CAMLprim value stub_xc_hvm_build_native(value xc_handle, value domid,
value mem_max_mib, value mem_start_mib, value image_name, value store_evtchn, value console_evtchn)
{
CAMLparam5(xc_handle, domid, mem_max_mib, mem_start_mib, image_name);
CAMLxparam2(store_evtchn, console_evtchn);
CAMLlocal1(result);
char *image_name_c = strdup(String_val(image_name));
char *error[256];
xc_interface *xch;
unsigned long store_mfn=0;
unsigned long console_mfn=0;
int r;
struct flags f;
/* The xenguest interface changed and was backported to XCP: */
#if defined(XENGUEST_HAS_HVM_BUILD_ARGS) || (__XEN_LATEST_INTERFACE_VERSION__ >= 0x00040200)
struct xc_hvm_build_args args;
#endif
get_flags(&f, _D(domid));
xch = _H(xc_handle);
configure_vcpus(xch, _D(domid), f);
configure_tsc(xch, _D(domid), f);
#if defined(XENGUEST_HAS_HVM_BUILD_ARGS) || (__XEN_LATEST_INTERFACE_VERSION__ >= 0x00040200)
args.mem_size = (uint64_t)Int_val(mem_max_mib) << 20;
args.mem_target = (uint64_t)Int_val(mem_start_mib) << 20;
args.mmio_size = f.mmio_size_mib << 20;
args.image_file_name = image_name_c;
#endif
caml_enter_blocking_section ();
#if defined(XENGUEST_HAS_HVM_BUILD_ARGS) || (__XEN_LATEST_INTERFACE_VERSION__ >= 0x00040200)
r = xc_hvm_build(xch, _D(domid), &args);
#else
r = xc_hvm_build_target_mem(xch, _D(domid),
Int_val(mem_max_mib),
Int_val(mem_start_mib),
image_name_c);
#endif
caml_leave_blocking_section ();
free(image_name_c);
if (r)
failwith_oss_xc(xch, "hvm_build");
r = hvm_build_set_params(xch, _D(domid), Int_val(store_evtchn), &store_mfn,
Int_val(console_evtchn), &console_mfn, f);
if (r)
failwith_oss_xc(xch, "hvm_build_params");
result = caml_alloc_tuple(2);
Store_field(result, 0, caml_copy_nativeint(store_mfn));
Store_field(result, 1, caml_copy_nativeint(console_mfn));
CAMLreturn(result);
}
int stub_xc_linux_build(int c_mem_max_mib, int mem_start_mib,
const char *image_name, const char *ramdisk_name,
const char *cmdline, const char *features,
int flags, int store_evtchn, int store_domid,
int console_evtchn, int console_domid,
unsigned long *store_mfn, unsigned long *console_mfn,
char *protocol)
{
int r;
struct xc_dom_image *dom;
struct flags f;
get_flags(&f);
xc_dom_loginit(xch);
dom = xc_dom_allocate(xch, cmdline, features);
if (!dom)
failwith_oss_xc("xc_dom_allocate");
/* The default image size limits are too large. */
xc_dom_kernel_max_size(dom, get_image_max_size("kernel"));
xc_dom_ramdisk_max_size(dom, get_image_max_size("ramdisk"));
configure_vcpus(f);
configure_tsc(f);
r = xc_dom_linux_build(xch, dom, domid, mem_start_mib,
image_name, ramdisk_name, flags,
store_evtchn, store_mfn,
console_evtchn, console_mfn);
if ( r )
failwith_oss_xc("xc_dom_linux_build");
r = construct_cpuid_policy(&f, false);
if ( r )
failwith_oss_xc("construct_cpuid_policy");
r = xc_dom_gnttab_seed(xch, domid,
*console_mfn, *store_mfn,
console_domid, store_domid);
if ( r )
failwith_oss_xc("xc_dom_gnttab_seed");
strncpy(protocol, xc_domain_get_native_protocol(xch, domid), 64);
free_flags(&f);
xc_dom_release(dom);
return 0;
}
static void configure_tsc(struct flags f)
{
int rc = xc_domain_set_tsc_info(xch, domid, f.tsc_mode, 0, 0, 0);
if (rc)
failwith_oss_xc("xc_domain_set_tsc_info");
}
CAMLprim value stub_xenguest_init()
{
xc_interface *xch;
xch = xc_interface_open(NULL, NULL, 0);
if (xch == NULL)
failwith_oss_xc(NULL, "xc_interface_open");
return (value)xch;
}
CAMLprim value stub_xc_domain_dumpcore(value handle, value domid, value file)
{
CAMLparam3(handle, domid, file);
int r;
r = xc_domain_dumpcore(_H(handle), _D(domid), String_val(file));
if (r)
failwith_oss_xc(_H(handle), "xc_domain_dumpcore");
CAMLreturn(Val_unit);
}
/* this is the slow version of resume for uncooperative domain,
* the fast version is available in close source xc */
int stub_xc_domain_resume_slow(void)
{
int r;
/* hard code fast to 0, we only want to expose the slow version here */
r = xc_domain_resume(xch, domid, 0);
if (r)
failwith_oss_xc("xc_domain_resume");
return 0;
}
static void configure_vcpus(struct flags f){
struct xen_domctl_sched_credit sdom;
int i, j, r, size, pcpus_supplied, min;
xc_cpumap_t cpumap;
size = xc_get_cpumap_size(xch) * 8; /* array is of uint8_t */
for (i=0; i<f.vcpus; i++){
if (f.vcpu_affinity[i]){ /* NULL means unset */
pcpus_supplied = strlen(f.vcpu_affinity[i]);
min = (pcpus_supplied < size)?pcpus_supplied:size;
cpumap = xc_cpumap_alloc(xch);
if (cpumap == NULL)
failwith_oss_xc("xc_cpumap_alloc");
for (j=0; j<min; j++) {
if (f.vcpu_affinity[i][j] == '1')
cpumap[j/8] |= 1 << (j&7);
}
r = xc_vcpu_setaffinity(xch, domid, i, cpumap, NULL,
XEN_VCPUAFFINITY_HARD);
free(cpumap);
if (r) {
failwith_oss_xc("xc_vcpu_setaffinity");
}
}
}
r = xc_sched_credit_domain_get(xch, domid, &sdom);
/* This should only happen when a different scheduler is set */
if (r) {
xg_info("Failed to get credit scheduler parameters: scheduler not enabled?\n");
return;
}
if (f.vcpu_weight != 0L) sdom.weight = f.vcpu_weight;
if (f.vcpu_cap != 0L) sdom.cap = f.vcpu_cap;
/* This shouldn't fail, if "get" above succeeds. This error is fatal
to highlight the need to investigate further. */
r = xc_sched_credit_domain_set(xch, domid, &sdom);
if (r)
failwith_oss_xc("xc_sched_credit_domain_set");
}
/* this is the slow version of resume for uncooperative domain,
* the fast version is available in close source xc */
CAMLprim value stub_xc_domain_resume_slow(value handle, value domid)
{
CAMLparam2(handle, domid);
int r;
/* hard code fast to 0, we only want to expose the slow version here */
r = xc_domain_resume(_H(handle), _D(domid), 0);
if (r)
failwith_oss_xc(_H(handle), "xc_domain_resume");
CAMLreturn(Val_unit);
}
CAMLprim value stub_xenguest_init()
{
xc_interface *xch;
openlog("xenguest", LOG_NDELAY, LOG_DAEMON);
xch = xc_interface_open(NULL, NULL, 0);
if (xch == NULL)
failwith_oss_xc(NULL, "xc_interface_open");
return (value)xch;
}
int stub_xc_domain_save(int fd, int max_iters, int max_factors,
int flags, int hvm)
{
int r;
struct save_callbacks callbacks =
{
.suspend = suspend_callback,
.switch_qemu_logdirty = switch_qemu_logdirty,
.data = NULL,
};
r = xc_domain_save(xch, fd, domid,
max_iters, max_factors,
flags, &callbacks, hvm);
if (r)
failwith_oss_xc("xc_domain_save");
return 0;
}
CAMLprim value stub_xc_domain_save(value handle, value fd, value domid,
value max_iters, value max_factors,
value flags, value hvm)
{
CAMLparam5(handle, fd, domid, max_iters, max_factors);
CAMLxparam2(flags, hvm);
struct save_callbacks callbacks;
uint32_t c_flags;
uint32_t c_domid;
int r;
unsigned long generation_id_addr;
c_flags = caml_convert_flag_list(flags, suspend_flag_list);
c_domid = _D(domid);
memset(&callbacks, 0, sizeof(callbacks));
callbacks.data = (void*) c_domid;
callbacks.suspend = dispatch_suspend;
callbacks.switch_qemu_logdirty = switch_qemu_logdirty;
caml_enter_blocking_section();
generation_id_addr = xenstore_get(c_domid, GENERATION_ID_ADDRESS);
r = xc_domain_save(_H(handle), Int_val(fd), c_domid,
Int_val(max_iters), Int_val(max_factors),
c_flags, &callbacks, Bool_val(hvm)
#ifdef XENGUEST_4_2
,generation_id_addr
#endif
);
caml_leave_blocking_section();
if (r)
failwith_oss_xc(_H(handle), "xc_domain_save");
CAMLreturn(Val_unit);
}
CAMLprim value stub_xc_domain_restore(value handle, value fd, value domid,
value store_evtchn, value store_domid,
value console_evtchn, value console_domid,
value hvm, value no_incr_generationid)
{
CAMLparam5(handle, fd, domid, store_evtchn, console_evtchn);
CAMLxparam1(hvm);
CAMLlocal1(result);
unsigned long store_mfn, console_mfn;
domid_t c_store_domid, c_console_domid;
unsigned long c_vm_generationid_addr;
char c_vm_generationid_addr_s[32];
unsigned int c_store_evtchn, c_console_evtchn;
int r;
size_t size, written;
struct flags f;
get_flags(&f,_D(domid));
c_store_evtchn = Int_val(store_evtchn);
c_store_domid = Int_val(store_domid);
c_console_evtchn = Int_val(console_evtchn);
c_console_domid = Int_val(console_domid);
#ifdef HVM_PARAM_VIRIDIAN
xc_set_hvm_param(_H(handle), _D(domid), HVM_PARAM_VIRIDIAN, f.viridian);
#endif
configure_vcpus(_H(handle), _D(domid), f);
caml_enter_blocking_section();
r = xc_domain_restore(_H(handle), Int_val(fd), _D(domid),
c_store_evtchn, &store_mfn,
#ifdef XENGUEST_4_2
c_store_domid,
#endif
c_console_evtchn, &console_mfn,
#ifdef XENGUEST_4_2
c_console_domid,
#endif
Bool_val(hvm), f.pae, 0 /*superpages*/
#ifdef XENGUEST_4_2
,
Bool_val(no_incr_generationid),
&c_vm_generationid_addr,
NULL /* restore_callbacks */
#endif
);
if (!r) {
size = sizeof(c_vm_generationid_addr_s) - 1; /* guarantee a NULL remains on the end */
written = snprintf(c_vm_generationid_addr_s, size, "0x%lx", c_vm_generationid_addr);
if (written < size)
r = xenstore_puts(_D(domid), c_vm_generationid_addr_s, GENERATION_ID_ADDRESS);
else {
syslog(LOG_ERR|LOG_DAEMON,"Failed to write %s (%d >= %d)", GENERATION_ID_ADDRESS, written, size);
r = 1;
}
}
caml_leave_blocking_section();
if (r)
failwith_oss_xc(_H(handle), "xc_domain_restore");
result = caml_alloc_tuple(2);
Store_field(result, 0, caml_copy_nativeint(store_mfn));
Store_field(result, 1, caml_copy_nativeint(console_mfn));
CAMLreturn(result);
}
CAMLprim value stub_xc_linux_build_native(value xc_handle, value domid,
value mem_max_mib, value mem_start_mib,
value image_name, value ramdisk_name,
value cmdline, value features,
value flags, value store_evtchn,
value console_evtchn)
{
CAMLparam5(xc_handle, domid, mem_max_mib, mem_start_mib, image_name);
CAMLxparam5(ramdisk_name, cmdline, features, flags, store_evtchn);
CAMLxparam1(console_evtchn);
CAMLlocal1(result);
unsigned long store_mfn;
unsigned long console_mfn;
int r;
struct xc_dom_image *dom;
char c_protocol[64];
/* Copy the ocaml values into c-land before dropping the mutex */
xc_interface *xch = _H(xc_handle);
unsigned int c_mem_start_mib = Int_val(mem_start_mib);
uint32_t c_domid = _D(domid);
char *c_image_name = strdup(String_val(image_name));
char *c_ramdisk_name = ramdisk_name == None_val ? NULL : strdup(String_val(Field(ramdisk_name, 0)));
unsigned long c_flags = Int_val(flags);
unsigned int c_store_evtchn = Int_val(store_evtchn);
unsigned int c_console_evtchn = Int_val(console_evtchn);
struct flags f;
get_flags(&f,c_domid);
xc_dom_loginit(xch);
dom = xc_dom_allocate(xch, String_val(cmdline), String_val(features));
if (!dom)
failwith_oss_xc(xch, "xc_dom_allocate");
configure_vcpus(xch, c_domid, f);
configure_tsc(xch, c_domid, f);
#ifdef XC_HAVE_DECOMPRESS_LIMITS
if ( xc_dom_kernel_max_size(dom, f.kernel_max_size) )
failwith_oss_xc(xch, "xc_dom_kernel_max_size");
if ( xc_dom_ramdisk_max_size(dom, f.ramdisk_max_size) )
failwith_oss_xc(xch, "xc_dom_ramdisk_max_size");
#else
if ( f.kernel_max_size || f.ramdisk_max_size ) {
syslog(LOG_WARNING|LOG_DAEMON,"Kernel/Ramdisk limits set, but no support compiled in");
}
#endif
caml_enter_blocking_section();
r = xc_dom_linux_build(xch, dom, c_domid, c_mem_start_mib,
c_image_name, c_ramdisk_name, c_flags,
c_store_evtchn, &store_mfn,
c_console_evtchn, &console_mfn);
caml_leave_blocking_section();
#ifndef XEN_UNSTABLE
strncpy(c_protocol, xc_dom_get_native_protocol(dom), 64);
#else
memset(c_protocol, '\0', 64);
#endif
free(c_image_name);
free(c_ramdisk_name);
xc_dom_release(dom);
if (r != 0)
failwith_oss_xc(xch, "xc_dom_linux_build");
result = caml_alloc_tuple(3);
Store_field(result, 0, caml_copy_nativeint(store_mfn));
Store_field(result, 1, caml_copy_nativeint(console_mfn));
Store_field(result, 2, caml_copy_string(c_protocol));
CAMLreturn(result);
}
int stub_xc_domain_restore(int fd, int store_evtchn, int console_evtchn,
int hvm,
unsigned long *store_mfn, unsigned long *console_mfn)
{
int r = 0;
struct flags f;
get_flags(&f);
if ( hvm )
{
/*
* We have to do this even in the domain restore case as XenServers
* prior to 6.0.2 did not create a viridian save record.
*/
if (f.viridian)
hvm_set_viridian_features(&f);
xc_set_hvm_param(xch, domid, HVM_PARAM_HPET_ENABLED, f.hpet);
#ifdef HAVE_CORES_PER_SOCKET
if ( f.cores_per_socket > 0 )
r = xc_domain_set_cores_per_socket(xch, domid, f.cores_per_socket);
#endif
if ( r )
failwith_oss_xc("xc_domain_set_cores_per_socket");
}
configure_vcpus(f);
r = xc_domain_restore(xch, fd, domid,
store_evtchn, store_mfn, 0,
console_evtchn, console_mfn, 0,
hvm, f.pae, 0, 0, NULL);
if ( r )
failwith_oss_xc("xc_domain_restore");
/*
* The legacy -> migration v2 code in XenServer 6.5 didn't combine the
* out-of-band HVM_PARAM_PAE_ENABLED into the converted stream, and
* xenguest didn't set it, as the v2 restore code was expected to.
*
* This causes xc_cpuid_apply_policy() to hide the PAE bit from the domain
* cpuid policy, which went unnoticed (and without incident, despite being
* a guest-visible change) until Xen-4.5 became stricter with its checks
* for when a guest writes to %cr4.
*
* The correct value is still available out-of-band, so clobber the result
* from the stream, in case the stream is from XenServer 6.5 and is a VM
* which hasn't rebooted and has a bad HVM PARAM in the v2 stream.
*/
if ( hvm )
xc_set_hvm_param(xch, domid, HVM_PARAM_PAE_ENABLED, f.pae);
r = construct_cpuid_policy(&f, hvm);
if ( r )
failwith_oss_xc("construct_cpuid_policy");
free_flags(&f);
if ( hvm )
{
r = set_genid();
if (r)
exit(1);
}
return 0;
}
int stub_xc_hvm_build_with_mem(uint64_t max_mem_mib, uint64_t max_start_mib,
const char *image)
{
uint64_t lowmem_end, highmem_start, highmem_end, mmio_start;
struct xc_hvm_build_args args = {
.mem_size = max_mem_mib << 20,
.mem_target = max_start_mib << 20,
.mmio_size = HVM_BELOW_4G_MMIO_LENGTH,
.image_file_name = image,
};
unsigned int i, j, nr = 0;
struct e820entry *e820;
int rc;
unsigned int nr_rdm_entries[MAX_RMRR_DEVICES] = {0};
unsigned int nr_rmrr_devs = 0;
struct xen_reserved_device_memory *xrdm[MAX_RMRR_DEVICES] = {0};
unsigned long rmrr_overlapped_ram = 0;
char *s;
if ( pci_passthrough_sbdf_list )
{
s = strtok(pci_passthrough_sbdf_list,",");
while ( s != NULL )
{
unsigned int seg, bus, device, func;
xg_info("Getting RMRRs for device '%s'\n",s);
if ( parse_pci_sbdf(s, &seg, &bus, &device, &func) )
{
if ( !get_rdm(seg, bus, (device << 3) + func,
&nr_rdm_entries[nr_rmrr_devs], &xrdm[nr_rmrr_devs]) )
nr_rmrr_devs++;
}
if ( nr_rmrr_devs == MAX_RMRR_DEVICES )
{
xg_err("Error: hit limit of %d RMRR devices for domain\n",
MAX_RMRR_DEVICES);
exit(1);
}
s = strtok (NULL, ",");
}
}
e820 = malloc(sizeof(*e820) * E820MAX);
if (!e820)
return -ENOMEM;
lowmem_end = args.mem_size;
highmem_end = highmem_start = 1ull << 32;
mmio_start = highmem_start - args.mmio_size;
if ( lowmem_end > mmio_start )
{
highmem_end = (1ull << 32) + (lowmem_end - mmio_start);
lowmem_end = mmio_start;
}
args.lowmem_end = lowmem_end;
args.highmem_end = highmem_end;
args.mmio_start = mmio_start;
/* Leave low 1MB to HVMLoader... */
e820[nr].addr = 0x100000u;
e820[nr].size = args.lowmem_end - 0x100000u;
e820[nr].type = E820_RAM;
nr++;
/* RDM mapping */
for (i = 0; i < nr_rmrr_devs; i++)
{
for (j = 0; j < nr_rdm_entries[i]; j++)
{
e820[nr].addr = xrdm[i][j].start_pfn << XC_PAGE_SHIFT;
e820[nr].size = xrdm[i][j].nr_pages << XC_PAGE_SHIFT;
e820[nr].type = E820_RESERVED;
xg_info("Adding RMRR 0x%lx size 0x%lx\n", e820[nr].addr, e820[nr].size);
if ( e820[nr].addr < args.lowmem_end ) {
rmrr_overlapped_ram += ( args.lowmem_end - e820[nr].addr );
args.lowmem_end = e820[nr].addr;
}
nr++;
}
}
e820[0].size -= rmrr_overlapped_ram;
args.highmem_end += rmrr_overlapped_ram;
args.mmio_size += rmrr_overlapped_ram;
args.mmio_start -= rmrr_overlapped_ram;
for (i = 0; i < nr_rmrr_devs; i++)
{
free(xrdm[i]);
}
if ( args.highmem_end > highmem_start )
{
e820[nr].addr = highmem_start;
e820[nr].size = args.highmem_end - e820[nr].addr;
e820[nr].type = E820_RAM;
nr++;
}
rc = xc_hvm_build(xch, domid, &args);
if (!rc)
rc = xc_domain_set_memory_map(xch, domid, e820, nr);
free(e820);
return rc;
}
int stub_xc_hvm_build(int mem_max_mib, int mem_start_mib, const char *image_name,
int store_evtchn, int store_domid,
int console_evtchn, int console_domid,
unsigned long *store_mfn, unsigned long *console_mfn)
{
int r;
struct flags f;
get_flags(&f);
configure_vcpus(f);
configure_tsc(f);
r = stub_xc_hvm_build_with_mem(mem_max_mib, mem_start_mib, image_name);
if ( r )
failwith_oss_xc("hvm_build");
r = hvm_build_set_params(store_evtchn, store_mfn,
console_evtchn, console_mfn, f);
if ( r )
failwith_oss_xc("hvm_build_params");
r = construct_cpuid_policy(&f, true);
if ( r )
failwith_oss_xc("construct_cpuid_policy");
r = xc_dom_gnttab_hvm_seed(xch, domid, *console_mfn, *store_mfn,
console_domid, store_domid);
if ( r )
failwith_oss_xc("xc_dom_gnttab_hvm_seed");
free_flags(&f);
return 0;
}
