xref: /kvmtool/pci.c (revision bb0d509bef09cc86ef8d084c32fe55fca85bcde2)
1 #include "kvm/devices.h"
2 #include "kvm/pci.h"
3 #include "kvm/ioport.h"
4 #include "kvm/irq.h"
5 #include "kvm/util.h"
6 #include "kvm/kvm.h"
7 
8 #include <linux/err.h>
9 #include <assert.h>
10 
11 static u32 pci_config_address_bits;
12 
13 /* This is within our PCI gap - in an unused area.
14  * Note this is a PCI *bus address*, is used to assign BARs etc.!
15  * (That's why it can still 32bit even with 64bit guests-- 64bit
16  * PCI isn't currently supported.)
17  */
18 static u32 io_space_blocks		= KVM_PCI_MMIO_AREA;
19 
20 /*
21  * BARs must be naturally aligned, so enforce this in the allocator.
22  */
23 u32 pci_get_io_space_block(u32 size)
24 {
25 	u32 block = ALIGN(io_space_blocks, size);
26 	io_space_blocks = block + size;
27 	return block;
28 }
29 
30 void *pci_find_cap(struct pci_device_header *hdr, u8 cap_type)
31 {
32 	u8 pos;
33 	struct pci_cap_hdr *cap;
34 
35 	pci_for_each_cap(pos, cap, hdr) {
36 		if (cap->type == cap_type)
37 			return cap;
38 	}
39 
40 	return NULL;
41 }
42 
43 void pci__assign_irq(struct device_header *dev_hdr)
44 {
45 	struct pci_device_header *pci_hdr = dev_hdr->data;
46 
47 	/*
48 	 * PCI supports only INTA#,B#,C#,D# per device.
49 	 *
50 	 * A#,B#,C#,D# are allowed for multifunctional devices so stick
51 	 * with A# for our single function devices.
52 	 */
53 	pci_hdr->irq_pin	= 1;
54 	pci_hdr->irq_line	= irq__alloc_line();
55 
56 	if (!pci_hdr->irq_type)
57 		pci_hdr->irq_type = IRQ_TYPE_EDGE_RISING;
58 }
59 
60 static void *pci_config_address_ptr(u16 port)
61 {
62 	unsigned long offset;
63 	void *base;
64 
65 	offset	= port - PCI_CONFIG_ADDRESS;
66 	base	= &pci_config_address_bits;
67 
68 	return base + offset;
69 }
70 
71 static bool pci_config_address_out(struct ioport *ioport, struct kvm_cpu *vcpu, u16 port, void *data, int size)
72 {
73 	void *p = pci_config_address_ptr(port);
74 
75 	memcpy(p, data, size);
76 
77 	return true;
78 }
79 
80 static bool pci_config_address_in(struct ioport *ioport, struct kvm_cpu *vcpu, u16 port, void *data, int size)
81 {
82 	void *p = pci_config_address_ptr(port);
83 
84 	memcpy(data, p, size);
85 
86 	return true;
87 }
88 
89 static struct ioport_operations pci_config_address_ops = {
90 	.io_in	= pci_config_address_in,
91 	.io_out	= pci_config_address_out,
92 };
93 
94 static bool pci_device_exists(u8 bus_number, u8 device_number, u8 function_number)
95 {
96 	union pci_config_address pci_config_address;
97 
98 	pci_config_address.w = ioport__read32(&pci_config_address_bits);
99 
100 	if (pci_config_address.bus_number != bus_number)
101 		return false;
102 
103 	if (pci_config_address.function_number != function_number)
104 		return false;
105 
106 	return !IS_ERR_OR_NULL(device__find_dev(DEVICE_BUS_PCI, device_number));
107 }
108 
109 static bool pci_config_data_out(struct ioport *ioport, struct kvm_cpu *vcpu, u16 port, void *data, int size)
110 {
111 	union pci_config_address pci_config_address;
112 
113 	pci_config_address.w = ioport__read32(&pci_config_address_bits);
114 	/*
115 	 * If someone accesses PCI configuration space offsets that are not
116 	 * aligned to 4 bytes, it uses ioports to signify that.
117 	 */
118 	pci_config_address.reg_offset = port - PCI_CONFIG_DATA;
119 
120 	pci__config_wr(vcpu->kvm, pci_config_address, data, size);
121 
122 	return true;
123 }
124 
125 static bool pci_config_data_in(struct ioport *ioport, struct kvm_cpu *vcpu, u16 port, void *data, int size)
126 {
127 	union pci_config_address pci_config_address;
128 
129 	pci_config_address.w = ioport__read32(&pci_config_address_bits);
130 	/*
131 	 * If someone accesses PCI configuration space offsets that are not
132 	 * aligned to 4 bytes, it uses ioports to signify that.
133 	 */
134 	pci_config_address.reg_offset = port - PCI_CONFIG_DATA;
135 
136 	pci__config_rd(vcpu->kvm, pci_config_address, data, size);
137 
138 	return true;
139 }
140 
141 static struct ioport_operations pci_config_data_ops = {
142 	.io_in	= pci_config_data_in,
143 	.io_out	= pci_config_data_out,
144 };
145 
146 void pci__config_wr(struct kvm *kvm, union pci_config_address addr, void *data, int size)
147 {
148 	void *base;
149 	u8 bar, offset;
150 	struct pci_device_header *pci_hdr;
151 	u8 dev_num = addr.device_number;
152 	u32 value = 0;
153 	u32 mask;
154 
155 	if (!pci_device_exists(addr.bus_number, dev_num, 0))
156 		return;
157 
158 	offset = addr.w & PCI_DEV_CFG_MASK;
159 	base = pci_hdr = device__find_dev(DEVICE_BUS_PCI, dev_num)->data;
160 
161 	if (pci_hdr->cfg_ops.write)
162 		pci_hdr->cfg_ops.write(kvm, pci_hdr, offset, data, size);
163 
164 	/*
165 	 * legacy hack: ignore writes to uninitialized regions (e.g. ROM BAR).
166 	 * Not very nice but has been working so far.
167 	 */
168 	if (*(u32 *)(base + offset) == 0)
169 		return;
170 
171 	bar = (offset - PCI_BAR_OFFSET(0)) / sizeof(u32);
172 
173 	/*
174 	 * If the kernel masks the BAR, it will expect to find the size of the
175 	 * BAR there next time it reads from it. After the kernel reads the
176 	 * size, it will write the address back.
177 	 */
178 	if (bar < 6) {
179 		if (pci_hdr->bar[bar] & PCI_BASE_ADDRESS_SPACE_IO)
180 			mask = (u32)PCI_BASE_ADDRESS_IO_MASK;
181 		else
182 			mask = (u32)PCI_BASE_ADDRESS_MEM_MASK;
183 		/*
184 		 * According to the PCI local bus specification REV 3.0:
185 		 * The number of upper bits that a device actually implements
186 		 * depends on how much of the address space the device will
187 		 * respond to. A device that wants a 1 MB memory address space
188 		 * (using a 32-bit base address register) would build the top
189 		 * 12 bits of the address register, hardwiring the other bits
190 		 * to 0.
191 		 *
192 		 * Furthermore, software can determine how much address space
193 		 * the device requires by writing a value of all 1's to the
194 		 * register and then reading the value back. The device will
195 		 * return 0's in all don't-care address bits, effectively
196 		 * specifying the address space required.
197 		 *
198 		 * Software computes the size of the address space with the
199 		 * formula S = ~B + 1, where S is the memory size and B is the
200 		 * value read from the BAR. This means that the BAR value that
201 		 * kvmtool should return is B = ~(S - 1).
202 		 */
203 		memcpy(&value, data, size);
204 		if (value == 0xffffffff)
205 			value = ~(pci_hdr->bar_size[bar] - 1);
206 		/* Preserve the special bits. */
207 		value = (value & mask) | (pci_hdr->bar[bar] & ~mask);
208 		memcpy(base + offset, &value, size);
209 	} else {
210 		memcpy(base + offset, data, size);
211 	}
212 }
213 
214 void pci__config_rd(struct kvm *kvm, union pci_config_address addr, void *data, int size)
215 {
216 	u8 offset;
217 	struct pci_device_header *pci_hdr;
218 	u8 dev_num = addr.device_number;
219 
220 	if (pci_device_exists(addr.bus_number, dev_num, 0)) {
221 		pci_hdr = device__find_dev(DEVICE_BUS_PCI, dev_num)->data;
222 		offset = addr.w & PCI_DEV_CFG_MASK;
223 
224 		if (pci_hdr->cfg_ops.read)
225 			pci_hdr->cfg_ops.read(kvm, pci_hdr, offset, data, size);
226 
227 		memcpy(data, (void *)pci_hdr + offset, size);
228 	} else {
229 		memset(data, 0xff, size);
230 	}
231 }
232 
233 static void pci_config_mmio_access(struct kvm_cpu *vcpu, u64 addr, u8 *data,
234 				   u32 len, u8 is_write, void *kvm)
235 {
236 	union pci_config_address cfg_addr;
237 
238 	addr			-= KVM_PCI_CFG_AREA;
239 	cfg_addr.w		= (u32)addr;
240 	cfg_addr.enable_bit	= 1;
241 
242 	if (is_write)
243 		pci__config_wr(kvm, cfg_addr, data, len);
244 	else
245 		pci__config_rd(kvm, cfg_addr, data, len);
246 }
247 
248 struct pci_device_header *pci__find_dev(u8 dev_num)
249 {
250 	struct device_header *hdr = device__find_dev(DEVICE_BUS_PCI, dev_num);
251 
252 	if (IS_ERR_OR_NULL(hdr))
253 		return NULL;
254 
255 	return hdr->data;
256 }
257 
258 int pci__init(struct kvm *kvm)
259 {
260 	int r;
261 
262 	r = ioport__register(kvm, PCI_CONFIG_DATA + 0, &pci_config_data_ops, 4, NULL);
263 	if (r < 0)
264 		return r;
265 
266 	r = ioport__register(kvm, PCI_CONFIG_ADDRESS + 0, &pci_config_address_ops, 4, NULL);
267 	if (r < 0)
268 		goto err_unregister_data;
269 
270 	r = kvm__register_mmio(kvm, KVM_PCI_CFG_AREA, PCI_CFG_SIZE, false,
271 			       pci_config_mmio_access, kvm);
272 	if (r < 0)
273 		goto err_unregister_addr;
274 
275 	return 0;
276 
277 err_unregister_addr:
278 	ioport__unregister(kvm, PCI_CONFIG_ADDRESS);
279 err_unregister_data:
280 	ioport__unregister(kvm, PCI_CONFIG_DATA);
281 	return r;
282 }
283 dev_base_init(pci__init);
284 
285 int pci__exit(struct kvm *kvm)
286 {
287 	ioport__unregister(kvm, PCI_CONFIG_DATA);
288 	ioport__unregister(kvm, PCI_CONFIG_ADDRESS);
289 
290 	return 0;
291 }
292 dev_base_exit(pci__exit);
293