1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/acpi.h>
3 #include <linux/cpufreq.h>
4 #include <linux/cpuidle.h>
5 #include <linux/export.h>
6 #include <linux/mm.h>
7
8 #include <xen/hvc-console.h>
9 #include <xen/acpi.h>
10
11 #include <asm/bootparam.h>
12 #include <asm/io_apic.h>
13 #include <asm/hypervisor.h>
14 #include <asm/e820/api.h>
15 #include <asm/setup.h>
16
17 #include <xen/xen.h>
18 #include <asm/xen/interface.h>
19 #include <asm/xen/hypercall.h>
20
21 #include <xen/interface/memory.h>
22
23 #include "xen-ops.h"
24
25 /*
26 * PVH variables.
27 *
28 * The variable xen_pvh needs to live in a data segment since it is used
29 * after startup_{32|64} is invoked, which will clear the .bss segment.
30 */
31 bool __ro_after_init xen_pvh;
32 EXPORT_SYMBOL_GPL(xen_pvh);
33
34 #ifdef CONFIG_XEN_DOM0
xen_pvh_setup_gsi(int gsi,int trigger,int polarity)35 int xen_pvh_setup_gsi(int gsi, int trigger, int polarity)
36 {
37 int ret;
38 struct physdev_setup_gsi setup_gsi;
39
40 setup_gsi.gsi = gsi;
41 setup_gsi.triggering = (trigger == ACPI_EDGE_SENSITIVE ? 0 : 1);
42 setup_gsi.polarity = (polarity == ACPI_ACTIVE_HIGH ? 0 : 1);
43
44 ret = HYPERVISOR_physdev_op(PHYSDEVOP_setup_gsi, &setup_gsi);
45 if (ret == -EEXIST) {
46 xen_raw_printk("Already setup the GSI :%d\n", gsi);
47 ret = 0;
48 } else if (ret)
49 xen_raw_printk("Fail to setup GSI (%d)!\n", gsi);
50
51 return ret;
52 }
53 EXPORT_SYMBOL_GPL(xen_pvh_setup_gsi);
54 #endif
55
56 /*
57 * Reserve e820 UNUSABLE regions to inflate the memory balloon.
58 *
59 * On PVH dom0 the host memory map is used, RAM regions available to dom0 are
60 * located as the same place as in the native memory map, but since dom0 gets
61 * less memory than the total amount of host RAM the ranges that can't be
62 * populated are converted from RAM -> UNUSABLE. Use such regions (up to the
63 * ratio signaled in EXTRA_MEM_RATIO) in order to inflate the balloon driver at
64 * boot. Doing so prevents the guest (even if just temporary) from using holes
65 * in the memory map in order to map grants or foreign addresses, and
66 * hopefully limits the risk of a clash with a device MMIO region. Ideally the
67 * hypervisor should notify us which memory ranges are suitable for creating
68 * foreign mappings, but that's not yet implemented.
69 */
pvh_reserve_extra_memory(void)70 static void __init pvh_reserve_extra_memory(void)
71 {
72 struct boot_params *bootp = &boot_params;
73 unsigned int i, ram_pages = 0, extra_pages;
74
75 for (i = 0; i < bootp->e820_entries; i++) {
76 struct boot_e820_entry *e = &bootp->e820_table[i];
77
78 if (e->type != E820_TYPE_RAM)
79 continue;
80 ram_pages += PFN_DOWN(e->addr + e->size) - PFN_UP(e->addr);
81 }
82
83 /* Max amount of extra memory. */
84 extra_pages = EXTRA_MEM_RATIO * ram_pages;
85
86 /*
87 * Convert UNUSABLE ranges to RAM and reserve them for foreign mapping
88 * purposes.
89 */
90 for (i = 0; i < bootp->e820_entries && extra_pages; i++) {
91 struct boot_e820_entry *e = &bootp->e820_table[i];
92 unsigned long pages;
93
94 if (e->type != E820_TYPE_UNUSABLE)
95 continue;
96
97 pages = min(extra_pages,
98 PFN_DOWN(e->addr + e->size) - PFN_UP(e->addr));
99
100 if (pages != (PFN_DOWN(e->addr + e->size) - PFN_UP(e->addr))) {
101 struct boot_e820_entry *next;
102
103 if (bootp->e820_entries ==
104 ARRAY_SIZE(bootp->e820_table))
105 /* No space left to split - skip region. */
106 continue;
107
108 /* Split entry. */
109 next = e + 1;
110 memmove(next, e,
111 (bootp->e820_entries - i) * sizeof(*e));
112 bootp->e820_entries++;
113 next->addr = PAGE_ALIGN(e->addr) + PFN_PHYS(pages);
114 e->size = next->addr - e->addr;
115 next->size -= e->size;
116 }
117 e->type = E820_TYPE_RAM;
118 extra_pages -= pages;
119
120 xen_add_extra_mem(PFN_UP(e->addr), pages);
121 }
122 }
123
pvh_arch_setup(void)124 static void __init pvh_arch_setup(void)
125 {
126 pvh_reserve_extra_memory();
127
128 if (xen_initial_domain()) {
129 xen_add_preferred_consoles();
130
131 /*
132 * Disable usage of CPU idle and frequency drivers: when
133 * running as hardware domain the exposed native ACPI tables
134 * causes idle and/or frequency drivers to attach and
135 * malfunction. It's Xen the entity that controls the idle and
136 * frequency states.
137 *
138 * For unprivileged domains the exposed ACPI tables are
139 * fabricated and don't contain such data.
140 */
141 disable_cpuidle();
142 disable_cpufreq();
143 WARN_ON(xen_set_default_idle());
144 }
145 }
146
xen_pvh_init(struct boot_params * boot_params)147 void __init xen_pvh_init(struct boot_params *boot_params)
148 {
149 xen_pvh = 1;
150 xen_domain_type = XEN_HVM_DOMAIN;
151 xen_start_flags = pvh_start_info.flags;
152
153 x86_init.oem.arch_setup = pvh_arch_setup;
154 x86_init.oem.banner = xen_banner;
155
156 xen_efi_init(boot_params);
157
158 if (xen_initial_domain()) {
159 struct xen_platform_op op = {
160 .cmd = XENPF_get_dom0_console,
161 };
162 int ret = HYPERVISOR_platform_op(&op);
163
164 if (ret > 0)
165 xen_init_vga(&op.u.dom0_console,
166 min(ret * sizeof(char),
167 sizeof(op.u.dom0_console)),
168 &boot_params->screen_info);
169 }
170 }
171
mem_map_via_hcall(struct boot_params * boot_params_p)172 void __init mem_map_via_hcall(struct boot_params *boot_params_p)
173 {
174 struct xen_memory_map memmap;
175 int rc;
176
177 memmap.nr_entries = ARRAY_SIZE(boot_params_p->e820_table);
178 set_xen_guest_handle(memmap.buffer, boot_params_p->e820_table);
179 rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap);
180 if (rc) {
181 xen_raw_printk("XENMEM_memory_map failed (%d)\n", rc);
182 BUG();
183 }
184 boot_params_p->e820_entries = memmap.nr_entries;
185 }
186