_hwirq field of 313e3c13e56SSuresh Warrier * an entry once it is set. 314e3c13e56SSuresh Warrier * 315e3c13e56SSuresh Warrier * We have also carefully ordered the stores in the writer 316e3c13e56SSuresh Warrier * and the loads here in the reader, so that if we find a matching 317e3c13e56SSuresh Warrier * hwirq here, the associated GSI and irq_desc fields are valid. 318e3c13e56SSuresh Warrier */ 319e3c13e56SSuresh Warrier for (i = 0; i < pimap->n_mapped; i++) { 320e3c13e56SSuresh Warrier if (xisr == pimap->mapped[i].r_hwirq) { 321e3c13e56SSuresh Warrier /* 322e3c13e56SSuresh Warrier * Order subsequent reads in the caller to serialize 323e3c13e56SSuresh Warrier * with the writer. 324e3c13e56SSuresh Warrier */ 325e3c13e56SSuresh Warrier smp_rmb(); 326e3c13e56SSuresh Warrier return &pimap->mapped[i]; 327e3c13e56SSuresh Warrier } 328e3c13e56SSuresh Warrier } 329e3c13e56SSuresh Warrier return NULL; 330e3c13e56SSuresh Warrier } 331e3c13e56SSuresh Warrier 332e3c13e56SSuresh Warrier /* 333e3c13e56SSuresh Warrier * If we have an interrupt that's not an IPI, check if we have a 334e3c13e56SSuresh Warrier * passthrough adapter and if so, check if this external interrupt 335e3c13e56SSuresh Warrier * is for the adapter. 336e3c13e56SSuresh Warrier * We will attempt to deliver the IRQ directly to the target VCPU's 337e3c13e56SSuresh Warrier * ICP, the virtual ICP (based on affinity - the xive value in ICS). 338e3c13e56SSuresh Warrier * 339e3c13e56SSuresh Warrier * If the delivery fails or if this is not for a passthrough adapter, 340e3c13e56SSuresh Warrier * return to the host to handle this interrupt. We earlier 341e3c13e56SSuresh Warrier * saved a copy of the XIRR in the PACA, it will be picked up by 342e3c13e56SSuresh Warrier * the host ICP driver. 343e3c13e56SSuresh Warrier */ 344f725758bSPaul Mackerras static int kvmppc_check_passthru(u32 xisr, __be32 xirr, bool *again) 345e3c13e56SSuresh Warrier { 346e3c13e56SSuresh Warrier struct kvmppc_passthru_irqmap *pimap; 347e3c13e56SSuresh Warrier struct kvmppc_irq_map *irq_map; 348e3c13e56SSuresh Warrier struct kvm_vcpu *vcpu; 349e3c13e56SSuresh Warrier 350e3c13e56SSuresh Warrier vcpu = local_paca->kvm_hstate.kvm_vcpu; 351e3c13e56SSuresh Warrier if (!vcpu) 352e3c13e56SSuresh Warrier return 1; 353e3c13e56SSuresh Warrier pimap = kvmppc_get_passthru_irqmap(vcpu->kvm); 354e3c13e56SSuresh Warrier if (!pimap) 355e3c13e56SSuresh Warrier return 1; 356e3c13e56SSuresh Warrier irq_map = get_irqmap(pimap, xisr); 357e3c13e56SSuresh Warrier if (!irq_map) 358e3c13e56SSuresh Warrier return 1; 359e3c13e56SSuresh Warrier 360e3c13e56SSuresh Warrier /* We're handling this interrupt, generic code doesn't need to */ 361e3c13e56SSuresh Warrier local_paca->kvm_hstate.saved_xirr = 0; 362e3c13e56SSuresh Warrier 363f725758bSPaul Mackerras return kvmppc_deliver_irq_passthru(vcpu, xirr, irq_map, pimap, again); 364e3c13e56SSuresh Warrier } 365e3c13e56SSuresh Warrier 366e3c13e56SSuresh Warrier #else 367e2702871SPaul Mackerras static inline int kvmppc_check_passthru(u32 xisr, __be32 xirr, bool *again) 368e3c13e56SSuresh Warrier { 369e3c13e56SSuresh Warrier return 1; 370e3c13e56SSuresh Warrier } 371e3c13e56SSuresh Warrier #endif 372e3c13e56SSuresh Warrier 37337f55d30SSuresh Warrier /* 37437f55d30SSuresh Warrier * Determine what sort of external interrupt is pending (if any). 37537f55d30SSuresh Warrier * Returns: 37637f55d30SSuresh Warrier * 0 if no interrupt is pending 37737f55d30SSuresh Warrier * 1 if an interrupt is pending that needs to be handled by the host 378f7af5209SSuresh Warrier * 2 Passthrough that needs completion in the host 37937f55d30SSuresh Warrier * -1 if there was a guest wakeup IPI (which has now been cleared) 380e3c13e56SSuresh Warrier * -2 if there is PCI passthrough external interrupt that was handled 38137f55d30SSuresh Warrier */ 382f725758bSPaul Mackerras static long kvmppc_read_one_intr(bool *again); 38337f55d30SSuresh Warrier 38437f55d30SSuresh Warrier long kvmppc_read_intr(void) 38537f55d30SSuresh Warrier { 386f725758bSPaul Mackerras long ret = 0; 387f725758bSPaul Mackerras long rc; 388f725758bSPaul Mackerras bool again; 389f725758bSPaul Mackerras 390f725758bSPaul Mackerras do { 391f725758bSPaul Mackerras again = false; 392f725758bSPaul Mackerras rc = kvmppc_read_one_intr(&again); 393f725758bSPaul Mackerras if (rc && (ret == 0 || rc > ret)) 394f725758bSPaul Mackerras ret = rc; 395f725758bSPaul Mackerras } while (again); 396f725758bSPaul Mackerras return ret; 397f725758bSPaul Mackerras } 398f725758bSPaul Mackerras 399f725758bSPaul Mackerras static long kvmppc_read_one_intr(bool *again) 400f725758bSPaul Mackerras { 40137f55d30SSuresh Warrier unsigned long xics_phys; 40237f55d30SSuresh Warrier u32 h_xirr; 40337f55d30SSuresh Warrier __be32 xirr; 40437f55d30SSuresh Warrier u32 xisr; 40537f55d30SSuresh Warrier u8 host_ipi; 406f725758bSPaul Mackerras int64_t rc; 40737f55d30SSuresh Warrier 40837f55d30SSuresh Warrier /* see if a host IPI is pending */ 40937f55d30SSuresh Warrier host_ipi = local_paca->kvm_hstate.host_ipi; 41037f55d30SSuresh Warrier if (host_ipi) 41137f55d30SSuresh Warrier return 1; 41237f55d30SSuresh Warrier 41337f55d30SSuresh Warrier /* Now read the interrupt from the ICP */ 41437f55d30SSuresh Warrier xics_phys = local_paca->kvm_hstate.xics_phys; 41553af3ba2SPaul Mackerras rc = 0; 416ab9bad0eSBenjamin Herrenschmidt if (!xics_phys) 41753af3ba2SPaul Mackerras rc = opal_int_get_xirr(&xirr, false); 41853af3ba2SPaul Mackerras else 41953af3ba2SPaul Mackerras xirr = _lwzcix(xics_phys + XICS_XIRR); 420f725758bSPaul Mackerras if (rc < 0) 42137f55d30SSuresh Warrier return 1; 42237f55d30SSuresh Warrier 42337f55d30SSuresh Warrier /* 42437f55d30SSuresh Warrier * Save XIRR for later. Since we get control in reverse endian 42537f55d30SSuresh Warrier * on LE systems, save it byte reversed and fetch it back in 42637f55d30SSuresh Warrier * host endian. Note that xirr is the value read from the 42737f55d30SSuresh Warrier * XIRR register, while h_xirr is the host endian version. 42837f55d30SSuresh Warrier */ 42937f55d30SSuresh Warrier h_xirr = be32_to_cpu(xirr); 43037f55d30SSuresh Warrier local_paca->kvm_hstate.saved_xirr = h_xirr; 43137f55d30SSuresh Warrier xisr = h_xirr & 0xffffff; 43237f55d30SSuresh Warrier /* 43337f55d30SSuresh Warrier * Ensure that the store/load complete to guarantee all side 43437f55d30SSuresh Warrier * effects of loading from XIRR has completed 43537f55d30SSuresh Warrier */ 43637f55d30SSuresh Warrier smp_mb(); 43737f55d30SSuresh Warrier 43837f55d30SSuresh Warrier /* if nothing pending in the ICP */ 43937f55d30SSuresh Warrier if (!xisr) 44037f55d30SSuresh Warrier return 0; 44137f55d30SSuresh Warrier 44237f55d30SSuresh Warrier /* We found something in the ICP... 44337f55d30SSuresh Warrier * 44437f55d30SSuresh Warrier * If it is an IPI, clear the MFRR and EOI it. 44537f55d30SSuresh Warrier */ 44637f55d30SSuresh Warrier if (xisr == XICS_IPI) { 44753af3ba2SPaul Mackerras rc = 0; 448ab9bad0eSBenjamin Herrenschmidt if (xics_phys) { 44937f55d30SSuresh Warrier _stbcix(xics_phys + XICS_MFRR, 0xff); 45037f55d30SSuresh Warrier _stwcix(xics_phys + XICS_XIRR, xirr); 451f725758bSPaul Mackerras } else { 452ab9bad0eSBenjamin Herrenschmidt opal_int_set_mfrr(hard_smp_processor_id(), 0xff); 453ab9bad0eSBenjamin Herrenschmidt rc = opal_int_eoi(h_xirr); 45453af3ba2SPaul Mackerras } 455f725758bSPaul Mackerras /* If rc > 0, there is another interrupt pending */ 456f725758bSPaul Mackerras *again = rc > 0; 457f725758bSPaul Mackerras 45837f55d30SSuresh Warrier /* 45937f55d30SSuresh Warrier * Need to ensure side effects of above stores 46037f55d30SSuresh Warrier * complete before proceeding. 46137f55d30SSuresh Warrier */ 46237f55d30SSuresh Warrier smp_mb(); 46337f55d30SSuresh Warrier 46437f55d30SSuresh Warrier /* 46537f55d30SSuresh Warrier * We need to re-check host IPI now in case it got set in the 46637f55d30SSuresh Warrier * meantime. If it's clear, we bounce the interrupt to the 46737f55d30SSuresh Warrier * guest 46837f55d30SSuresh Warrier */ 46937f55d30SSuresh Warrier host_ipi = local_paca->kvm_hstate.host_ipi; 47037f55d30SSuresh Warrier if (unlikely(host_ipi != 0)) { 47137f55d30SSuresh Warrier /* We raced with the host, 47237f55d30SSuresh Warrier * we need to resend that IPI, bummer 47337f55d30SSuresh Warrier */ 474ab9bad0eSBenjamin Herrenschmidt if (xics_phys) 47537f55d30SSuresh Warrier _stbcix(xics_phys + XICS_MFRR, IPI_PRIORITY); 476f725758bSPaul Mackerras else 477ab9bad0eSBenjamin Herrenschmidt opal_int_set_mfrr(hard_smp_processor_id(), 478f725758bSPaul Mackerras IPI_PRIORITY); 47937f55d30SSuresh Warrier /* Let side effects complete */ 48037f55d30SSuresh Warrier smp_mb(); 48137f55d30SSuresh Warrier return 1; 48237f55d30SSuresh Warrier } 48337f55d30SSuresh Warrier 48437f55d30SSuresh Warrier /* OK, it's an IPI for us */ 48537f55d30SSuresh Warrier local_paca->kvm_hstate.saved_xirr = 0; 48637f55d30SSuresh Warrier return -1; 48737f55d30SSuresh Warrier } 48837f55d30SSuresh Warrier 489f725758bSPaul Mackerras return kvmppc_check_passthru(xisr, xirr, again); 49037f55d30SSuresh Warrier } 491