xref: /openbsd/usr.sbin/vmd/pci.c (revision 4cfece93)
1 /*	$OpenBSD: pci.c,v 1.28 2019/10/16 02:47:34 mlarkin Exp $	*/
2 
3 /*
4  * Copyright (c) 2015 Mike Larkin <mlarkin@openbsd.org>
5  *
6  * Permission to use, copy, modify, and distribute this software for any
7  * purpose with or without fee is hereby granted, provided that the above
8  * copyright notice and this permission notice appear in all copies.
9  *
10  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17  */
18 
19 #include <sys/types.h>
20 
21 #include <dev/pci/pcireg.h>
22 #include <dev/pci/pcidevs.h>
23 #include <dev/pv/virtioreg.h>
24 #include <machine/vmmvar.h>
25 
26 #include <string.h>
27 #include <unistd.h>
28 #include "vmd.h"
29 #include "pci.h"
30 #include "vmm.h"
31 #include "i8259.h"
32 #include "atomicio.h"
33 
34 struct pci pci;
35 
36 extern char *__progname;
37 
38 /* PIC IRQs, assigned to devices in order */
39 const uint8_t pci_pic_irqs[PCI_MAX_PIC_IRQS] = {3, 5, 6, 7, 9, 10, 11, 12,
40     14, 15};
41 
42 /*
43  * pci_add_bar
44  *
45  * Adds a BAR for the PCI device 'id'. On access, 'barfn' will be
46  * called, and passed 'cookie' as an identifier.
47  *
48  * BARs are fixed size, meaning all I/O BARs requested have the
49  * same size and all MMIO BARs have the same size.
50  *
51  * Parameters:
52  *  id: PCI device to add the BAR to (local count, eg if id == 4,
53  *      this BAR is to be added to the VM's 5th PCI device)
54  *  type: type of the BAR to add (PCI_MAPREG_TYPE_xxx)
55  *  barfn: callback function invoked on BAR access
56  *  cookie: cookie passed to barfn on access
57  *
58  * Returns 0 if the BAR was added successfully, 1 otherwise.
59  */
60 int
61 pci_add_bar(uint8_t id, uint32_t type, void *barfn, void *cookie)
62 {
63 	uint8_t bar_reg_idx, bar_ct;
64 
65 	/* Check id */
66 	if (id >= pci.pci_dev_ct)
67 		return (1);
68 
69 	/* Can only add PCI_MAX_BARS BARs to any device */
70 	bar_ct = pci.pci_devices[id].pd_bar_ct;
71 	if (bar_ct >= PCI_MAX_BARS)
72 		return (1);
73 
74 	/* Compute BAR address and add */
75 	bar_reg_idx = (PCI_MAPREG_START + (bar_ct * 4)) / 4;
76 	if (type == PCI_MAPREG_TYPE_MEM) {
77 		if (pci.pci_next_mmio_bar >= VMM_PCI_MMIO_BAR_END)
78 			return (1);
79 
80 		pci.pci_devices[id].pd_cfg_space[bar_reg_idx] =
81 		    PCI_MAPREG_MEM_ADDR(pci.pci_next_mmio_bar);
82 		pci.pci_next_mmio_bar += VMM_PCI_MMIO_BAR_SIZE;
83 		pci.pci_devices[id].pd_barfunc[bar_ct] = barfn;
84 		pci.pci_devices[id].pd_bar_cookie[bar_ct] = cookie;
85 		pci.pci_devices[id].pd_bartype[bar_ct] = PCI_BAR_TYPE_MMIO;
86 		pci.pci_devices[id].pd_barsize[bar_ct] = VMM_PCI_MMIO_BAR_SIZE;
87 		pci.pci_devices[id].pd_bar_ct++;
88 	} else if (type == PCI_MAPREG_TYPE_IO) {
89 		if (pci.pci_next_io_bar >= VMM_PCI_IO_BAR_END)
90 			return (1);
91 
92 		pci.pci_devices[id].pd_cfg_space[bar_reg_idx] =
93 		    PCI_MAPREG_IO_ADDR(pci.pci_next_io_bar) |
94 		    PCI_MAPREG_TYPE_IO;
95 		pci.pci_next_io_bar += VMM_PCI_IO_BAR_SIZE;
96 		pci.pci_devices[id].pd_barfunc[bar_ct] = barfn;
97 		pci.pci_devices[id].pd_bar_cookie[bar_ct] = cookie;
98 		DPRINTF("%s: adding pci bar cookie for dev %d bar %d = %p",
99 		    __progname, id, bar_ct, cookie);
100 		pci.pci_devices[id].pd_bartype[bar_ct] = PCI_BAR_TYPE_IO;
101 		pci.pci_devices[id].pd_barsize[bar_ct] = VMM_PCI_IO_BAR_SIZE;
102 		pci.pci_devices[id].pd_bar_ct++;
103 	}
104 
105 	return (0);
106 }
107 
108 int
109 pci_set_bar_fn(uint8_t id, uint8_t bar_ct, void *barfn, void *cookie)
110 {
111 	/* Check id */
112 	if (id >= pci.pci_dev_ct)
113 		return (1);
114 
115 	if (bar_ct >= PCI_MAX_BARS)
116 		return (1);
117 
118 	pci.pci_devices[id].pd_barfunc[bar_ct] = barfn;
119 	pci.pci_devices[id].pd_bar_cookie[bar_ct] = cookie;
120 
121 	return (0);
122 }
123 
124 /*
125  * pci_get_dev_irq
126  *
127  * Returns the IRQ for the specified PCI device
128  *
129  * Parameters:
130  *  id: PCI device id to return IRQ for
131  *
132  * Return values:
133  *  The IRQ for the device, or 0xff if no device IRQ assigned
134  */
135 uint8_t
136 pci_get_dev_irq(uint8_t id)
137 {
138 	if (pci.pci_devices[id].pd_int)
139 		return pci.pci_devices[id].pd_irq;
140 	else
141 		return 0xFF;
142 }
143 
144 /*
145  * pci_add_device
146  *
147  * Adds a PCI device to the guest VM defined by the supplied parameters.
148  *
149  * Parameters:
150  *  id: the new PCI device ID (0 .. PCI_CONFIG_MAX_DEV)
151  *  vid: PCI VID of the new device
152  *  pid: PCI PID of the new device
153  *  class: PCI 'class' of the new device
154  *  subclass: PCI 'subclass' of the new device
155  *  subsys_vid: subsystem VID of the new device
156  *  subsys_id: subsystem ID of the new device
157  *  irq_needed: 1 if an IRQ should be assigned to this PCI device, 0 otherwise
158  *  csfunc: PCI config space callback function when the guest VM accesses
159  *      CS of this PCI device
160  *
161  * Return values:
162  *  0: the PCI device was added successfully. The PCI device ID is in 'id'.
163  *  1: the PCI device addition failed.
164  */
165 int
166 pci_add_device(uint8_t *id, uint16_t vid, uint16_t pid, uint8_t class,
167     uint8_t subclass, uint16_t subsys_vid, uint16_t subsys_id,
168     uint8_t irq_needed, pci_cs_fn_t csfunc)
169 {
170 	/* Exceeded max devices? */
171 	if (pci.pci_dev_ct >= PCI_CONFIG_MAX_DEV)
172 		return (1);
173 
174 	/* Exceeded max IRQs? */
175 	/* XXX we could share IRQs ... */
176 	if (pci.pci_next_pic_irq >= PCI_MAX_PIC_IRQS && irq_needed)
177 		return (1);
178 
179 	*id = pci.pci_dev_ct;
180 
181 	pci.pci_devices[*id].pd_vid = vid;
182 	pci.pci_devices[*id].pd_did = pid;
183 	pci.pci_devices[*id].pd_class = class;
184 	pci.pci_devices[*id].pd_subclass = subclass;
185 	pci.pci_devices[*id].pd_subsys_vid = subsys_vid;
186 	pci.pci_devices[*id].pd_subsys_id = subsys_id;
187 
188 	pci.pci_devices[*id].pd_csfunc = csfunc;
189 
190 	if (irq_needed) {
191 		pci.pci_devices[*id].pd_irq =
192 		    pci_pic_irqs[pci.pci_next_pic_irq];
193 		pci.pci_devices[*id].pd_int = 1;
194 		pci.pci_next_pic_irq++;
195 		DPRINTF("assigned irq %d to pci dev %d",
196 		    pci.pci_devices[*id].pd_irq, *id);
197 		pic_set_elcr(pci.pci_devices[*id].pd_irq, 1);
198 	}
199 
200 	pci.pci_dev_ct ++;
201 
202 	return (0);
203 }
204 
205 /*
206  * pci_init
207  *
208  * Initializes the PCI subsystem for the VM by adding a PCI host bridge
209  * as the first PCI device.
210  */
211 void
212 pci_init(void)
213 {
214 	uint8_t id;
215 
216 	memset(&pci, 0, sizeof(pci));
217 	pci.pci_next_mmio_bar = VMM_PCI_MMIO_BAR_BASE;
218 	pci.pci_next_io_bar = VMM_PCI_IO_BAR_BASE;
219 
220 	if (pci_add_device(&id, PCI_VENDOR_OPENBSD, PCI_PRODUCT_OPENBSD_PCHB,
221 	    PCI_CLASS_BRIDGE, PCI_SUBCLASS_BRIDGE_HOST,
222 	    PCI_VENDOR_OPENBSD, 0, 0, NULL)) {
223 		log_warnx("%s: can't add PCI host bridge", __progname);
224 		return;
225 	}
226 }
227 
228 void
229 pci_handle_address_reg(struct vm_run_params *vrp)
230 {
231 	struct vm_exit *vei = vrp->vrp_exit;
232 
233 	/*
234 	 * vei_dir == VEI_DIR_OUT : out instruction
235 	 *
236 	 * The guest wrote to the address register.
237 	 */
238 	if (vei->vei.vei_dir == VEI_DIR_OUT) {
239 		get_input_data(vei, &pci.pci_addr_reg);
240 	} else {
241 		/*
242 		 * vei_dir == VEI_DIR_IN : in instruction
243 		 *
244 		 * The guest read the address register
245 		 */
246 		set_return_data(vei, pci.pci_addr_reg);
247 	}
248 }
249 
250 uint8_t
251 pci_handle_io(struct vm_run_params *vrp)
252 {
253 	int i, j, k, l;
254 	uint16_t reg, b_hi, b_lo;
255 	pci_iobar_fn_t fn;
256 	struct vm_exit *vei = vrp->vrp_exit;
257 	uint8_t intr, dir;
258 
259 	k = -1;
260 	l = -1;
261 	reg = vei->vei.vei_port;
262 	dir = vei->vei.vei_dir;
263 	intr = 0xFF;
264 
265 	for (i = 0 ; i < pci.pci_dev_ct ; i++) {
266 		for (j = 0 ; j < pci.pci_devices[i].pd_bar_ct; j++) {
267 			b_lo = PCI_MAPREG_IO_ADDR(pci.pci_devices[i].pd_bar[j]);
268 			b_hi = b_lo + VMM_PCI_IO_BAR_SIZE;
269 			if (reg >= b_lo && reg < b_hi) {
270 				if (pci.pci_devices[i].pd_barfunc[j]) {
271 					k = j;
272 					l = i;
273 				}
274 			}
275 		}
276 	}
277 
278 	if (k >= 0 && l >= 0) {
279 		fn = (pci_iobar_fn_t)pci.pci_devices[l].pd_barfunc[k];
280 		if (fn(vei->vei.vei_dir, reg -
281 		    PCI_MAPREG_IO_ADDR(pci.pci_devices[l].pd_bar[k]),
282 		    &vei->vei.vei_data, &intr,
283 		    pci.pci_devices[l].pd_bar_cookie[k],
284 		    vei->vei.vei_size)) {
285 			log_warnx("%s: pci i/o access function failed",
286 			    __progname);
287 		}
288 	} else {
289 		DPRINTF("%s: no pci i/o function for reg 0x%llx (dir=%d "
290 		    "guest %%rip=0x%llx", __progname, (uint64_t)reg, dir,
291 		    vei->vrs.vrs_gprs[VCPU_REGS_RIP]);
292 		/* Reads from undefined ports return 0xFF */
293 		if (dir == VEI_DIR_IN)
294 			set_return_data(vei, 0xFFFFFFFF);
295 	}
296 
297 	if (intr != 0xFF) {
298 		intr = pci.pci_devices[l].pd_irq;
299 	}
300 
301 	return (intr);
302 }
303 
304 void
305 pci_handle_data_reg(struct vm_run_params *vrp)
306 {
307 	struct vm_exit *vei = vrp->vrp_exit;
308 	uint8_t b, d, f, o, baridx, ofs, sz;
309 	int ret;
310 	pci_cs_fn_t csfunc;
311 
312 	/* abort if the address register is wack */
313 	if (!(pci.pci_addr_reg & PCI_MODE1_ENABLE)) {
314 		/* if read, return FFs */
315 		if (vei->vei.vei_dir == VEI_DIR_IN)
316 			set_return_data(vei, 0xFFFFFFFF);
317 		log_warnx("invalid address register during pci read: "
318 		    "0x%llx", (uint64_t)pci.pci_addr_reg);
319 		return;
320 	}
321 
322 	/* I/Os to 0xCFC..0xCFF are permitted */
323 	ofs = vei->vei.vei_port - 0xCFC;
324 	sz = vei->vei.vei_size;
325 
326 	b = (pci.pci_addr_reg >> 16) & 0xff;
327 	d = (pci.pci_addr_reg >> 11) & 0x1f;
328 	f = (pci.pci_addr_reg >> 8) & 0x7;
329 	o = (pci.pci_addr_reg & 0xfc);
330 
331 	csfunc = pci.pci_devices[d].pd_csfunc;
332 	if (csfunc != NULL) {
333 		ret = csfunc(vei->vei.vei_dir, (o / 4), &vei->vei.vei_data);
334 		if (ret)
335 			log_warnx("cfg space access function failed for "
336 			    "pci device %d", d);
337 		return;
338 	}
339 
340 	/* No config space function, fallback to default simple r/w impl. */
341 
342 	o += ofs;
343 
344 	/*
345 	 * vei_dir == VEI_DIR_OUT : out instruction
346 	 *
347 	 * The guest wrote to the config space location denoted by the current
348 	 * value in the address register.
349 	 */
350 	if (vei->vei.vei_dir == VEI_DIR_OUT) {
351 		if ((o >= 0x10 && o <= 0x24) &&
352 		    vei->vei.vei_data == 0xffffffff) {
353 			/*
354 			 * Compute BAR index:
355 			 * o = 0x10 -> baridx = 0
356 			 * o = 0x14 -> baridx = 1
357 			 * o = 0x18 -> baridx = 2
358 			 * o = 0x1c -> baridx = 3
359 			 * o = 0x20 -> baridx = 4
360 			 * o = 0x24 -> baridx = 5
361 			 */
362 			baridx = (o / 4) - 4;
363 			if (baridx < pci.pci_devices[d].pd_bar_ct)
364 				vei->vei.vei_data = 0xfffff000;
365 			else
366 				vei->vei.vei_data = 0;
367 		}
368 
369 		/* IOBAR registers must have bit 0 set */
370 		if (o >= 0x10 && o <= 0x24) {
371 			baridx = (o / 4) - 4;
372 			if (baridx < pci.pci_devices[d].pd_bar_ct &&
373 			    pci.pci_devices[d].pd_bartype[baridx] ==
374 			    PCI_BAR_TYPE_IO)
375 				vei->vei.vei_data |= 1;
376 		}
377 
378 		/*
379 		 * Discard writes to "option rom base address" as none of our
380 		 * emulated devices have PCI option roms. Accept any other
381 		 * writes and copy data to config space registers.
382 		 */
383 		if (o != PCI_EXROMADDR_0)
384 			get_input_data(vei,
385 			    &pci.pci_devices[d].pd_cfg_space[o / 4]);
386 	} else {
387 		/*
388 		 * vei_dir == VEI_DIR_IN : in instruction
389 		 *
390 		 * The guest read from the config space location determined by
391 		 * the current value in the address register.
392 		 */
393 		if (d > pci.pci_dev_ct || b > 0 || f > 0)
394 			set_return_data(vei, 0xFFFFFFFF);
395 		else {
396 			switch (sz) {
397 			case 4:
398 				set_return_data(vei,
399 				    pci.pci_devices[d].pd_cfg_space[o / 4]);
400 				break;
401 			case 2:
402 				if (ofs == 0)
403 					set_return_data(vei, pci.pci_devices[d].
404 					    pd_cfg_space[o / 4]);
405 				else
406 					set_return_data(vei, pci.pci_devices[d].
407 					    pd_cfg_space[o / 4] >> 16);
408 				break;
409 			case 1:
410 				set_return_data(vei, pci.pci_devices[d].
411 				    pd_cfg_space[o / 4] >> (ofs * 8));
412 				break;
413 			}
414 		}
415 	}
416 }
417 
418 int
419 pci_dump(int fd)
420 {
421 	log_debug("%s: sending pci", __func__);
422 	if (atomicio(vwrite, fd, &pci, sizeof(pci)) != sizeof(pci)) {
423 		log_warnx("%s: error writing pci to fd", __func__);
424 		return (-1);
425 	}
426 	return (0);
427 }
428 
429 int
430 pci_restore(int fd)
431 {
432 	log_debug("%s: receiving pci", __func__);
433 	if (atomicio(read, fd, &pci, sizeof(pci)) != sizeof(pci)) {
434 		log_warnx("%s: error reading pci from fd", __func__);
435 		return (-1);
436 	}
437 	return (0);
438 }
439