xref: /linux/arch/x86/hyperv/hv_proc.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/types.h>
3 #include <linux/vmalloc.h>
4 #include <linux/mm.h>
5 #include <linux/clockchips.h>
6 #include <linux/hyperv.h>
7 #include <linux/slab.h>
8 #include <linux/cpuhotplug.h>
9 #include <linux/minmax.h>
10 #include <asm/hypervisor.h>
11 #include <asm/mshyperv.h>
12 #include <asm/apic.h>
13 
14 #include <asm/trace/hyperv.h>
15 
16 /*
17  * See struct hv_deposit_memory. The first u64 is partition ID, the rest
18  * are GPAs.
19  */
20 #define HV_DEPOSIT_MAX (HV_HYP_PAGE_SIZE / sizeof(u64) - 1)
21 
22 /* Deposits exact number of pages. Must be called with interrupts enabled.  */
23 int hv_call_deposit_pages(int node, u64 partition_id, u32 num_pages)
24 {
25 	struct page **pages, *page;
26 	int *counts;
27 	int num_allocations;
28 	int i, j, page_count;
29 	int order;
30 	u64 status;
31 	int ret;
32 	u64 base_pfn;
33 	struct hv_deposit_memory *input_page;
34 	unsigned long flags;
35 
36 	if (num_pages > HV_DEPOSIT_MAX)
37 		return -E2BIG;
38 	if (!num_pages)
39 		return 0;
40 
41 	/* One buffer for page pointers and counts */
42 	page = alloc_page(GFP_KERNEL);
43 	if (!page)
44 		return -ENOMEM;
45 	pages = page_address(page);
46 
47 	counts = kcalloc(HV_DEPOSIT_MAX, sizeof(int), GFP_KERNEL);
48 	if (!counts) {
49 		free_page((unsigned long)pages);
50 		return -ENOMEM;
51 	}
52 
53 	/* Allocate all the pages before disabling interrupts */
54 	i = 0;
55 
56 	while (num_pages) {
57 		/* Find highest order we can actually allocate */
58 		order = 31 - __builtin_clz(num_pages);
59 
60 		while (1) {
61 			pages[i] = alloc_pages_node(node, GFP_KERNEL, order);
62 			if (pages[i])
63 				break;
64 			if (!order) {
65 				ret = -ENOMEM;
66 				num_allocations = i;
67 				goto err_free_allocations;
68 			}
69 			--order;
70 		}
71 
72 		split_page(pages[i], order);
73 		counts[i] = 1 << order;
74 		num_pages -= counts[i];
75 		i++;
76 	}
77 	num_allocations = i;
78 
79 	local_irq_save(flags);
80 
81 	input_page = *this_cpu_ptr(hyperv_pcpu_input_arg);
82 
83 	input_page->partition_id = partition_id;
84 
85 	/* Populate gpa_page_list - these will fit on the input page */
86 	for (i = 0, page_count = 0; i < num_allocations; ++i) {
87 		base_pfn = page_to_pfn(pages[i]);
88 		for (j = 0; j < counts[i]; ++j, ++page_count)
89 			input_page->gpa_page_list[page_count] = base_pfn + j;
90 	}
91 	status = hv_do_rep_hypercall(HVCALL_DEPOSIT_MEMORY,
92 				     page_count, 0, input_page, NULL);
93 	local_irq_restore(flags);
94 	if (!hv_result_success(status)) {
95 		pr_err("Failed to deposit pages: %lld\n", status);
96 		ret = hv_result(status);
97 		goto err_free_allocations;
98 	}
99 
100 	ret = 0;
101 	goto free_buf;
102 
103 err_free_allocations:
104 	for (i = 0; i < num_allocations; ++i) {
105 		base_pfn = page_to_pfn(pages[i]);
106 		for (j = 0; j < counts[i]; ++j)
107 			__free_page(pfn_to_page(base_pfn + j));
108 	}
109 
110 free_buf:
111 	free_page((unsigned long)pages);
112 	kfree(counts);
113 	return ret;
114 }
115 
116 int hv_call_add_logical_proc(int node, u32 lp_index, u32 apic_id)
117 {
118 	struct hv_input_add_logical_processor *input;
119 	struct hv_output_add_logical_processor *output;
120 	u64 status;
121 	unsigned long flags;
122 	int ret = HV_STATUS_SUCCESS;
123 
124 	/*
125 	 * When adding a logical processor, the hypervisor may return
126 	 * HV_STATUS_INSUFFICIENT_MEMORY. When that happens, we deposit more
127 	 * pages and retry.
128 	 */
129 	do {
130 		local_irq_save(flags);
131 
132 		input = *this_cpu_ptr(hyperv_pcpu_input_arg);
133 		/* We don't do anything with the output right now */
134 		output = *this_cpu_ptr(hyperv_pcpu_output_arg);
135 
136 		input->lp_index = lp_index;
137 		input->apic_id = apic_id;
138 		input->proximity_domain_info = hv_numa_node_to_pxm_info(node);
139 		status = hv_do_hypercall(HVCALL_ADD_LOGICAL_PROCESSOR,
140 					 input, output);
141 		local_irq_restore(flags);
142 
143 		if (hv_result(status) != HV_STATUS_INSUFFICIENT_MEMORY) {
144 			if (!hv_result_success(status)) {
145 				pr_err("%s: cpu %u apic ID %u, %lld\n", __func__,
146 				       lp_index, apic_id, status);
147 				ret = hv_result(status);
148 			}
149 			break;
150 		}
151 		ret = hv_call_deposit_pages(node, hv_current_partition_id, 1);
152 	} while (!ret);
153 
154 	return ret;
155 }
156 
157 int hv_call_create_vp(int node, u64 partition_id, u32 vp_index, u32 flags)
158 {
159 	struct hv_create_vp *input;
160 	u64 status;
161 	unsigned long irq_flags;
162 	int ret = HV_STATUS_SUCCESS;
163 
164 	/* Root VPs don't seem to need pages deposited */
165 	if (partition_id != hv_current_partition_id) {
166 		/* The value 90 is empirically determined. It may change. */
167 		ret = hv_call_deposit_pages(node, partition_id, 90);
168 		if (ret)
169 			return ret;
170 	}
171 
172 	do {
173 		local_irq_save(irq_flags);
174 
175 		input = *this_cpu_ptr(hyperv_pcpu_input_arg);
176 
177 		input->partition_id = partition_id;
178 		input->vp_index = vp_index;
179 		input->flags = flags;
180 		input->subnode_type = HvSubnodeAny;
181 		input->proximity_domain_info = hv_numa_node_to_pxm_info(node);
182 		status = hv_do_hypercall(HVCALL_CREATE_VP, input, NULL);
183 		local_irq_restore(irq_flags);
184 
185 		if (hv_result(status) != HV_STATUS_INSUFFICIENT_MEMORY) {
186 			if (!hv_result_success(status)) {
187 				pr_err("%s: vcpu %u, lp %u, %lld\n", __func__,
188 				       vp_index, flags, status);
189 				ret = hv_result(status);
190 			}
191 			break;
192 		}
193 		ret = hv_call_deposit_pages(node, partition_id, 1);
194 
195 	} while (!ret);
196 
197 	return ret;
198 }
199 
200