xref: /kvmtool/pci.c (revision 6ea32ebdb84b1ba9d72459029eb92363efd4ffd6)
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 	if (size > 4)
123 		size = 4;
124 
125 	pci_config_address.w = ioport__read32(&pci_config_address_bits);
126 	/*
127 	 * If someone accesses PCI configuration space offsets that are not
128 	 * aligned to 4 bytes, it uses ioports to signify that.
129 	 */
130 	pci_config_address.reg_offset = port - PCI_CONFIG_DATA;
131 
132 	pci__config_wr(vcpu->kvm, pci_config_address, data, size);
133 
134 	return true;
135 }
136 
137 static bool pci_config_data_in(struct ioport *ioport, struct kvm_cpu *vcpu, u16 port, void *data, int size)
138 {
139 	union pci_config_address pci_config_address;
140 
141 	if (size > 4)
142 		size = 4;
143 
144 	pci_config_address.w = ioport__read32(&pci_config_address_bits);
145 	/*
146 	 * If someone accesses PCI configuration space offsets that are not
147 	 * aligned to 4 bytes, it uses ioports to signify that.
148 	 */
149 	pci_config_address.reg_offset = port - PCI_CONFIG_DATA;
150 
151 	pci__config_rd(vcpu->kvm, pci_config_address, data, size);
152 
153 	return true;
154 }
155 
156 static struct ioport_operations pci_config_data_ops = {
157 	.io_in	= pci_config_data_in,
158 	.io_out	= pci_config_data_out,
159 };
160 
161 void pci__config_wr(struct kvm *kvm, union pci_config_address addr, void *data, int size)
162 {
163 	void *base;
164 	u8 bar, offset;
165 	struct pci_device_header *pci_hdr;
166 	u8 dev_num = addr.device_number;
167 	u32 value = 0;
168 	u32 mask;
169 
170 	if (!pci_device_exists(addr.bus_number, dev_num, 0))
171 		return;
172 
173 	offset = addr.w & PCI_DEV_CFG_MASK;
174 	base = pci_hdr = device__find_dev(DEVICE_BUS_PCI, dev_num)->data;
175 
176 	if (pci_hdr->cfg_ops.write)
177 		pci_hdr->cfg_ops.write(kvm, pci_hdr, offset, data, size);
178 
179 	/*
180 	 * legacy hack: ignore writes to uninitialized regions (e.g. ROM BAR).
181 	 * Not very nice but has been working so far.
182 	 */
183 	if (*(u32 *)(base + offset) == 0)
184 		return;
185 
186 	bar = (offset - PCI_BAR_OFFSET(0)) / sizeof(u32);
187 
188 	/*
189 	 * If the kernel masks the BAR, it will expect to find the size of the
190 	 * BAR there next time it reads from it. After the kernel reads the
191 	 * size, it will write the address back.
192 	 */
193 	if (bar < 6) {
194 		if (pci__bar_is_io(pci_hdr, bar))
195 			mask = (u32)PCI_BASE_ADDRESS_IO_MASK;
196 		else
197 			mask = (u32)PCI_BASE_ADDRESS_MEM_MASK;
198 		/*
199 		 * According to the PCI local bus specification REV 3.0:
200 		 * The number of upper bits that a device actually implements
201 		 * depends on how much of the address space the device will
202 		 * respond to. A device that wants a 1 MB memory address space
203 		 * (using a 32-bit base address register) would build the top
204 		 * 12 bits of the address register, hardwiring the other bits
205 		 * to 0.
206 		 *
207 		 * Furthermore, software can determine how much address space
208 		 * the device requires by writing a value of all 1's to the
209 		 * register and then reading the value back. The device will
210 		 * return 0's in all don't-care address bits, effectively
211 		 * specifying the address space required.
212 		 *
213 		 * Software computes the size of the address space with the
214 		 * formula S = ~B + 1, where S is the memory size and B is the
215 		 * value read from the BAR. This means that the BAR value that
216 		 * kvmtool should return is B = ~(S - 1).
217 		 */
218 		memcpy(&value, data, size);
219 		if (value == 0xffffffff)
220 			value = ~(pci__bar_size(pci_hdr, bar) - 1);
221 		/* Preserve the special bits. */
222 		value = (value & mask) | (pci_hdr->bar[bar] & ~mask);
223 		memcpy(base + offset, &value, size);
224 	} else {
225 		memcpy(base + offset, data, size);
226 	}
227 }
228 
229 void pci__config_rd(struct kvm *kvm, union pci_config_address addr, void *data, int size)
230 {
231 	u8 offset;
232 	struct pci_device_header *pci_hdr;
233 	u8 dev_num = addr.device_number;
234 
235 	if (pci_device_exists(addr.bus_number, dev_num, 0)) {
236 		pci_hdr = device__find_dev(DEVICE_BUS_PCI, dev_num)->data;
237 		offset = addr.w & PCI_DEV_CFG_MASK;
238 
239 		if (pci_hdr->cfg_ops.read)
240 			pci_hdr->cfg_ops.read(kvm, pci_hdr, offset, data, size);
241 
242 		memcpy(data, (void *)pci_hdr + offset, size);
243 	} else {
244 		memset(data, 0xff, size);
245 	}
246 }
247 
248 static void pci_config_mmio_access(struct kvm_cpu *vcpu, u64 addr, u8 *data,
249 				   u32 len, u8 is_write, void *kvm)
250 {
251 	union pci_config_address cfg_addr;
252 
253 	addr			-= KVM_PCI_CFG_AREA;
254 	cfg_addr.w		= (u32)addr;
255 	cfg_addr.enable_bit	= 1;
256 
257 	if (len > 4)
258 		len = 4;
259 
260 	if (is_write)
261 		pci__config_wr(kvm, cfg_addr, data, len);
262 	else
263 		pci__config_rd(kvm, cfg_addr, data, len);
264 }
265 
266 struct pci_device_header *pci__find_dev(u8 dev_num)
267 {
268 	struct device_header *hdr = device__find_dev(DEVICE_BUS_PCI, dev_num);
269 
270 	if (IS_ERR_OR_NULL(hdr))
271 		return NULL;
272 
273 	return hdr->data;
274 }
275 
276 int pci__init(struct kvm *kvm)
277 {
278 	int r;
279 
280 	r = ioport__register(kvm, PCI_CONFIG_DATA + 0, &pci_config_data_ops, 4, NULL);
281 	if (r < 0)
282 		return r;
283 
284 	r = ioport__register(kvm, PCI_CONFIG_ADDRESS + 0, &pci_config_address_ops, 4, NULL);
285 	if (r < 0)
286 		goto err_unregister_data;
287 
288 	r = kvm__register_mmio(kvm, KVM_PCI_CFG_AREA, PCI_CFG_SIZE, false,
289 			       pci_config_mmio_access, kvm);
290 	if (r < 0)
291 		goto err_unregister_addr;
292 
293 	return 0;
294 
295 err_unregister_addr:
296 	ioport__unregister(kvm, PCI_CONFIG_ADDRESS);
297 err_unregister_data:
298 	ioport__unregister(kvm, PCI_CONFIG_DATA);
299 	return r;
300 }
301 dev_base_init(pci__init);
302 
303 int pci__exit(struct kvm *kvm)
304 {
305 	ioport__unregister(kvm, PCI_CONFIG_DATA);
306 	ioport__unregister(kvm, PCI_CONFIG_ADDRESS);
307 
308 	return 0;
309 }
310 dev_base_exit(pci__exit);
311