xref: /kvmtool/pci.c (revision 2f6384f924d709196a7c25f10eb097dd888e84bf)
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 mmio_blocks			= KVM_PCI_MMIO_AREA;
19 static u16 io_port_blocks		= PCI_IOPORT_START;
20 
21 u16 pci_get_io_port_block(u32 size)
22 {
23 	u16 port = ALIGN(io_port_blocks, PCI_IO_SIZE);
24 
25 	io_port_blocks = port + size;
26 	return port;
27 }
28 
29 /*
30  * BARs must be naturally aligned, so enforce this in the allocator.
31  */
32 u32 pci_get_mmio_block(u32 size)
33 {
34 	u32 block = ALIGN(mmio_blocks, size);
35 	mmio_blocks = block + size;
36 	return block;
37 }
38 
39 void *pci_find_cap(struct pci_device_header *hdr, u8 cap_type)
40 {
41 	u8 pos;
42 	struct pci_cap_hdr *cap;
43 
44 	pci_for_each_cap(pos, cap, hdr) {
45 		if (cap->type == cap_type)
46 			return cap;
47 	}
48 
49 	return NULL;
50 }
51 
52 int pci__assign_irq(struct pci_device_header *pci_hdr)
53 {
54 	/*
55 	 * PCI supports only INTA#,B#,C#,D# per device.
56 	 *
57 	 * A#,B#,C#,D# are allowed for multifunctional devices so stick
58 	 * with A# for our single function devices.
59 	 */
60 	pci_hdr->irq_pin	= 1;
61 	pci_hdr->irq_line	= irq__alloc_line();
62 
63 	if (!pci_hdr->irq_type)
64 		pci_hdr->irq_type = IRQ_TYPE_EDGE_RISING;
65 
66 	return pci_hdr->irq_line;
67 }
68 
69 static void *pci_config_address_ptr(u16 port)
70 {
71 	unsigned long offset;
72 	void *base;
73 
74 	offset	= port - PCI_CONFIG_ADDRESS;
75 	base	= &pci_config_address_bits;
76 
77 	return base + offset;
78 }
79 
80 static bool pci_config_address_out(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(p, data, size);
85 
86 	return true;
87 }
88 
89 static bool pci_config_address_in(struct ioport *ioport, struct kvm_cpu *vcpu, u16 port, void *data, int size)
90 {
91 	void *p = pci_config_address_ptr(port);
92 
93 	memcpy(data, p, size);
94 
95 	return true;
96 }
97 
98 static struct ioport_operations pci_config_address_ops = {
99 	.io_in	= pci_config_address_in,
100 	.io_out	= pci_config_address_out,
101 };
102 
103 static bool pci_device_exists(u8 bus_number, u8 device_number, u8 function_number)
104 {
105 	union pci_config_address pci_config_address;
106 
107 	pci_config_address.w = ioport__read32(&pci_config_address_bits);
108 
109 	if (pci_config_address.bus_number != bus_number)
110 		return false;
111 
112 	if (pci_config_address.function_number != function_number)
113 		return false;
114 
115 	return !IS_ERR_OR_NULL(device__find_dev(DEVICE_BUS_PCI, device_number));
116 }
117 
118 static bool pci_config_data_out(struct ioport *ioport, struct kvm_cpu *vcpu, u16 port, void *data, int size)
119 {
120 	union pci_config_address pci_config_address;
121 
122 	pci_config_address.w = ioport__read32(&pci_config_address_bits);
123 	/*
124 	 * If someone accesses PCI configuration space offsets that are not
125 	 * aligned to 4 bytes, it uses ioports to signify that.
126 	 */
127 	pci_config_address.reg_offset = port - PCI_CONFIG_DATA;
128 
129 	pci__config_wr(vcpu->kvm, pci_config_address, data, size);
130 
131 	return true;
132 }
133 
134 static bool pci_config_data_in(struct ioport *ioport, struct kvm_cpu *vcpu, u16 port, void *data, int size)
135 {
136 	union pci_config_address pci_config_address;
137 
138 	pci_config_address.w = ioport__read32(&pci_config_address_bits);
139 	/*
140 	 * If someone accesses PCI configuration space offsets that are not
141 	 * aligned to 4 bytes, it uses ioports to signify that.
142 	 */
143 	pci_config_address.reg_offset = port - PCI_CONFIG_DATA;
144 
145 	pci__config_rd(vcpu->kvm, pci_config_address, data, size);
146 
147 	return true;
148 }
149 
150 static struct ioport_operations pci_config_data_ops = {
151 	.io_in	= pci_config_data_in,
152 	.io_out	= pci_config_data_out,
153 };
154 
155 void pci__config_wr(struct kvm *kvm, union pci_config_address addr, void *data, int size)
156 {
157 	void *base;
158 	u8 bar, offset;
159 	struct pci_device_header *pci_hdr;
160 	u8 dev_num = addr.device_number;
161 	u32 value = 0;
162 	u32 mask;
163 
164 	if (!pci_device_exists(addr.bus_number, dev_num, 0))
165 		return;
166 
167 	offset = addr.w & PCI_DEV_CFG_MASK;
168 	base = pci_hdr = device__find_dev(DEVICE_BUS_PCI, dev_num)->data;
169 
170 	if (pci_hdr->cfg_ops.write)
171 		pci_hdr->cfg_ops.write(kvm, pci_hdr, offset, data, size);
172 
173 	/*
174 	 * legacy hack: ignore writes to uninitialized regions (e.g. ROM BAR).
175 	 * Not very nice but has been working so far.
176 	 */
177 	if (*(u32 *)(base + offset) == 0)
178 		return;
179 
180 	bar = (offset - PCI_BAR_OFFSET(0)) / sizeof(u32);
181 
182 	/*
183 	 * If the kernel masks the BAR, it will expect to find the size of the
184 	 * BAR there next time it reads from it. After the kernel reads the
185 	 * size, it will write the address back.
186 	 */
187 	if (bar < 6) {
188 		if (pci__bar_is_io(pci_hdr, bar))
189 			mask = (u32)PCI_BASE_ADDRESS_IO_MASK;
190 		else
191 			mask = (u32)PCI_BASE_ADDRESS_MEM_MASK;
192 		/*
193 		 * According to the PCI local bus specification REV 3.0:
194 		 * The number of upper bits that a device actually implements
195 		 * depends on how much of the address space the device will
196 		 * respond to. A device that wants a 1 MB memory address space
197 		 * (using a 32-bit base address register) would build the top
198 		 * 12 bits of the address register, hardwiring the other bits
199 		 * to 0.
200 		 *
201 		 * Furthermore, software can determine how much address space
202 		 * the device requires by writing a value of all 1's to the
203 		 * register and then reading the value back. The device will
204 		 * return 0's in all don't-care address bits, effectively
205 		 * specifying the address space required.
206 		 *
207 		 * Software computes the size of the address space with the
208 		 * formula S = ~B + 1, where S is the memory size and B is the
209 		 * value read from the BAR. This means that the BAR value that
210 		 * kvmtool should return is B = ~(S - 1).
211 		 */
212 		memcpy(&value, data, size);
213 		if (value == 0xffffffff)
214 			value = ~(pci__bar_size(pci_hdr, bar) - 1);
215 		/* Preserve the special bits. */
216 		value = (value & mask) | (pci_hdr->bar[bar] & ~mask);
217 		memcpy(base + offset, &value, size);
218 	} else {
219 		memcpy(base + offset, data, size);
220 	}
221 }
222 
223 void pci__config_rd(struct kvm *kvm, union pci_config_address addr, void *data, int size)
224 {
225 	u8 offset;
226 	struct pci_device_header *pci_hdr;
227 	u8 dev_num = addr.device_number;
228 
229 	if (pci_device_exists(addr.bus_number, dev_num, 0)) {
230 		pci_hdr = device__find_dev(DEVICE_BUS_PCI, dev_num)->data;
231 		offset = addr.w & PCI_DEV_CFG_MASK;
232 
233 		if (pci_hdr->cfg_ops.read)
234 			pci_hdr->cfg_ops.read(kvm, pci_hdr, offset, data, size);
235 
236 		memcpy(data, (void *)pci_hdr + offset, size);
237 	} else {
238 		memset(data, 0xff, size);
239 	}
240 }
241 
242 static void pci_config_mmio_access(struct kvm_cpu *vcpu, u64 addr, u8 *data,
243 				   u32 len, u8 is_write, void *kvm)
244 {
245 	union pci_config_address cfg_addr;
246 
247 	addr			-= KVM_PCI_CFG_AREA;
248 	cfg_addr.w		= (u32)addr;
249 	cfg_addr.enable_bit	= 1;
250 
251 	if (is_write)
252 		pci__config_wr(kvm, cfg_addr, data, len);
253 	else
254 		pci__config_rd(kvm, cfg_addr, data, len);
255 }
256 
257 struct pci_device_header *pci__find_dev(u8 dev_num)
258 {
259 	struct device_header *hdr = device__find_dev(DEVICE_BUS_PCI, dev_num);
260 
261 	if (IS_ERR_OR_NULL(hdr))
262 		return NULL;
263 
264 	return hdr->data;
265 }
266 
267 int pci__init(struct kvm *kvm)
268 {
269 	int r;
270 
271 	r = ioport__register(kvm, PCI_CONFIG_DATA + 0, &pci_config_data_ops, 4, NULL);
272 	if (r < 0)
273 		return r;
274 
275 	r = ioport__register(kvm, PCI_CONFIG_ADDRESS + 0, &pci_config_address_ops, 4, NULL);
276 	if (r < 0)
277 		goto err_unregister_data;
278 
279 	r = kvm__register_mmio(kvm, KVM_PCI_CFG_AREA, PCI_CFG_SIZE, false,
280 			       pci_config_mmio_access, kvm);
281 	if (r < 0)
282 		goto err_unregister_addr;
283 
284 	return 0;
285 
286 err_unregister_addr:
287 	ioport__unregister(kvm, PCI_CONFIG_ADDRESS);
288 err_unregister_data:
289 	ioport__unregister(kvm, PCI_CONFIG_DATA);
290 	return r;
291 }
292 dev_base_init(pci__init);
293 
294 int pci__exit(struct kvm *kvm)
295 {
296 	ioport__unregister(kvm, PCI_CONFIG_DATA);
297 	ioport__unregister(kvm, PCI_CONFIG_ADDRESS);
298 
299 	return 0;
300 }
301 dev_base_exit(pci__exit);
302