1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * tools/testing/selftests/kvm/include/perf_test_util.h
4 *
5 * Copyright (C) 2020, Google LLC.
6 */
7
8 #ifndef SELFTEST_KVM_PERF_TEST_UTIL_H
9 #define SELFTEST_KVM_PERF_TEST_UTIL_H
10
11 #include "kvm_util.h"
12 #include "processor.h"
13
14 #define MAX_VCPUS 512
15
16 #define PAGE_SHIFT_4K 12
17 #define PTES_PER_4K_PT 512
18
19 #define TEST_MEM_SLOT_INDEX 1
20
21 /* Default guest test virtual memory offset */
22 #define DEFAULT_GUEST_TEST_MEM 0xc0000000
23
24 #define DEFAULT_PER_VCPU_MEM_SIZE (1 << 30) /* 1G */
25
26 /*
27 * Guest physical memory offset of the testing memory slot.
28 * This will be set to the topmost valid physical address minus
29 * the test memory size.
30 */
31 static uint64_t guest_test_phys_mem;
32
33 /*
34 * Guest virtual memory offset of the testing memory slot.
35 * Must not conflict with identity mapped test code.
36 */
37 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
38 static uint64_t guest_percpu_mem_size = DEFAULT_PER_VCPU_MEM_SIZE;
39
40 /* Number of VCPUs for the test */
41 static int nr_vcpus = 1;
42
43 struct vcpu_args {
44 uint64_t gva;
45 uint64_t pages;
46
47 /* Only used by the host userspace part of the vCPU thread */
48 int vcpu_id;
49 };
50
51 struct perf_test_args {
52 struct kvm_vm *vm;
53 uint64_t host_page_size;
54 uint64_t guest_page_size;
55 int wr_fract;
56
57 struct vcpu_args vcpu_args[MAX_VCPUS];
58 };
59
60 static struct perf_test_args perf_test_args;
61
62 /*
63 * Continuously write to the first 8 bytes of each page in the
64 * specified region.
65 */
guest_code(uint32_t vcpu_id)66 static void guest_code(uint32_t vcpu_id)
67 {
68 struct vcpu_args *vcpu_args = &perf_test_args.vcpu_args[vcpu_id];
69 uint64_t gva;
70 uint64_t pages;
71 int i;
72
73 /* Make sure vCPU args data structure is not corrupt. */
74 GUEST_ASSERT(vcpu_args->vcpu_id == vcpu_id);
75
76 gva = vcpu_args->gva;
77 pages = vcpu_args->pages;
78
79 while (true) {
80 for (i = 0; i < pages; i++) {
81 uint64_t addr = gva + (i * perf_test_args.guest_page_size);
82
83 if (i % perf_test_args.wr_fract == 0)
84 *(uint64_t *)addr = 0x0123456789ABCDEF;
85 else
86 READ_ONCE(*(uint64_t *)addr);
87 }
88
89 GUEST_SYNC(1);
90 }
91 }
92
create_vm(enum vm_guest_mode mode,int vcpus,uint64_t vcpu_memory_bytes)93 static struct kvm_vm *create_vm(enum vm_guest_mode mode, int vcpus,
94 uint64_t vcpu_memory_bytes)
95 {
96 struct kvm_vm *vm;
97 uint64_t pages = DEFAULT_GUEST_PHY_PAGES;
98 uint64_t guest_num_pages;
99
100 /* Account for a few pages per-vCPU for stacks */
101 pages += DEFAULT_STACK_PGS * vcpus;
102
103 /*
104 * Reserve twice the ammount of memory needed to map the test region and
105 * the page table / stacks region, at 4k, for page tables. Do the
106 * calculation with 4K page size: the smallest of all archs. (e.g., 64K
107 * page size guest will need even less memory for page tables).
108 */
109 pages += (2 * pages) / PTES_PER_4K_PT;
110 pages += ((2 * vcpus * vcpu_memory_bytes) >> PAGE_SHIFT_4K) /
111 PTES_PER_4K_PT;
112 pages = vm_adjust_num_guest_pages(mode, pages);
113
114 pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
115
116 vm = vm_create(mode, pages, O_RDWR);
117 kvm_vm_elf_load(vm, program_invocation_name, 0, 0);
118 #ifdef __x86_64__
119 vm_create_irqchip(vm);
120 #endif
121
122 perf_test_args.vm = vm;
123 perf_test_args.guest_page_size = vm_get_page_size(vm);
124 perf_test_args.host_page_size = getpagesize();
125
126 TEST_ASSERT(vcpu_memory_bytes % perf_test_args.guest_page_size == 0,
127 "Guest memory size is not guest page size aligned.");
128
129 guest_num_pages = (vcpus * vcpu_memory_bytes) /
130 perf_test_args.guest_page_size;
131 guest_num_pages = vm_adjust_num_guest_pages(mode, guest_num_pages);
132
133 /*
134 * If there should be more memory in the guest test region than there
135 * can be pages in the guest, it will definitely cause problems.
136 */
137 TEST_ASSERT(guest_num_pages < vm_get_max_gfn(vm),
138 "Requested more guest memory than address space allows.\n"
139 " guest pages: %lx max gfn: %x vcpus: %d wss: %lx]\n",
140 guest_num_pages, vm_get_max_gfn(vm), vcpus,
141 vcpu_memory_bytes);
142
143 TEST_ASSERT(vcpu_memory_bytes % perf_test_args.host_page_size == 0,
144 "Guest memory size is not host page size aligned.");
145
146 guest_test_phys_mem = (vm_get_max_gfn(vm) - guest_num_pages) *
147 perf_test_args.guest_page_size;
148 guest_test_phys_mem &= ~(perf_test_args.host_page_size - 1);
149
150 #ifdef __s390x__
151 /* Align to 1M (segment size) */
152 guest_test_phys_mem &= ~((1 << 20) - 1);
153 #endif
154
155 pr_info("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
156
157 /* Add an extra memory slot for testing */
158 vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
159 guest_test_phys_mem,
160 TEST_MEM_SLOT_INDEX,
161 guest_num_pages, 0);
162
163 /* Do mapping for the demand paging memory slot */
164 virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages, 0);
165
166 ucall_init(vm, NULL);
167
168 return vm;
169 }
170
add_vcpus(struct kvm_vm * vm,int vcpus,uint64_t vcpu_memory_bytes)171 static void add_vcpus(struct kvm_vm *vm, int vcpus, uint64_t vcpu_memory_bytes)
172 {
173 vm_paddr_t vcpu_gpa;
174 struct vcpu_args *vcpu_args;
175 int vcpu_id;
176
177 for (vcpu_id = 0; vcpu_id < vcpus; vcpu_id++) {
178 vcpu_args = &perf_test_args.vcpu_args[vcpu_id];
179
180 vm_vcpu_add_default(vm, vcpu_id, guest_code);
181
182 #ifdef __x86_64__
183 vcpu_set_cpuid(vm, vcpu_id, kvm_get_supported_cpuid());
184 #endif
185
186 vcpu_args->vcpu_id = vcpu_id;
187 vcpu_args->gva = guest_test_virt_mem +
188 (vcpu_id * vcpu_memory_bytes);
189 vcpu_args->pages = vcpu_memory_bytes /
190 perf_test_args.guest_page_size;
191
192 vcpu_gpa = guest_test_phys_mem + (vcpu_id * vcpu_memory_bytes);
193 pr_debug("Added VCPU %d with test mem gpa [%lx, %lx)\n",
194 vcpu_id, vcpu_gpa, vcpu_gpa + vcpu_memory_bytes);
195 }
196 }
197
198 #endif /* SELFTEST_KVM_PERF_TEST_UTIL_H */
199