]>
Commit | Line | Data |
---|---|---|
2e04ef76 RR |
1 | /*P:100 |
2 | * This is the Launcher code, a simple program which lays out the "physical" | |
3 | * memory for the new Guest by mapping the kernel image and the virtual | |
4 | * devices, then opens /dev/lguest to tell the kernel about the Guest and | |
5 | * control it. | |
6 | :*/ | |
8ca47e00 RR |
7 | #define _LARGEFILE64_SOURCE |
8 | #define _GNU_SOURCE | |
9 | #include <stdio.h> | |
10 | #include <string.h> | |
11 | #include <unistd.h> | |
12 | #include <err.h> | |
13 | #include <stdint.h> | |
14 | #include <stdlib.h> | |
15 | #include <elf.h> | |
16 | #include <sys/mman.h> | |
6649bb7a | 17 | #include <sys/param.h> |
8ca47e00 RR |
18 | #include <sys/types.h> |
19 | #include <sys/stat.h> | |
20 | #include <sys/wait.h> | |
659a0e66 | 21 | #include <sys/eventfd.h> |
8ca47e00 RR |
22 | #include <fcntl.h> |
23 | #include <stdbool.h> | |
24 | #include <errno.h> | |
25 | #include <ctype.h> | |
26 | #include <sys/socket.h> | |
27 | #include <sys/ioctl.h> | |
28 | #include <sys/time.h> | |
29 | #include <time.h> | |
30 | #include <netinet/in.h> | |
31 | #include <net/if.h> | |
32 | #include <linux/sockios.h> | |
33 | #include <linux/if_tun.h> | |
34 | #include <sys/uio.h> | |
35 | #include <termios.h> | |
36 | #include <getopt.h> | |
17cbca2b RR |
37 | #include <assert.h> |
38 | #include <sched.h> | |
a586d4f6 RR |
39 | #include <limits.h> |
40 | #include <stddef.h> | |
a161883a | 41 | #include <signal.h> |
8aeb36e8 PS |
42 | #include <pwd.h> |
43 | #include <grp.h> | |
c565650b | 44 | #include <sys/user.h> |
d7fbf6e9 | 45 | #include <linux/pci_regs.h> |
8aeb36e8 | 46 | |
927cfb97 RR |
47 | #ifndef VIRTIO_F_ANY_LAYOUT |
48 | #define VIRTIO_F_ANY_LAYOUT 27 | |
49 | #endif | |
50 | ||
2e04ef76 | 51 | /*L:110 |
9f54288d | 52 | * We can ignore the 43 include files we need for this program, but I do want |
2e04ef76 | 53 | * to draw attention to the use of kernel-style types. |
db24e8c2 RR |
54 | * |
55 | * As Linus said, "C is a Spartan language, and so should your naming be." I | |
56 | * like these abbreviations, so we define them here. Note that u64 is always | |
57 | * unsigned long long, which works on all Linux systems: this means that we can | |
2e04ef76 RR |
58 | * use %llu in printf for any u64. |
59 | */ | |
db24e8c2 RR |
60 | typedef unsigned long long u64; |
61 | typedef uint32_t u32; | |
62 | typedef uint16_t u16; | |
63 | typedef uint8_t u8; | |
dde79789 | 64 | /*:*/ |
8ca47e00 | 65 | |
eb39f833 | 66 | #define VIRTIO_CONFIG_NO_LEGACY |
93153077 | 67 | #define VIRTIO_PCI_NO_LEGACY |
50516547 | 68 | #define VIRTIO_BLK_NO_LEGACY |
93153077 RR |
69 | |
70 | /* Use in-kernel ones, which defines VIRTIO_F_VERSION_1 */ | |
71 | #include "../../include/uapi/linux/virtio_config.h" | |
bf6d4034 | 72 | #include "../../include/uapi/linux/virtio_net.h" |
50516547 | 73 | #include "../../include/uapi/linux/virtio_blk.h" |
e8330d9b | 74 | #include "../../include/uapi/linux/virtio_console.h" |
0d5b5d39 | 75 | #include "../../include/uapi/linux/virtio_rng.h" |
e6dc0418 | 76 | #include <linux/virtio_ring.h> |
93153077 | 77 | #include "../../include/uapi/linux/virtio_pci.h" |
e6dc0418 RR |
78 | #include <asm/bootparam.h> |
79 | #include "../../include/linux/lguest_launcher.h" | |
80 | ||
8ca47e00 RR |
81 | #define BRIDGE_PFX "bridge:" |
82 | #ifndef SIOCBRADDIF | |
83 | #define SIOCBRADDIF 0x89a2 /* add interface to bridge */ | |
84 | #endif | |
3c6b5bfa RR |
85 | /* We can have up to 256 pages for devices. */ |
86 | #define DEVICE_PAGES 256 | |
0f0c4fab RR |
87 | /* This will occupy 3 pages: it must be a power of 2. */ |
88 | #define VIRTQUEUE_NUM 256 | |
8ca47e00 | 89 | |
2e04ef76 RR |
90 | /*L:120 |
91 | * verbose is both a global flag and a macro. The C preprocessor allows | |
92 | * this, and although I wouldn't recommend it, it works quite nicely here. | |
93 | */ | |
8ca47e00 RR |
94 | static bool verbose; |
95 | #define verbose(args...) \ | |
96 | do { if (verbose) printf(args); } while(0) | |
dde79789 RR |
97 | /*:*/ |
98 | ||
3c6b5bfa RR |
99 | /* The pointer to the start of guest memory. */ |
100 | static void *guest_base; | |
101 | /* The maximum guest physical address allowed, and maximum possible. */ | |
0a6bcc18 | 102 | static unsigned long guest_limit, guest_max, guest_mmio; |
56739c80 RR |
103 | /* The /dev/lguest file descriptor. */ |
104 | static int lguest_fd; | |
8ca47e00 | 105 | |
e3283fa0 GOC |
106 | /* a per-cpu variable indicating whose vcpu is currently running */ |
107 | static unsigned int __thread cpu_id; | |
108 | ||
6a54f9ab RR |
109 | /* 5 bit device number in the PCI_CONFIG_ADDR => 32 only */ |
110 | #define MAX_PCI_DEVICES 32 | |
111 | ||
dde79789 | 112 | /* This is our list of devices. */ |
1842f23c | 113 | struct device_list { |
17cbca2b RR |
114 | /* Counter to assign interrupt numbers. */ |
115 | unsigned int next_irq; | |
116 | ||
117 | /* Counter to print out convenient device numbers. */ | |
118 | unsigned int device_num; | |
119 | ||
6a54f9ab RR |
120 | /* PCI devices. */ |
121 | struct device *pci[MAX_PCI_DEVICES]; | |
8ca47e00 RR |
122 | }; |
123 | ||
17cbca2b RR |
124 | /* The list of Guest devices, based on command line arguments. */ |
125 | static struct device_list devices; | |
126 | ||
93153077 RR |
127 | struct virtio_pci_cfg_cap { |
128 | struct virtio_pci_cap cap; | |
b2ce1ea4 | 129 | u32 pci_cfg_data; /* Data for BAR access. */ |
93153077 RR |
130 | }; |
131 | ||
132 | struct virtio_pci_mmio { | |
133 | struct virtio_pci_common_cfg cfg; | |
134 | u16 notify; | |
135 | u8 isr; | |
136 | u8 padding; | |
137 | /* Device-specific configuration follows this. */ | |
138 | }; | |
139 | ||
d7fbf6e9 RR |
140 | /* This is the layout (little-endian) of the PCI config space. */ |
141 | struct pci_config { | |
142 | u16 vendor_id, device_id; | |
143 | u16 command, status; | |
144 | u8 revid, prog_if, subclass, class; | |
145 | u8 cacheline_size, lat_timer, header_type, bist; | |
146 | u32 bar[6]; | |
147 | u32 cardbus_cis_ptr; | |
148 | u16 subsystem_vendor_id, subsystem_device_id; | |
149 | u32 expansion_rom_addr; | |
150 | u8 capabilities, reserved1[3]; | |
151 | u32 reserved2; | |
152 | u8 irq_line, irq_pin, min_grant, max_latency; | |
93153077 RR |
153 | |
154 | /* Now, this is the linked capability list. */ | |
155 | struct virtio_pci_cap common; | |
156 | struct virtio_pci_notify_cap notify; | |
157 | struct virtio_pci_cap isr; | |
158 | struct virtio_pci_cap device; | |
93153077 | 159 | struct virtio_pci_cfg_cap cfg_access; |
d7fbf6e9 RR |
160 | }; |
161 | ||
dde79789 | 162 | /* The device structure describes a single device. */ |
1842f23c | 163 | struct device { |
17cbca2b RR |
164 | /* The name of this device, for --verbose. */ |
165 | const char *name; | |
8ca47e00 | 166 | |
17cbca2b RR |
167 | /* Any queues attached to this device */ |
168 | struct virtqueue *vq; | |
8ca47e00 | 169 | |
659a0e66 RR |
170 | /* Is it operational */ |
171 | bool running; | |
a007a751 | 172 | |
d39a6785 RR |
173 | /* Has it written FEATURES_OK but not re-checked it? */ |
174 | bool wrote_features_ok; | |
175 | ||
d7fbf6e9 RR |
176 | /* PCI configuration */ |
177 | union { | |
178 | struct pci_config config; | |
179 | u32 config_words[sizeof(struct pci_config) / sizeof(u32)]; | |
180 | }; | |
181 | ||
93153077 RR |
182 | /* Features we offer, and those accepted. */ |
183 | u64 features, features_accepted; | |
184 | ||
d7fbf6e9 RR |
185 | /* Device-specific config hangs off the end of this. */ |
186 | struct virtio_pci_mmio *mmio; | |
187 | ||
6a54f9ab RR |
188 | /* PCI MMIO resources (all in BAR0) */ |
189 | size_t mmio_size; | |
190 | u32 mmio_addr; | |
191 | ||
8ca47e00 RR |
192 | /* Device-specific data. */ |
193 | void *priv; | |
194 | }; | |
195 | ||
17cbca2b | 196 | /* The virtqueue structure describes a queue attached to a device. */ |
1842f23c | 197 | struct virtqueue { |
17cbca2b RR |
198 | struct virtqueue *next; |
199 | ||
200 | /* Which device owns me. */ | |
201 | struct device *dev; | |
202 | ||
17cbca2b RR |
203 | /* The actual ring of buffers. */ |
204 | struct vring vring; | |
205 | ||
93153077 RR |
206 | /* The information about this virtqueue (we only use queue_size on) */ |
207 | struct virtio_pci_common_cfg pci_config; | |
208 | ||
17cbca2b RR |
209 | /* Last available index we saw. */ |
210 | u16 last_avail_idx; | |
211 | ||
95c517c0 RR |
212 | /* How many are used since we sent last irq? */ |
213 | unsigned int pending_used; | |
214 | ||
659a0e66 RR |
215 | /* Eventfd where Guest notifications arrive. */ |
216 | int eventfd; | |
20887611 | 217 | |
659a0e66 RR |
218 | /* Function for the thread which is servicing this virtqueue. */ |
219 | void (*service)(struct virtqueue *vq); | |
220 | pid_t thread; | |
17cbca2b RR |
221 | }; |
222 | ||
ec04b13f BR |
223 | /* Remember the arguments to the program so we can "reboot" */ |
224 | static char **main_args; | |
225 | ||
659a0e66 RR |
226 | /* The original tty settings to restore on exit. */ |
227 | static struct termios orig_term; | |
228 | ||
2e04ef76 RR |
229 | /* |
230 | * We have to be careful with barriers: our devices are all run in separate | |
f7027c63 | 231 | * threads and so we need to make sure that changes visible to the Guest happen |
2e04ef76 RR |
232 | * in precise order. |
233 | */ | |
f7027c63 | 234 | #define wmb() __asm__ __volatile__("" : : : "memory") |
0d69a65e RR |
235 | #define rmb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory") |
236 | #define mb() __asm__ __volatile__("lock; addl $0,0(%%esp)" : : : "memory") | |
17cbca2b | 237 | |
b5111790 RR |
238 | /* Wrapper for the last available index. Makes it easier to change. */ |
239 | #define lg_last_avail(vq) ((vq)->last_avail_idx) | |
240 | ||
2e04ef76 RR |
241 | /* |
242 | * The virtio configuration space is defined to be little-endian. x86 is | |
243 | * little-endian too, but it's nice to be explicit so we have these helpers. | |
244 | */ | |
17cbca2b RR |
245 | #define cpu_to_le16(v16) (v16) |
246 | #define cpu_to_le32(v32) (v32) | |
247 | #define cpu_to_le64(v64) (v64) | |
248 | #define le16_to_cpu(v16) (v16) | |
249 | #define le32_to_cpu(v32) (v32) | |
a586d4f6 | 250 | #define le64_to_cpu(v64) (v64) |
17cbca2b | 251 | |
28fd6d7f RR |
252 | /* Is this iovec empty? */ |
253 | static bool iov_empty(const struct iovec iov[], unsigned int num_iov) | |
254 | { | |
255 | unsigned int i; | |
256 | ||
257 | for (i = 0; i < num_iov; i++) | |
258 | if (iov[i].iov_len) | |
259 | return false; | |
260 | return true; | |
261 | } | |
262 | ||
263 | /* Take len bytes from the front of this iovec. */ | |
c0316a94 RR |
264 | static void iov_consume(struct iovec iov[], unsigned num_iov, |
265 | void *dest, unsigned len) | |
28fd6d7f RR |
266 | { |
267 | unsigned int i; | |
268 | ||
269 | for (i = 0; i < num_iov; i++) { | |
270 | unsigned int used; | |
271 | ||
272 | used = iov[i].iov_len < len ? iov[i].iov_len : len; | |
c0316a94 RR |
273 | if (dest) { |
274 | memcpy(dest, iov[i].iov_base, used); | |
275 | dest += used; | |
276 | } | |
28fd6d7f RR |
277 | iov[i].iov_base += used; |
278 | iov[i].iov_len -= used; | |
279 | len -= used; | |
280 | } | |
c0316a94 RR |
281 | if (len != 0) |
282 | errx(1, "iovec too short!"); | |
28fd6d7f RR |
283 | } |
284 | ||
2e04ef76 RR |
285 | /*L:100 |
286 | * The Launcher code itself takes us out into userspace, that scary place where | |
287 | * pointers run wild and free! Unfortunately, like most userspace programs, | |
288 | * it's quite boring (which is why everyone likes to hack on the kernel!). | |
289 | * Perhaps if you make up an Lguest Drinking Game at this point, it will get | |
290 | * you through this section. Or, maybe not. | |
3c6b5bfa RR |
291 | * |
292 | * The Launcher sets up a big chunk of memory to be the Guest's "physical" | |
293 | * memory and stores it in "guest_base". In other words, Guest physical == | |
294 | * Launcher virtual with an offset. | |
295 | * | |
296 | * This can be tough to get your head around, but usually it just means that we | |
a33f3224 | 297 | * use these trivial conversion functions when the Guest gives us its |
2e04ef76 RR |
298 | * "physical" addresses: |
299 | */ | |
3c6b5bfa RR |
300 | static void *from_guest_phys(unsigned long addr) |
301 | { | |
302 | return guest_base + addr; | |
303 | } | |
304 | ||
305 | static unsigned long to_guest_phys(const void *addr) | |
306 | { | |
307 | return (addr - guest_base); | |
308 | } | |
309 | ||
dde79789 RR |
310 | /*L:130 |
311 | * Loading the Kernel. | |
312 | * | |
313 | * We start with couple of simple helper routines. open_or_die() avoids | |
2e04ef76 RR |
314 | * error-checking code cluttering the callers: |
315 | */ | |
8ca47e00 RR |
316 | static int open_or_die(const char *name, int flags) |
317 | { | |
318 | int fd = open(name, flags); | |
319 | if (fd < 0) | |
320 | err(1, "Failed to open %s", name); | |
321 | return fd; | |
322 | } | |
323 | ||
3c6b5bfa RR |
324 | /* map_zeroed_pages() takes a number of pages. */ |
325 | static void *map_zeroed_pages(unsigned int num) | |
8ca47e00 | 326 | { |
3c6b5bfa RR |
327 | int fd = open_or_die("/dev/zero", O_RDONLY); |
328 | void *addr; | |
8ca47e00 | 329 | |
2e04ef76 RR |
330 | /* |
331 | * We use a private mapping (ie. if we write to the page, it will be | |
5230ff0c PS |
332 | * copied). We allocate an extra two pages PROT_NONE to act as guard |
333 | * pages against read/write attempts that exceed allocated space. | |
2e04ef76 | 334 | */ |
5230ff0c PS |
335 | addr = mmap(NULL, getpagesize() * (num+2), |
336 | PROT_NONE, MAP_PRIVATE, fd, 0); | |
337 | ||
3c6b5bfa | 338 | if (addr == MAP_FAILED) |
af901ca1 | 339 | err(1, "Mmapping %u pages of /dev/zero", num); |
a91d74a3 | 340 | |
5230ff0c PS |
341 | if (mprotect(addr + getpagesize(), getpagesize() * num, |
342 | PROT_READ|PROT_WRITE) == -1) | |
343 | err(1, "mprotect rw %u pages failed", num); | |
344 | ||
a91d74a3 RR |
345 | /* |
346 | * One neat mmap feature is that you can close the fd, and it | |
347 | * stays mapped. | |
348 | */ | |
34bdaab4 | 349 | close(fd); |
3c6b5bfa | 350 | |
5230ff0c PS |
351 | /* Return address after PROT_NONE page */ |
352 | return addr + getpagesize(); | |
3c6b5bfa RR |
353 | } |
354 | ||
0a6bcc18 RR |
355 | /* Get some bytes which won't be mapped into the guest. */ |
356 | static unsigned long get_mmio_region(size_t size) | |
357 | { | |
358 | unsigned long addr = guest_mmio; | |
359 | size_t i; | |
360 | ||
361 | if (!size) | |
362 | return addr; | |
363 | ||
364 | /* Size has to be a power of 2 (and multiple of 16) */ | |
365 | for (i = 1; i < size; i <<= 1); | |
366 | ||
367 | guest_mmio += i; | |
368 | ||
369 | return addr; | |
370 | } | |
371 | ||
2e04ef76 RR |
372 | /* |
373 | * This routine is used to load the kernel or initrd. It tries mmap, but if | |
6649bb7a | 374 | * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), |
2e04ef76 RR |
375 | * it falls back to reading the memory in. |
376 | */ | |
6649bb7a RM |
377 | static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) |
378 | { | |
379 | ssize_t r; | |
380 | ||
2e04ef76 RR |
381 | /* |
382 | * We map writable even though for some segments are marked read-only. | |
6649bb7a RM |
383 | * The kernel really wants to be writable: it patches its own |
384 | * instructions. | |
385 | * | |
386 | * MAP_PRIVATE means that the page won't be copied until a write is | |
387 | * done to it. This allows us to share untouched memory between | |
2e04ef76 RR |
388 | * Guests. |
389 | */ | |
5230ff0c | 390 | if (mmap(addr, len, PROT_READ|PROT_WRITE, |
6649bb7a RM |
391 | MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) |
392 | return; | |
393 | ||
394 | /* pread does a seek and a read in one shot: saves a few lines. */ | |
395 | r = pread(fd, addr, len, offset); | |
396 | if (r != len) | |
397 | err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); | |
398 | } | |
399 | ||
2e04ef76 RR |
400 | /* |
401 | * This routine takes an open vmlinux image, which is in ELF, and maps it into | |
dde79789 RR |
402 | * the Guest memory. ELF = Embedded Linking Format, which is the format used |
403 | * by all modern binaries on Linux including the kernel. | |
404 | * | |
405 | * The ELF headers give *two* addresses: a physical address, and a virtual | |
47436aa4 RR |
406 | * address. We use the physical address; the Guest will map itself to the |
407 | * virtual address. | |
dde79789 | 408 | * |
2e04ef76 RR |
409 | * We return the starting address. |
410 | */ | |
47436aa4 | 411 | static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) |
8ca47e00 | 412 | { |
8ca47e00 RR |
413 | Elf32_Phdr phdr[ehdr->e_phnum]; |
414 | unsigned int i; | |
8ca47e00 | 415 | |
2e04ef76 RR |
416 | /* |
417 | * Sanity checks on the main ELF header: an x86 executable with a | |
418 | * reasonable number of correctly-sized program headers. | |
419 | */ | |
8ca47e00 RR |
420 | if (ehdr->e_type != ET_EXEC |
421 | || ehdr->e_machine != EM_386 | |
422 | || ehdr->e_phentsize != sizeof(Elf32_Phdr) | |
423 | || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) | |
424 | errx(1, "Malformed elf header"); | |
425 | ||
2e04ef76 RR |
426 | /* |
427 | * An ELF executable contains an ELF header and a number of "program" | |
dde79789 | 428 | * headers which indicate which parts ("segments") of the program to |
2e04ef76 RR |
429 | * load where. |
430 | */ | |
dde79789 RR |
431 | |
432 | /* We read in all the program headers at once: */ | |
8ca47e00 RR |
433 | if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) |
434 | err(1, "Seeking to program headers"); | |
435 | if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) | |
436 | err(1, "Reading program headers"); | |
437 | ||
2e04ef76 RR |
438 | /* |
439 | * Try all the headers: there are usually only three. A read-only one, | |
440 | * a read-write one, and a "note" section which we don't load. | |
441 | */ | |
8ca47e00 | 442 | for (i = 0; i < ehdr->e_phnum; i++) { |
dde79789 | 443 | /* If this isn't a loadable segment, we ignore it */ |
8ca47e00 RR |
444 | if (phdr[i].p_type != PT_LOAD) |
445 | continue; | |
446 | ||
447 | verbose("Section %i: size %i addr %p\n", | |
448 | i, phdr[i].p_memsz, (void *)phdr[i].p_paddr); | |
449 | ||
6649bb7a | 450 | /* We map this section of the file at its physical address. */ |
3c6b5bfa | 451 | map_at(elf_fd, from_guest_phys(phdr[i].p_paddr), |
6649bb7a | 452 | phdr[i].p_offset, phdr[i].p_filesz); |
8ca47e00 RR |
453 | } |
454 | ||
814a0e5c RR |
455 | /* The entry point is given in the ELF header. */ |
456 | return ehdr->e_entry; | |
8ca47e00 RR |
457 | } |
458 | ||
2e04ef76 RR |
459 | /*L:150 |
460 | * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed | |
461 | * to jump into it and it will unpack itself. We used to have to perform some | |
462 | * hairy magic because the unpacking code scared me. | |
dde79789 | 463 | * |
5bbf89fc RR |
464 | * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote |
465 | * a small patch to jump over the tricky bits in the Guest, so now we just read | |
2e04ef76 RR |
466 | * the funky header so we know where in the file to load, and away we go! |
467 | */ | |
47436aa4 | 468 | static unsigned long load_bzimage(int fd) |
8ca47e00 | 469 | { |
43d33b21 | 470 | struct boot_params boot; |
5bbf89fc RR |
471 | int r; |
472 | /* Modern bzImages get loaded at 1M. */ | |
473 | void *p = from_guest_phys(0x100000); | |
474 | ||
2e04ef76 RR |
475 | /* |
476 | * Go back to the start of the file and read the header. It should be | |
395cf969 | 477 | * a Linux boot header (see Documentation/x86/boot.txt) |
2e04ef76 | 478 | */ |
5bbf89fc | 479 | lseek(fd, 0, SEEK_SET); |
43d33b21 | 480 | read(fd, &boot, sizeof(boot)); |
5bbf89fc | 481 | |
43d33b21 RR |
482 | /* Inside the setup_hdr, we expect the magic "HdrS" */ |
483 | if (memcmp(&boot.hdr.header, "HdrS", 4) != 0) | |
5bbf89fc RR |
484 | errx(1, "This doesn't look like a bzImage to me"); |
485 | ||
43d33b21 RR |
486 | /* Skip over the extra sectors of the header. */ |
487 | lseek(fd, (boot.hdr.setup_sects+1) * 512, SEEK_SET); | |
5bbf89fc RR |
488 | |
489 | /* Now read everything into memory. in nice big chunks. */ | |
490 | while ((r = read(fd, p, 65536)) > 0) | |
491 | p += r; | |
492 | ||
43d33b21 RR |
493 | /* Finally, code32_start tells us where to enter the kernel. */ |
494 | return boot.hdr.code32_start; | |
8ca47e00 RR |
495 | } |
496 | ||
2e04ef76 RR |
497 | /*L:140 |
498 | * Loading the kernel is easy when it's a "vmlinux", but most kernels | |
e1e72965 | 499 | * come wrapped up in the self-decompressing "bzImage" format. With a little |
2e04ef76 RR |
500 | * work, we can load those, too. |
501 | */ | |
47436aa4 | 502 | static unsigned long load_kernel(int fd) |
8ca47e00 RR |
503 | { |
504 | Elf32_Ehdr hdr; | |
505 | ||
dde79789 | 506 | /* Read in the first few bytes. */ |
8ca47e00 RR |
507 | if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr)) |
508 | err(1, "Reading kernel"); | |
509 | ||
dde79789 | 510 | /* If it's an ELF file, it starts with "\177ELF" */ |
8ca47e00 | 511 | if (memcmp(hdr.e_ident, ELFMAG, SELFMAG) == 0) |
47436aa4 | 512 | return map_elf(fd, &hdr); |
8ca47e00 | 513 | |
a6bd8e13 | 514 | /* Otherwise we assume it's a bzImage, and try to load it. */ |
47436aa4 | 515 | return load_bzimage(fd); |
8ca47e00 RR |
516 | } |
517 | ||
2e04ef76 RR |
518 | /* |
519 | * This is a trivial little helper to align pages. Andi Kleen hated it because | |
dde79789 RR |
520 | * it calls getpagesize() twice: "it's dumb code." |
521 | * | |
522 | * Kernel guys get really het up about optimization, even when it's not | |
2e04ef76 RR |
523 | * necessary. I leave this code as a reaction against that. |
524 | */ | |
8ca47e00 RR |
525 | static inline unsigned long page_align(unsigned long addr) |
526 | { | |
dde79789 | 527 | /* Add upwards and truncate downwards. */ |
8ca47e00 RR |
528 | return ((addr + getpagesize()-1) & ~(getpagesize()-1)); |
529 | } | |
530 | ||
2e04ef76 RR |
531 | /*L:180 |
532 | * An "initial ram disk" is a disk image loaded into memory along with the | |
533 | * kernel which the kernel can use to boot from without needing any drivers. | |
534 | * Most distributions now use this as standard: the initrd contains the code to | |
535 | * load the appropriate driver modules for the current machine. | |
dde79789 RR |
536 | * |
537 | * Importantly, James Morris works for RedHat, and Fedora uses initrds for its | |
2e04ef76 RR |
538 | * kernels. He sent me this (and tells me when I break it). |
539 | */ | |
8ca47e00 RR |
540 | static unsigned long load_initrd(const char *name, unsigned long mem) |
541 | { | |
542 | int ifd; | |
543 | struct stat st; | |
544 | unsigned long len; | |
8ca47e00 RR |
545 | |
546 | ifd = open_or_die(name, O_RDONLY); | |
dde79789 | 547 | /* fstat() is needed to get the file size. */ |
8ca47e00 RR |
548 | if (fstat(ifd, &st) < 0) |
549 | err(1, "fstat() on initrd '%s'", name); | |
550 | ||
2e04ef76 RR |
551 | /* |
552 | * We map the initrd at the top of memory, but mmap wants it to be | |
553 | * page-aligned, so we round the size up for that. | |
554 | */ | |
8ca47e00 | 555 | len = page_align(st.st_size); |
3c6b5bfa | 556 | map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); |
2e04ef76 RR |
557 | /* |
558 | * Once a file is mapped, you can close the file descriptor. It's a | |
559 | * little odd, but quite useful. | |
560 | */ | |
8ca47e00 | 561 | close(ifd); |
6649bb7a | 562 | verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); |
dde79789 RR |
563 | |
564 | /* We return the initrd size. */ | |
8ca47e00 RR |
565 | return len; |
566 | } | |
e1e72965 | 567 | /*:*/ |
8ca47e00 | 568 | |
2e04ef76 RR |
569 | /* |
570 | * Simple routine to roll all the commandline arguments together with spaces | |
571 | * between them. | |
572 | */ | |
8ca47e00 RR |
573 | static void concat(char *dst, char *args[]) |
574 | { | |
575 | unsigned int i, len = 0; | |
576 | ||
577 | for (i = 0; args[i]; i++) { | |
1ef36fa6 PB |
578 | if (i) { |
579 | strcat(dst+len, " "); | |
580 | len++; | |
581 | } | |
8ca47e00 | 582 | strcpy(dst+len, args[i]); |
1ef36fa6 | 583 | len += strlen(args[i]); |
8ca47e00 RR |
584 | } |
585 | /* In case it's empty. */ | |
586 | dst[len] = '\0'; | |
587 | } | |
588 | ||
2e04ef76 RR |
589 | /*L:185 |
590 | * This is where we actually tell the kernel to initialize the Guest. We | |
e1e72965 | 591 | * saw the arguments it expects when we looked at initialize() in lguest_user.c: |
58a24566 | 592 | * the base of Guest "physical" memory, the top physical page to allow and the |
2e04ef76 RR |
593 | * entry point for the Guest. |
594 | */ | |
56739c80 | 595 | static void tell_kernel(unsigned long start) |
8ca47e00 | 596 | { |
511801dc JS |
597 | unsigned long args[] = { LHREQ_INITIALIZE, |
598 | (unsigned long)guest_base, | |
7313d521 | 599 | guest_limit / getpagesize(), start, |
0a6bcc18 RR |
600 | (guest_mmio+getpagesize()-1) / getpagesize() }; |
601 | verbose("Guest: %p - %p (%#lx, MMIO %#lx)\n", | |
602 | guest_base, guest_base + guest_limit, | |
603 | guest_limit, guest_mmio); | |
56739c80 RR |
604 | lguest_fd = open_or_die("/dev/lguest", O_RDWR); |
605 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
8ca47e00 | 606 | err(1, "Writing to /dev/lguest"); |
8ca47e00 | 607 | } |
dde79789 | 608 | /*:*/ |
8ca47e00 | 609 | |
a91d74a3 | 610 | /*L:200 |
dde79789 RR |
611 | * Device Handling. |
612 | * | |
e1e72965 | 613 | * When the Guest gives us a buffer, it sends an array of addresses and sizes. |
dde79789 | 614 | * We need to make sure it's not trying to reach into the Launcher itself, so |
e1e72965 | 615 | * we have a convenient routine which checks it and exits with an error message |
dde79789 RR |
616 | * if something funny is going on: |
617 | */ | |
8ca47e00 RR |
618 | static void *_check_pointer(unsigned long addr, unsigned int size, |
619 | unsigned int line) | |
620 | { | |
2e04ef76 | 621 | /* |
5230ff0c PS |
622 | * Check if the requested address and size exceeds the allocated memory, |
623 | * or addr + size wraps around. | |
2e04ef76 | 624 | */ |
5230ff0c | 625 | if ((addr + size) > guest_limit || (addr + size) < addr) |
17cbca2b | 626 | errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); |
2e04ef76 RR |
627 | /* |
628 | * We return a pointer for the caller's convenience, now we know it's | |
629 | * safe to use. | |
630 | */ | |
3c6b5bfa | 631 | return from_guest_phys(addr); |
8ca47e00 | 632 | } |
dde79789 | 633 | /* A macro which transparently hands the line number to the real function. */ |
8ca47e00 RR |
634 | #define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) |
635 | ||
2e04ef76 RR |
636 | /* |
637 | * Each buffer in the virtqueues is actually a chain of descriptors. This | |
e1e72965 | 638 | * function returns the next descriptor in the chain, or vq->vring.num if we're |
2e04ef76 RR |
639 | * at the end. |
640 | */ | |
d1f0132e MM |
641 | static unsigned next_desc(struct vring_desc *desc, |
642 | unsigned int i, unsigned int max) | |
17cbca2b RR |
643 | { |
644 | unsigned int next; | |
645 | ||
646 | /* If this descriptor says it doesn't chain, we're done. */ | |
d1f0132e MM |
647 | if (!(desc[i].flags & VRING_DESC_F_NEXT)) |
648 | return max; | |
17cbca2b RR |
649 | |
650 | /* Check they're not leading us off end of descriptors. */ | |
d1f0132e | 651 | next = desc[i].next; |
17cbca2b RR |
652 | /* Make sure compiler knows to grab that: we don't want it changing! */ |
653 | wmb(); | |
654 | ||
d1f0132e | 655 | if (next >= max) |
17cbca2b RR |
656 | errx(1, "Desc next is %u", next); |
657 | ||
658 | return next; | |
659 | } | |
660 | ||
a91d74a3 RR |
661 | /* |
662 | * This actually sends the interrupt for this virtqueue, if we've used a | |
663 | * buffer. | |
664 | */ | |
38bc2b8c RR |
665 | static void trigger_irq(struct virtqueue *vq) |
666 | { | |
d9028eda | 667 | unsigned long buf[] = { LHREQ_IRQ, vq->dev->config.irq_line }; |
38bc2b8c | 668 | |
95c517c0 RR |
669 | /* Don't inform them if nothing used. */ |
670 | if (!vq->pending_used) | |
671 | return; | |
672 | vq->pending_used = 0; | |
673 | ||
d39a6785 RR |
674 | /* |
675 | * 2.4.7.1: | |
676 | * | |
677 | * If the VIRTIO_F_EVENT_IDX feature bit is not negotiated: | |
678 | * The driver MUST set flags to 0 or 1. | |
679 | */ | |
680 | if (vq->vring.avail->flags > 1) | |
681 | errx(1, "%s: avail->flags = %u\n", | |
682 | vq->dev->name, vq->vring.avail->flags); | |
683 | ||
684 | /* | |
685 | * 2.4.7.2: | |
686 | * | |
687 | * If the VIRTIO_F_EVENT_IDX feature bit is not negotiated: | |
688 | * | |
689 | * - The device MUST ignore the used_event value. | |
690 | * - After the device writes a descriptor index into the used ring: | |
691 | * - If flags is 1, the device SHOULD NOT send an interrupt. | |
692 | * - If flags is 0, the device MUST send an interrupt. | |
693 | */ | |
ca60a42c | 694 | if (vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT) { |
990c91f0 | 695 | return; |
ca60a42c | 696 | } |
38bc2b8c | 697 | |
8dc425ff RR |
698 | /* |
699 | * 4.1.4.5.1: | |
700 | * | |
701 | * If MSI-X capability is disabled, the device MUST set the Queue | |
702 | * Interrupt bit in ISR status before sending a virtqueue notification | |
703 | * to the driver. | |
704 | */ | |
d9028eda | 705 | vq->dev->mmio->isr = 0x1; |
93153077 | 706 | |
38bc2b8c RR |
707 | /* Send the Guest an interrupt tell them we used something up. */ |
708 | if (write(lguest_fd, buf, sizeof(buf)) != 0) | |
d9028eda | 709 | err(1, "Triggering irq %i", vq->dev->config.irq_line); |
38bc2b8c RR |
710 | } |
711 | ||
2e04ef76 | 712 | /* |
a91d74a3 | 713 | * This looks in the virtqueue for the first available buffer, and converts |
17cbca2b RR |
714 | * it to an iovec for convenient access. Since descriptors consist of some |
715 | * number of output then some number of input descriptors, it's actually two | |
716 | * iovecs, but we pack them into one and note how many of each there were. | |
717 | * | |
a91d74a3 | 718 | * This function waits if necessary, and returns the descriptor number found. |
2e04ef76 | 719 | */ |
659a0e66 RR |
720 | static unsigned wait_for_vq_desc(struct virtqueue *vq, |
721 | struct iovec iov[], | |
722 | unsigned int *out_num, unsigned int *in_num) | |
17cbca2b | 723 | { |
d1f0132e MM |
724 | unsigned int i, head, max; |
725 | struct vring_desc *desc; | |
659a0e66 RR |
726 | u16 last_avail = lg_last_avail(vq); |
727 | ||
d39a6785 RR |
728 | /* |
729 | * 2.4.7.1: | |
730 | * | |
731 | * The driver MUST handle spurious interrupts from the device. | |
732 | * | |
733 | * That's why this is a while loop. | |
734 | */ | |
735 | ||
a91d74a3 | 736 | /* There's nothing available? */ |
659a0e66 RR |
737 | while (last_avail == vq->vring.avail->idx) { |
738 | u64 event; | |
739 | ||
a91d74a3 RR |
740 | /* |
741 | * Since we're about to sleep, now is a good time to tell the | |
742 | * Guest about what we've used up to now. | |
743 | */ | |
38bc2b8c RR |
744 | trigger_irq(vq); |
745 | ||
b60da13f RR |
746 | /* OK, now we need to know about added descriptors. */ |
747 | vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; | |
748 | ||
2e04ef76 RR |
749 | /* |
750 | * They could have slipped one in as we were doing that: make | |
751 | * sure it's written, then check again. | |
752 | */ | |
b60da13f RR |
753 | mb(); |
754 | if (last_avail != vq->vring.avail->idx) { | |
755 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
756 | break; | |
757 | } | |
758 | ||
659a0e66 RR |
759 | /* Nothing new? Wait for eventfd to tell us they refilled. */ |
760 | if (read(vq->eventfd, &event, sizeof(event)) != sizeof(event)) | |
761 | errx(1, "Event read failed?"); | |
b60da13f RR |
762 | |
763 | /* We don't need to be notified again. */ | |
764 | vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; | |
659a0e66 | 765 | } |
17cbca2b RR |
766 | |
767 | /* Check it isn't doing very strange things with descriptor numbers. */ | |
b5111790 | 768 | if ((u16)(vq->vring.avail->idx - last_avail) > vq->vring.num) |
17cbca2b | 769 | errx(1, "Guest moved used index from %u to %u", |
b5111790 | 770 | last_avail, vq->vring.avail->idx); |
17cbca2b | 771 | |
8fd9a636 RR |
772 | /* |
773 | * Make sure we read the descriptor number *after* we read the ring | |
774 | * update; don't let the cpu or compiler change the order. | |
775 | */ | |
776 | rmb(); | |
777 | ||
2e04ef76 RR |
778 | /* |
779 | * Grab the next descriptor number they're advertising, and increment | |
780 | * the index we've seen. | |
781 | */ | |
b5111790 RR |
782 | head = vq->vring.avail->ring[last_avail % vq->vring.num]; |
783 | lg_last_avail(vq)++; | |
17cbca2b RR |
784 | |
785 | /* If their number is silly, that's a fatal mistake. */ | |
786 | if (head >= vq->vring.num) | |
787 | errx(1, "Guest says index %u is available", head); | |
788 | ||
789 | /* When we start there are none of either input nor output. */ | |
790 | *out_num = *in_num = 0; | |
791 | ||
d1f0132e MM |
792 | max = vq->vring.num; |
793 | desc = vq->vring.desc; | |
17cbca2b | 794 | i = head; |
d1f0132e | 795 | |
8fd9a636 RR |
796 | /* |
797 | * We have to read the descriptor after we read the descriptor number, | |
798 | * but there's a data dependency there so the CPU shouldn't reorder | |
799 | * that: no rmb() required. | |
800 | */ | |
801 | ||
3afe3e0f RR |
802 | do { |
803 | /* | |
804 | * If this is an indirect entry, then this buffer contains a | |
805 | * descriptor table which we handle as if it's any normal | |
806 | * descriptor chain. | |
807 | */ | |
808 | if (desc[i].flags & VRING_DESC_F_INDIRECT) { | |
d39a6785 RR |
809 | /* 2.4.5.3.1: |
810 | * | |
811 | * The driver MUST NOT set the VIRTQ_DESC_F_INDIRECT | |
812 | * flag unless the VIRTIO_F_INDIRECT_DESC feature was | |
813 | * negotiated. | |
814 | */ | |
815 | if (!(vq->dev->features_accepted & | |
816 | (1<<VIRTIO_RING_F_INDIRECT_DESC))) | |
817 | errx(1, "%s: vq indirect not negotiated", | |
818 | vq->dev->name); | |
819 | ||
820 | /* | |
821 | * 2.4.5.3.1: | |
822 | * | |
823 | * The driver MUST NOT set the VIRTQ_DESC_F_INDIRECT | |
824 | * flag within an indirect descriptor (ie. only one | |
825 | * table per descriptor). | |
826 | */ | |
827 | if (desc != vq->vring.desc) | |
828 | errx(1, "%s: Indirect within indirect", | |
829 | vq->dev->name); | |
830 | ||
831 | /* | |
832 | * Proposed update VIRTIO-134 spells this out: | |
833 | * | |
834 | * A driver MUST NOT set both VIRTQ_DESC_F_INDIRECT | |
835 | * and VIRTQ_DESC_F_NEXT in flags. | |
836 | */ | |
837 | if (desc[i].flags & VRING_DESC_F_NEXT) | |
838 | errx(1, "%s: indirect and next together", | |
839 | vq->dev->name); | |
840 | ||
3afe3e0f RR |
841 | if (desc[i].len % sizeof(struct vring_desc)) |
842 | errx(1, "Invalid size for indirect buffer table"); | |
d39a6785 RR |
843 | /* |
844 | * 2.4.5.3.2: | |
845 | * | |
846 | * The device MUST ignore the write-only flag | |
847 | * (flags&VIRTQ_DESC_F_WRITE) in the descriptor that | |
848 | * refers to an indirect table. | |
849 | * | |
850 | * We ignore it here: :) | |
851 | */ | |
d1f0132e | 852 | |
3afe3e0f RR |
853 | max = desc[i].len / sizeof(struct vring_desc); |
854 | desc = check_pointer(desc[i].addr, desc[i].len); | |
855 | i = 0; | |
d39a6785 RR |
856 | |
857 | /* 2.4.5.3.1: | |
858 | * | |
859 | * A driver MUST NOT create a descriptor chain longer | |
860 | * than the Queue Size of the device. | |
861 | */ | |
862 | if (max > vq->pci_config.queue_size) | |
863 | errx(1, "%s: indirect has too many entries", | |
864 | vq->dev->name); | |
3afe3e0f | 865 | } |
d1f0132e | 866 | |
17cbca2b | 867 | /* Grab the first descriptor, and check it's OK. */ |
d1f0132e | 868 | iov[*out_num + *in_num].iov_len = desc[i].len; |
17cbca2b | 869 | iov[*out_num + *in_num].iov_base |
d1f0132e | 870 | = check_pointer(desc[i].addr, desc[i].len); |
17cbca2b | 871 | /* If this is an input descriptor, increment that count. */ |
d1f0132e | 872 | if (desc[i].flags & VRING_DESC_F_WRITE) |
17cbca2b RR |
873 | (*in_num)++; |
874 | else { | |
2e04ef76 RR |
875 | /* |
876 | * If it's an output descriptor, they're all supposed | |
877 | * to come before any input descriptors. | |
878 | */ | |
17cbca2b RR |
879 | if (*in_num) |
880 | errx(1, "Descriptor has out after in"); | |
881 | (*out_num)++; | |
882 | } | |
883 | ||
884 | /* If we've got too many, that implies a descriptor loop. */ | |
d1f0132e | 885 | if (*out_num + *in_num > max) |
17cbca2b | 886 | errx(1, "Looped descriptor"); |
d1f0132e | 887 | } while ((i = next_desc(desc, i, max)) != max); |
dde79789 | 888 | |
17cbca2b | 889 | return head; |
8ca47e00 RR |
890 | } |
891 | ||
2e04ef76 | 892 | /* |
a91d74a3 RR |
893 | * After we've used one of their buffers, we tell the Guest about it. Sometime |
894 | * later we'll want to send them an interrupt using trigger_irq(); note that | |
895 | * wait_for_vq_desc() does that for us if it has to wait. | |
2e04ef76 | 896 | */ |
17cbca2b | 897 | static void add_used(struct virtqueue *vq, unsigned int head, int len) |
8ca47e00 | 898 | { |
17cbca2b RR |
899 | struct vring_used_elem *used; |
900 | ||
2e04ef76 RR |
901 | /* |
902 | * The virtqueue contains a ring of used buffers. Get a pointer to the | |
903 | * next entry in that used ring. | |
904 | */ | |
17cbca2b RR |
905 | used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; |
906 | used->id = head; | |
907 | used->len = len; | |
908 | /* Make sure buffer is written before we update index. */ | |
909 | wmb(); | |
910 | vq->vring.used->idx++; | |
95c517c0 | 911 | vq->pending_used++; |
8ca47e00 RR |
912 | } |
913 | ||
17cbca2b | 914 | /* And here's the combo meal deal. Supersize me! */ |
56739c80 | 915 | static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len) |
8ca47e00 | 916 | { |
17cbca2b | 917 | add_used(vq, head, len); |
56739c80 | 918 | trigger_irq(vq); |
8ca47e00 RR |
919 | } |
920 | ||
e1e72965 RR |
921 | /* |
922 | * The Console | |
923 | * | |
2e04ef76 RR |
924 | * We associate some data with the console for our exit hack. |
925 | */ | |
1842f23c | 926 | struct console_abort { |
dde79789 | 927 | /* How many times have they hit ^C? */ |
8ca47e00 | 928 | int count; |
dde79789 | 929 | /* When did they start? */ |
8ca47e00 RR |
930 | struct timeval start; |
931 | }; | |
932 | ||
dde79789 | 933 | /* This is the routine which handles console input (ie. stdin). */ |
659a0e66 | 934 | static void console_input(struct virtqueue *vq) |
8ca47e00 | 935 | { |
8ca47e00 | 936 | int len; |
17cbca2b | 937 | unsigned int head, in_num, out_num; |
659a0e66 RR |
938 | struct console_abort *abort = vq->dev->priv; |
939 | struct iovec iov[vq->vring.num]; | |
56ae43df | 940 | |
a91d74a3 | 941 | /* Make sure there's a descriptor available. */ |
659a0e66 | 942 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
56ae43df | 943 | if (out_num) |
17cbca2b | 944 | errx(1, "Output buffers in console in queue?"); |
8ca47e00 | 945 | |
a91d74a3 | 946 | /* Read into it. This is where we usually wait. */ |
659a0e66 | 947 | len = readv(STDIN_FILENO, iov, in_num); |
8ca47e00 | 948 | if (len <= 0) { |
659a0e66 | 949 | /* Ran out of input? */ |
8ca47e00 | 950 | warnx("Failed to get console input, ignoring console."); |
2e04ef76 RR |
951 | /* |
952 | * For simplicity, dying threads kill the whole Launcher. So | |
953 | * just nap here. | |
954 | */ | |
659a0e66 RR |
955 | for (;;) |
956 | pause(); | |
8ca47e00 RR |
957 | } |
958 | ||
a91d74a3 | 959 | /* Tell the Guest we used a buffer. */ |
659a0e66 | 960 | add_used_and_trigger(vq, head, len); |
8ca47e00 | 961 | |
2e04ef76 RR |
962 | /* |
963 | * Three ^C within one second? Exit. | |
dde79789 | 964 | * |
659a0e66 RR |
965 | * This is such a hack, but works surprisingly well. Each ^C has to |
966 | * be in a buffer by itself, so they can't be too fast. But we check | |
967 | * that we get three within about a second, so they can't be too | |
2e04ef76 RR |
968 | * slow. |
969 | */ | |
659a0e66 | 970 | if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { |
8ca47e00 | 971 | abort->count = 0; |
659a0e66 RR |
972 | return; |
973 | } | |
8ca47e00 | 974 | |
659a0e66 RR |
975 | abort->count++; |
976 | if (abort->count == 1) | |
977 | gettimeofday(&abort->start, NULL); | |
978 | else if (abort->count == 3) { | |
979 | struct timeval now; | |
980 | gettimeofday(&now, NULL); | |
981 | /* Kill all Launcher processes with SIGINT, like normal ^C */ | |
982 | if (now.tv_sec <= abort->start.tv_sec+1) | |
983 | kill(0, SIGINT); | |
984 | abort->count = 0; | |
985 | } | |
8ca47e00 RR |
986 | } |
987 | ||
659a0e66 RR |
988 | /* This is the routine which handles console output (ie. stdout). */ |
989 | static void console_output(struct virtqueue *vq) | |
8ca47e00 | 990 | { |
17cbca2b | 991 | unsigned int head, out, in; |
17cbca2b RR |
992 | struct iovec iov[vq->vring.num]; |
993 | ||
a91d74a3 | 994 | /* We usually wait in here, for the Guest to give us something. */ |
659a0e66 RR |
995 | head = wait_for_vq_desc(vq, iov, &out, &in); |
996 | if (in) | |
997 | errx(1, "Input buffers in console output queue?"); | |
a91d74a3 RR |
998 | |
999 | /* writev can return a partial write, so we loop here. */ | |
659a0e66 RR |
1000 | while (!iov_empty(iov, out)) { |
1001 | int len = writev(STDOUT_FILENO, iov, out); | |
e0377e25 SA |
1002 | if (len <= 0) { |
1003 | warn("Write to stdout gave %i (%d)", len, errno); | |
1004 | break; | |
1005 | } | |
c0316a94 | 1006 | iov_consume(iov, out, NULL, len); |
17cbca2b | 1007 | } |
a91d74a3 RR |
1008 | |
1009 | /* | |
1010 | * We're finished with that buffer: if we're going to sleep, | |
1011 | * wait_for_vq_desc() will prod the Guest with an interrupt. | |
1012 | */ | |
38bc2b8c | 1013 | add_used(vq, head, 0); |
a161883a RR |
1014 | } |
1015 | ||
e1e72965 RR |
1016 | /* |
1017 | * The Network | |
1018 | * | |
1019 | * Handling output for network is also simple: we get all the output buffers | |
659a0e66 | 1020 | * and write them to /dev/net/tun. |
a6bd8e13 | 1021 | */ |
659a0e66 RR |
1022 | struct net_info { |
1023 | int tunfd; | |
1024 | }; | |
1025 | ||
1026 | static void net_output(struct virtqueue *vq) | |
8ca47e00 | 1027 | { |
659a0e66 RR |
1028 | struct net_info *net_info = vq->dev->priv; |
1029 | unsigned int head, out, in; | |
17cbca2b | 1030 | struct iovec iov[vq->vring.num]; |
a161883a | 1031 | |
a91d74a3 | 1032 | /* We usually wait in here for the Guest to give us a packet. */ |
659a0e66 RR |
1033 | head = wait_for_vq_desc(vq, iov, &out, &in); |
1034 | if (in) | |
1035 | errx(1, "Input buffers in net output queue?"); | |
a91d74a3 RR |
1036 | /* |
1037 | * Send the whole thing through to /dev/net/tun. It expects the exact | |
1038 | * same format: what a coincidence! | |
1039 | */ | |
659a0e66 | 1040 | if (writev(net_info->tunfd, iov, out) < 0) |
e0377e25 | 1041 | warnx("Write to tun failed (%d)?", errno); |
a91d74a3 RR |
1042 | |
1043 | /* | |
1044 | * Done with that one; wait_for_vq_desc() will send the interrupt if | |
1045 | * all packets are processed. | |
1046 | */ | |
38bc2b8c | 1047 | add_used(vq, head, 0); |
8ca47e00 RR |
1048 | } |
1049 | ||
a91d74a3 RR |
1050 | /* |
1051 | * Handling network input is a bit trickier, because I've tried to optimize it. | |
1052 | * | |
1053 | * First we have a helper routine which tells is if from this file descriptor | |
1054 | * (ie. the /dev/net/tun device) will block: | |
1055 | */ | |
4a8962e2 RR |
1056 | static bool will_block(int fd) |
1057 | { | |
1058 | fd_set fdset; | |
1059 | struct timeval zero = { 0, 0 }; | |
1060 | FD_ZERO(&fdset); | |
1061 | FD_SET(fd, &fdset); | |
1062 | return select(fd+1, &fdset, NULL, NULL, &zero) != 1; | |
1063 | } | |
1064 | ||
a91d74a3 RR |
1065 | /* |
1066 | * This handles packets coming in from the tun device to our Guest. Like all | |
1067 | * service routines, it gets called again as soon as it returns, so you don't | |
1068 | * see a while(1) loop here. | |
1069 | */ | |
659a0e66 | 1070 | static void net_input(struct virtqueue *vq) |
8ca47e00 | 1071 | { |
8ca47e00 | 1072 | int len; |
659a0e66 RR |
1073 | unsigned int head, out, in; |
1074 | struct iovec iov[vq->vring.num]; | |
1075 | struct net_info *net_info = vq->dev->priv; | |
1076 | ||
a91d74a3 RR |
1077 | /* |
1078 | * Get a descriptor to write an incoming packet into. This will also | |
1079 | * send an interrupt if they're out of descriptors. | |
1080 | */ | |
659a0e66 RR |
1081 | head = wait_for_vq_desc(vq, iov, &out, &in); |
1082 | if (out) | |
1083 | errx(1, "Output buffers in net input queue?"); | |
4a8962e2 | 1084 | |
a91d74a3 RR |
1085 | /* |
1086 | * If it looks like we'll block reading from the tun device, send them | |
1087 | * an interrupt. | |
1088 | */ | |
4a8962e2 RR |
1089 | if (vq->pending_used && will_block(net_info->tunfd)) |
1090 | trigger_irq(vq); | |
1091 | ||
a91d74a3 RR |
1092 | /* |
1093 | * Read in the packet. This is where we normally wait (when there's no | |
1094 | * incoming network traffic). | |
1095 | */ | |
659a0e66 | 1096 | len = readv(net_info->tunfd, iov, in); |
8ca47e00 | 1097 | if (len <= 0) |
e0377e25 | 1098 | warn("Failed to read from tun (%d).", errno); |
a91d74a3 RR |
1099 | |
1100 | /* | |
1101 | * Mark that packet buffer as used, but don't interrupt here. We want | |
1102 | * to wait until we've done as much work as we can. | |
1103 | */ | |
4a8962e2 | 1104 | add_used(vq, head, len); |
659a0e66 | 1105 | } |
a91d74a3 | 1106 | /*:*/ |
dde79789 | 1107 | |
a91d74a3 | 1108 | /* This is the helper to create threads: run the service routine in a loop. */ |
659a0e66 RR |
1109 | static int do_thread(void *_vq) |
1110 | { | |
1111 | struct virtqueue *vq = _vq; | |
17cbca2b | 1112 | |
659a0e66 RR |
1113 | for (;;) |
1114 | vq->service(vq); | |
1115 | return 0; | |
1116 | } | |
17cbca2b | 1117 | |
2e04ef76 RR |
1118 | /* |
1119 | * When a child dies, we kill our entire process group with SIGTERM. This | |
1120 | * also has the side effect that the shell restores the console for us! | |
1121 | */ | |
659a0e66 RR |
1122 | static void kill_launcher(int signal) |
1123 | { | |
1124 | kill(0, SIGTERM); | |
8ca47e00 RR |
1125 | } |
1126 | ||
d2dbdac3 RR |
1127 | static void reset_vq_pci_config(struct virtqueue *vq) |
1128 | { | |
1129 | vq->pci_config.queue_size = VIRTQUEUE_NUM; | |
1130 | vq->pci_config.queue_enable = 0; | |
1131 | } | |
1132 | ||
659a0e66 | 1133 | static void reset_device(struct device *dev) |
56ae43df | 1134 | { |
659a0e66 RR |
1135 | struct virtqueue *vq; |
1136 | ||
1137 | verbose("Resetting device %s\n", dev->name); | |
1138 | ||
1139 | /* Clear any features they've acked. */ | |
d9028eda | 1140 | dev->features_accepted = 0; |
659a0e66 RR |
1141 | |
1142 | /* We're going to be explicitly killing threads, so ignore them. */ | |
1143 | signal(SIGCHLD, SIG_IGN); | |
1144 | ||
d2dbdac3 RR |
1145 | /* |
1146 | * 4.1.4.3.1: | |
1147 | * | |
1148 | * The device MUST present a 0 in queue_enable on reset. | |
1149 | * | |
1150 | * This means we set it here, and reset the saved ones in every vq. | |
1151 | */ | |
1152 | dev->mmio->cfg.queue_enable = 0; | |
1153 | ||
d9028eda | 1154 | /* Get rid of the virtqueue threads */ |
659a0e66 | 1155 | for (vq = dev->vq; vq; vq = vq->next) { |
d2dbdac3 RR |
1156 | vq->last_avail_idx = 0; |
1157 | reset_vq_pci_config(vq); | |
659a0e66 RR |
1158 | if (vq->thread != (pid_t)-1) { |
1159 | kill(vq->thread, SIGTERM); | |
1160 | waitpid(vq->thread, NULL, 0); | |
1161 | vq->thread = (pid_t)-1; | |
1162 | } | |
659a0e66 RR |
1163 | } |
1164 | dev->running = false; | |
d39a6785 | 1165 | dev->wrote_features_ok = false; |
659a0e66 RR |
1166 | |
1167 | /* Now we care if threads die. */ | |
1168 | signal(SIGCHLD, (void *)kill_launcher); | |
56ae43df RR |
1169 | } |
1170 | ||
d9028eda | 1171 | static void cleanup_devices(void) |
6e5aa7ef | 1172 | { |
659a0e66 | 1173 | unsigned int i; |
659a0e66 | 1174 | |
d9028eda RR |
1175 | for (i = 1; i < MAX_PCI_DEVICES; i++) { |
1176 | struct device *d = devices.pci[i]; | |
1177 | if (!d) | |
1178 | continue; | |
1179 | reset_device(d); | |
659a0e66 | 1180 | } |
6e5aa7ef | 1181 | |
659a0e66 RR |
1182 | /* If we saved off the original terminal settings, restore them now. */ |
1183 | if (orig_term.c_lflag & (ISIG|ICANON|ECHO)) | |
1184 | tcsetattr(STDIN_FILENO, TCSANOW, &orig_term); | |
1185 | } | |
6e5aa7ef | 1186 | |
d7fbf6e9 RR |
1187 | /*L:217 |
1188 | * We do PCI. This is mainly done to let us test the kernel virtio PCI | |
1189 | * code. | |
1190 | */ | |
1191 | ||
8e709469 RR |
1192 | /* Linux expects a PCI host bridge: ours is a dummy, and first on the bus. */ |
1193 | static struct device pci_host_bridge; | |
1194 | ||
1195 | static void init_pci_host_bridge(void) | |
1196 | { | |
1197 | pci_host_bridge.name = "PCI Host Bridge"; | |
1198 | pci_host_bridge.config.class = 0x06; /* bridge */ | |
1199 | pci_host_bridge.config.subclass = 0; /* host bridge */ | |
1200 | devices.pci[0] = &pci_host_bridge; | |
1201 | } | |
1202 | ||
d7fbf6e9 RR |
1203 | /* The IO ports used to read the PCI config space. */ |
1204 | #define PCI_CONFIG_ADDR 0xCF8 | |
1205 | #define PCI_CONFIG_DATA 0xCFC | |
1206 | ||
1207 | /* | |
1208 | * Not really portable, but does help readability: this is what the Guest | |
1209 | * writes to the PCI_CONFIG_ADDR IO port. | |
1210 | */ | |
1211 | union pci_config_addr { | |
1212 | struct { | |
1213 | unsigned mbz: 2; | |
1214 | unsigned offset: 6; | |
1215 | unsigned funcnum: 3; | |
1216 | unsigned devnum: 5; | |
1217 | unsigned busnum: 8; | |
1218 | unsigned reserved: 7; | |
1219 | unsigned enabled : 1; | |
1220 | } bits; | |
1221 | u32 val; | |
1222 | }; | |
1223 | ||
1224 | /* | |
1225 | * We cache what they wrote to the address port, so we know what they're | |
1226 | * talking about when they access the data port. | |
1227 | */ | |
1228 | static union pci_config_addr pci_config_addr; | |
1229 | ||
1230 | static struct device *find_pci_device(unsigned int index) | |
1231 | { | |
1232 | return devices.pci[index]; | |
1233 | } | |
1234 | ||
1235 | /* PCI can do 1, 2 and 4 byte reads; we handle that here. */ | |
1236 | static void ioread(u16 off, u32 v, u32 mask, u32 *val) | |
1237 | { | |
1238 | assert(off < 4); | |
1239 | assert(mask == 0xFF || mask == 0xFFFF || mask == 0xFFFFFFFF); | |
1240 | *val = (v >> (off * 8)) & mask; | |
1241 | } | |
1242 | ||
1243 | /* PCI can do 1, 2 and 4 byte writes; we handle that here. */ | |
1244 | static void iowrite(u16 off, u32 v, u32 mask, u32 *dst) | |
1245 | { | |
1246 | assert(off < 4); | |
1247 | assert(mask == 0xFF || mask == 0xFFFF || mask == 0xFFFFFFFF); | |
1248 | *dst &= ~(mask << (off * 8)); | |
1249 | *dst |= (v & mask) << (off * 8); | |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * Where PCI_CONFIG_DATA accesses depends on the previous write to | |
1254 | * PCI_CONFIG_ADDR. | |
1255 | */ | |
1256 | static struct device *dev_and_reg(u32 *reg) | |
1257 | { | |
1258 | if (!pci_config_addr.bits.enabled) | |
1259 | return NULL; | |
1260 | ||
1261 | if (pci_config_addr.bits.funcnum != 0) | |
1262 | return NULL; | |
1263 | ||
1264 | if (pci_config_addr.bits.busnum != 0) | |
1265 | return NULL; | |
1266 | ||
1267 | if (pci_config_addr.bits.offset * 4 >= sizeof(struct pci_config)) | |
1268 | return NULL; | |
1269 | ||
1270 | *reg = pci_config_addr.bits.offset; | |
1271 | return find_pci_device(pci_config_addr.bits.devnum); | |
1272 | } | |
1273 | ||
59eba788 RR |
1274 | /* |
1275 | * We can get invalid combinations of values while they're writing, so we | |
1276 | * only fault if they try to write with some invalid bar/offset/length. | |
1277 | */ | |
1278 | static bool valid_bar_access(struct device *d, | |
1279 | struct virtio_pci_cfg_cap *cfg_access) | |
1280 | { | |
1281 | /* We only have 1 bar (BAR0) */ | |
1282 | if (cfg_access->cap.bar != 0) | |
1283 | return false; | |
1284 | ||
1285 | /* Check it's within BAR0. */ | |
1286 | if (cfg_access->cap.offset >= d->mmio_size | |
1287 | || cfg_access->cap.offset + cfg_access->cap.length > d->mmio_size) | |
1288 | return false; | |
1289 | ||
1290 | /* Check length is 1, 2 or 4. */ | |
1291 | if (cfg_access->cap.length != 1 | |
1292 | && cfg_access->cap.length != 2 | |
1293 | && cfg_access->cap.length != 4) | |
1294 | return false; | |
1295 | ||
c97eb679 RR |
1296 | /* |
1297 | * 4.1.4.7.2: | |
1298 | * | |
1299 | * The driver MUST NOT write a cap.offset which is not a multiple of | |
1300 | * cap.length (ie. all accesses MUST be aligned). | |
1301 | */ | |
59eba788 RR |
1302 | if (cfg_access->cap.offset % cfg_access->cap.length != 0) |
1303 | return false; | |
1304 | ||
1305 | /* Return pointer into word in BAR0. */ | |
1306 | return true; | |
1307 | } | |
1308 | ||
d7fbf6e9 RR |
1309 | /* Is this accessing the PCI config address port?. */ |
1310 | static bool is_pci_addr_port(u16 port) | |
1311 | { | |
1312 | return port >= PCI_CONFIG_ADDR && port < PCI_CONFIG_ADDR + 4; | |
1313 | } | |
1314 | ||
1315 | static bool pci_addr_iowrite(u16 port, u32 mask, u32 val) | |
1316 | { | |
1317 | iowrite(port - PCI_CONFIG_ADDR, val, mask, | |
1318 | &pci_config_addr.val); | |
1319 | verbose("PCI%s: %#x/%x: bus %u dev %u func %u reg %u\n", | |
1320 | pci_config_addr.bits.enabled ? "" : " DISABLED", | |
1321 | val, mask, | |
1322 | pci_config_addr.bits.busnum, | |
1323 | pci_config_addr.bits.devnum, | |
1324 | pci_config_addr.bits.funcnum, | |
1325 | pci_config_addr.bits.offset); | |
1326 | return true; | |
1327 | } | |
1328 | ||
1329 | static void pci_addr_ioread(u16 port, u32 mask, u32 *val) | |
1330 | { | |
1331 | ioread(port - PCI_CONFIG_ADDR, pci_config_addr.val, mask, val); | |
1332 | } | |
1333 | ||
1334 | /* Is this accessing the PCI config data port?. */ | |
1335 | static bool is_pci_data_port(u16 port) | |
1336 | { | |
1337 | return port >= PCI_CONFIG_DATA && port < PCI_CONFIG_DATA + 4; | |
1338 | } | |
1339 | ||
59eba788 RR |
1340 | static void emulate_mmio_write(struct device *d, u32 off, u32 val, u32 mask); |
1341 | ||
d7fbf6e9 RR |
1342 | static bool pci_data_iowrite(u16 port, u32 mask, u32 val) |
1343 | { | |
1344 | u32 reg, portoff; | |
1345 | struct device *d = dev_and_reg(®); | |
1346 | ||
1347 | /* Complain if they don't belong to a device. */ | |
1348 | if (!d) | |
1349 | return false; | |
1350 | ||
1351 | /* They can do 1 byte writes, etc. */ | |
1352 | portoff = port - PCI_CONFIG_DATA; | |
1353 | ||
1354 | /* | |
1355 | * PCI uses a weird way to determine the BAR size: the OS | |
1356 | * writes all 1's, and sees which ones stick. | |
1357 | */ | |
1358 | if (&d->config_words[reg] == &d->config.bar[0]) { | |
1359 | int i; | |
1360 | ||
1361 | iowrite(portoff, val, mask, &d->config.bar[0]); | |
1362 | for (i = 0; (1 << i) < d->mmio_size; i++) | |
1363 | d->config.bar[0] &= ~(1 << i); | |
1364 | return true; | |
1365 | } else if ((&d->config_words[reg] > &d->config.bar[0] | |
1366 | && &d->config_words[reg] <= &d->config.bar[6]) | |
1367 | || &d->config_words[reg] == &d->config.expansion_rom_addr) { | |
1368 | /* Allow writing to any other BAR, or expansion ROM */ | |
1369 | iowrite(portoff, val, mask, &d->config_words[reg]); | |
1370 | return true; | |
1371 | /* We let them overide latency timer and cacheline size */ | |
1372 | } else if (&d->config_words[reg] == (void *)&d->config.cacheline_size) { | |
1373 | /* Only let them change the first two fields. */ | |
1374 | if (mask == 0xFFFFFFFF) | |
1375 | mask = 0xFFFF; | |
1376 | iowrite(portoff, val, mask, &d->config_words[reg]); | |
1377 | return true; | |
1378 | } else if (&d->config_words[reg] == (void *)&d->config.command | |
1379 | && mask == 0xFFFF) { | |
1380 | /* Ignore command writes. */ | |
1381 | return true; | |
59eba788 RR |
1382 | } else if (&d->config_words[reg] |
1383 | == (void *)&d->config.cfg_access.cap.bar | |
1384 | || &d->config_words[reg] | |
1385 | == &d->config.cfg_access.cap.length | |
1386 | || &d->config_words[reg] | |
1387 | == &d->config.cfg_access.cap.offset) { | |
1388 | ||
1389 | /* | |
1390 | * The VIRTIO_PCI_CAP_PCI_CFG capability | |
1391 | * provides a backdoor to access the MMIO | |
1392 | * regions without mapping them. Weird, but | |
1393 | * useful. | |
1394 | */ | |
1395 | iowrite(portoff, val, mask, &d->config_words[reg]); | |
1396 | return true; | |
b2ce1ea4 | 1397 | } else if (&d->config_words[reg] == &d->config.cfg_access.pci_cfg_data) { |
59eba788 RR |
1398 | u32 write_mask; |
1399 | ||
8dc425ff RR |
1400 | /* |
1401 | * 4.1.4.7.1: | |
1402 | * | |
1403 | * Upon detecting driver write access to pci_cfg_data, the | |
1404 | * device MUST execute a write access at offset cap.offset at | |
1405 | * BAR selected by cap.bar using the first cap.length bytes | |
1406 | * from pci_cfg_data. | |
1407 | */ | |
1408 | ||
59eba788 RR |
1409 | /* Must be bar 0 */ |
1410 | if (!valid_bar_access(d, &d->config.cfg_access)) | |
1411 | return false; | |
1412 | ||
b2ce1ea4 | 1413 | iowrite(portoff, val, mask, &d->config.cfg_access.pci_cfg_data); |
59eba788 RR |
1414 | |
1415 | /* | |
1416 | * Now emulate a write. The mask we use is set by | |
1417 | * len, *not* this write! | |
1418 | */ | |
1419 | write_mask = (1ULL<<(8*d->config.cfg_access.cap.length)) - 1; | |
1420 | verbose("Window writing %#x/%#x to bar %u, offset %u len %u\n", | |
b2ce1ea4 | 1421 | d->config.cfg_access.pci_cfg_data, write_mask, |
59eba788 RR |
1422 | d->config.cfg_access.cap.bar, |
1423 | d->config.cfg_access.cap.offset, | |
1424 | d->config.cfg_access.cap.length); | |
1425 | ||
1426 | emulate_mmio_write(d, d->config.cfg_access.cap.offset, | |
b2ce1ea4 RR |
1427 | d->config.cfg_access.pci_cfg_data, |
1428 | write_mask); | |
59eba788 | 1429 | return true; |
d7fbf6e9 RR |
1430 | } |
1431 | ||
c97eb679 RR |
1432 | /* |
1433 | * 4.1.4.1: | |
1434 | * | |
1435 | * The driver MUST NOT write into any field of the capability | |
1436 | * structure, with the exception of those with cap_type | |
1437 | * VIRTIO_PCI_CAP_PCI_CFG... | |
1438 | */ | |
d7fbf6e9 RR |
1439 | return false; |
1440 | } | |
1441 | ||
59eba788 RR |
1442 | static u32 emulate_mmio_read(struct device *d, u32 off, u32 mask); |
1443 | ||
d7fbf6e9 RR |
1444 | static void pci_data_ioread(u16 port, u32 mask, u32 *val) |
1445 | { | |
1446 | u32 reg; | |
1447 | struct device *d = dev_and_reg(®); | |
1448 | ||
1449 | if (!d) | |
1450 | return; | |
59eba788 RR |
1451 | |
1452 | /* Read through the PCI MMIO access window is special */ | |
b2ce1ea4 | 1453 | if (&d->config_words[reg] == &d->config.cfg_access.pci_cfg_data) { |
59eba788 RR |
1454 | u32 read_mask; |
1455 | ||
8dc425ff RR |
1456 | /* |
1457 | * 4.1.4.7.1: | |
1458 | * | |
1459 | * Upon detecting driver read access to pci_cfg_data, the | |
1460 | * device MUST execute a read access of length cap.length at | |
1461 | * offset cap.offset at BAR selected by cap.bar and store the | |
1462 | * first cap.length bytes in pci_cfg_data. | |
1463 | */ | |
59eba788 RR |
1464 | /* Must be bar 0 */ |
1465 | if (!valid_bar_access(d, &d->config.cfg_access)) | |
1466 | errx(1, "Invalid cfg_access to bar%u, offset %u len %u", | |
1467 | d->config.cfg_access.cap.bar, | |
1468 | d->config.cfg_access.cap.offset, | |
1469 | d->config.cfg_access.cap.length); | |
1470 | ||
1471 | /* | |
1472 | * Read into the window. The mask we use is set by | |
1473 | * len, *not* this read! | |
1474 | */ | |
1475 | read_mask = (1ULL<<(8*d->config.cfg_access.cap.length))-1; | |
b2ce1ea4 | 1476 | d->config.cfg_access.pci_cfg_data |
59eba788 RR |
1477 | = emulate_mmio_read(d, |
1478 | d->config.cfg_access.cap.offset, | |
1479 | read_mask); | |
1480 | verbose("Window read %#x/%#x from bar %u, offset %u len %u\n", | |
b2ce1ea4 | 1481 | d->config.cfg_access.pci_cfg_data, read_mask, |
59eba788 RR |
1482 | d->config.cfg_access.cap.bar, |
1483 | d->config.cfg_access.cap.offset, | |
1484 | d->config.cfg_access.cap.length); | |
1485 | } | |
d7fbf6e9 RR |
1486 | ioread(port - PCI_CONFIG_DATA, d->config_words[reg], mask, val); |
1487 | } | |
1488 | ||
c565650b RR |
1489 | /*L:216 |
1490 | * This is where we emulate a handful of Guest instructions. It's ugly | |
1491 | * and we used to do it in the kernel but it grew over time. | |
1492 | */ | |
1493 | ||
1494 | /* | |
1495 | * We use the ptrace syscall's pt_regs struct to talk about registers | |
1496 | * to lguest: these macros convert the names to the offsets. | |
1497 | */ | |
1498 | #define getreg(name) getreg_off(offsetof(struct user_regs_struct, name)) | |
1499 | #define setreg(name, val) \ | |
1500 | setreg_off(offsetof(struct user_regs_struct, name), (val)) | |
1501 | ||
1502 | static u32 getreg_off(size_t offset) | |
1503 | { | |
1504 | u32 r; | |
1505 | unsigned long args[] = { LHREQ_GETREG, offset }; | |
1506 | ||
1507 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) | |
1508 | err(1, "Getting register %u", offset); | |
1509 | if (pread(lguest_fd, &r, sizeof(r), cpu_id) != sizeof(r)) | |
1510 | err(1, "Reading register %u", offset); | |
1511 | ||
1512 | return r; | |
1513 | } | |
1514 | ||
1515 | static void setreg_off(size_t offset, u32 val) | |
1516 | { | |
1517 | unsigned long args[] = { LHREQ_SETREG, offset, val }; | |
1518 | ||
1519 | if (pwrite(lguest_fd, args, sizeof(args), cpu_id) < 0) | |
1520 | err(1, "Setting register %u", offset); | |
1521 | } | |
1522 | ||
6a54f9ab RR |
1523 | /* Get register by instruction encoding */ |
1524 | static u32 getreg_num(unsigned regnum, u32 mask) | |
1525 | { | |
1526 | /* 8 bit ops use regnums 4-7 for high parts of word */ | |
1527 | if (mask == 0xFF && (regnum & 0x4)) | |
1528 | return getreg_num(regnum & 0x3, 0xFFFF) >> 8; | |
1529 | ||
1530 | switch (regnum) { | |
1531 | case 0: return getreg(eax) & mask; | |
1532 | case 1: return getreg(ecx) & mask; | |
1533 | case 2: return getreg(edx) & mask; | |
1534 | case 3: return getreg(ebx) & mask; | |
1535 | case 4: return getreg(esp) & mask; | |
1536 | case 5: return getreg(ebp) & mask; | |
1537 | case 6: return getreg(esi) & mask; | |
1538 | case 7: return getreg(edi) & mask; | |
1539 | } | |
1540 | abort(); | |
1541 | } | |
1542 | ||
1543 | /* Set register by instruction encoding */ | |
1544 | static void setreg_num(unsigned regnum, u32 val, u32 mask) | |
1545 | { | |
1546 | /* Don't try to set bits out of range */ | |
1547 | assert(~(val & ~mask)); | |
1548 | ||
1549 | /* 8 bit ops use regnums 4-7 for high parts of word */ | |
1550 | if (mask == 0xFF && (regnum & 0x4)) { | |
1551 | /* Construct the 16 bits we want. */ | |
1552 | val = (val << 8) | getreg_num(regnum & 0x3, 0xFF); | |
1553 | setreg_num(regnum & 0x3, val, 0xFFFF); | |
1554 | return; | |
1555 | } | |
1556 | ||
1557 | switch (regnum) { | |
1558 | case 0: setreg(eax, val | (getreg(eax) & ~mask)); return; | |
1559 | case 1: setreg(ecx, val | (getreg(ecx) & ~mask)); return; | |
1560 | case 2: setreg(edx, val | (getreg(edx) & ~mask)); return; | |
1561 | case 3: setreg(ebx, val | (getreg(ebx) & ~mask)); return; | |
1562 | case 4: setreg(esp, val | (getreg(esp) & ~mask)); return; | |
1563 | case 5: setreg(ebp, val | (getreg(ebp) & ~mask)); return; | |
1564 | case 6: setreg(esi, val | (getreg(esi) & ~mask)); return; | |
1565 | case 7: setreg(edi, val | (getreg(edi) & ~mask)); return; | |
1566 | } | |
1567 | abort(); | |
1568 | } | |
1569 | ||
1570 | /* Get bytes of displacement appended to instruction, from r/m encoding */ | |
1571 | static u32 insn_displacement_len(u8 mod_reg_rm) | |
1572 | { | |
1573 | /* Switch on the mod bits */ | |
1574 | switch (mod_reg_rm >> 6) { | |
1575 | case 0: | |
1576 | /* If mod == 0, and r/m == 101, 16-bit displacement follows */ | |
1577 | if ((mod_reg_rm & 0x7) == 0x5) | |
1578 | return 2; | |
1579 | /* Normally, mod == 0 means no literal displacement */ | |
1580 | return 0; | |
1581 | case 1: | |
1582 | /* One byte displacement */ | |
1583 | return 1; | |
1584 | case 2: | |
1585 | /* Four byte displacement */ | |
1586 | return 4; | |
1587 | case 3: | |
1588 | /* Register mode */ | |
1589 | return 0; | |
1590 | } | |
1591 | abort(); | |
1592 | } | |
1593 | ||
c565650b RR |
1594 | static void emulate_insn(const u8 insn[]) |
1595 | { | |
1596 | unsigned long args[] = { LHREQ_TRAP, 13 }; | |
1597 | unsigned int insnlen = 0, in = 0, small_operand = 0, byte_access; | |
1598 | unsigned int eax, port, mask; | |
1599 | /* | |
d7fbf6e9 | 1600 | * Default is to return all-ones on IO port reads, which traditionally |
c565650b RR |
1601 | * means "there's nothing there". |
1602 | */ | |
1603 | u32 val = 0xFFFFFFFF; | |
1604 | ||
1605 | /* | |
1606 | * This must be the Guest kernel trying to do something, not userspace! | |
1607 | * The bottom two bits of the CS segment register are the privilege | |
1608 | * level. | |
1609 | */ | |
1610 | if ((getreg(xcs) & 3) != 0x1) | |
1611 | goto no_emulate; | |
1612 | ||
1613 | /* Decoding x86 instructions is icky. */ | |
1614 | ||
1615 | /* | |
1616 | * Around 2.6.33, the kernel started using an emulation for the | |
1617 | * cmpxchg8b instruction in early boot on many configurations. This | |
1618 | * code isn't paravirtualized, and it tries to disable interrupts. | |
1619 | * Ignore it, which will Mostly Work. | |
1620 | */ | |
1621 | if (insn[insnlen] == 0xfa) { | |
1622 | /* "cli", or Clear Interrupt Enable instruction. Skip it. */ | |
1623 | insnlen = 1; | |
1624 | goto skip_insn; | |
1625 | } | |
1626 | ||
1627 | /* | |
1628 | * 0x66 is an "operand prefix". It means a 16, not 32 bit in/out. | |
1629 | */ | |
1630 | if (insn[insnlen] == 0x66) { | |
1631 | small_operand = 1; | |
1632 | /* The instruction is 1 byte so far, read the next byte. */ | |
1633 | insnlen = 1; | |
1634 | } | |
1635 | ||
1636 | /* If the lower bit isn't set, it's a single byte access */ | |
1637 | byte_access = !(insn[insnlen] & 1); | |
1638 | ||
1639 | /* | |
1640 | * Now we can ignore the lower bit and decode the 4 opcodes | |
1641 | * we need to emulate. | |
1642 | */ | |
1643 | switch (insn[insnlen] & 0xFE) { | |
1644 | case 0xE4: /* in <next byte>,%al */ | |
1645 | port = insn[insnlen+1]; | |
1646 | insnlen += 2; | |
1647 | in = 1; | |
1648 | break; | |
1649 | case 0xEC: /* in (%dx),%al */ | |
1650 | port = getreg(edx) & 0xFFFF; | |
1651 | insnlen += 1; | |
1652 | in = 1; | |
1653 | break; | |
1654 | case 0xE6: /* out %al,<next byte> */ | |
1655 | port = insn[insnlen+1]; | |
1656 | insnlen += 2; | |
1657 | break; | |
1658 | case 0xEE: /* out %al,(%dx) */ | |
1659 | port = getreg(edx) & 0xFFFF; | |
1660 | insnlen += 1; | |
1661 | break; | |
1662 | default: | |
1663 | /* OK, we don't know what this is, can't emulate. */ | |
1664 | goto no_emulate; | |
1665 | } | |
1666 | ||
1667 | /* Set a mask of the 1, 2 or 4 bytes, depending on size of IO */ | |
1668 | if (byte_access) | |
1669 | mask = 0xFF; | |
1670 | else if (small_operand) | |
1671 | mask = 0xFFFF; | |
1672 | else | |
1673 | mask = 0xFFFFFFFF; | |
1674 | ||
1675 | /* | |
1676 | * If it was an "IN" instruction, they expect the result to be read | |
1677 | * into %eax, so we change %eax. | |
1678 | */ | |
1679 | eax = getreg(eax); | |
1680 | ||
1681 | if (in) { | |
d7fbf6e9 RR |
1682 | /* This is the PS/2 keyboard status; 1 means ready for output */ |
1683 | if (port == 0x64) | |
1684 | val = 1; | |
1685 | else if (is_pci_addr_port(port)) | |
1686 | pci_addr_ioread(port, mask, &val); | |
1687 | else if (is_pci_data_port(port)) | |
1688 | pci_data_ioread(port, mask, &val); | |
1689 | ||
c565650b RR |
1690 | /* Clear the bits we're about to read */ |
1691 | eax &= ~mask; | |
1692 | /* Copy bits in from val. */ | |
1693 | eax |= val & mask; | |
1694 | /* Now update the register. */ | |
1695 | setreg(eax, eax); | |
d7fbf6e9 RR |
1696 | } else { |
1697 | if (is_pci_addr_port(port)) { | |
1698 | if (!pci_addr_iowrite(port, mask, eax)) | |
1699 | goto bad_io; | |
1700 | } else if (is_pci_data_port(port)) { | |
1701 | if (!pci_data_iowrite(port, mask, eax)) | |
1702 | goto bad_io; | |
1703 | } | |
1704 | /* There are many other ports, eg. CMOS clock, serial | |
1705 | * and parallel ports, so we ignore them all. */ | |
c565650b RR |
1706 | } |
1707 | ||
1708 | verbose("IO %s of %x to %u: %#08x\n", | |
1709 | in ? "IN" : "OUT", mask, port, eax); | |
1710 | skip_insn: | |
1711 | /* Finally, we've "done" the instruction, so move past it. */ | |
1712 | setreg(eip, getreg(eip) + insnlen); | |
1713 | return; | |
1714 | ||
d7fbf6e9 RR |
1715 | bad_io: |
1716 | warnx("Attempt to %s port %u (%#x mask)", | |
1717 | in ? "read from" : "write to", port, mask); | |
1718 | ||
c565650b RR |
1719 | no_emulate: |
1720 | /* Inject trap into Guest. */ | |
1721 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
1722 | err(1, "Reinjecting trap 13 for fault at %#x", getreg(eip)); | |
1723 | } | |
1724 | ||
6a54f9ab RR |
1725 | static struct device *find_mmio_region(unsigned long paddr, u32 *off) |
1726 | { | |
1727 | unsigned int i; | |
1728 | ||
1729 | for (i = 1; i < MAX_PCI_DEVICES; i++) { | |
1730 | struct device *d = devices.pci[i]; | |
1731 | ||
1732 | if (!d) | |
1733 | continue; | |
1734 | if (paddr < d->mmio_addr) | |
1735 | continue; | |
1736 | if (paddr >= d->mmio_addr + d->mmio_size) | |
1737 | continue; | |
1738 | *off = paddr - d->mmio_addr; | |
1739 | return d; | |
1740 | } | |
1741 | return NULL; | |
1742 | } | |
1743 | ||
93153077 RR |
1744 | /* FIXME: Use vq array. */ |
1745 | static struct virtqueue *vq_by_num(struct device *d, u32 num) | |
1746 | { | |
1747 | struct virtqueue *vq = d->vq; | |
1748 | ||
1749 | while (num-- && vq) | |
1750 | vq = vq->next; | |
1751 | ||
1752 | return vq; | |
1753 | } | |
1754 | ||
1755 | static void save_vq_config(const struct virtio_pci_common_cfg *cfg, | |
1756 | struct virtqueue *vq) | |
1757 | { | |
1758 | vq->pci_config = *cfg; | |
1759 | } | |
1760 | ||
1761 | static void restore_vq_config(struct virtio_pci_common_cfg *cfg, | |
1762 | struct virtqueue *vq) | |
1763 | { | |
1764 | /* Only restore the per-vq part */ | |
1765 | size_t off = offsetof(struct virtio_pci_common_cfg, queue_size); | |
1766 | ||
1767 | memcpy((void *)cfg + off, (void *)&vq->pci_config + off, | |
1768 | sizeof(*cfg) - off); | |
1769 | } | |
1770 | ||
1771 | /* | |
d761b032 RR |
1772 | * 4.1.4.3.2: |
1773 | * | |
1774 | * The driver MUST configure the other virtqueue fields before | |
1775 | * enabling the virtqueue with queue_enable. | |
1776 | * | |
93153077 RR |
1777 | * When they enable the virtqueue, we check that their setup is valid. |
1778 | */ | |
d761b032 | 1779 | static void check_virtqueue(struct device *d, struct virtqueue *vq) |
93153077 | 1780 | { |
93153077 RR |
1781 | /* Because lguest is 32 bit, all the descriptor high bits must be 0 */ |
1782 | if (vq->pci_config.queue_desc_hi | |
1783 | || vq->pci_config.queue_avail_hi | |
1784 | || vq->pci_config.queue_used_hi) | |
1785 | errx(1, "%s: invalid 64-bit queue address", d->name); | |
1786 | ||
d39a6785 RR |
1787 | /* |
1788 | * 2.4.1: | |
1789 | * | |
1790 | * The driver MUST ensure that the physical address of the first byte | |
1791 | * of each virtqueue part is a multiple of the specified alignment | |
1792 | * value in the above table. | |
1793 | */ | |
1794 | if (vq->pci_config.queue_desc_lo % 16 | |
1795 | || vq->pci_config.queue_avail_lo % 2 | |
1796 | || vq->pci_config.queue_used_lo % 4) | |
1797 | errx(1, "%s: invalid alignment in queue addresses", d->name); | |
1798 | ||
93153077 RR |
1799 | /* Initialize the virtqueue and check they're all in range. */ |
1800 | vq->vring.num = vq->pci_config.queue_size; | |
1801 | vq->vring.desc = check_pointer(vq->pci_config.queue_desc_lo, | |
1802 | sizeof(*vq->vring.desc) * vq->vring.num); | |
1803 | vq->vring.avail = check_pointer(vq->pci_config.queue_avail_lo, | |
1804 | sizeof(*vq->vring.avail) | |
1805 | + (sizeof(vq->vring.avail->ring[0]) | |
1806 | * vq->vring.num)); | |
1807 | vq->vring.used = check_pointer(vq->pci_config.queue_used_lo, | |
1808 | sizeof(*vq->vring.used) | |
1809 | + (sizeof(vq->vring.used->ring[0]) | |
1810 | * vq->vring.num)); | |
d39a6785 RR |
1811 | |
1812 | /* | |
1813 | * 2.4.9.1: | |
1814 | * | |
1815 | * The driver MUST initialize flags in the used ring to 0 | |
1816 | * when allocating the used ring. | |
1817 | */ | |
1818 | if (vq->vring.used->flags != 0) | |
1819 | errx(1, "%s: invalid initial used.flags %#x", | |
1820 | d->name, vq->vring.used->flags); | |
d761b032 | 1821 | } |
93153077 | 1822 | |
d761b032 RR |
1823 | static void start_virtqueue(struct virtqueue *vq) |
1824 | { | |
1825 | /* | |
1826 | * Create stack for thread. Since the stack grows upwards, we point | |
1827 | * the stack pointer to the end of this region. | |
1828 | */ | |
1829 | char *stack = malloc(32768); | |
93153077 RR |
1830 | |
1831 | /* Create a zero-initialized eventfd. */ | |
1832 | vq->eventfd = eventfd(0, 0); | |
1833 | if (vq->eventfd < 0) | |
1834 | err(1, "Creating eventfd"); | |
1835 | ||
1836 | /* | |
1837 | * CLONE_VM: because it has to access the Guest memory, and SIGCHLD so | |
1838 | * we get a signal if it dies. | |
1839 | */ | |
1840 | vq->thread = clone(do_thread, stack + 32768, CLONE_VM | SIGCHLD, vq); | |
1841 | if (vq->thread == (pid_t)-1) | |
1842 | err(1, "Creating clone"); | |
1843 | } | |
1844 | ||
d761b032 RR |
1845 | static void start_virtqueues(struct device *d) |
1846 | { | |
1847 | struct virtqueue *vq; | |
1848 | ||
1849 | for (vq = d->vq; vq; vq = vq->next) { | |
1850 | if (vq->pci_config.queue_enable) | |
1851 | start_virtqueue(vq); | |
1852 | } | |
1853 | } | |
1854 | ||
6a54f9ab RR |
1855 | static void emulate_mmio_write(struct device *d, u32 off, u32 val, u32 mask) |
1856 | { | |
93153077 RR |
1857 | struct virtqueue *vq; |
1858 | ||
1859 | switch (off) { | |
1860 | case offsetof(struct virtio_pci_mmio, cfg.device_feature_select): | |
8dc425ff RR |
1861 | /* |
1862 | * 4.1.4.3.1: | |
1863 | * | |
1864 | * The device MUST present the feature bits it is offering in | |
1865 | * device_feature, starting at bit device_feature_select ∗ 32 | |
1866 | * for any device_feature_select written by the driver | |
1867 | */ | |
93153077 RR |
1868 | if (val == 0) |
1869 | d->mmio->cfg.device_feature = d->features; | |
1870 | else if (val == 1) | |
1871 | d->mmio->cfg.device_feature = (d->features >> 32); | |
1872 | else | |
1873 | d->mmio->cfg.device_feature = 0; | |
d39a6785 | 1874 | goto feature_write_through32; |
93153077 RR |
1875 | case offsetof(struct virtio_pci_mmio, cfg.guest_feature_select): |
1876 | if (val > 1) | |
1877 | errx(1, "%s: Unexpected driver select %u", | |
1878 | d->name, val); | |
d39a6785 | 1879 | goto feature_write_through32; |
93153077 RR |
1880 | case offsetof(struct virtio_pci_mmio, cfg.guest_feature): |
1881 | if (d->mmio->cfg.guest_feature_select == 0) { | |
1882 | d->features_accepted &= ~((u64)0xFFFFFFFF); | |
1883 | d->features_accepted |= val; | |
1884 | } else { | |
1885 | assert(d->mmio->cfg.guest_feature_select == 1); | |
53aceb49 | 1886 | d->features_accepted &= 0xFFFFFFFF; |
93153077 RR |
1887 | d->features_accepted |= ((u64)val) << 32; |
1888 | } | |
d39a6785 RR |
1889 | /* |
1890 | * 2.2.1: | |
1891 | * | |
1892 | * The driver MUST NOT accept a feature which the device did | |
1893 | * not offer | |
1894 | */ | |
93153077 RR |
1895 | if (d->features_accepted & ~d->features) |
1896 | errx(1, "%s: over-accepted features %#llx of %#llx", | |
1897 | d->name, d->features_accepted, d->features); | |
d39a6785 RR |
1898 | goto feature_write_through32; |
1899 | case offsetof(struct virtio_pci_mmio, cfg.device_status): { | |
1900 | u8 prev; | |
1901 | ||
93153077 | 1902 | verbose("%s: device status -> %#x\n", d->name, val); |
8dc425ff RR |
1903 | /* |
1904 | * 4.1.4.3.1: | |
1905 | * | |
1906 | * The device MUST reset when 0 is written to device_status, | |
1907 | * and present a 0 in device_status once that is done. | |
1908 | */ | |
d39a6785 | 1909 | if (val == 0) { |
d9028eda | 1910 | reset_device(d); |
d39a6785 RR |
1911 | goto write_through8; |
1912 | } | |
1913 | ||
1914 | /* 2.1.1: The driver MUST NOT clear a device status bit. */ | |
1915 | if (d->mmio->cfg.device_status & ~val) | |
1916 | errx(1, "%s: unset of device status bit %#x -> %#x", | |
1917 | d->name, d->mmio->cfg.device_status, val); | |
d761b032 RR |
1918 | |
1919 | /* | |
1920 | * 2.1.2: | |
1921 | * | |
1922 | * The device MUST NOT consume buffers or notify the driver | |
1923 | * before DRIVER_OK. | |
1924 | */ | |
1925 | if (val & VIRTIO_CONFIG_S_DRIVER_OK | |
1926 | && !(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK)) | |
1927 | start_virtqueues(d); | |
1928 | ||
d39a6785 RR |
1929 | /* |
1930 | * 3.1.1: | |
1931 | * | |
1932 | * The driver MUST follow this sequence to initialize a device: | |
1933 | * - Reset the device. | |
1934 | * - Set the ACKNOWLEDGE status bit: the guest OS has | |
1935 | * notice the device. | |
1936 | * - Set the DRIVER status bit: the guest OS knows how | |
1937 | * to drive the device. | |
1938 | * - Read device feature bits, and write the subset | |
1939 | * of feature bits understood by the OS and driver | |
1940 | * to the device. During this step the driver MAY | |
1941 | * read (but MUST NOT write) the device-specific | |
1942 | * configuration fields to check that it can | |
1943 | * support the device before accepting it. | |
1944 | * - Set the FEATURES_OK status bit. The driver | |
1945 | * MUST not accept new feature bits after this | |
1946 | * step. | |
1947 | * - Re-read device status to ensure the FEATURES_OK | |
1948 | * bit is still set: otherwise, the device does | |
1949 | * not support our subset of features and the | |
1950 | * device is unusable. | |
1951 | * - Perform device-specific setup, including | |
1952 | * discovery of virtqueues for the device, | |
1953 | * optional per-bus setup, reading and possibly | |
1954 | * writing the device’s virtio configuration | |
1955 | * space, and population of virtqueues. | |
1956 | * - Set the DRIVER_OK status bit. At this point the | |
1957 | * device is “live”. | |
1958 | */ | |
1959 | prev = 0; | |
1960 | switch (val & ~d->mmio->cfg.device_status) { | |
1961 | case VIRTIO_CONFIG_S_DRIVER_OK: | |
1962 | prev |= VIRTIO_CONFIG_S_FEATURES_OK; /* fall thru */ | |
1963 | case VIRTIO_CONFIG_S_FEATURES_OK: | |
1964 | prev |= VIRTIO_CONFIG_S_DRIVER; /* fall thru */ | |
1965 | case VIRTIO_CONFIG_S_DRIVER: | |
1966 | prev |= VIRTIO_CONFIG_S_ACKNOWLEDGE; /* fall thru */ | |
1967 | case VIRTIO_CONFIG_S_ACKNOWLEDGE: | |
1968 | break; | |
1969 | default: | |
1970 | errx(1, "%s: unknown device status bit %#x -> %#x", | |
1971 | d->name, d->mmio->cfg.device_status, val); | |
1972 | } | |
1973 | if (d->mmio->cfg.device_status != prev) | |
1974 | errx(1, "%s: unexpected status transition %#x -> %#x", | |
1975 | d->name, d->mmio->cfg.device_status, val); | |
1976 | ||
1977 | /* If they just wrote FEATURES_OK, we make sure they read */ | |
1978 | switch (val & ~d->mmio->cfg.device_status) { | |
1979 | case VIRTIO_CONFIG_S_FEATURES_OK: | |
1980 | d->wrote_features_ok = true; | |
1981 | break; | |
1982 | case VIRTIO_CONFIG_S_DRIVER_OK: | |
1983 | if (d->wrote_features_ok) | |
1984 | errx(1, "%s: did not re-read FEATURES_OK", | |
1985 | d->name); | |
1986 | break; | |
1987 | } | |
93153077 | 1988 | goto write_through8; |
d39a6785 | 1989 | } |
93153077 RR |
1990 | case offsetof(struct virtio_pci_mmio, cfg.queue_select): |
1991 | vq = vq_by_num(d, val); | |
8dc425ff RR |
1992 | /* |
1993 | * 4.1.4.3.1: | |
1994 | * | |
1995 | * The device MUST present a 0 in queue_size if the virtqueue | |
1996 | * corresponding to the current queue_select is unavailable. | |
1997 | */ | |
93153077 RR |
1998 | if (!vq) { |
1999 | d->mmio->cfg.queue_size = 0; | |
2000 | goto write_through16; | |
2001 | } | |
2002 | /* Save registers for old vq, if it was a valid vq */ | |
2003 | if (d->mmio->cfg.queue_size) | |
2004 | save_vq_config(&d->mmio->cfg, | |
2005 | vq_by_num(d, d->mmio->cfg.queue_select)); | |
2006 | /* Restore the registers for the queue they asked for */ | |
2007 | restore_vq_config(&d->mmio->cfg, vq); | |
2008 | goto write_through16; | |
2009 | case offsetof(struct virtio_pci_mmio, cfg.queue_size): | |
c97eb679 RR |
2010 | /* |
2011 | * 4.1.4.3.2: | |
2012 | * | |
2013 | * The driver MUST NOT write a value which is not a power of 2 | |
2014 | * to queue_size. | |
2015 | */ | |
93153077 RR |
2016 | if (val & (val-1)) |
2017 | errx(1, "%s: invalid queue size %u\n", d->name, val); | |
2018 | if (d->mmio->cfg.queue_enable) | |
2019 | errx(1, "%s: changing queue size on live device", | |
2020 | d->name); | |
2021 | goto write_through16; | |
2022 | case offsetof(struct virtio_pci_mmio, cfg.queue_msix_vector): | |
2023 | errx(1, "%s: attempt to set MSIX vector to %u", | |
2024 | d->name, val); | |
d39a6785 RR |
2025 | case offsetof(struct virtio_pci_mmio, cfg.queue_enable): { |
2026 | struct virtqueue *vq = vq_by_num(d, d->mmio->cfg.queue_select); | |
2027 | ||
c97eb679 RR |
2028 | /* |
2029 | * 4.1.4.3.2: | |
2030 | * | |
2031 | * The driver MUST NOT write a 0 to queue_enable. | |
2032 | */ | |
93153077 RR |
2033 | if (val != 1) |
2034 | errx(1, "%s: setting queue_enable to %u", d->name, val); | |
d39a6785 | 2035 | |
c97eb679 | 2036 | /* |
d39a6785 | 2037 | * 3.1.1: |
c97eb679 | 2038 | * |
d39a6785 RR |
2039 | * 7. Perform device-specific setup, including discovery of |
2040 | * virtqueues for the device, optional per-bus setup, | |
2041 | * reading and possibly writing the device’s virtio | |
2042 | * configuration space, and population of virtqueues. | |
2043 | * 8. Set the DRIVER_OK status bit. | |
2044 | * | |
2045 | * All our devices require all virtqueues to be enabled, so | |
2046 | * they should have done that before setting DRIVER_OK. | |
c97eb679 | 2047 | */ |
d39a6785 RR |
2048 | if (d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK) |
2049 | errx(1, "%s: enabling vs after DRIVER_OK", d->name); | |
2050 | ||
2051 | d->mmio->cfg.queue_enable = val; | |
2052 | save_vq_config(&d->mmio->cfg, vq); | |
2053 | check_virtqueue(d, vq); | |
93153077 | 2054 | goto write_through16; |
d39a6785 | 2055 | } |
93153077 RR |
2056 | case offsetof(struct virtio_pci_mmio, cfg.queue_notify_off): |
2057 | errx(1, "%s: attempt to write to queue_notify_off", d->name); | |
2058 | case offsetof(struct virtio_pci_mmio, cfg.queue_desc_lo): | |
2059 | case offsetof(struct virtio_pci_mmio, cfg.queue_desc_hi): | |
2060 | case offsetof(struct virtio_pci_mmio, cfg.queue_avail_lo): | |
2061 | case offsetof(struct virtio_pci_mmio, cfg.queue_avail_hi): | |
2062 | case offsetof(struct virtio_pci_mmio, cfg.queue_used_lo): | |
2063 | case offsetof(struct virtio_pci_mmio, cfg.queue_used_hi): | |
c97eb679 RR |
2064 | /* |
2065 | * 4.1.4.3.2: | |
2066 | * | |
2067 | * The driver MUST configure the other virtqueue fields before | |
2068 | * enabling the virtqueue with queue_enable. | |
2069 | */ | |
93153077 RR |
2070 | if (d->mmio->cfg.queue_enable) |
2071 | errx(1, "%s: changing queue on live device", | |
2072 | d->name); | |
d39a6785 RR |
2073 | |
2074 | /* | |
2075 | * 3.1.1: | |
2076 | * | |
2077 | * The driver MUST follow this sequence to initialize a device: | |
2078 | *... | |
2079 | * 5. Set the FEATURES_OK status bit. The driver MUST not | |
2080 | * accept new feature bits after this step. | |
2081 | */ | |
2082 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_FEATURES_OK)) | |
2083 | errx(1, "%s: enabling vs before FEATURES_OK", d->name); | |
2084 | ||
2085 | /* | |
2086 | * 6. Re-read device status to ensure the FEATURES_OK bit is | |
2087 | * still set... | |
2088 | */ | |
2089 | if (d->wrote_features_ok) | |
2090 | errx(1, "%s: didn't re-read FEATURES_OK before setup", | |
2091 | d->name); | |
2092 | ||
93153077 RR |
2093 | goto write_through32; |
2094 | case offsetof(struct virtio_pci_mmio, notify): | |
2095 | vq = vq_by_num(d, val); | |
2096 | if (!vq) | |
2097 | errx(1, "Invalid vq notification on %u", val); | |
2098 | /* Notify the process handling this vq by adding 1 to eventfd */ | |
2099 | write(vq->eventfd, "\1\0\0\0\0\0\0\0", 8); | |
2100 | goto write_through16; | |
2101 | case offsetof(struct virtio_pci_mmio, isr): | |
2102 | errx(1, "%s: Unexpected write to isr", d->name); | |
e8330d9b RR |
2103 | /* Weird corner case: write to emerg_wr of console */ |
2104 | case sizeof(struct virtio_pci_mmio) | |
2105 | + offsetof(struct virtio_console_config, emerg_wr): | |
2106 | if (strcmp(d->name, "console") == 0) { | |
2107 | char c = val; | |
2108 | write(STDOUT_FILENO, &c, 1); | |
2109 | goto write_through32; | |
2110 | } | |
2111 | /* Fall through... */ | |
93153077 | 2112 | default: |
c97eb679 RR |
2113 | /* |
2114 | * 4.1.4.3.2: | |
2115 | * | |
2116 | * The driver MUST NOT write to device_feature, num_queues, | |
2117 | * config_generation or queue_notify_off. | |
2118 | */ | |
93153077 RR |
2119 | errx(1, "%s: Unexpected write to offset %u", d->name, off); |
2120 | } | |
2121 | ||
d39a6785 RR |
2122 | feature_write_through32: |
2123 | /* | |
2124 | * 3.1.1: | |
2125 | * | |
2126 | * The driver MUST follow this sequence to initialize a device: | |
2127 | *... | |
2128 | * - Set the DRIVER status bit: the guest OS knows how | |
2129 | * to drive the device. | |
2130 | * - Read device feature bits, and write the subset | |
2131 | * of feature bits understood by the OS and driver | |
2132 | * to the device. | |
2133 | *... | |
2134 | * - Set the FEATURES_OK status bit. The driver MUST not | |
2135 | * accept new feature bits after this step. | |
2136 | */ | |
2137 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER)) | |
2138 | errx(1, "%s: feature write before VIRTIO_CONFIG_S_DRIVER", | |
2139 | d->name); | |
2140 | if (d->mmio->cfg.device_status & VIRTIO_CONFIG_S_FEATURES_OK) | |
2141 | errx(1, "%s: feature write after VIRTIO_CONFIG_S_FEATURES_OK", | |
2142 | d->name); | |
c97eb679 RR |
2143 | |
2144 | /* | |
2145 | * 4.1.3.1: | |
2146 | * | |
2147 | * The driver MUST access each field using the “natural” access | |
2148 | * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses for | |
2149 | * 16-bit fields and 8-bit accesses for 8-bit fields. | |
2150 | */ | |
93153077 RR |
2151 | write_through32: |
2152 | if (mask != 0xFFFFFFFF) { | |
2153 | errx(1, "%s: non-32-bit write to offset %u (%#x)", | |
2154 | d->name, off, getreg(eip)); | |
2155 | return; | |
2156 | } | |
2157 | memcpy((char *)d->mmio + off, &val, 4); | |
2158 | return; | |
2159 | ||
2160 | write_through16: | |
2161 | if (mask != 0xFFFF) | |
2162 | errx(1, "%s: non-16-bit (%#x) write to offset %u (%#x)", | |
2163 | d->name, mask, off, getreg(eip)); | |
2164 | memcpy((char *)d->mmio + off, &val, 2); | |
2165 | return; | |
2166 | ||
2167 | write_through8: | |
2168 | if (mask != 0xFF) | |
2169 | errx(1, "%s: non-8-bit write to offset %u (%#x)", | |
2170 | d->name, off, getreg(eip)); | |
2171 | memcpy((char *)d->mmio + off, &val, 1); | |
2172 | return; | |
6a54f9ab RR |
2173 | } |
2174 | ||
2175 | static u32 emulate_mmio_read(struct device *d, u32 off, u32 mask) | |
2176 | { | |
93153077 RR |
2177 | u8 isr; |
2178 | u32 val = 0; | |
2179 | ||
2180 | switch (off) { | |
2181 | case offsetof(struct virtio_pci_mmio, cfg.device_feature_select): | |
2182 | case offsetof(struct virtio_pci_mmio, cfg.device_feature): | |
2183 | case offsetof(struct virtio_pci_mmio, cfg.guest_feature_select): | |
2184 | case offsetof(struct virtio_pci_mmio, cfg.guest_feature): | |
d39a6785 RR |
2185 | /* |
2186 | * 3.1.1: | |
2187 | * | |
2188 | * The driver MUST follow this sequence to initialize a device: | |
2189 | *... | |
2190 | * - Set the DRIVER status bit: the guest OS knows how | |
2191 | * to drive the device. | |
2192 | * - Read device feature bits, and write the subset | |
2193 | * of feature bits understood by the OS and driver | |
2194 | * to the device. | |
2195 | */ | |
2196 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER)) | |
2197 | errx(1, "%s: feature read before VIRTIO_CONFIG_S_DRIVER", | |
2198 | d->name); | |
93153077 RR |
2199 | goto read_through32; |
2200 | case offsetof(struct virtio_pci_mmio, cfg.msix_config): | |
2201 | errx(1, "%s: read of msix_config", d->name); | |
2202 | case offsetof(struct virtio_pci_mmio, cfg.num_queues): | |
2203 | goto read_through16; | |
2204 | case offsetof(struct virtio_pci_mmio, cfg.device_status): | |
d39a6785 RR |
2205 | /* As they did read, any write of FEATURES_OK is now fine. */ |
2206 | d->wrote_features_ok = false; | |
2207 | goto read_through8; | |
93153077 | 2208 | case offsetof(struct virtio_pci_mmio, cfg.config_generation): |
8dc425ff RR |
2209 | /* |
2210 | * 4.1.4.3.1: | |
2211 | * | |
2212 | * The device MUST present a changed config_generation after | |
2213 | * the driver has read a device-specific configuration value | |
2214 | * which has changed since any part of the device-specific | |
2215 | * configuration was last read. | |
2216 | * | |
2217 | * This is simple: none of our devices change config, so this | |
2218 | * is always 0. | |
2219 | */ | |
93153077 RR |
2220 | goto read_through8; |
2221 | case offsetof(struct virtio_pci_mmio, notify): | |
d39a6785 RR |
2222 | /* |
2223 | * 3.1.1: | |
2224 | * | |
2225 | * The driver MUST NOT notify the device before setting | |
2226 | * DRIVER_OK. | |
2227 | */ | |
2228 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER_OK)) | |
2229 | errx(1, "%s: notify before VIRTIO_CONFIG_S_DRIVER_OK", | |
2230 | d->name); | |
93153077 RR |
2231 | goto read_through16; |
2232 | case offsetof(struct virtio_pci_mmio, isr): | |
2233 | if (mask != 0xFF) | |
2234 | errx(1, "%s: non-8-bit read from offset %u (%#x)", | |
2235 | d->name, off, getreg(eip)); | |
93153077 | 2236 | isr = d->mmio->isr; |
8dc425ff RR |
2237 | /* |
2238 | * 4.1.4.5.1: | |
2239 | * | |
2240 | * The device MUST reset ISR status to 0 on driver read. | |
2241 | */ | |
93153077 RR |
2242 | d->mmio->isr = 0; |
2243 | return isr; | |
2244 | case offsetof(struct virtio_pci_mmio, padding): | |
2245 | errx(1, "%s: read from padding (%#x)", | |
2246 | d->name, getreg(eip)); | |
2247 | default: | |
2248 | /* Read from device config space, beware unaligned overflow */ | |
2249 | if (off > d->mmio_size - 4) | |
2250 | errx(1, "%s: read past end (%#x)", | |
2251 | d->name, getreg(eip)); | |
d39a6785 RR |
2252 | |
2253 | /* | |
2254 | * 3.1.1: | |
2255 | * The driver MUST follow this sequence to initialize a device: | |
2256 | *... | |
2257 | * 3. Set the DRIVER status bit: the guest OS knows how to | |
2258 | * drive the device. | |
2259 | * 4. Read device feature bits, and write the subset of | |
2260 | * feature bits understood by the OS and driver to the | |
2261 | * device. During this step the driver MAY read (but MUST NOT | |
2262 | * write) the device-specific configuration fields to check | |
2263 | * that it can support the device before accepting it. | |
2264 | */ | |
2265 | if (!(d->mmio->cfg.device_status & VIRTIO_CONFIG_S_DRIVER)) | |
2266 | errx(1, "%s: config read before VIRTIO_CONFIG_S_DRIVER", | |
2267 | d->name); | |
2268 | ||
93153077 RR |
2269 | if (mask == 0xFFFFFFFF) |
2270 | goto read_through32; | |
2271 | else if (mask == 0xFFFF) | |
2272 | goto read_through16; | |
2273 | else | |
2274 | goto read_through8; | |
2275 | } | |
2276 | ||
c97eb679 RR |
2277 | /* |
2278 | * 4.1.3.1: | |
2279 | * | |
2280 | * The driver MUST access each field using the “natural” access | |
2281 | * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses for | |
2282 | * 16-bit fields and 8-bit accesses for 8-bit fields. | |
2283 | */ | |
93153077 RR |
2284 | read_through32: |
2285 | if (mask != 0xFFFFFFFF) | |
2286 | errx(1, "%s: non-32-bit read to offset %u (%#x)", | |
2287 | d->name, off, getreg(eip)); | |
2288 | memcpy(&val, (char *)d->mmio + off, 4); | |
2289 | return val; | |
2290 | ||
2291 | read_through16: | |
2292 | if (mask != 0xFFFF) | |
2293 | errx(1, "%s: non-16-bit read to offset %u (%#x)", | |
2294 | d->name, off, getreg(eip)); | |
2295 | memcpy(&val, (char *)d->mmio + off, 2); | |
2296 | return val; | |
2297 | ||
2298 | read_through8: | |
2299 | if (mask != 0xFF) | |
2300 | errx(1, "%s: non-8-bit read to offset %u (%#x)", | |
2301 | d->name, off, getreg(eip)); | |
2302 | memcpy(&val, (char *)d->mmio + off, 1); | |
2303 | return val; | |
6a54f9ab RR |
2304 | } |
2305 | ||
2306 | static void emulate_mmio(unsigned long paddr, const u8 *insn) | |
2307 | { | |
2308 | u32 val, off, mask = 0xFFFFFFFF, insnlen = 0; | |
2309 | struct device *d = find_mmio_region(paddr, &off); | |
2310 | unsigned long args[] = { LHREQ_TRAP, 14 }; | |
2311 | ||
2312 | if (!d) { | |
2313 | warnx("MMIO touching %#08lx (not a device)", paddr); | |
2314 | goto reinject; | |
2315 | } | |
2316 | ||
2317 | /* Prefix makes it a 16 bit op */ | |
2318 | if (insn[0] == 0x66) { | |
2319 | mask = 0xFFFF; | |
2320 | insnlen++; | |
2321 | } | |
2322 | ||
2323 | /* iowrite */ | |
2324 | if (insn[insnlen] == 0x89) { | |
2325 | /* Next byte is r/m byte: bits 3-5 are register. */ | |
2326 | val = getreg_num((insn[insnlen+1] >> 3) & 0x7, mask); | |
2327 | emulate_mmio_write(d, off, val, mask); | |
2328 | insnlen += 2 + insn_displacement_len(insn[insnlen+1]); | |
2329 | } else if (insn[insnlen] == 0x8b) { /* ioread */ | |
2330 | /* Next byte is r/m byte: bits 3-5 are register. */ | |
2331 | val = emulate_mmio_read(d, off, mask); | |
2332 | setreg_num((insn[insnlen+1] >> 3) & 0x7, val, mask); | |
2333 | insnlen += 2 + insn_displacement_len(insn[insnlen+1]); | |
2334 | } else if (insn[0] == 0x88) { /* 8-bit iowrite */ | |
2335 | mask = 0xff; | |
2336 | /* Next byte is r/m byte: bits 3-5 are register. */ | |
2337 | val = getreg_num((insn[1] >> 3) & 0x7, mask); | |
2338 | emulate_mmio_write(d, off, val, mask); | |
2339 | insnlen = 2 + insn_displacement_len(insn[1]); | |
2340 | } else if (insn[0] == 0x8a) { /* 8-bit ioread */ | |
2341 | mask = 0xff; | |
2342 | val = emulate_mmio_read(d, off, mask); | |
2343 | setreg_num((insn[1] >> 3) & 0x7, val, mask); | |
2344 | insnlen = 2 + insn_displacement_len(insn[1]); | |
2345 | } else { | |
2346 | warnx("Unknown MMIO instruction touching %#08lx:" | |
2347 | " %02x %02x %02x %02x at %u", | |
2348 | paddr, insn[0], insn[1], insn[2], insn[3], getreg(eip)); | |
2349 | reinject: | |
2350 | /* Inject trap into Guest. */ | |
2351 | if (write(lguest_fd, args, sizeof(args)) < 0) | |
2352 | err(1, "Reinjecting trap 14 for fault at %#x", | |
2353 | getreg(eip)); | |
2354 | return; | |
2355 | } | |
2356 | ||
2357 | /* Finally, we've "done" the instruction, so move past it. */ | |
2358 | setreg(eip, getreg(eip) + insnlen); | |
2359 | } | |
c565650b | 2360 | |
dde79789 RR |
2361 | /*L:190 |
2362 | * Device Setup | |
2363 | * | |
2364 | * All devices need a descriptor so the Guest knows it exists, and a "struct | |
2365 | * device" so the Launcher can keep track of it. We have common helper | |
a6bd8e13 RR |
2366 | * routines to allocate and manage them. |
2367 | */ | |
93153077 RR |
2368 | static void add_pci_virtqueue(struct device *dev, |
2369 | void (*service)(struct virtqueue *)) | |
2370 | { | |
2371 | struct virtqueue **i, *vq = malloc(sizeof(*vq)); | |
2372 | ||
2373 | /* Initialize the virtqueue */ | |
2374 | vq->next = NULL; | |
2375 | vq->last_avail_idx = 0; | |
2376 | vq->dev = dev; | |
2377 | ||
2378 | /* | |
2379 | * This is the routine the service thread will run, and its Process ID | |
2380 | * once it's running. | |
2381 | */ | |
2382 | vq->service = service; | |
2383 | vq->thread = (pid_t)-1; | |
2384 | ||
2385 | /* Initialize the configuration. */ | |
d2dbdac3 | 2386 | reset_vq_pci_config(vq); |
93153077 RR |
2387 | vq->pci_config.queue_notify_off = 0; |
2388 | ||
2389 | /* Add one to the number of queues */ | |
2390 | vq->dev->mmio->cfg.num_queues++; | |
2391 | ||
93153077 RR |
2392 | /* |
2393 | * Add to tail of list, so dev->vq is first vq, dev->vq->next is | |
2394 | * second. | |
2395 | */ | |
2396 | for (i = &dev->vq; *i; i = &(*i)->next); | |
2397 | *i = vq; | |
2398 | } | |
2399 | ||
d9028eda | 2400 | /* The Guest accesses the feature bits via the PCI common config MMIO region */ |
93153077 RR |
2401 | static void add_pci_feature(struct device *dev, unsigned bit) |
2402 | { | |
2403 | dev->features |= (1ULL << bit); | |
2404 | } | |
2405 | ||
93153077 RR |
2406 | /* For devices with no config. */ |
2407 | static void no_device_config(struct device *dev) | |
2408 | { | |
2409 | dev->mmio_addr = get_mmio_region(dev->mmio_size); | |
2410 | ||
2411 | dev->config.bar[0] = dev->mmio_addr; | |
2412 | /* Bottom 4 bits must be zero */ | |
2413 | assert(~(dev->config.bar[0] & 0xF)); | |
2414 | } | |
2415 | ||
2416 | /* This puts the device config into BAR0 */ | |
2417 | static void set_device_config(struct device *dev, const void *conf, size_t len) | |
2418 | { | |
2419 | /* Set up BAR 0 */ | |
2420 | dev->mmio_size += len; | |
2421 | dev->mmio = realloc(dev->mmio, dev->mmio_size); | |
2422 | memcpy(dev->mmio + 1, conf, len); | |
2423 | ||
8dc425ff RR |
2424 | /* |
2425 | * 4.1.4.6: | |
2426 | * | |
2427 | * The device MUST present at least one VIRTIO_PCI_CAP_DEVICE_CFG | |
2428 | * capability for any device type which has a device-specific | |
2429 | * configuration. | |
2430 | */ | |
93153077 RR |
2431 | /* Hook up device cfg */ |
2432 | dev->config.cfg_access.cap.cap_next | |
2433 | = offsetof(struct pci_config, device); | |
2434 | ||
8dc425ff RR |
2435 | /* |
2436 | * 4.1.4.6.1: | |
2437 | * | |
2438 | * The offset for the device-specific configuration MUST be 4-byte | |
2439 | * aligned. | |
2440 | */ | |
2441 | assert(dev->config.cfg_access.cap.cap_next % 4 == 0); | |
2442 | ||
93153077 RR |
2443 | /* Fix up device cfg field length. */ |
2444 | dev->config.device.length = len; | |
2445 | ||
2446 | /* The rest is the same as the no-config case */ | |
2447 | no_device_config(dev); | |
2448 | } | |
2449 | ||
2450 | static void init_cap(struct virtio_pci_cap *cap, size_t caplen, int type, | |
2451 | size_t bar_offset, size_t bar_bytes, u8 next) | |
2452 | { | |
2453 | cap->cap_vndr = PCI_CAP_ID_VNDR; | |
2454 | cap->cap_next = next; | |
2455 | cap->cap_len = caplen; | |
2456 | cap->cfg_type = type; | |
2457 | cap->bar = 0; | |
2458 | memset(cap->padding, 0, sizeof(cap->padding)); | |
2459 | cap->offset = bar_offset; | |
2460 | cap->length = bar_bytes; | |
2461 | } | |
2462 | ||
2463 | /* | |
2464 | * This sets up the pci_config structure, as defined in the virtio 1.0 | |
2465 | * standard (and PCI standard). | |
2466 | */ | |
2467 | static void init_pci_config(struct pci_config *pci, u16 type, | |
2468 | u8 class, u8 subclass) | |
2469 | { | |
2470 | size_t bar_offset, bar_len; | |
2471 | ||
8dc425ff RR |
2472 | /* |
2473 | * 4.1.4.4.1: | |
2474 | * | |
2475 | * The device MUST either present notify_off_multiplier as an even | |
2476 | * power of 2, or present notify_off_multiplier as 0. | |
d39a6785 RR |
2477 | * |
2478 | * 2.1.2: | |
2479 | * | |
2480 | * The device MUST initialize device status to 0 upon reset. | |
8dc425ff | 2481 | */ |
93153077 RR |
2482 | memset(pci, 0, sizeof(*pci)); |
2483 | ||
2484 | /* 4.1.2.1: Devices MUST have the PCI Vendor ID 0x1AF4 */ | |
2485 | pci->vendor_id = 0x1AF4; | |
2486 | /* 4.1.2.1: ... PCI Device ID calculated by adding 0x1040 ... */ | |
2487 | pci->device_id = 0x1040 + type; | |
2488 | ||
2489 | /* | |
2490 | * PCI have specific codes for different types of devices. | |
2491 | * Linux doesn't care, but it's a good clue for people looking | |
2492 | * at the device. | |
93153077 RR |
2493 | */ |
2494 | pci->class = class; | |
2495 | pci->subclass = subclass; | |
2496 | ||
2497 | /* | |
8dc425ff RR |
2498 | * 4.1.2.1: |
2499 | * | |
2500 | * Non-transitional devices SHOULD have a PCI Revision ID of 1 or | |
2501 | * higher | |
93153077 RR |
2502 | */ |
2503 | pci->revid = 1; | |
2504 | ||
2505 | /* | |
8dc425ff RR |
2506 | * 4.1.2.1: |
2507 | * | |
2508 | * Non-transitional devices SHOULD have a PCI Subsystem Device ID of | |
2509 | * 0x40 or higher. | |
93153077 RR |
2510 | */ |
2511 | pci->subsystem_device_id = 0x40; | |
2512 | ||
2513 | /* We use our dummy interrupt controller, and irq_line is the irq */ | |
2514 | pci->irq_line = devices.next_irq++; | |
2515 | pci->irq_pin = 0; | |
2516 | ||
2517 | /* Support for extended capabilities. */ | |
2518 | pci->status = (1 << 4); | |
2519 | ||
2520 | /* Link them in. */ | |
8dc425ff RR |
2521 | /* |
2522 | * 4.1.4.3.1: | |
2523 | * | |
2524 | * The device MUST present at least one common configuration | |
2525 | * capability. | |
2526 | */ | |
93153077 RR |
2527 | pci->capabilities = offsetof(struct pci_config, common); |
2528 | ||
8dc425ff RR |
2529 | /* 4.1.4.3.1 ... offset MUST be 4-byte aligned. */ |
2530 | assert(pci->capabilities % 4 == 0); | |
2531 | ||
93153077 RR |
2532 | bar_offset = offsetof(struct virtio_pci_mmio, cfg); |
2533 | bar_len = sizeof(((struct virtio_pci_mmio *)0)->cfg); | |
2534 | init_cap(&pci->common, sizeof(pci->common), VIRTIO_PCI_CAP_COMMON_CFG, | |
2535 | bar_offset, bar_len, | |
2536 | offsetof(struct pci_config, notify)); | |
2537 | ||
8dc425ff RR |
2538 | /* |
2539 | * 4.1.4.4.1: | |
2540 | * | |
2541 | * The device MUST present at least one notification capability. | |
2542 | */ | |
93153077 RR |
2543 | bar_offset += bar_len; |
2544 | bar_len = sizeof(((struct virtio_pci_mmio *)0)->notify); | |
8dc425ff RR |
2545 | |
2546 | /* | |
2547 | * 4.1.4.4.1: | |
2548 | * | |
2549 | * The cap.offset MUST be 2-byte aligned. | |
2550 | */ | |
2551 | assert(pci->common.cap_next % 2 == 0); | |
2552 | ||
93153077 | 2553 | /* FIXME: Use a non-zero notify_off, for per-queue notification? */ |
8dc425ff RR |
2554 | /* |
2555 | * 4.1.4.4.1: | |
2556 | * | |
2557 | * The value cap.length presented by the device MUST be at least 2 and | |
2558 | * MUST be large enough to support queue notification offsets for all | |
2559 | * supported queues in all possible configurations. | |
2560 | */ | |
2561 | assert(bar_len >= 2); | |
2562 | ||
93153077 RR |
2563 | init_cap(&pci->notify.cap, sizeof(pci->notify), |
2564 | VIRTIO_PCI_CAP_NOTIFY_CFG, | |
2565 | bar_offset, bar_len, | |
2566 | offsetof(struct pci_config, isr)); | |
2567 | ||
2568 | bar_offset += bar_len; | |
2569 | bar_len = sizeof(((struct virtio_pci_mmio *)0)->isr); | |
8dc425ff RR |
2570 | /* |
2571 | * 4.1.4.5.1: | |
2572 | * | |
2573 | * The device MUST present at least one VIRTIO_PCI_CAP_ISR_CFG | |
2574 | * capability. | |
2575 | */ | |
93153077 RR |
2576 | init_cap(&pci->isr, sizeof(pci->isr), |
2577 | VIRTIO_PCI_CAP_ISR_CFG, | |
2578 | bar_offset, bar_len, | |
2579 | offsetof(struct pci_config, cfg_access)); | |
2580 | ||
8dc425ff RR |
2581 | /* |
2582 | * 4.1.4.7.1: | |
2583 | * | |
2584 | * The device MUST present at least one VIRTIO_PCI_CAP_PCI_CFG | |
2585 | * capability. | |
2586 | */ | |
93153077 RR |
2587 | /* This doesn't have any presence in the BAR */ |
2588 | init_cap(&pci->cfg_access.cap, sizeof(pci->cfg_access), | |
2589 | VIRTIO_PCI_CAP_PCI_CFG, | |
2590 | 0, 0, 0); | |
2591 | ||
2592 | bar_offset += bar_len + sizeof(((struct virtio_pci_mmio *)0)->padding); | |
2593 | assert(bar_offset == sizeof(struct virtio_pci_mmio)); | |
2594 | ||
2595 | /* | |
2596 | * This gets sewn in and length set in set_device_config(). | |
2597 | * Some devices don't have a device configuration interface, so | |
2598 | * we never expose this if we don't call set_device_config(). | |
2599 | */ | |
2600 | init_cap(&pci->device, sizeof(pci->device), VIRTIO_PCI_CAP_DEVICE_CFG, | |
2601 | bar_offset, 0, 0); | |
2602 | } | |
2603 | ||
2e04ef76 | 2604 | /* |
d9028eda RR |
2605 | * This routine does all the creation and setup of a new device, but we don't |
2606 | * actually place the MMIO region until we know the size (if any) of the | |
2607 | * device-specific config. And we don't actually start the service threads | |
2608 | * until later. | |
a6bd8e13 | 2609 | * |
2e04ef76 RR |
2610 | * See what I mean about userspace being boring? |
2611 | */ | |
93153077 RR |
2612 | static struct device *new_pci_device(const char *name, u16 type, |
2613 | u8 class, u8 subclass) | |
2614 | { | |
2615 | struct device *dev = malloc(sizeof(*dev)); | |
2616 | ||
2617 | /* Now we populate the fields one at a time. */ | |
93153077 RR |
2618 | dev->name = name; |
2619 | dev->vq = NULL; | |
93153077 | 2620 | dev->running = false; |
d39a6785 | 2621 | dev->wrote_features_ok = false; |
93153077 RR |
2622 | dev->mmio_size = sizeof(struct virtio_pci_mmio); |
2623 | dev->mmio = calloc(1, dev->mmio_size); | |
2624 | dev->features = (u64)1 << VIRTIO_F_VERSION_1; | |
2625 | dev->features_accepted = 0; | |
2626 | ||
d9028eda | 2627 | if (devices.device_num + 1 >= MAX_PCI_DEVICES) |
93153077 RR |
2628 | errx(1, "Can only handle 31 PCI devices"); |
2629 | ||
2630 | init_pci_config(&dev->config, type, class, subclass); | |
2631 | assert(!devices.pci[devices.device_num+1]); | |
2632 | devices.pci[++devices.device_num] = dev; | |
2633 | ||
2634 | return dev; | |
2635 | } | |
2636 | ||
2e04ef76 RR |
2637 | /* |
2638 | * Our first setup routine is the console. It's a fairly simple device, but | |
2639 | * UNIX tty handling makes it uglier than it could be. | |
2640 | */ | |
17cbca2b | 2641 | static void setup_console(void) |
8ca47e00 RR |
2642 | { |
2643 | struct device *dev; | |
e8330d9b | 2644 | struct virtio_console_config conf; |
8ca47e00 | 2645 | |
dde79789 | 2646 | /* If we can save the initial standard input settings... */ |
8ca47e00 RR |
2647 | if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { |
2648 | struct termios term = orig_term; | |
2e04ef76 RR |
2649 | /* |
2650 | * Then we turn off echo, line buffering and ^C etc: We want a | |
2651 | * raw input stream to the Guest. | |
2652 | */ | |
8ca47e00 RR |
2653 | term.c_lflag &= ~(ISIG|ICANON|ECHO); |
2654 | tcsetattr(STDIN_FILENO, TCSANOW, &term); | |
8ca47e00 RR |
2655 | } |
2656 | ||
ebff0113 | 2657 | dev = new_pci_device("console", VIRTIO_ID_CONSOLE, 0x07, 0x00); |
659a0e66 | 2658 | |
dde79789 | 2659 | /* We store the console state in dev->priv, and initialize it. */ |
8ca47e00 RR |
2660 | dev->priv = malloc(sizeof(struct console_abort)); |
2661 | ((struct console_abort *)dev->priv)->count = 0; | |
8ca47e00 | 2662 | |
2e04ef76 RR |
2663 | /* |
2664 | * The console needs two virtqueues: the input then the output. When | |
56ae43df RR |
2665 | * they put something the input queue, we make sure we're listening to |
2666 | * stdin. When they put something in the output queue, we write it to | |
2e04ef76 RR |
2667 | * stdout. |
2668 | */ | |
ebff0113 RR |
2669 | add_pci_virtqueue(dev, console_input); |
2670 | add_pci_virtqueue(dev, console_output); | |
2671 | ||
e8330d9b RR |
2672 | /* We need a configuration area for the emerg_wr early writes. */ |
2673 | add_pci_feature(dev, VIRTIO_CONSOLE_F_EMERG_WRITE); | |
2674 | set_device_config(dev, &conf, sizeof(conf)); | |
17cbca2b | 2675 | |
ebff0113 | 2676 | verbose("device %u: console\n", devices.device_num); |
8ca47e00 | 2677 | } |
17cbca2b | 2678 | /*:*/ |
8ca47e00 | 2679 | |
2e04ef76 RR |
2680 | /*M:010 |
2681 | * Inter-guest networking is an interesting area. Simplest is to have a | |
17cbca2b RR |
2682 | * --sharenet=<name> option which opens or creates a named pipe. This can be |
2683 | * used to send packets to another guest in a 1:1 manner. | |
dde79789 | 2684 | * |
9f54288d | 2685 | * More sophisticated is to use one of the tools developed for project like UML |
17cbca2b | 2686 | * to do networking. |
dde79789 | 2687 | * |
17cbca2b RR |
2688 | * Faster is to do virtio bonding in kernel. Doing this 1:1 would be |
2689 | * completely generic ("here's my vring, attach to your vring") and would work | |
2690 | * for any traffic. Of course, namespace and permissions issues need to be | |
2691 | * dealt with. A more sophisticated "multi-channel" virtio_net.c could hide | |
2692 | * multiple inter-guest channels behind one interface, although it would | |
2693 | * require some manner of hotplugging new virtio channels. | |
2694 | * | |
9f54288d | 2695 | * Finally, we could use a virtio network switch in the kernel, ie. vhost. |
2e04ef76 | 2696 | :*/ |
8ca47e00 RR |
2697 | |
2698 | static u32 str2ip(const char *ipaddr) | |
2699 | { | |
dec6a2be | 2700 | unsigned int b[4]; |
8ca47e00 | 2701 | |
dec6a2be MM |
2702 | if (sscanf(ipaddr, "%u.%u.%u.%u", &b[0], &b[1], &b[2], &b[3]) != 4) |
2703 | errx(1, "Failed to parse IP address '%s'", ipaddr); | |
2704 | return (b[0] << 24) | (b[1] << 16) | (b[2] << 8) | b[3]; | |
2705 | } | |
2706 | ||
2707 | static void str2mac(const char *macaddr, unsigned char mac[6]) | |
2708 | { | |
2709 | unsigned int m[6]; | |
2710 | if (sscanf(macaddr, "%02x:%02x:%02x:%02x:%02x:%02x", | |
2711 | &m[0], &m[1], &m[2], &m[3], &m[4], &m[5]) != 6) | |
2712 | errx(1, "Failed to parse mac address '%s'", macaddr); | |
2713 | mac[0] = m[0]; | |
2714 | mac[1] = m[1]; | |
2715 | mac[2] = m[2]; | |
2716 | mac[3] = m[3]; | |
2717 | mac[4] = m[4]; | |
2718 | mac[5] = m[5]; | |
8ca47e00 RR |
2719 | } |
2720 | ||
2e04ef76 RR |
2721 | /* |
2722 | * This code is "adapted" from libbridge: it attaches the Host end of the | |
dde79789 RR |
2723 | * network device to the bridge device specified by the command line. |
2724 | * | |
2725 | * This is yet another James Morris contribution (I'm an IP-level guy, so I | |
2e04ef76 RR |
2726 | * dislike bridging), and I just try not to break it. |
2727 | */ | |
8ca47e00 RR |
2728 | static void add_to_bridge(int fd, const char *if_name, const char *br_name) |
2729 | { | |
2730 | int ifidx; | |
2731 | struct ifreq ifr; | |
2732 | ||
2733 | if (!*br_name) | |
2734 | errx(1, "must specify bridge name"); | |
2735 | ||
2736 | ifidx = if_nametoindex(if_name); | |
2737 | if (!ifidx) | |
2738 | errx(1, "interface %s does not exist!", if_name); | |
2739 | ||
2740 | strncpy(ifr.ifr_name, br_name, IFNAMSIZ); | |
dec6a2be | 2741 | ifr.ifr_name[IFNAMSIZ-1] = '\0'; |
8ca47e00 RR |
2742 | ifr.ifr_ifindex = ifidx; |
2743 | if (ioctl(fd, SIOCBRADDIF, &ifr) < 0) | |
2744 | err(1, "can't add %s to bridge %s", if_name, br_name); | |
2745 | } | |
2746 | ||
2e04ef76 RR |
2747 | /* |
2748 | * This sets up the Host end of the network device with an IP address, brings | |
dde79789 | 2749 | * it up so packets will flow, the copies the MAC address into the hwaddr |
2e04ef76 RR |
2750 | * pointer. |
2751 | */ | |
dec6a2be | 2752 | static void configure_device(int fd, const char *tapif, u32 ipaddr) |
8ca47e00 RR |
2753 | { |
2754 | struct ifreq ifr; | |
f846619e | 2755 | struct sockaddr_in sin; |
8ca47e00 RR |
2756 | |
2757 | memset(&ifr, 0, sizeof(ifr)); | |
dec6a2be MM |
2758 | strcpy(ifr.ifr_name, tapif); |
2759 | ||
2760 | /* Don't read these incantations. Just cut & paste them like I did! */ | |
f846619e RR |
2761 | sin.sin_family = AF_INET; |
2762 | sin.sin_addr.s_addr = htonl(ipaddr); | |
2763 | memcpy(&ifr.ifr_addr, &sin, sizeof(sin)); | |
8ca47e00 | 2764 | if (ioctl(fd, SIOCSIFADDR, &ifr) != 0) |
dec6a2be | 2765 | err(1, "Setting %s interface address", tapif); |
8ca47e00 RR |
2766 | ifr.ifr_flags = IFF_UP; |
2767 | if (ioctl(fd, SIOCSIFFLAGS, &ifr) != 0) | |
dec6a2be MM |
2768 | err(1, "Bringing interface %s up", tapif); |
2769 | } | |
2770 | ||
dec6a2be | 2771 | static int get_tun_device(char tapif[IFNAMSIZ]) |
8ca47e00 | 2772 | { |
8ca47e00 | 2773 | struct ifreq ifr; |
bf6d4034 | 2774 | int vnet_hdr_sz; |
dec6a2be MM |
2775 | int netfd; |
2776 | ||
2777 | /* Start with this zeroed. Messy but sure. */ | |
2778 | memset(&ifr, 0, sizeof(ifr)); | |
8ca47e00 | 2779 | |
2e04ef76 RR |
2780 | /* |
2781 | * We open the /dev/net/tun device and tell it we want a tap device. A | |
dde79789 RR |
2782 | * tap device is like a tun device, only somehow different. To tell |
2783 | * the truth, I completely blundered my way through this code, but it | |
2e04ef76 RR |
2784 | * works now! |
2785 | */ | |
8ca47e00 | 2786 | netfd = open_or_die("/dev/net/tun", O_RDWR); |
398f187d | 2787 | ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; |
8ca47e00 RR |
2788 | strcpy(ifr.ifr_name, "tap%d"); |
2789 | if (ioctl(netfd, TUNSETIFF, &ifr) != 0) | |
2790 | err(1, "configuring /dev/net/tun"); | |
dec6a2be | 2791 | |
398f187d RR |
2792 | if (ioctl(netfd, TUNSETOFFLOAD, |
2793 | TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) | |
2794 | err(1, "Could not set features for tun device"); | |
2795 | ||
2e04ef76 RR |
2796 | /* |
2797 | * We don't need checksums calculated for packets coming in this | |
2798 | * device: trust us! | |
2799 | */ | |
8ca47e00 RR |
2800 | ioctl(netfd, TUNSETNOCSUM, 1); |
2801 | ||
bf6d4034 RR |
2802 | /* |
2803 | * In virtio before 1.0 (aka legacy virtio), we added a 16-bit | |
2804 | * field at the end of the network header iff | |
2805 | * VIRTIO_NET_F_MRG_RXBUF was negotiated. For virtio 1.0, | |
2806 | * that became the norm, but we need to tell the tun device | |
2807 | * about our expanded header (which is called | |
2808 | * virtio_net_hdr_mrg_rxbuf in the legacy system). | |
2809 | */ | |
2810 | vnet_hdr_sz = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2811 | if (ioctl(netfd, TUNSETVNETHDRSZ, &vnet_hdr_sz) != 0) | |
2812 | err(1, "Setting tun header size to %u", vnet_hdr_sz); | |
2813 | ||
dec6a2be MM |
2814 | memcpy(tapif, ifr.ifr_name, IFNAMSIZ); |
2815 | return netfd; | |
2816 | } | |
2817 | ||
2e04ef76 RR |
2818 | /*L:195 |
2819 | * Our network is a Host<->Guest network. This can either use bridging or | |
dec6a2be MM |
2820 | * routing, but the principle is the same: it uses the "tun" device to inject |
2821 | * packets into the Host as if they came in from a normal network card. We | |
2e04ef76 RR |
2822 | * just shunt packets between the Guest and the tun device. |
2823 | */ | |
dec6a2be MM |
2824 | static void setup_tun_net(char *arg) |
2825 | { | |
2826 | struct device *dev; | |
659a0e66 RR |
2827 | struct net_info *net_info = malloc(sizeof(*net_info)); |
2828 | int ipfd; | |
dec6a2be MM |
2829 | u32 ip = INADDR_ANY; |
2830 | bool bridging = false; | |
2831 | char tapif[IFNAMSIZ], *p; | |
2832 | struct virtio_net_config conf; | |
2833 | ||
659a0e66 | 2834 | net_info->tunfd = get_tun_device(tapif); |
dec6a2be | 2835 | |
17cbca2b | 2836 | /* First we create a new network device. */ |
bf6d4034 | 2837 | dev = new_pci_device("net", VIRTIO_ID_NET, 0x02, 0x00); |
659a0e66 | 2838 | dev->priv = net_info; |
dde79789 | 2839 | |
2e04ef76 | 2840 | /* Network devices need a recv and a send queue, just like console. */ |
bf6d4034 RR |
2841 | add_pci_virtqueue(dev, net_input); |
2842 | add_pci_virtqueue(dev, net_output); | |
8ca47e00 | 2843 | |
2e04ef76 RR |
2844 | /* |
2845 | * We need a socket to perform the magic network ioctls to bring up the | |
2846 | * tap interface, connect to the bridge etc. Any socket will do! | |
2847 | */ | |
8ca47e00 RR |
2848 | ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); |
2849 | if (ipfd < 0) | |
2850 | err(1, "opening IP socket"); | |
2851 | ||
dde79789 | 2852 | /* If the command line was --tunnet=bridge:<name> do bridging. */ |
8ca47e00 | 2853 | if (!strncmp(BRIDGE_PFX, arg, strlen(BRIDGE_PFX))) { |
dec6a2be MM |
2854 | arg += strlen(BRIDGE_PFX); |
2855 | bridging = true; | |
2856 | } | |
2857 | ||
2858 | /* A mac address may follow the bridge name or IP address */ | |
2859 | p = strchr(arg, ':'); | |
2860 | if (p) { | |
2861 | str2mac(p+1, conf.mac); | |
bf6d4034 | 2862 | add_pci_feature(dev, VIRTIO_NET_F_MAC); |
dec6a2be | 2863 | *p = '\0'; |
dec6a2be MM |
2864 | } |
2865 | ||
2866 | /* arg is now either an IP address or a bridge name */ | |
2867 | if (bridging) | |
2868 | add_to_bridge(ipfd, tapif, arg); | |
2869 | else | |
8ca47e00 RR |
2870 | ip = str2ip(arg); |
2871 | ||
dec6a2be MM |
2872 | /* Set up the tun device. */ |
2873 | configure_device(ipfd, tapif, ip); | |
8ca47e00 | 2874 | |
398f187d | 2875 | /* Expect Guest to handle everything except UFO */ |
bf6d4034 RR |
2876 | add_pci_feature(dev, VIRTIO_NET_F_CSUM); |
2877 | add_pci_feature(dev, VIRTIO_NET_F_GUEST_CSUM); | |
2878 | add_pci_feature(dev, VIRTIO_NET_F_GUEST_TSO4); | |
2879 | add_pci_feature(dev, VIRTIO_NET_F_GUEST_TSO6); | |
2880 | add_pci_feature(dev, VIRTIO_NET_F_GUEST_ECN); | |
2881 | add_pci_feature(dev, VIRTIO_NET_F_HOST_TSO4); | |
2882 | add_pci_feature(dev, VIRTIO_NET_F_HOST_TSO6); | |
2883 | add_pci_feature(dev, VIRTIO_NET_F_HOST_ECN); | |
d1f0132e | 2884 | /* We handle indirect ring entries */ |
bf6d4034 RR |
2885 | add_pci_feature(dev, VIRTIO_RING_F_INDIRECT_DESC); |
2886 | set_device_config(dev, &conf, sizeof(conf)); | |
8ca47e00 | 2887 | |
a586d4f6 | 2888 | /* We don't need the socket any more; setup is done. */ |
8ca47e00 RR |
2889 | close(ipfd); |
2890 | ||
dec6a2be MM |
2891 | if (bridging) |
2892 | verbose("device %u: tun %s attached to bridge: %s\n", | |
2893 | devices.device_num, tapif, arg); | |
2894 | else | |
2895 | verbose("device %u: tun %s: %s\n", | |
2896 | devices.device_num, tapif, arg); | |
8ca47e00 | 2897 | } |
a91d74a3 | 2898 | /*:*/ |
17cbca2b | 2899 | |
e1e72965 | 2900 | /* This hangs off device->priv. */ |
1842f23c | 2901 | struct vblk_info { |
17cbca2b RR |
2902 | /* The size of the file. */ |
2903 | off64_t len; | |
2904 | ||
2905 | /* The file descriptor for the file. */ | |
2906 | int fd; | |
2907 | ||
17cbca2b RR |
2908 | }; |
2909 | ||
e1e72965 RR |
2910 | /*L:210 |
2911 | * The Disk | |
2912 | * | |
a91d74a3 RR |
2913 | * The disk only has one virtqueue, so it only has one thread. It is really |
2914 | * simple: the Guest asks for a block number and we read or write that position | |
2915 | * in the file. | |
2916 | * | |
2917 | * Before we serviced each virtqueue in a separate thread, that was unacceptably | |
2918 | * slow: the Guest waits until the read is finished before running anything | |
2919 | * else, even if it could have been doing useful work. | |
2920 | * | |
2921 | * We could have used async I/O, except it's reputed to suck so hard that | |
2922 | * characters actually go missing from your code when you try to use it. | |
e1e72965 | 2923 | */ |
659a0e66 | 2924 | static void blk_request(struct virtqueue *vq) |
17cbca2b | 2925 | { |
659a0e66 | 2926 | struct vblk_info *vblk = vq->dev->priv; |
17cbca2b | 2927 | unsigned int head, out_num, in_num, wlen; |
c0316a94 | 2928 | int ret, i; |
cb38fa23 | 2929 | u8 *in; |
c0316a94 | 2930 | struct virtio_blk_outhdr out; |
659a0e66 | 2931 | struct iovec iov[vq->vring.num]; |
17cbca2b RR |
2932 | off64_t off; |
2933 | ||
a91d74a3 RR |
2934 | /* |
2935 | * Get the next request, where we normally wait. It triggers the | |
2936 | * interrupt to acknowledge previously serviced requests (if any). | |
2937 | */ | |
659a0e66 | 2938 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
17cbca2b | 2939 | |
c0316a94 RR |
2940 | /* Copy the output header from the front of the iov (adjusts iov) */ |
2941 | iov_consume(iov, out_num, &out, sizeof(out)); | |
2942 | ||
2943 | /* Find and trim end of iov input array, for our status byte. */ | |
2944 | in = NULL; | |
2945 | for (i = out_num + in_num - 1; i >= out_num; i--) { | |
2946 | if (iov[i].iov_len > 0) { | |
2947 | in = iov[i].iov_base + iov[i].iov_len - 1; | |
2948 | iov[i].iov_len--; | |
2949 | break; | |
2950 | } | |
2951 | } | |
2952 | if (!in) | |
2953 | errx(1, "Bad virtblk cmd with no room for status"); | |
17cbca2b | 2954 | |
a91d74a3 RR |
2955 | /* |
2956 | * For historical reasons, block operations are expressed in 512 byte | |
2957 | * "sectors". | |
2958 | */ | |
c0316a94 | 2959 | off = out.sector * 512; |
17cbca2b | 2960 | |
50516547 | 2961 | if (out.type & VIRTIO_BLK_T_OUT) { |
2e04ef76 RR |
2962 | /* |
2963 | * Write | |
2964 | * | |
2965 | * Move to the right location in the block file. This can fail | |
2966 | * if they try to write past end. | |
2967 | */ | |
17cbca2b | 2968 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
c0316a94 | 2969 | err(1, "Bad seek to sector %llu", out.sector); |
17cbca2b | 2970 | |
c0316a94 RR |
2971 | ret = writev(vblk->fd, iov, out_num); |
2972 | verbose("WRITE to sector %llu: %i\n", out.sector, ret); | |
17cbca2b | 2973 | |
2e04ef76 RR |
2974 | /* |
2975 | * Grr... Now we know how long the descriptor they sent was, we | |
17cbca2b | 2976 | * make sure they didn't try to write over the end of the block |
2e04ef76 RR |
2977 | * file (possibly extending it). |
2978 | */ | |
17cbca2b RR |
2979 | if (ret > 0 && off + ret > vblk->len) { |
2980 | /* Trim it back to the correct length */ | |
2981 | ftruncate64(vblk->fd, vblk->len); | |
2982 | /* Die, bad Guest, die. */ | |
2983 | errx(1, "Write past end %llu+%u", off, ret); | |
2984 | } | |
7bc9fdda TH |
2985 | |
2986 | wlen = sizeof(*in); | |
2987 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); | |
c0316a94 | 2988 | } else if (out.type & VIRTIO_BLK_T_FLUSH) { |
7bc9fdda TH |
2989 | /* Flush */ |
2990 | ret = fdatasync(vblk->fd); | |
2991 | verbose("FLUSH fdatasync: %i\n", ret); | |
1200e646 | 2992 | wlen = sizeof(*in); |
cb38fa23 | 2993 | *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); |
17cbca2b | 2994 | } else { |
2e04ef76 RR |
2995 | /* |
2996 | * Read | |
2997 | * | |
2998 | * Move to the right location in the block file. This can fail | |
2999 | * if they try to read past end. | |
3000 | */ | |
17cbca2b | 3001 | if (lseek64(vblk->fd, off, SEEK_SET) != off) |
c0316a94 | 3002 | err(1, "Bad seek to sector %llu", out.sector); |
17cbca2b | 3003 | |
c0316a94 | 3004 | ret = readv(vblk->fd, iov + out_num, in_num); |
17cbca2b | 3005 | if (ret >= 0) { |
1200e646 | 3006 | wlen = sizeof(*in) + ret; |
cb38fa23 | 3007 | *in = VIRTIO_BLK_S_OK; |
17cbca2b | 3008 | } else { |
1200e646 | 3009 | wlen = sizeof(*in); |
cb38fa23 | 3010 | *in = VIRTIO_BLK_S_IOERR; |
17cbca2b RR |
3011 | } |
3012 | } | |
3013 | ||
a91d74a3 | 3014 | /* Finished that request. */ |
38bc2b8c | 3015 | add_used(vq, head, wlen); |
17cbca2b RR |
3016 | } |
3017 | ||
e1e72965 | 3018 | /*L:198 This actually sets up a virtual block device. */ |
17cbca2b RR |
3019 | static void setup_block_file(const char *filename) |
3020 | { | |
17cbca2b RR |
3021 | struct device *dev; |
3022 | struct vblk_info *vblk; | |
a586d4f6 | 3023 | struct virtio_blk_config conf; |
17cbca2b | 3024 | |
50516547 RR |
3025 | /* Create the device. */ |
3026 | dev = new_pci_device("block", VIRTIO_ID_BLOCK, 0x01, 0x80); | |
17cbca2b | 3027 | |
e1e72965 | 3028 | /* The device has one virtqueue, where the Guest places requests. */ |
50516547 | 3029 | add_pci_virtqueue(dev, blk_request); |
17cbca2b RR |
3030 | |
3031 | /* Allocate the room for our own bookkeeping */ | |
3032 | vblk = dev->priv = malloc(sizeof(*vblk)); | |
3033 | ||
3034 | /* First we open the file and store the length. */ | |
3035 | vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); | |
3036 | vblk->len = lseek64(vblk->fd, 0, SEEK_END); | |
3037 | ||
3038 | /* Tell Guest how many sectors this device has. */ | |
a586d4f6 | 3039 | conf.capacity = cpu_to_le64(vblk->len / 512); |
17cbca2b | 3040 | |
2e04ef76 RR |
3041 | /* |
3042 | * Tell Guest not to put in too many descriptors at once: two are used | |
3043 | * for the in and out elements. | |
3044 | */ | |
50516547 | 3045 | add_pci_feature(dev, VIRTIO_BLK_F_SEG_MAX); |
a586d4f6 RR |
3046 | conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); |
3047 | ||
50516547 | 3048 | set_device_config(dev, &conf, sizeof(struct virtio_blk_config)); |
17cbca2b | 3049 | |
17cbca2b | 3050 | verbose("device %u: virtblock %llu sectors\n", |
50516547 | 3051 | devices.device_num, le64_to_cpu(conf.capacity)); |
17cbca2b | 3052 | } |
28fd6d7f | 3053 | |
2e04ef76 | 3054 | /*L:211 |
a454bb36 | 3055 | * Our random number generator device reads from /dev/urandom into the Guest's |
28fd6d7f | 3056 | * input buffers. The usual case is that the Guest doesn't want random numbers |
a454bb36 | 3057 | * and so has no buffers although /dev/urandom is still readable, whereas |
28fd6d7f RR |
3058 | * console is the reverse. |
3059 | * | |
2e04ef76 RR |
3060 | * The same logic applies, however. |
3061 | */ | |
3062 | struct rng_info { | |
3063 | int rfd; | |
3064 | }; | |
3065 | ||
659a0e66 | 3066 | static void rng_input(struct virtqueue *vq) |
28fd6d7f RR |
3067 | { |
3068 | int len; | |
3069 | unsigned int head, in_num, out_num, totlen = 0; | |
659a0e66 RR |
3070 | struct rng_info *rng_info = vq->dev->priv; |
3071 | struct iovec iov[vq->vring.num]; | |
28fd6d7f RR |
3072 | |
3073 | /* First we need a buffer from the Guests's virtqueue. */ | |
659a0e66 | 3074 | head = wait_for_vq_desc(vq, iov, &out_num, &in_num); |
28fd6d7f RR |
3075 | if (out_num) |
3076 | errx(1, "Output buffers in rng?"); | |
3077 | ||
2e04ef76 | 3078 | /* |
a91d74a3 RR |
3079 | * Just like the console write, we loop to cover the whole iovec. |
3080 | * In this case, short reads actually happen quite a bit. | |
2e04ef76 | 3081 | */ |
28fd6d7f | 3082 | while (!iov_empty(iov, in_num)) { |
659a0e66 | 3083 | len = readv(rng_info->rfd, iov, in_num); |
28fd6d7f | 3084 | if (len <= 0) |
a454bb36 | 3085 | err(1, "Read from /dev/urandom gave %i", len); |
c0316a94 | 3086 | iov_consume(iov, in_num, NULL, len); |
28fd6d7f RR |
3087 | totlen += len; |
3088 | } | |
3089 | ||
3090 | /* Tell the Guest about the new input. */ | |
38bc2b8c | 3091 | add_used(vq, head, totlen); |
28fd6d7f RR |
3092 | } |
3093 | ||
2e04ef76 RR |
3094 | /*L:199 |
3095 | * This creates a "hardware" random number device for the Guest. | |
3096 | */ | |
28fd6d7f RR |
3097 | static void setup_rng(void) |
3098 | { | |
3099 | struct device *dev; | |
659a0e66 | 3100 | struct rng_info *rng_info = malloc(sizeof(*rng_info)); |
28fd6d7f | 3101 | |
a454bb36 RR |
3102 | /* Our device's private info simply contains the /dev/urandom fd. */ |
3103 | rng_info->rfd = open_or_die("/dev/urandom", O_RDONLY); | |
28fd6d7f | 3104 | |
2e04ef76 | 3105 | /* Create the new device. */ |
0d5b5d39 | 3106 | dev = new_pci_device("rng", VIRTIO_ID_RNG, 0xff, 0); |
659a0e66 | 3107 | dev->priv = rng_info; |
28fd6d7f RR |
3108 | |
3109 | /* The device has one virtqueue, where the Guest places inbufs. */ | |
0d5b5d39 | 3110 | add_pci_virtqueue(dev, rng_input); |
28fd6d7f | 3111 | |
0d5b5d39 RR |
3112 | /* We don't have any configuration space */ |
3113 | no_device_config(dev); | |
3114 | ||
3115 | verbose("device %u: rng\n", devices.device_num); | |
28fd6d7f | 3116 | } |
a6bd8e13 | 3117 | /* That's the end of device setup. */ |
ec04b13f | 3118 | |
a6bd8e13 | 3119 | /*L:230 Reboot is pretty easy: clean up and exec() the Launcher afresh. */ |
ec04b13f BR |
3120 | static void __attribute__((noreturn)) restart_guest(void) |
3121 | { | |
3122 | unsigned int i; | |
3123 | ||
2e04ef76 RR |
3124 | /* |
3125 | * Since we don't track all open fds, we simply close everything beyond | |
3126 | * stderr. | |
3127 | */ | |
ec04b13f BR |
3128 | for (i = 3; i < FD_SETSIZE; i++) |
3129 | close(i); | |
8c79873d | 3130 | |
659a0e66 RR |
3131 | /* Reset all the devices (kills all threads). */ |
3132 | cleanup_devices(); | |
3133 | ||
ec04b13f BR |
3134 | execv(main_args[0], main_args); |
3135 | err(1, "Could not exec %s", main_args[0]); | |
3136 | } | |
8ca47e00 | 3137 | |
2e04ef76 RR |
3138 | /*L:220 |
3139 | * Finally we reach the core of the Launcher which runs the Guest, serves | |
3140 | * its input and output, and finally, lays it to rest. | |
3141 | */ | |
56739c80 | 3142 | static void __attribute__((noreturn)) run_guest(void) |
8ca47e00 RR |
3143 | { |
3144 | for (;;) { | |
69a09dc1 | 3145 | struct lguest_pending notify; |
8ca47e00 RR |
3146 | int readval; |
3147 | ||
3148 | /* We read from the /dev/lguest device to run the Guest. */ | |
69a09dc1 | 3149 | readval = pread(lguest_fd, ¬ify, sizeof(notify), cpu_id); |
69a09dc1 | 3150 | if (readval == sizeof(notify)) { |
00f8d546 | 3151 | if (notify.trap == 13) { |
c565650b RR |
3152 | verbose("Emulating instruction at %#x\n", |
3153 | getreg(eip)); | |
3154 | emulate_insn(notify.insn); | |
6a54f9ab RR |
3155 | } else if (notify.trap == 14) { |
3156 | verbose("Emulating MMIO at %#x\n", | |
3157 | getreg(eip)); | |
3158 | emulate_mmio(notify.addr, notify.insn); | |
69a09dc1 RR |
3159 | } else |
3160 | errx(1, "Unknown trap %i addr %#08x\n", | |
3161 | notify.trap, notify.addr); | |
dde79789 | 3162 | /* ENOENT means the Guest died. Reading tells us why. */ |
8ca47e00 RR |
3163 | } else if (errno == ENOENT) { |
3164 | char reason[1024] = { 0 }; | |
e3283fa0 | 3165 | pread(lguest_fd, reason, sizeof(reason)-1, cpu_id); |
8ca47e00 | 3166 | errx(1, "%s", reason); |
ec04b13f BR |
3167 | /* ERESTART means that we need to reboot the guest */ |
3168 | } else if (errno == ERESTART) { | |
3169 | restart_guest(); | |
659a0e66 RR |
3170 | /* Anything else means a bug or incompatible change. */ |
3171 | } else | |
8ca47e00 | 3172 | err(1, "Running guest failed"); |
8ca47e00 RR |
3173 | } |
3174 | } | |
a6bd8e13 | 3175 | /*L:240 |
e1e72965 RR |
3176 | * This is the end of the Launcher. The good news: we are over halfway |
3177 | * through! The bad news: the most fiendish part of the code still lies ahead | |
3178 | * of us. | |
dde79789 | 3179 | * |
e1e72965 RR |
3180 | * Are you ready? Take a deep breath and join me in the core of the Host, in |
3181 | * "make Host". | |
2e04ef76 | 3182 | :*/ |
8ca47e00 RR |
3183 | |
3184 | static struct option opts[] = { | |
3185 | { "verbose", 0, NULL, 'v' }, | |
8ca47e00 RR |
3186 | { "tunnet", 1, NULL, 't' }, |
3187 | { "block", 1, NULL, 'b' }, | |
28fd6d7f | 3188 | { "rng", 0, NULL, 'r' }, |
8ca47e00 | 3189 | { "initrd", 1, NULL, 'i' }, |
8aeb36e8 PS |
3190 | { "username", 1, NULL, 'u' }, |
3191 | { "chroot", 1, NULL, 'c' }, | |
8ca47e00 RR |
3192 | { NULL }, |
3193 | }; | |
3194 | static void usage(void) | |
3195 | { | |
3196 | errx(1, "Usage: lguest [--verbose] " | |
dec6a2be | 3197 | "[--tunnet=(<ipaddr>:<macaddr>|bridge:<bridgename>:<macaddr>)\n" |
8ca47e00 RR |
3198 | "|--block=<filename>|--initrd=<filename>]...\n" |
3199 | "<mem-in-mb> vmlinux [args...]"); | |
3200 | } | |
3201 | ||
3c6b5bfa | 3202 | /*L:105 The main routine is where the real work begins: */ |
8ca47e00 RR |
3203 | int main(int argc, char *argv[]) |
3204 | { | |
2e04ef76 | 3205 | /* Memory, code startpoint and size of the (optional) initrd. */ |
58a24566 | 3206 | unsigned long mem = 0, start, initrd_size = 0; |
56739c80 RR |
3207 | /* Two temporaries. */ |
3208 | int i, c; | |
3c6b5bfa | 3209 | /* The boot information for the Guest. */ |
43d33b21 | 3210 | struct boot_params *boot; |
dde79789 | 3211 | /* If they specify an initrd file to load. */ |
8ca47e00 RR |
3212 | const char *initrd_name = NULL; |
3213 | ||
8aeb36e8 PS |
3214 | /* Password structure for initgroups/setres[gu]id */ |
3215 | struct passwd *user_details = NULL; | |
3216 | ||
3217 | /* Directory to chroot to */ | |
3218 | char *chroot_path = NULL; | |
3219 | ||
ec04b13f BR |
3220 | /* Save the args: we "reboot" by execing ourselves again. */ |
3221 | main_args = argv; | |
ec04b13f | 3222 | |
2e04ef76 | 3223 | /* |
d9028eda RR |
3224 | * First we initialize the device list. We remember next interrupt |
3225 | * number to use for devices (1: remember that 0 is used by the timer). | |
2e04ef76 | 3226 | */ |
17cbca2b | 3227 | devices.next_irq = 1; |
8ca47e00 | 3228 | |
a91d74a3 | 3229 | /* We're CPU 0. In fact, that's the only CPU possible right now. */ |
e3283fa0 | 3230 | cpu_id = 0; |
a91d74a3 | 3231 | |
2e04ef76 RR |
3232 | /* |
3233 | * We need to know how much memory so we can set up the device | |
dde79789 RR |
3234 | * descriptor and memory pages for the devices as we parse the command |
3235 | * line. So we quickly look through the arguments to find the amount | |
2e04ef76 RR |
3236 | * of memory now. |
3237 | */ | |
6570c459 RR |
3238 | for (i = 1; i < argc; i++) { |
3239 | if (argv[i][0] != '-') { | |
3c6b5bfa | 3240 | mem = atoi(argv[i]) * 1024 * 1024; |
2e04ef76 RR |
3241 | /* |
3242 | * We start by mapping anonymous pages over all of | |
3c6b5bfa RR |
3243 | * guest-physical memory range. This fills it with 0, |
3244 | * and ensures that the Guest won't be killed when it | |
2e04ef76 RR |
3245 | * tries to access it. |
3246 | */ | |
3c6b5bfa RR |
3247 | guest_base = map_zeroed_pages(mem / getpagesize() |
3248 | + DEVICE_PAGES); | |
3249 | guest_limit = mem; | |
0a6bcc18 | 3250 | guest_max = guest_mmio = mem + DEVICE_PAGES*getpagesize(); |
6570c459 RR |
3251 | break; |
3252 | } | |
3253 | } | |
dde79789 | 3254 | |
713e3f72 RR |
3255 | /* We always have a console device, and it's always device 1. */ |
3256 | setup_console(); | |
3257 | ||
dde79789 | 3258 | /* The options are fairly straight-forward */ |
8ca47e00 RR |
3259 | while ((c = getopt_long(argc, argv, "v", opts, NULL)) != EOF) { |
3260 | switch (c) { | |
3261 | case 'v': | |
3262 | verbose = true; | |
3263 | break; | |
8ca47e00 | 3264 | case 't': |
17cbca2b | 3265 | setup_tun_net(optarg); |
8ca47e00 RR |
3266 | break; |
3267 | case 'b': | |
17cbca2b | 3268 | setup_block_file(optarg); |
8ca47e00 | 3269 | break; |
28fd6d7f RR |
3270 | case 'r': |
3271 | setup_rng(); | |
3272 | break; | |
8ca47e00 RR |
3273 | case 'i': |
3274 | initrd_name = optarg; | |
3275 | break; | |
8aeb36e8 PS |
3276 | case 'u': |
3277 | user_details = getpwnam(optarg); | |
3278 | if (!user_details) | |
3279 | err(1, "getpwnam failed, incorrect username?"); | |
3280 | break; | |
3281 | case 'c': | |
3282 | chroot_path = optarg; | |
3283 | break; | |
8ca47e00 RR |
3284 | default: |
3285 | warnx("Unknown argument %s", argv[optind]); | |
3286 | usage(); | |
3287 | } | |
3288 | } | |
2e04ef76 RR |
3289 | /* |
3290 | * After the other arguments we expect memory and kernel image name, | |
3291 | * followed by command line arguments for the kernel. | |
3292 | */ | |
8ca47e00 RR |
3293 | if (optind + 2 > argc) |
3294 | usage(); | |
3295 | ||
3c6b5bfa RR |
3296 | verbose("Guest base is at %p\n", guest_base); |
3297 | ||
8e709469 RR |
3298 | /* Initialize the (fake) PCI host bridge device. */ |
3299 | init_pci_host_bridge(); | |
3300 | ||
8ca47e00 | 3301 | /* Now we load the kernel */ |
47436aa4 | 3302 | start = load_kernel(open_or_die(argv[optind+1], O_RDONLY)); |
8ca47e00 | 3303 | |
3c6b5bfa RR |
3304 | /* Boot information is stashed at physical address 0 */ |
3305 | boot = from_guest_phys(0); | |
3306 | ||
dde79789 | 3307 | /* Map the initrd image if requested (at top of physical memory) */ |
8ca47e00 RR |
3308 | if (initrd_name) { |
3309 | initrd_size = load_initrd(initrd_name, mem); | |
2e04ef76 RR |
3310 | /* |
3311 | * These are the location in the Linux boot header where the | |
3312 | * start and size of the initrd are expected to be found. | |
3313 | */ | |
43d33b21 RR |
3314 | boot->hdr.ramdisk_image = mem - initrd_size; |
3315 | boot->hdr.ramdisk_size = initrd_size; | |
dde79789 | 3316 | /* The bootloader type 0xFF means "unknown"; that's OK. */ |
43d33b21 | 3317 | boot->hdr.type_of_loader = 0xFF; |
8ca47e00 RR |
3318 | } |
3319 | ||
2e04ef76 RR |
3320 | /* |
3321 | * The Linux boot header contains an "E820" memory map: ours is a | |
3322 | * simple, single region. | |
3323 | */ | |
43d33b21 RR |
3324 | boot->e820_entries = 1; |
3325 | boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); | |
2e04ef76 RR |
3326 | /* |
3327 | * The boot header contains a command line pointer: we put the command | |
3328 | * line after the boot header. | |
3329 | */ | |
43d33b21 | 3330 | boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); |
e1e72965 | 3331 | /* We use a simple helper to copy the arguments separated by spaces. */ |
43d33b21 | 3332 | concat((char *)(boot + 1), argv+optind+2); |
dde79789 | 3333 | |
e22a5398 RR |
3334 | /* Set kernel alignment to 16M (CONFIG_PHYSICAL_ALIGN) */ |
3335 | boot->hdr.kernel_alignment = 0x1000000; | |
3336 | ||
814a0e5c | 3337 | /* Boot protocol version: 2.07 supports the fields for lguest. */ |
43d33b21 | 3338 | boot->hdr.version = 0x207; |
814a0e5c RR |
3339 | |
3340 | /* The hardware_subarch value of "1" tells the Guest it's an lguest. */ | |
43d33b21 | 3341 | boot->hdr.hardware_subarch = 1; |
814a0e5c | 3342 | |
43d33b21 RR |
3343 | /* Tell the entry path not to try to reload segment registers. */ |
3344 | boot->hdr.loadflags |= KEEP_SEGMENTS; | |
8ca47e00 | 3345 | |
9f54288d | 3346 | /* We tell the kernel to initialize the Guest. */ |
56739c80 | 3347 | tell_kernel(start); |
dde79789 | 3348 | |
a91d74a3 | 3349 | /* Ensure that we terminate if a device-servicing child dies. */ |
659a0e66 RR |
3350 | signal(SIGCHLD, kill_launcher); |
3351 | ||
3352 | /* If we exit via err(), this kills all the threads, restores tty. */ | |
3353 | atexit(cleanup_devices); | |
8ca47e00 | 3354 | |
8aeb36e8 PS |
3355 | /* If requested, chroot to a directory */ |
3356 | if (chroot_path) { | |
3357 | if (chroot(chroot_path) != 0) | |
3358 | err(1, "chroot(\"%s\") failed", chroot_path); | |
3359 | ||
3360 | if (chdir("/") != 0) | |
3361 | err(1, "chdir(\"/\") failed"); | |
3362 | ||
3363 | verbose("chroot done\n"); | |
3364 | } | |
3365 | ||
3366 | /* If requested, drop privileges */ | |
3367 | if (user_details) { | |
3368 | uid_t u; | |
3369 | gid_t g; | |
3370 | ||
3371 | u = user_details->pw_uid; | |
3372 | g = user_details->pw_gid; | |
3373 | ||
3374 | if (initgroups(user_details->pw_name, g) != 0) | |
3375 | err(1, "initgroups failed"); | |
3376 | ||
3377 | if (setresgid(g, g, g) != 0) | |
3378 | err(1, "setresgid failed"); | |
3379 | ||
3380 | if (setresuid(u, u, u) != 0) | |
3381 | err(1, "setresuid failed"); | |
3382 | ||
3383 | verbose("Dropping privileges completed\n"); | |
3384 | } | |
3385 | ||
dde79789 | 3386 | /* Finally, run the Guest. This doesn't return. */ |
56739c80 | 3387 | run_guest(); |
8ca47e00 | 3388 | } |
f56a384e RR |
3389 | /*:*/ |
3390 | ||
3391 | /*M:999 | |
3392 | * Mastery is done: you now know everything I do. | |
3393 | * | |
3394 | * But surely you have seen code, features and bugs in your wanderings which | |
3395 | * you now yearn to attack? That is the real game, and I look forward to you | |
3396 | * patching and forking lguest into the Your-Name-Here-visor. | |
3397 | * | |
3398 | * Farewell, and good coding! | |
3399 | * Rusty Russell. | |
3400 | */ |