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