dune: add a dune command
authorRonald G. Minnich <rminnich@gmail.com>
Mon, 21 Nov 2016 17:03:50 +0000 (09:03 -0800)
committerBarret Rhoden <brho@cs.berkeley.edu>
Tue, 22 Nov 2016 00:23:55 +0000 (19:23 -0500)
This is much like the Stanford dune system in that it is
designed to run simple non-kernels that support user
mode programs. It lets us show the ease of implementation
of such a command in the Akaros VM model.

To start, this is just a clone of vmrunkernel.

Change-Id: I2ac0fdddd3e834e6d9ea06d75c166a60d1fb4775
Signed-off-by: Ronald G. Minnich <rminnich@gmail.com>
Signed-off-by: Barret Rhoden <brho@cs.berkeley.edu>
tests/Makefile
tests/dune/Makefrag [new file with mode: 0644]
tests/dune/dune.c [new file with mode: 0644]

index 3127f37..204a3de 100644 (file)
@@ -26,6 +26,7 @@ TESTS_EXECS_CPP  = $(patsubst $(TESTS_DIR)/%.cc, \
 
 include $(TESTS_DIR)/openmp/Makefrag
 include $(TESTS_DIR)/vmm/Makefrag
+include $(TESTS_DIR)/dune/Makefrag
 
 TESTS_EXECS = $(TESTS_EXECS_C) $(TESTS_EXECS_CPP)
 
diff --git a/tests/dune/Makefrag b/tests/dune/Makefrag
new file mode 100644 (file)
index 0000000..3014ec3
--- /dev/null
@@ -0,0 +1,25 @@
+DUNE_TESTS_DIR = $(TESTS_DIR)/dune
+
+DUNE_TESTS_CFLAGS += $(CFLAGS_TESTS) -Wunused -Werror
+
+ALL_DUNE_TEST_FILES := $(wildcard $(DUNE_TESTS_DIR)/*.c)
+
+DUNE_TESTS_LDLIBS := $(TESTS_LDLIBS) -lelf
+
+DUNE_TESTS_SRCS := $(ALL_DUNE_TEST_FILES)
+
+DUNE_TESTS_LDDEPENDS := $(DUNE_TESTS_DIR)/%.c 
+
+TESTS_EXECS_C  += $(patsubst $(DUNE_TESTS_DIR)/%.c, \
+                      $(OBJDIR)/$(DUNE_TESTS_DIR)/%, \
+                      $(DUNE_TESTS_SRCS))
+
+STATIC := $(findstring static,$(DUNE_TESTS_CFLAGS))
+$(OBJDIR)/$(DUNE_TESTS_DIR)/%: $(DUNE_TESTS_LDDEPENDS)
+       @echo + cc [DUNE_TESTS] $<
+       @mkdir -p $(@D)
+       $(Q)$(CC) $(DUNE_TESTS_CFLAGS) -o $@ $< $(DUNE_TESTS_LDLIBS)
+       @if [ "$(STATIC)" != "static" ]; then \
+               $(OBJDUMP) -S $@ > $@.asm; \
+               $(NM) -n $@ > $@.sym; \
+       fi
diff --git a/tests/dune/dune.c b/tests/dune/dune.c
new file mode 100644 (file)
index 0000000..7e5844a
--- /dev/null
@@ -0,0 +1,860 @@
+#include <stdio.h>
+#include <pthread.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <parlib/arch/arch.h>
+#include <parlib/ros_debug.h>
+#include <unistd.h>
+#include <gelf.h>
+#include <errno.h>
+#include <libelf.h>
+#include <dirent.h>
+#include <stdlib.h>
+#include <string.h>
+#include <ros/syscall.h>
+#include <sys/mman.h>
+#include <vmm/vmm.h>
+#include <vmm/acpi/acpi.h>
+#include <vmm/acpi/vmm_simple_dsdt.h>
+#include <ros/arch/mmu.h>
+#include <ros/arch/membar.h>
+#include <ros/vmm.h>
+#include <parlib/uthread.h>
+#include <vmm/linux_bootparam.h>
+#include <getopt.h>
+
+#include <vmm/virtio.h>
+#include <vmm/virtio_blk.h>
+#include <vmm/virtio_mmio.h>
+#include <vmm/virtio_ids.h>
+#include <vmm/virtio_config.h>
+#include <vmm/virtio_console.h>
+#include <vmm/virtio_net.h>
+#include <vmm/virtio_lguest_console.h>
+
+#include <vmm/sched.h>
+#include <sys/eventfd.h>
+#include <sys/uio.h>
+
+struct virtual_machine local_vm, *vm = &local_vm;
+
+struct vmm_gpcore_init gpci;
+
+/* By 1999, you could just scan the hardware
+ * and work it out. But 2005, that was no longer possible. How sad.
+ * so we have to fake acpi to make it all work.
+ * This will be copied to memory at 0xe0000, so the kernel can find it.
+ */
+
+/* assume they're all 256 bytes long just to make it easy.
+ * Just have pointers that point to aligned things.
+ */
+
+struct acpi_table_rsdp rsdp = {
+       .signature = ACPI_SIG_RSDP,
+       .oem_id = "AKAROS",
+       .revision = 2,
+       .length = 36,
+};
+
+struct acpi_table_xsdt xsdt = {
+       .header = {
+               .signature = ACPI_SIG_DSDT,
+               .revision = 2,
+               .oem_id = "AKAROS",
+               .oem_table_id = "ALPHABET",
+               .oem_revision = 0,
+               .asl_compiler_id = "RON ",
+               .asl_compiler_revision = 0,
+       },
+};
+struct acpi_table_fadt fadt = {
+       .header = {
+               .signature = ACPI_SIG_FADT,
+               .revision = 2,
+               .oem_id = "AKAROS",
+               .oem_table_id = "ALPHABET",
+               .oem_revision = 0,
+               .asl_compiler_id = "RON ",
+               .asl_compiler_revision = 0,
+       },
+};
+
+
+/* This has to be dropped into memory, then the other crap just follows it.
+ */
+struct acpi_table_madt madt = {
+       .header = {
+               .signature = ACPI_SIG_MADT,
+               .revision = 2,
+               .oem_id = "AKAROS",
+               .oem_table_id = "ALPHABET",
+               .oem_revision = 0,
+               .asl_compiler_id = "RON ",
+               .asl_compiler_revision = 0,
+       },
+
+       .address = 0xfee00000ULL,
+       .flags = 0,
+};
+
+struct acpi_madt_local_apic Apic0 = {.header = {.type = ACPI_MADT_TYPE_LOCAL_APIC, .length = sizeof(struct acpi_madt_local_apic)},
+                                    .processor_id = 0, .id = 0, .lapic_flags = 1};
+struct acpi_madt_io_apic Apic1 = {.header = {.type = ACPI_MADT_TYPE_IO_APIC, .length = sizeof(struct acpi_madt_io_apic)},
+                                 .id = 0, .address = 0xfec00000, .global_irq_base = 0};
+struct acpi_madt_local_x2apic X2Apic0 = {
+       .header = {
+               .type = ACPI_MADT_TYPE_LOCAL_X2APIC,
+               .length = sizeof(struct acpi_madt_local_x2apic)
+       },
+       .local_apic_id = 0,
+       .uid = 0
+};
+
+struct acpi_madt_interrupt_override isor[] = {
+       /* From the ACPI Specification Version 6.1:
+        * For example, if your machine has the ISA Programmable Interrupt Timer
+        * (PIT) connected to ISA IRQ 0, but in APIC mode, it is connected to I/O
+        * APIC interrupt input 2, then you would need an Interrupt Source Override
+        * where the source entry is ‘0’ and the Global System Interrupt is ‘2.’
+        */
+};
+
+
+/* this test will run the "kernel" in the negative address space. We hope. */
+void *low1m;
+volatile int shared = 0;
+volatile int quit = 0;
+
+/* total hack. If the vm runs away we want to get control again. */
+unsigned int maxresume = (unsigned int) -1;
+
+#define MiB 0x100000ull
+#define GiB (1ull << 30)
+#define MinMemory (16*MiB)
+void *kernel;
+unsigned long long memsize = GiB;
+uintptr_t memstart = MinMemory;
+uintptr_t stack;
+unsigned long long *p512, *p1, *p2m;
+
+void **my_retvals;
+int nr_threads = 4;
+int debug = 0;
+int resumeprompt = 0;
+/* unlike Linux, this shared struct is for both host and guest. */
+//     struct virtqueue *constoguest =
+//             vring_new_virtqueue(0, 512, 8192, 0, inpages, NULL, NULL, "test");
+
+void vapic_status_dump(FILE *f, void *vapic);
+
+#if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 1)
+#error "Get a gcc newer than 4.4.0"
+#else
+#define BITOP_ADDR(x) "+m" (*(volatile long *) (x))
+#endif
+
+#define LOCK_PREFIX "lock "
+#define ADDR                           BITOP_ADDR(addr)
+static inline int test_and_set_bit(int nr, volatile unsigned long *addr);
+static int default_nic = 1;
+
+pthread_t timerthread_struct;
+
+void timer_thread(void *arg)
+{
+       uint8_t vector;
+       uint32_t initial_count;
+       while (1) {
+               vector = ((uint32_t *)gpci.vapic_addr)[0x32] & 0xff;
+               initial_count = ((uint32_t *)gpci.vapic_addr)[0x38];
+               if (vector && initial_count)
+                       vmm_interrupt_guest(vm, 0, vector);
+               uthread_usleep(100000);
+       }
+       fprintf(stderr, "SENDING TIMER\n");
+}
+
+
+// FIXME.
+volatile int consdata = 0;
+
+/* TODO: pass a core id to poke_guest */
+static void virtio_poke_guest(uint8_t vec)
+{
+       vmm_interrupt_guest(vm, 0, vec);
+}
+
+static struct virtio_mmio_dev cons_mmio_dev = {
+       .poke_guest = virtio_poke_guest,
+};
+
+static struct virtio_console_config cons_cfg;
+static struct virtio_console_config cons_cfg_d;
+
+static struct virtio_vq_dev cons_vqdev = {
+       .name = "console",
+       .dev_id = VIRTIO_ID_CONSOLE,
+       .dev_feat =
+           (1ULL << VIRTIO_F_VERSION_1) | (1 << VIRTIO_RING_F_INDIRECT_DESC),
+       .num_vqs = 2,
+       .cfg = &cons_cfg,
+       .cfg_d = &cons_cfg_d,
+       .cfg_sz = sizeof(struct virtio_console_config),
+       .transport_dev = &cons_mmio_dev,
+       .vqs = {
+                       {
+                               .name = "cons_receiveq",
+                               .qnum_max = 64,
+                               .srv_fn = cons_receiveq_fn,
+                               .vqdev = &cons_vqdev
+                       },
+                       {
+                               .name = "cons_transmitq",
+                               .qnum_max = 64,
+                               .srv_fn = cons_transmitq_fn,
+                               .vqdev = &cons_vqdev
+                       },
+               }
+};
+
+static struct virtio_mmio_dev net_mmio_dev = {
+       .poke_guest = virtio_poke_guest,
+};
+
+static struct virtio_net_config net_cfg = {
+       .max_virtqueue_pairs = 1
+};
+static struct virtio_net_config net_cfg_d = {
+       .max_virtqueue_pairs = 1
+};
+
+static struct virtio_vq_dev net_vqdev = {
+       .name = "network",
+       .dev_id = VIRTIO_ID_NET,
+       .dev_feat = (1ULL << VIRTIO_F_VERSION_1 | 1 << VIRTIO_NET_F_MAC),
+
+       .num_vqs = 2,
+       .cfg = &net_cfg,
+       .cfg_d = &net_cfg_d,
+       .cfg_sz = sizeof(struct virtio_net_config),
+       .transport_dev = &net_mmio_dev,
+       .vqs = {
+               {
+                       .name = "net_receiveq",
+                       .qnum_max = 64,
+                       .srv_fn = net_receiveq_fn,
+                       .vqdev = &net_vqdev
+               },
+               {
+                       .name = "net_transmitq",
+                       .qnum_max = 64,
+                       .srv_fn = net_transmitq_fn,
+                       .vqdev = &net_vqdev
+               },
+       }
+};
+
+static struct virtio_mmio_dev blk_mmio_dev = {
+       .poke_guest = virtio_poke_guest,
+};
+
+static struct virtio_blk_config blk_cfg = {
+};
+
+static struct virtio_blk_config blk_cfg_d = {
+};
+
+static struct virtio_vq_dev blk_vqdev = {
+       .name = "block",
+       .dev_id = VIRTIO_ID_BLOCK,
+       .dev_feat = (1ULL << VIRTIO_F_VERSION_1),
+
+       .num_vqs = 1,
+       .cfg = &blk_cfg,
+       .cfg_d = &blk_cfg_d,
+       .cfg_sz = sizeof(struct virtio_blk_config),
+       .transport_dev = &blk_mmio_dev,
+       .vqs = {
+               {
+                       .name = "blk_request",
+                       .qnum_max = 64,
+                       .srv_fn = blk_request,
+                       .vqdev = &blk_vqdev
+               },
+       }
+};
+
+void lowmem() {
+       __asm__ __volatile__ (".section .lowmem, \"aw\"\n\tlow: \n\t.=0x1000\n\t.align 0x100000\n\t.previous\n");
+}
+
+static uint8_t acpi_tb_checksum(uint8_t *buffer, uint32_t length)
+{
+       uint8_t sum = 0;
+       uint8_t *end = buffer + length;
+       fprintf(stderr, "tbchecksum %p for %d", buffer, length);
+       while (buffer < end) {
+               if (end - buffer < 2)
+                       fprintf(stderr, "%02x\n", sum);
+               sum = (uint8_t)(sum + *(buffer++));
+       }
+       fprintf(stderr, " is %02x\n", sum);
+       return (sum);
+}
+
+static void gencsum(uint8_t *target, void *data, int len)
+{
+       uint8_t csum;
+       // blast target to zero so it does not get counted
+       // (it might be in the struct we checksum) And, yes, it is, goodness.
+       fprintf(stderr, "gencsum %p target %p source %d bytes\n", target, data, len);
+       *target = 0;
+       csum  = acpi_tb_checksum((uint8_t *)data, len);
+       *target = ~csum + 1;
+       fprintf(stderr, "Cmoputed is %02x\n", *target);
+}
+
+static inline int test_and_set_bit(int nr, volatile unsigned long *addr)
+{
+       int oldbit;
+
+       asm volatile(LOCK_PREFIX "bts %2,%1\n\t"
+                    "sbb %0,%0" : "=r" (oldbit), ADDR : "Ir" (nr) : "memory");
+
+       return oldbit;
+}
+
+/* load_kernel loads an ELF file as a kernel. */
+uintptr_t
+load_kernel(char *filename)
+{
+       Elf64_Ehdr *ehdr;
+       Elf *elf;
+       size_t phnum = 0;
+       Elf64_Phdr *hdrs;
+       int fd;
+
+       elf_version(EV_CURRENT);
+       fd = open(filename, O_RDONLY);
+       if (fd < 0) {
+               fprintf(stderr, "Can't open %s: %r\n", filename);
+               return 0;
+       }
+
+       elf = elf_begin(fd, ELF_C_READ, NULL);
+       if (elf == NULL) {
+               fprintf(stderr, "%s: cannot read %s ELF file.\n", __func__, filename);
+               close(fd);
+               return 0;
+       }
+
+       ehdr = elf64_getehdr(elf);
+       if (ehdr == NULL) {
+               fprintf(stderr, "%s: cannot get exec header of %s.\n",
+                       __func__, filename);
+               goto fail;
+       }
+       fprintf(stderr, "%s ELF entry point is %p\n", filename,
+               (void *)ehdr->e_entry);
+
+       if (elf_getphdrnum(elf, &phnum) < 0) {
+               fprintf(stderr, "%s: cannot get program header num of %s.\n",
+                       __func__, filename);
+               goto fail;
+       }
+       fprintf(stderr, "%s has %p program headers\n", filename, phnum);
+
+       hdrs = elf64_getphdr(elf);
+       if (hdrs == NULL) {
+               fprintf(stderr, "%s: cannot get program headers of %s.\n",
+                       __func__, filename);
+               goto fail;
+       }
+
+       for (int i = 0; i < phnum; i++) {
+               size_t tot;
+               Elf64_Phdr *h = &hdrs[i];
+               uintptr_t pa;
+
+               fprintf(stderr,
+                       "%d: type 0x%lx flags 0x%lx  offset 0x%lx vaddr 0x%lx paddr 0x%lx size 0x%lx  memsz 0x%lx align 0x%lx\n",
+                       i,
+                       h->p_type,              /* Segment type */
+                       h->p_flags,             /* Segment flags */
+                       h->p_offset,            /* Segment file offset */
+                       h->p_vaddr,             /* Segment virtual address */
+                       h->p_paddr,             /* Segment physical address */
+                       h->p_filesz,            /* Segment size in file */
+                       h->p_memsz,             /* Segment size in memory */
+                       h->p_align              /* Segment alignment */);
+               if (h->p_type != PT_LOAD)
+                       continue;
+               if ((h->p_flags & (PF_R | PF_W | PF_X)) == 0)
+                       continue;
+
+               pa = h->p_paddr;
+               fprintf(stderr,
+                       "Read header %d @offset %p to %p (elf PA is %p) %d bytes:",
+                       i, h->p_offset, pa, h->p_paddr, h->p_filesz);
+               tot = 0;
+               while (tot < h->p_filesz) {
+                       int amt = pread(fd, (void *)(pa + tot), h->p_filesz - tot,
+                                       h->p_offset + tot);
+                       if (amt < 1)
+                               break;
+                       tot += amt;
+               }
+               fprintf(stderr, "read a total of %d bytes\n", tot);
+               if (tot < h->p_filesz) {
+                       fprintf(stderr, "%s: got %d bytes, wanted %d bytes\n",
+                               filename, tot, h->p_filesz);
+                       goto fail;
+               }
+       }
+
+       close(fd);
+       elf_end(elf);
+       return ehdr->e_entry;
+ fail:
+       close(fd);
+       elf_end(elf);
+       return 0;
+}
+
+int main(int argc, char **argv)
+{
+       struct boot_params *bp;
+       char cmdline_default[512] = {0};
+       char *cmdline_extra = "\0";
+       char *cmdline;
+       void *a = (void *)0xe0000;
+       struct acpi_table_rsdp *r;
+       struct acpi_table_fadt *f;
+       struct acpi_table_madt *m;
+       struct acpi_table_xsdt *x;
+       int vmmflags = 0; // Disabled probably forever. VMM_VMCALL_PRINTF;
+       uint64_t entry = 0;
+       int ret;
+       uint8_t csum;
+       void *a_page;
+       struct vm_trapframe *vm_tf;
+       uint64_t tsc_freq_khz;
+       char *cmdlinep;
+       int cmdlinesz, len, cmdline_fd;
+       char *disk_image_file = NULL;
+       int c;
+       struct stat stat_result;
+       int num_read;
+       int option_index;
+       static struct option long_options[] = {
+               {"debug",         no_argument,       0, 'd'},
+               {"vmm_vmcall",    no_argument,       0, 'v'},
+               {"maxresume",     required_argument, 0, 'R'},
+               {"memsize",       required_argument, 0, 'm'},
+               {"memstart",      required_argument, 0, 'M'},
+               {"stack",         required_argument, 0, 'S'},
+               {"cmdline_extra", required_argument, 0, 'c'},
+               {"greedy",        no_argument,       0, 'g'},
+               {"scp",           no_argument,       0, 's'},
+               {"image_file",    required_argument, 0, 'f'},
+               {"cmdline",       required_argument, 0, 'k'},
+               {"nic",           required_argument, 0, 'n'},
+               {"help",          no_argument,       0, 'h'},
+               {0, 0, 0, 0}
+       };
+
+       fprintf(stderr, "%p %p %p %p\n", PGSIZE, PGSHIFT, PML1_SHIFT,
+                       PML1_PTE_REACH);
+
+       if ((uintptr_t)__procinfo.program_end >= MinMemory) {
+               fprintf(stderr,
+                       "Panic: vmrunkernel binary extends into guest memory\n");
+               exit(1);
+       }
+
+       vm->low4k = malloc(PGSIZE);
+       memset(vm->low4k, 0xff, PGSIZE);
+       vm->low4k[0x40e] = 0;
+       vm->low4k[0x40f] = 0;
+
+       //Place mmap(Gan)
+       a_page = mmap((void *)0xfee00000, PGSIZE, PROT_READ | PROT_WRITE,
+                     MAP_POPULATE | MAP_ANONYMOUS, -1, 0);
+       fprintf(stderr, "a_page mmap pointer %p\n", a_page);
+
+       if (a_page != (void *)0xfee00000) {
+               perror("Could not mmap APIC");
+               exit(1);
+       }
+       if (((uint64_t)a_page & 0xfff) != 0) {
+               perror("APIC page mapping is not page aligned");
+               exit(1);
+       }
+
+       ((uint32_t *)a_page)[0x30/4] = 0x01060015;
+       //((uint32_t *)a_page)[0x30/4] = 0xDEADBEEF;
+
+       while ((c = getopt_long(argc, argv, "dvm:M:S:c:gsf:k:n:hR:", long_options,
+                               &option_index)) != -1) {
+               switch (c) {
+                       case 'd':
+                               debug++;
+                               break;
+                       case 'v':
+                               vmmflags |= VMM_VMCALL_PRINTF;
+                               break;
+                       case 'm':
+                               memsize = strtoull(optarg, 0, 0);
+                               break;
+                       case 'M':
+                               memstart = strtoull(optarg, 0, 0);
+                               break;
+                       case 'S':
+                               stack = strtoull(optarg, 0, 0);
+                               break;
+                       case 'R':
+                               maxresume = strtoull(optarg, 0, 0);
+                               break;
+                       case 'c':
+                               cmdline_extra = optarg;
+                       case 'g':       /* greedy */
+                               parlib_never_yield = TRUE;
+                               break;
+                       case 's':       /* scp */
+                               parlib_wants_to_be_mcp = FALSE;
+                               break;
+                       case 'f':       /* file to pass to blk_init */
+                               disk_image_file = optarg;
+                               break;
+                       case 'k':       /* specify file to get cmdline args from */
+                               cmdline_fd = open(optarg, O_RDONLY);
+                               if (cmdline_fd < 0) {
+                                       fprintf(stderr, "failed to open file: %s\n", optarg);
+                                       exit(1);
+                               }
+                               if (stat(optarg, &stat_result) == -1) {
+                                       fprintf(stderr, "stat of %s failed\n", optarg);
+                                       exit(1);
+                               }
+                               len = stat_result.st_size;
+                               if (len > 512) {
+                                       fprintf(stderr, "command line options exceed 512 bytes!");
+                                       exit(1);
+                               }
+                               num_read = read(cmdline_fd, cmdline_default, len);
+                               if (num_read != len) {
+                                       fprintf(stderr, "read failed len was : %d, num_read was: %d\n",
+                                               len, num_read);
+                                       exit(1);
+                               }
+                               close(cmdline_fd);
+                               break;
+                       case 'n':
+                               default_nic = strtoull(optarg, 0, 0);
+                               break;
+                       case 'h':
+                       default:
+                               // Sadly, the getopt_long struct does
+                               // not have a pointer to help text.
+                               for (int i = 0;
+                                   i < sizeof(long_options)/sizeof(long_options[0]) - 1;
+                                   i++) {
+                                       struct option *l = &long_options[i];
+
+                                       fprintf(stderr, "%s or %c%s\n", l->name, l->val,
+                                               l->has_arg ? " <arg>" : "");
+                               }
+                               exit(0);
+               }
+       }
+       if (strlen(cmdline_default) == 0) {
+               fprintf(stderr, "WARNING: No command line parameter file specified.\n");
+       }
+       argc -= optind;
+       argv += optind;
+       if (argc < 1) {
+               fprintf(stderr, "Usage: %s vmimage [-n (no vmcall printf)]\n", argv[0]);
+               exit(1);
+       }
+
+       if ((uintptr_t)(memstart + memsize) >= (uintptr_t)BRK_START) {
+               fprintf(stderr,
+                       "memstart 0x%lx memsize 0x%lx -> 0x%lx is too large; overlaps BRK_START at %p\n",
+                       memstart, memsize, memstart + memsize, BRK_START);
+               exit(1);
+       }
+
+       kernel = mmap((void *)memstart, memsize,
+                     PROT_READ | PROT_WRITE | PROT_EXEC,
+                     MAP_POPULATE | MAP_ANONYMOUS, -1, 0);
+       if (kernel != (void *)memstart) {
+               fprintf(stderr, "Could not mmap 0x%lx bytes at 0x%lx\n",
+                       memsize, memstart);
+               exit(1);
+       }
+
+       entry = load_kernel(argv[0]);
+       if (entry == 0) {
+               fprintf(stderr, "Unable to load kernel %s\n", argv[0]);
+               exit(1);
+       }
+
+
+       // The low 1m so we can fill in bullshit like ACPI. */
+       // And, sorry, due to the STUPID format of the RSDP for now we need the low 1M.
+       low1m = mmap((int*)4096, MiB-4096, PROT_READ | PROT_WRITE,
+                    MAP_ANONYMOUS, -1, 0);
+       if (low1m != (void *)4096) {
+               perror("Unable to mmap low 1m");
+               exit(1);
+       }
+       r = a;
+       fprintf(stderr, "install rsdp to %p\n", r);
+       *r = rsdp;
+       a += sizeof(*r);
+       r->xsdt_physical_address = (uint64_t)a;
+       gencsum(&r->checksum, r, ACPI_RSDP_CHECKSUM_LENGTH);
+       if ((csum = acpi_tb_checksum((uint8_t *) r, ACPI_RSDP_CHECKSUM_LENGTH)) != 0) {
+               fprintf(stderr, "RSDP has bad checksum; summed to %x\n", csum);
+               exit(1);
+       }
+
+       /* Check extended checksum if table version >= 2 */
+       gencsum(&r->extended_checksum, r, ACPI_RSDP_XCHECKSUM_LENGTH);
+       if ((rsdp.revision >= 2) &&
+           (acpi_tb_checksum((uint8_t *) r, ACPI_RSDP_XCHECKSUM_LENGTH) != 0)) {
+               fprintf(stderr, "RSDP has bad checksum v2\n");
+               exit(1);
+       }
+
+       /* just leave a bunch of space for the xsdt. */
+       /* we need to zero the area since it has pointers. */
+       x = a;
+       a += sizeof(*x) + 8*sizeof(void *);
+       memset(x, 0, a - (void *)x);
+       fprintf(stderr, "install xsdt to %p\n", x);
+       *x = xsdt;
+       x->table_offset_entry[0] = 0;
+       x->table_offset_entry[1] = 0;
+       x->header.length = a - (void *)x;
+
+       f = a;
+       fprintf(stderr, "install fadt to %p\n", f);
+       *f = fadt;
+       x->table_offset_entry[0] = (uint64_t)f; // fadt MUST be first in xsdt!
+       a += sizeof(*f);
+       f->header.length = a - (void *)f;
+
+       f->Xdsdt = (uint64_t) a;
+       fprintf(stderr, "install dsdt to %p\n", a);
+       memcpy(a, &DSDT_DSDTTBL_Header, 36);
+       a += 36;
+
+       gencsum(&f->header.checksum, f, f->header.length);
+       if (acpi_tb_checksum((uint8_t *)f, f->header.length) != 0) {
+               fprintf(stderr, "fadt has bad checksum v2\n");
+               exit(1);
+       }
+
+       m = a;
+       *m = madt;
+       x->table_offset_entry[3] = (uint64_t) m;
+       a += sizeof(*m);
+       fprintf(stderr, "install madt to %p\n", m);
+       memmove(a, &Apic0, sizeof(Apic0));
+       a += sizeof(Apic0);
+       memmove(a, &Apic1, sizeof(Apic1));
+       a += sizeof(Apic1);
+       memmove(a, &X2Apic0, sizeof(X2Apic0));
+       a += sizeof(X2Apic0);
+       memmove(a, &isor, sizeof(isor));
+       a += sizeof(isor);
+       m->header.length = a - (void *)m;
+
+       gencsum(&m->header.checksum, m, m->header.length);
+       if (acpi_tb_checksum((uint8_t *) m, m->header.length) != 0) {
+               fprintf(stderr, "madt has bad checksum v2\n");
+               exit(1);
+       }
+
+       gencsum(&x->header.checksum, x, x->header.length);
+       if ((csum = acpi_tb_checksum((uint8_t *) x, x->header.length)) != 0) {
+               fprintf(stderr, "XSDT has bad checksum; summed to %x\n", csum);
+               exit(1);
+       }
+
+       fprintf(stderr, "allchecksums ok\n");
+
+       hexdump(stdout, r, a-(void *)r);
+
+       a = (void *)(((unsigned long)a + 0xfff) & ~0xfff);
+       gpci.posted_irq_desc = a;
+       memset(a, 0, 4096);
+       a += 4096;
+       gpci.vapic_addr = a;
+       memset(a, 0, 4096);
+       ((uint32_t *)a)[0x30/4] = 0x01060014;
+       // set up apic values? do we need to?
+       // qemu does this.
+       //((uint8_t *)a)[4] = 1;
+       a += 4096;
+       gpci.apic_addr = (void*)0xfee00000;
+
+       /* Allocate memory for, and zero the bootparams
+        * page before writing to it, or Linux thinks
+        * we're talking crazy.
+        */
+       a += 4096;
+       bp = a;
+       memset(bp, 0, 4096);
+
+       /* Put the e820 memory region information in the boot_params */
+       bp->e820_entries = 5;
+       int e820i = 0;
+
+       /* Give it just a tiny bit of memory -- 60k -- at low memory. */
+       bp->e820_map[e820i].addr = 0;
+       bp->e820_map[e820i].size = 4 * 1024;
+       bp->e820_map[e820i++].type = E820_RESERVED;
+
+       bp->e820_map[e820i].addr = 4 * 1024;
+       bp->e820_map[e820i].size = 64 * 1024 - 4 * 1024;
+       bp->e820_map[e820i++].type = E820_RAM;
+
+       bp->e820_map[e820i].addr = 64 * 1024;
+       bp->e820_map[e820i].size = memstart - 64 * 1024;
+       bp->e820_map[e820i++].type = E820_RESERVED;
+
+       bp->e820_map[e820i].addr = memstart;
+       bp->e820_map[e820i].size = memsize;
+       bp->e820_map[e820i++].type = E820_RAM;
+
+       bp->e820_map[e820i].addr = 0xf0000000;
+       bp->e820_map[e820i].size = 0x10000000;
+       bp->e820_map[e820i++].type = E820_RESERVED;
+
+       /* The MMIO address of the console device is really the address of an
+        * unbacked EPT page: accesses to this page will cause a page fault that
+        * traps to the host, which will examine the fault, see it was for the
+        * known MMIO address, and fulfill the MMIO read or write on the guest's
+        * behalf accordingly. We place the virtio space at 512 GB higher than the
+        * guest physical memory to avoid a full page table walk. */
+       uint64_t virtio_mmio_base_addr = ROUNDUP((bp->e820_map[e820i - 1].addr +
+                                                 bp->e820_map[e820i - 1].size),
+                                                512 * GiB);
+
+       cons_mmio_dev.addr =
+           virtio_mmio_base_addr + PGSIZE * VIRTIO_MMIO_CONSOLE_DEV;
+       cons_mmio_dev.vqdev = &cons_vqdev;
+       vm->virtio_mmio_devices[VIRTIO_MMIO_CONSOLE_DEV] = &cons_mmio_dev;
+
+       net_mmio_dev.addr =
+           virtio_mmio_base_addr + PGSIZE * VIRTIO_MMIO_NETWORK_DEV;
+       net_mmio_dev.vqdev = &net_vqdev;
+       vm->virtio_mmio_devices[VIRTIO_MMIO_NETWORK_DEV] = &net_mmio_dev;
+
+       if (disk_image_file != NULL) {
+               blk_mmio_dev.addr =
+                   virtio_mmio_base_addr + PGSIZE * VIRTIO_MMIO_BLOCK_DEV;
+               blk_mmio_dev.vqdev = &blk_vqdev;
+               vm->virtio_mmio_devices[VIRTIO_MMIO_BLOCK_DEV] = &blk_mmio_dev;
+               blk_init_fn(&blk_vqdev, disk_image_file);
+       }
+       net_init_fn(&net_vqdev, default_nic);
+
+       /* Set the kernel command line parameters */
+       a += 4096;
+       cmdline = a;
+       a += 4096;
+       bp->hdr.cmd_line_ptr = (uintptr_t) cmdline;
+
+       tsc_freq_khz = get_tsc_freq()/1000;
+       len = snprintf(cmdline, 4096, "%s tscfreq=%lld %s", cmdline_default,
+                      tsc_freq_khz, cmdline_extra);
+
+       cmdlinesz = 4096 - len;
+       cmdlinep = cmdline + len;
+
+       for (int i = 0; i < VIRTIO_MMIO_MAX_NUM_DEV; i++) {
+               if (vm->virtio_mmio_devices[i] == NULL)
+                       continue;
+
+               /* Append all the virtio mmio base addresses. */
+
+                       /* Since the lower number irqs are no longer being used, the irqs
+                        * can now be assigned starting from 0.
+                        */
+                       vm->virtio_mmio_devices[i]->irq = i;
+                       len = snprintf(cmdlinep, cmdlinesz,
+                              " virtio_mmio.device=1K@0x%llx:%lld",
+                              vm->virtio_mmio_devices[i]->addr,
+                              vm->virtio_mmio_devices[i]->irq);
+               if (len >= cmdlinesz) {
+                       fprintf(stderr, "Too many arguments to the linux command line.");
+                       exit(1);
+               }
+               cmdlinesz -= len;
+               cmdlinep += len;
+       }
+
+       vm->nr_gpcs = 1;
+       vm->gpcis = &gpci;
+       ret = vmm_init(vm, vmmflags);
+       assert(!ret);
+
+       /* Allocate 3 pages for page table pages: a page of 512 GiB
+        * PTEs with only one entry filled to point to a page of 1 GiB
+        * PTEs; a page of 1 GiB PTEs with only one entry filled to
+        * point to a page of 2 MiB PTEs; and a page of 2 MiB PTEs,
+        * all of which may be filled. For now, we don't handle
+        * starting addresses not aligned on 512 GiB boundaries or
+        * sizes > GiB */
+       ret = posix_memalign((void **)&p512, PGSIZE, 3 * PGSIZE);
+       if (ret) {
+               perror("ptp alloc");
+               exit(1);
+       }
+
+       /* Set up a 1:1 ("identity") page mapping from guest virtual
+        * to guest physical using the (host virtual)
+        * `kerneladdress`. This mapping may be used for only a short
+        * time, until the guest sets up its own page tables. Be aware
+        * that the values stored in the table are physical addresses.
+        * This is subtle and mistakes are easily disguised due to the
+        * identity mapping, so take care when manipulating these
+        * mappings. */
+       p1 = &p512[NPTENTRIES];
+       p2m = &p512[2 * NPTENTRIES];
+
+       fprintf(stderr, "Map %p for %zu bytes\n", memstart, memsize);
+       /* TODO: fix this nested loop so it's correct for more than
+        * one GiB. */
+       for(uintptr_t p4 = memstart; p4 < memstart + memsize;
+           p4 += PML4_PTE_REACH) {
+               p512[PML4(p4)] = (uint64_t)p1 | PTE_KERN_RW;
+               for (uintptr_t p3 = p4; p3 < memstart + memsize;
+                    p3 += PML3_PTE_REACH) {
+                       p1[PML3(p3)] = (uint64_t)p2m | PTE_KERN_RW;
+                       for (uintptr_t p2 = p3; p2 < memstart + memsize; p2 += PML2_PTE_REACH) {
+                               p2m[PML2(p2)] =
+                                       (uint64_t)(p2) | PTE_KERN_RW | PTE_PS;
+                       }
+               }
+       }
+
+       fprintf(stderr, "p512 %p p512[0] is 0x%lx p1 %p p1[0] is 0x%x\n", p512, p512[0], p1, p1[0]);
+
+       vmm_run_task(vm, timer_thread, 0);
+
+       vm_tf = gth_to_vmtf(vm->gths[0]);
+       vm_tf->tf_cr3 = (uint64_t) p512;
+       vm_tf->tf_rip = entry;
+       vm_tf->tf_rsp = stack;
+       vm_tf->tf_rsi = (uint64_t) bp;
+       start_guest_thread(vm->gths[0]);
+
+       uthread_sleep_forever();
+       return 0;
+}