startup.c revision 1de082f7b7fd4b6629e14b0f9b8f94f6c0bda3c2
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright 2009 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
#include <sys/sysmacros.h>
#include <sys/autoconf.h>
#include <sys/privregs.h>
#include <sys/bootconf.h>
#include <sys/ndi_impldefs.h>
#include <sys/ddidmareq.h>
#include <vm/kboot_mmu.h>
#include <vm/seg_kmem.h>
#include <sys/vm_machparam.h>
#include <sys/archsystm.h>
#include <sys/machsystm.h>
#include <sys/smp_impldefs.h>
#include <sys/x86_archext.h>
#include <sys/segments.h>
#include <sys/kobj_lex.h>
#include <sys/cpc_impl.h>
#include <sys/cpu_module.h>
#include <sys/debug_info.h>
#include <sys/bootinfo.h>
#include <sys/ddi_timer.h>
#include <sys/systeminfo.h>
#include <sys/multiboot.h>
#ifdef __xpv
#include <sys/hypervisor.h>
#include <sys/evtchn_impl.h>
#include <sys/xpv_panic.h>
extern void xen_late_startup(void);
struct xen_evt_data cpu0_evt_data;
#endif /* __xpv */
extern void progressbar_init(void);
extern void progressbar_start(void);
extern void brand_init(void);
extern void pcf_init(void);
extern void pg_init(void);
extern int size_pse_array(pgcnt_t, int);
#if defined(_SOFT_HOSTID)
static int32_t set_soft_hostid(void);
static char hostid_file[] = "/etc/hostid";
#endif
void *gfx_devinfo_list;
/*
* XXX make declaration below "static" when drivers no longer use this
* interface.
*/
/*
* segkp
*/
extern int segkp_fromheap;
static void kvm_init(void);
static void startup_init(void);
static void startup_memlist(void);
static void startup_kmem(void);
static void startup_modules(void);
static void startup_vm(void);
static void startup_end(void);
static void layout_kernel_va(void);
#if !defined(__xpv)
void (*rootnex_iommu_init)(void) = NULL;
#endif
/*
* Declare these as initialized data so we can patch them.
*/
#ifdef __i386
/*
* Due to virtual address space limitations running in 32 bit mode, restrict
* the amount of physical memory configured to a max of PHYSMEM pages (16g).
*
* If the physical max memory size of 64g were allowed to be configured, the
* size of user virtual address space will be less than 1g. A limited user
* address space greatly reduces the range of applications that can run.
*
* If more physical memory than PHYSMEM is required, users should preferably
* run in 64 bit mode which has far looser virtual address space limitations.
*
* than PHYSMEM is required in 32 bit mode, physmem can be set to the desired
* value or to 0 (to configure all available memory) via eeprom(1M). kernelbase
* should also be carefully tuned to balance out the need of the user
* application while minimizing the risk of kernel heap exhaustion due to
* kernelbase being set too high.
*/
#define PHYSMEM 0x400000
#else /* __amd64 */
/*
* For now we can handle memory with physical addresses up to about
* 64 Terabytes. This keeps the kernel above the VA hole, leaving roughly
* half the VA space for seg_kpm. When systems get bigger than 64TB this
* code will need revisiting. There is an implicit assumption that there
* are no *huge* holes in the physical address space too.
*/
#define PHYSMEM PHYSMEM_MAX64
#define AMD64_VA_HOLE_END 0xFFFF800000000000ul
#endif /* __amd64 */
char *kobj_file_buf;
int kobj_file_bufsize; /* set in /etc/system */
/* Global variables for MP support. Used in mp_startup */
caddr_t rm_platter_va = 0;
int auto_lpg_disable = 1;
/*
* Some CPUs have holes in the middle of the 64-bit virtual address range.
*/
/*
* kpm mapping window
*/
static int kpm_desired;
#ifdef __amd64
#endif
/*
* Configuration parameters set at boot time.
*/
char bootblock_fstype[16];
char kern_bootargs[OBP_MAXPATHLEN];
char kern_bootfile[OBP_MAXPATHLEN];
/*
* ZFS zio segment. This allows us to exclude large portions of ZFS data that
* gets cached in kmem caches on the heap. If this is set to zero, we allocate
* zio buffers from their own segment, otherwise they are allocated from the
* heap. The optimization of allocating zio buffers from their own segment is
* only valid on 64-bit kernels.
*/
#if defined(__amd64)
int segzio_fromheap = 0;
#else
int segzio_fromheap = 1;
#endif
/*
* new memory fragmentations are possible in startup() due to BOP_ALLOCs. this
* depends on number of BOP_ALLOC calls made and requested size, memory size
* combination and whether boot.bin memory needs to be freed.
*/
#define POSS_NEW_FRAGMENTS 12
/*
* VM data structures
*/
long page_hashsz; /* Size of page hash table (power of two) */
int pse_shift; /* log2(pse_table_size) */
#if defined(__amd64)
#else
#endif
#if defined(__amd64)
#else
#endif
/*
* VA range available to the debugger
*/
struct memseg *memseg_base;
struct vnode unused_pages_vp;
#define FOURGB 0x100000000LL
/*
* kphysm_init returns the number of pages that were processed
*/
/*
* a couple useful roundup macros
*/
#define ROUND_UP_PAGE(x) \
#define ROUND_UP_LPAGE(x) \
#define ROUND_UP_4MEG(x) \
#define ROUND_UP_TOPLEVEL(x) \
/*
* 32-bit Kernel's Virtual memory layout.
* +-----------------------+
* | |
* 0xFFC00000 -|-----------------------|- ARGSBASE
* | debugger |
* 0xFF800000 -|-----------------------|- SEGDEBUGBASE
* | Kernel Data |
* 0xFEC00000 -|-----------------------|
* | Kernel Text |
* 0xFE800000 -|-----------------------|- KERNEL_TEXT (0xFB400000 on Xen)
* |--- GDT ---|- GDT page (GDT_VA)
* |--- debug info ---|- debug info (DEBUG_INFO_VA)
* | |
* | page_t structures |
* | memsegs, memlists, |
* | page hash, etc. |
* --- -|-----------------------|- ekernelheap, valloc_base (floating)
* | | (segkp is just an arena in the heap)
* | |
* | kvseg |
* | |
* | |
* --- -|-----------------------|- kernelheap (floating)
* | Segkmap |
* 0xC3002000 -|-----------------------|- segmap_start (floating)
* | Red Zone |
* 0xC3000000 -|-----------------------|- kernelbase / userlimit (floating)
* | | ||
* | Shared objects | \/
* | |
* : :
* | user data |
* |-----------------------|
* | user text |
* 0x08048000 -|-----------------------|
* | user stack |
* : :
* | invalid |
* 0x00000000 +-----------------------+
*
*
* 64-bit Kernel's Virtual memory layout. (assuming 64 bit app)
* +-----------------------+
* | |
* 0xFFFFFFFF.FFC00000 |-----------------------|- ARGSBASE
* | debugger (?) |
* 0xFFFFFFFF.FF800000 |-----------------------|- SEGDEBUGBASE
* | unused |
* +-----------------------+
* | Kernel Data |
* 0xFFFFFFFF.FBC00000 |-----------------------|
* | Kernel Text |
* 0xFFFFFFFF.FB800000 |-----------------------|- KERNEL_TEXT
* |--- GDT ---|- GDT page (GDT_VA)
* |--- debug info ---|- debug info (DEBUG_INFO_VA)
* | |
* | Core heap | (used for loadable modules)
* 0xFFFFFFFF.C0000000 |-----------------------|- core_base / ekernelheap
* | Kernel |
* | heap |
* 0xFFFFFXXX.XXX00000 |-----------------------|- kernelheap (floating)
* | segmap |
* 0xFFFFFXXX.XXX00000 |-----------------------|- segmap_start (floating)
* | device mappings |
* 0xFFFFFXXX.XXX00000 |-----------------------|- toxic_addr (floating)
* | segzio |
* 0xFFFFFXXX.XXX00000 |-----------------------|- segzio_base (floating)
* | segkp |
* --- |-----------------------|- segkp_base (floating)
* | page_t structures | valloc_base + valloc_sz
* | memsegs, memlists, |
* | page hash, etc. |
* 0xFFFFFF00.00000000 |-----------------------|- valloc_base (lower if > 1TB)
* | segkpm |
* 0xFFFFFE00.00000000 |-----------------------|
* | Red Zone |
* 0xFFFFFD80.00000000 |-----------------------|- KERNELBASE (lower if > 1TB)
* | User stack |- User space memory
* | |
* | shared objects, etc | (grows downwards)
* : :
* | |
* 0xFFFF8000.00000000 |-----------------------|
* | |
* | VA Hole / unused |
* | |
* 0x00008000.00000000 |-----------------------|
* | |
* | |
* : :
* | user heap | (grows upwards)
* | |
* | user data |
* |-----------------------|
* | user text |
* 0x00000000.04000000 |-----------------------|
* | invalid |
* 0x00000000.00000000 +-----------------------+
*
* A 32 bit app on the 64 bit kernel sees the same layout as on the 32 bit
* kernel, except that userlimit is raised to 0xfe000000
*
* Floating values:
*
* valloc_base: start of the kernel's memory management/tracking data
* structures. This region contains page_t structures for
* physical memory, memsegs, memlists, and the page hash.
*
* core_base: start of the kernel's "core" heap area on 64-bit systems.
* This area is intended to be used for global data as well as for module
* restricted to a 2GB range, allowing every address within it to be
* accessed using rip-relative addressing
*
* ekernelheap: end of kernelheap and start of segmap.
*
* kernelheap: start of kernel heap. On 32-bit systems, this starts right
* above a red zone that separates the user's address space from the
* kernel's. On 64-bit systems, it sits above segkp and segkpm.
*
* segmap_start: start of segmap. The length of segmap can be modified
* through eeprom. The default length is 16MB on 32-bit systems and 64MB
* on 64-bit systems.
*
* kernelbase: On a 32-bit kernel the default value of 0xd4000000 will be
* decreased by 2X the size required for page_t. This allows the kernel
* heap to grow in size with physical memory. With sizeof(page_t) == 80
* bytes, the following shows the values of kernelbase and kernel heap
* sizes for different memory configurations (assuming default segmap and
* segkp sizes).
*
* mem size for kernelbase kernel heap
* size page_t's size
* ---- --------- ---------- -----------
* 1gb 0x01400000 0xd1800000 684MB
* 2gb 0x02800000 0xcf000000 704MB
* 4gb 0x05000000 0xca000000 744MB
* 6gb 0x07800000 0xc5000000 784MB
* 8gb 0x0a000000 0xc0000000 824MB
* 16gb 0x14000000 0xac000000 984MB
* 32gb 0x28000000 0x84000000 1304MB
* 64gb 0x50000000 0x34000000 1944MB (*)
*
* kernelbase is less than the abi minimum of 0xc0000000 for memory
* configurations above 8gb.
*
* (*) support for memory configurations above 32gb will require manual tuning
* of kernelbase to balance out the need of user applications.
*/
/* real-time-clock initialization parameters */
extern time_t process_rtc_config_file(void);
int segmapfreelists;
/*
* List of bootstrap pages. We mark these as allocated in startup.
* release_bootstrap() will free them when we're completely done with
* the bootstrap.
*/
/*
* boot time pages that have a vnode from the ramdisk will keep that forever.
*/
/*
* Lower 64K
*/
static int lower_pages_count = 0;
struct system_hardware system_hardware;
/*
* Enable some debugging messages concerning memory usage...
*/
static void
{
}
}
/*
* XX64 need a comment here.. are these just default values, surely
* we read the "cpuid" type information to figure this out.
*/
int l2cache_sz = 0x80000;
int l2cache_linesz = 0x40;
int l2cache_assoc = 1;
/*
* on 64 bit we use a predifined VA range for mapping devices in the kernel
* on 32 bit the mappings are intermixed in the heap, so we use a bit map
*/
#ifdef __amd64
#else /* __i386 */
#endif /* __i386 */
/*
* Simple boot time debug facilities
*/
static char *prm_dbg_str[] = {
"%s:%d: '%s' is 0x%x\n",
"%s:%d: '%s' is 0x%llx\n"
};
int prom_debug;
#define PRM_DEBUG(q) if (prom_debug) \
#define PRM_POINT(q) if (prom_debug) \
/*
* This structure is used to keep track of the intial allocations
* done in startup_memlist(). The value of NUM_ALLOCATIONS needs to
* be >= the number of ADD_TO_ALLOCATIONS() executed in the code.
*/
#define NUM_ALLOCATIONS 8
int num_allocations = 0;
struct {
void **al_ptr;
if (num_allocations == NUM_ALLOCATIONS) \
panic("too many ADD_TO_ALLOCATIONS()"); \
++num_allocations; \
}
/*
* Allocate all the initial memory needed by the page allocator.
*/
static void
perform_allocations(void)
{
int i;
int valloc_align;
panic("BOP_ALLOC() failed");
for (i = 0; i < num_allocations; ++i) {
}
}
/*
* Our world looks like this at startup time.
*
* In a 32-bit OS, boot loads the kernel text at 0xfe800000 and kernel data
* at 0xfec00000. On a 64-bit OS, kernel text and data are loaded at
* 0xffffffff.fe800000 and 0xffffffff.fec00000 respectively. Those
* addresses are fixed in the binary at link time.
*
* On the text page:
*
* On the data page:
*
* Machine-dependent startup code
*/
void
startup(void)
{
#if !defined(__xpv)
extern void startup_bios_disk(void);
extern void startup_pci_bios(void);
extern int post_fastreboot;
#endif
extern cpuset_t cpu_ready_set;
/*
* Make sure that nobody tries to use sekpm until we have
* initialized it properly.
*/
#if defined(__amd64)
kpm_desired = 1;
#endif
kpm_enable = 0;
#if defined(__xpv) /* XXPV fix me! */
{
extern int segvn_use_regions;
segvn_use_regions = 0;
}
#endif
startup_init();
#if defined(__xpv)
#endif
startup_kmem();
startup_vm();
#if !defined(__xpv)
if (!post_fastreboot)
#endif
#if defined(__xpv)
#endif
#if !defined(__xpv)
if (!post_fastreboot)
#endif
startup_end();
}
static void
{
PRM_POINT("startup_init() starting...");
/*
* Complete the extraction of cpuid data
*/
/*
* Check for prom_debug in boot environment
*/
++prom_debug;
PRM_POINT("prom_debug found in boot enviroment");
}
/*
* Collect node, cpu and memory configuration information.
*/
/*
* Halt if this is an unsupported processor.
*/
printf("\n486 processor (\"%s\") detected.\n",
CPU->cpu_brandstr);
halt("This processor is not supported by this release "
"of Solaris.");
}
PRM_POINT("startup_init() done");
}
/*
* everything mapped above KERNEL_TEXT) pages from phys_avail. Note it
* also filters out physical page zero. There is some reliance on the
* boot loader allocating only a few contiguous physical memory chunks.
*/
static void
{
if (prom_debug)
/*
* page zero is required for BIOS.. never make it available
*/
if (*addr == 0) {
*addr += MMU_PAGESIZE;
*size -= MMU_PAGESIZE;
}
/*
* First we trim from the front of the range. Since kbm_probe()
* to the list until no changes are seen. This deals with the case
* where page "p" is mapped at v, page "p + PAGESIZE" is mapped at w
* but w < v.
*/
do {
change = 0;
for (va = KERNEL_TEXT;
change = 1;
*addr += MMU_PAGESIZE;
*size -= MMU_PAGESIZE;
len -= MMU_PAGESIZE;
}
}
}
if (change && prom_debug)
} while (change);
/*
* Trim pages from the end of the range.
*/
for (va = KERNEL_TEXT;
}
if (prom_debug)
}
static void
kpm_init()
{
struct segkpm_crargs b;
/*
* These variables were all designed for sfmmu in which segkpm is
* mapped using a single pagesize - either 8KB or 4MB. On x86, we
* might use 2+ page sizes on a single machine, so none of these
* variables have a single correct value. They are set up as if we
* always use a 4KB pagesize, which should do no harm. In the long
* run, we should get rid of KPM's assumption that only a single
* pagesize is used.
*/
kpmp2pshft = 0;
kpmpnpgs = 1;
PRM_POINT("about to create segkpm");
panic("cannot attach segkpm");
b.nvcolors = 1;
panic("segkpm_create segkpm");
}
/*
* The debug info page provides enough information to allow external
* inspectors (e.g. when running under a hypervisor) to bootstrap
* themselves into allowing full-blown kernel debugging.
*/
static void
init_debug_info(void)
{
#ifndef __lint
#endif
panic("BOP_ALLOC() failed");
}
/*
* Build the memlists and other kernel essential memory system data structures.
* This is everything at valloc_base.
*/
static void
startup_memlist(void)
{
int memblocks;
int rsvdmemblocks;
extern void startup_build_mem_nodes(struct memlist *);
/* XX64 fix these - they should be in include files */
extern void page_coloring_setup(caddr_t);
PRM_POINT("startup_memlist() starting...");
/*
*/
/*
* Examine the boot loader physical memory map to find out:
* - total memory in system - physinstalled
* - the max physical address - physmax
* - the number of discontiguous segments of memory.
*/
if (prom_debug)
print_memlist("boot physinstalled",
&physinstalled, &memblocks);
/*
* Examine the bios reserved memory to find out:
* - the number of discontiguous segments of memory.
*/
if (prom_debug)
print_memlist("boot reserved mem",
&rsvd_pgcnt, &rsvdmemblocks);
/*
* Initialize hat's mmu parameters.
* Check for enforce-prot-exec in boot environment. It's used to
* The default is to enforce PROT_EXEC on processors that support NX.
* Boot seems to round up the "len", but 8 seems to be big enough.
*/
mmu_init();
#ifdef __i386
/*
* physmax is lowered if there is more memory than can be
*/
if (PFN_ABOVE64G(physmax)) {
}
} else {
if (PFN_ABOVE4G(physmax)) {
}
}
#endif
char value[8];
if (len < 8)
else
}
/*
* We will need page_t's for every page in the system, except for
* memory mapped at or above above the start of the kernel text segment.
*
* pages above e_modtext are attributed to kernel debugger (obp_pages)
*/
obp_pages = 0;
va = KERNEL_TEXT;
}
/*
* If physmem is patched to be non-zero, use it instead of the computed
* value unless it is larger than the actual amount of memory on hand.
*/
}
/*
* We now compute the sizes of all the initial allocations for
* structures the kernel needs in order do kmem_alloc(). These
* include:
* memsegs
* memlists
* page hash table
* page_t's
* page coloring data structs
*/
/*
* Reserve space for memlists. There's no real good way to know exactly
* how much room we'll need, but this should be a good upper bound.
*/
(memblocks + POSS_NEW_FRAGMENTS));
/*
* Reserve space for bios reserved memlists.
*/
/*
* The page structure hash table size is a power of 2
* such that the average hash chain length is PAGE_HASHAVELEN.
*/
/*
* Set aside room for the page structures themselves.
*/
/*
* determine l2 cache info and memory size for page coloring
*/
(void) getl2cacheinfo(CPU,
/*
* Allocate the array that protects pp->p_selock.
*/
#if defined(__amd64)
/*
* The default values of VALLOC_BASE and SEGKPM_BASE should work
* for values of physmax up to 1 Terabyte. They need adjusting when
* memory is at addresses above 1 TB.
*/
/* Round to largest possible pagesize for now */
/* make sure we leave some space for user apps above hole */
if (segkpm_base > SEGKPM_BASE)
}
#else /* __i386 */
#endif /* __i386 */
/*
* do all the initial allocations
*/
/*
* Build phys_install and phys_avail in kernel memspace.
* - phys_install should be all memory in the system.
* - phys_avail is phys_install minus any memory mapped before this
* point above KERNEL_TEXT.
*/
panic("physinstalled was too big!");
if (prom_debug)
PRM_POINT("Building phys_avail:\n");
panic("physavail was too big!");
if (prom_debug)
/*
* Build bios reserved memspace
*/
panic("bios_rsvd was too big!");
if (prom_debug)
/*
* setup page coloring
*/
page_lock_init(); /* currently a no-op */
/*
* free page list counters
*/
(void) page_ctrs_alloc(page_ctrs_mem);
/*
* Size the pcf array based on the number of cpus in the box at
* boot time.
*/
pcf_init();
/*
* Initialize the page structures from the memory lists.
*/
PRM_POINT("Calling kphysm_init()...");
PRM_POINT("kphysm_init() done");
/*
* Now that page_t's have been initialized, remove all the
* initial allocation pages from the kernel free page lists.
*/
PRM_POINT("startup_memlist() done");
#if defined(__amd64)
extern size_t textrepl_size_thresh;
}
#endif
}
/*
* Layout the kernel's part of address space and initialize kmem allocator.
*/
static void
startup_kmem(void)
{
extern void page_set_colorequiv_arr(void);
const char *fmt = "?features: %b\n";
PRM_POINT("startup_kmem() starting...");
#if defined(__amd64)
"systems.");
#else /* __i386 */
/*
* We configure kernelbase based on:
*
* 1. user specified kernelbase via eeprom command. Value cannot exceed
* KERNELBASE_MAX. we large page align eprom_kernelbase
*
* 2. Default to KERNELBASE and adjust to 2X less the size for page_t.
* On large memory systems we must lower kernelbase to allow
* enough room for page_t's for all of memory.
*
* The value set here, might be changed a little later.
*/
if (eprom_kernelbase) {
if (kernelbase > KERNELBASE_MAX)
} else {
}
core_size = 0;
#endif /* __i386 */
#if defined(__i386)
segkp_fromheap = 1;
#endif /* __i386 */
ekernelheap = (char *)core_base;
/*
* Now that we know the real value of kernelbase,
* update variables that were initialized with a value of
*
* XXX The problem with this sort of hackery is that the
* compiler just may feel like putting the const declarations
* (in param.c) into the .text section. Perhaps they should
* just be declared as variables there?
*/
#if defined(__amd64)
#else
#endif
#if defined(__i386)
/*
* If segmap is too large we can push the bottom of the kernel heap
* higher than the base. Or worse, it could exceed the top of the
* VA space entirely, causing it to wrap around.
*/
panic("too little address space available for kernelheap,"
" use eeprom for lower kernelbase or smaller segmapsize");
#endif /* __i386 */
/*
* Initialize the kernel heap. Note 3rd argument must be > 1st.
*/
#if defined(__xpv)
/*
* Link pending events struct into cpu struct
*/
#endif
/*
* Initialize kernel memory allocator.
*/
kmem_init();
/*
* Factor in colorequiv to check additional 'equivalent' bins
*/
/*
* print this out early so that we know what's going on
*/
/*
* Initialize bp_mapin().
*/
/*
* orig_npages is non-zero if physmem has been configured for less
* than the available memory.
*/
if (orig_npages) {
}
#if defined(__i386)
"System using 0x%lx",
#endif
#ifdef KERNELBASE_ABI_MIN
}
#endif
#ifdef __xpv
/*
* Some of the xen start information has to be relocated up
* into the kernel's permanent address space.
*/
PRM_POINT("calling xen_relocate_start_info()");
PRM_POINT("xen_relocate_start_info() done");
/*
* (Update the vcpu pointer in our cpu structure to point into
* the relocated shared info.)
*/
#endif
PRM_POINT("startup_kmem() done");
}
#ifndef __xpv
/*
* If we have detected that we are running in an HVM environment, we need
* to prepend the PV driver directory to the module search path.
*/
#define HVM_MOD_DIR "/platform/i86hvm/kernel"
static void
{
int newlen;
/*
* We are about to resync with krtld. krtld will reset its
* internal module search path iff Solaris has set default_path.
* We want to be sure we're prepending this new directory to the
* right search path.
*/
}
#endif
static void
startup_modules(void)
{
int cnt;
extern void prom_setup(void);
int32_t v, h;
char d[11];
char *cp;
PRM_POINT("startup_modules() starting...");
#ifndef __xpv
/*
* Initialize ten-micro second timer so that drivers will
* not get short changed in their init phase. This was
* not getting called until clkinit which, on fast cpu's
* caused the drv_usecwait to be way too short.
*/
microfind();
if (get_hwenv() == HW_XEN_HVM)
#endif
/*
* Read the GMT lag from /etc/rtc_config.
*/
/*
* Calculate default settings of system parameters based upon
*/
param_calc(0);
mod_setup();
/*
* Initialize system parameters.
*/
param_init();
/*
* Initialize the default brands
*/
brand_init();
/*
* maxmem is the amount of physical memory we're playing with.
*/
/*
* Initialize segment management stuff.
*/
seg_init();
halt("Can't load specfs");
halt("Can't load devfs");
halt("Can't load dev");
dispinit();
/*
* This is needed here to initialize hw_serial[] for cluster booting.
*/
if ((h = set_soft_hostid()) == HW_INVALID_HOSTID) {
} else {
d[cnt] = (char)(v % 10);
v /= 10;
if (v == 0)
break;
}
*cp = 0;
}
/* Read cluster configuration data. */
clconf_init();
#if defined(__xpv)
ec_init();
gnttab_init();
(void) xs_early_init();
#endif /* __xpv */
/*
* Create a kernel device tree. First, create rootnex and
* then invoke bus specific code to probe devices.
*/
setup_ddi();
/*
* Set up the CPU module subsystem for the boot cpu in the native
* case, and all physical cpu resource in the xpv dom0 case.
* Modifies the device tree, so this must be done after
* setup_ddi().
*/
#ifdef __xpv
/*
* If paravirtualized and on dom0 then we initialize all physical
* cpu handles now; if paravirtualized on a domU then do not
* initialize.
*/
if (DOMAIN_IS_INITDOMAIN(xen_info)) {
(x86_feature & X86_MCA))
}
}
#else
/*
* Initialize a handle for the boot cpu - others will initialize
* as they startup. Do not do this if we know we are in an HVM domU.
*/
if ((get_hwenv() != HW_XEN_HVM) &&
(x86_feature & X86_MCA))
#endif /* __xpv */
/*
*/
PRM_POINT("startup_modules: calling prom_setup...");
prom_setup();
PRM_POINT("startup_modules: done");
/*
* Load all platform specific modules
*/
PRM_POINT("startup_modules: calling psm_modload...");
psm_modload();
PRM_POINT("startup_modules() done");
}
/*
* claim a "setaside" boot page for use in the kernel
*/
page_t *
{
if (PP_ISBOOTPAGES(pp)) {
else
} else {
/*
* htable_attach() expects a base pagesize page
*/
}
return (pp);
}
/*
* Walk through the pagetables looking for pages mapped in by boot. If the
* setaside flag is set the pages are expected to be returned to the
* kernel later in boot, so we add them to the bootpages list.
*/
static void
{
pgcnt_t boot_protect_cnt = 0;
panic("0x%lx byte mapping at 0x%p exceeds boot's "
while (len > 0) {
if (setaside == 0)
panic("Unexpected mapping by boot. "
"addr=%p pfn=%lx\n",
}
}
++pfn;
len -= MMU_PAGESIZE;
va += MMU_PAGESIZE;
}
}
}
/*
*
*/
static void
layout_kernel_va(void)
{
PRM_POINT("layout_kernel_va() starting...");
/*
* Establish the final size of the kernel's heap, size of segmap,
* segkp, etc.
*/
#if defined(__amd64)
panic("not enough room for kpm!");
/*
* By default we create a seg_kp in 64 bit kernels, it's a little
* faster to access than embedding it in the heap.
*/
if (!segkp_fromheap) {
/*
* determine size of segkp
*/
sz = SEGKPDEFSIZE;
"segkpsize has been reset to %ld pages",
}
}
/*
* segzio is used for ZFS cached data. It uses a distinct VA
* segment (from kernel heap) so that we can easily tell not to
* include it in kernel crash dumps on 64 bit kernels. The trick is
* to give it lots of VA, but not constrain the kernel heap.
* We scale the size of segzio linearly with physmem up to
* SEGZIOMAXSIZE. Above that amount it scales at 50% of physmem.
*/
if (segzio_fromheap) {
segziosize = 0;
} else {
if (size < SEGZIOMINSIZE)
if (size > SEGZIOMAXSIZE) {
if (physmem_size > size)
}
}
/*
* Put the range of VA for device mappings next, kmdb knows to not
* grep in this range of addresses.
*/
#else /* __i386 */
#endif /* __i386 */
/*
* Users can change segmapsize through eeprom. If the variable
* is tuned through eeprom, there is no upper bound on the
* size of segmap.
*/
#if defined(__i386)
/*
* 32-bit systems don't have segkpm or segkp, so segmap appears at
* the bottom of the kernel's address range. Set aside space for a
* small red zone just below the start of segmap.
*/
#endif
PRM_POINT("layout_kernel_va() done...");
}
/*
* Finish initializing the VM system, now that we are no longer
* relying on the boot time memory allocators.
*/
static void
startup_vm(void)
{
struct segmap_crargs a;
extern int use_brk_lpg, use_stk_lpg;
PRM_POINT("startup_vm() starting...");
/*
* Initialize the hat layer.
*/
hat_init();
/*
* Do final allocations of HAT data structures that need to
* be allocated before quiescing the boot loader.
*/
PRM_POINT("Calling hat_kern_alloc()...");
PRM_POINT("hat_kern_alloc() done");
#ifndef __xpv
/*
* Setup Page Attribute Table
*/
pat_sync();
#endif
/*
* The next two loops are done in distinct steps in order
* to be sure that any page that is doubly mapped (both above
* KERNEL_TEXT and below kernelbase) is dealt with correctly.
* Note this may never happen, but it might someday.
*/
PRM_POINT("Protecting boot pages");
/*
* Protect any pages mapped above KERNEL_TEXT that somehow have
* page_t's. This can only happen if something weird allocated
*/
/*
* Before we can take over memory allocation/mapping from the boot
* loader we must remove from our free page lists any boot allocated
* pages that stay mapped until release_bootstrap().
*/
/*
* Switch to running on regular HAT (not boot_mmu)
*/
PRM_POINT("Calling hat_kern_setup()...");
/*
* It is no longer safe to call BOP_ALLOC(), so make sure we don't.
*/
PRM_POINT("hat_kern_setup() done");
/*
* Initialize VM system
*/
PRM_POINT("Calling kvm_init()...");
kvm_init();
PRM_POINT("kvm_init() done");
/*
* Tell kmdb that the VM system is now working
*/
#if defined(__xpv)
/*
* Populate the I/O pool on domain 0
*/
if (DOMAIN_IS_INITDOMAIN(xen_info)) {
extern long populate_io_pool(void);
long init_io_pool_cnt;
PRM_POINT("Populating reserve I/O page pool");
}
#endif
/*
* Mangle the brand string etc.
*/
#if defined(__amd64)
/*
*/
#else /* __i386 */
/*
* allocate the bit map that tracks toxic pages
*/
#endif /* __i386 */
/*
* Now that we've got more VA, as well as the ability to allocate from
* it, tell the debugger.
*/
/*
* The following code installs a special page fault handler (#pf)
* to work around a pentium bug.
*/
if (x86_type == X86_TYPE_P5) {
panic("failed to install pentium_pftrap");
}
#endif /* !__amd64 */
#if !defined(__xpv)
/*
* Map page pfn=0 for drivers, such as kd, that need to pick up
* parameters left there by controllers/BIOS.
*/
PRM_POINT("setup up p0_va");
#endif
/*
* disable automatic large pages for small memory systems or
* when the disable flag is set.
*
* Do not yet consider page sizes larger than 2m/4m.
*/
}
use_brk_lpg = 0;
use_stk_lpg = 0;
}
PRM_POINT("Calling hat_init_finish()...");
PRM_POINT("hat_init_finish() done");
/*
* Initialize the segkp segment type.
*/
PRM_POINT("Attaching segkp");
if (segkp_fromheap) {
segkp) < 0) {
panic("startup: cannot attach segkp");
/*NOTREACHED*/
}
PRM_POINT("Doing segkp_create()");
if (segkp_create(segkp) != 0) {
panic("startup: segkp_create failed");
/*NOTREACHED*/
}
/*
* kpm segment
*/
segmap_kpm = 0;
if (kpm_desired) {
kpm_init();
kpm_enable = 1;
}
/*
* Now create segmap segment.
*/
panic("cannot attach segmap");
/*NOTREACHED*/
}
a.shmsize = 0;
a.nfreelist = segmapfreelists;
panic("segmap_create segmap");
segdev_init();
#if defined(__xpv)
if (DOMAIN_IS_INITDOMAIN(xen_info))
#endif
pmem_init();
PRM_POINT("startup_vm() done");
}
/*
* Load a tod module for the non-standard tod part found on this system.
*/
static void
load_tod_module(char *todmod)
{
halt("Can't load TOD module");
}
static void
startup_end(void)
{
int i;
extern void setx86isalist(void);
extern void cpu_event_init(void);
PRM_POINT("startup_end() starting...");
/*
* Perform tasks that get done after most of the VM
* initialization has been done but before the clock
* and other devices get started.
*/
kern_setup1();
/*
* Perform CPC initialization for this CPU.
*/
/*
* Initialize cpu event framework.
*/
#if defined(OPTERON_WORKAROUND_6323525)
#endif
/*
* If needed, load TOD module now so that ddi_get_time(9F) etc. work
*/
if (tod_module_name != NULL) {
PRM_POINT("load_tod_module()");
}
#if defined(__xpv)
/*
* Forceload interposing TOD module for the hypervisor.
*/
PRM_POINT("load_tod_module()");
load_tod_module("xpvtod");
#endif
/*
* Configure the system.
*/
PRM_POINT("Calling configure()...");
configure(); /* set up devices */
PRM_POINT("configure() done");
/*
* Set the isa_list string to the defined instruction sets we
* support.
*/
psm_install();
/*
* We're done with bootops. We don't unmap the bootstrap yet because
* we're still using bootsvcs.
*/
PRM_POINT("NULLing out bootops");
#if defined(__xpv)
xs_domu_init();
#endif
#if !defined(__xpv)
if (rootnex_iommu_init != NULL) {
}
#endif
PRM_POINT("Enabling interrupts");
(*picinitf)();
sti();
#if defined(__xpv)
#endif
/*
* Register these software interrupts for ddi timer.
* Software interrupts up to the level 10 are supported.
*/
for (i = DDI_IPL_1; i <= DDI_IPL_10; i++) {
}
#if !defined(__xpv)
PRM_POINT("No AMD IOMMU present\n");
} else if (ddi_hold_installed_driver(ddi_name_to_major(
"amd_iommu")) == NULL) {
prom_printf("ERROR: failed to attach AMD IOMMU\n");
}
#endif
PRM_POINT("startup_end() done");
}
/*
* Don't remove the following 2 variables. They are necessary
*/
void
post_startup(void)
{
extern void cpupm_init(cpu_t *);
extern void cpu_event_init_cpu(cpu_t *);
/*
* Set the system wide, processor-specific flags to be passed
* to userland via the aux vector for performance hints and
* instruction set extensions.
*/
bind_hwcap();
#ifdef __xpv
if (DOMAIN_IS_INITDOMAIN(xen_info))
#endif
{
/*
* Load the System Management BIOS into the global ksmbios
* handle, if an SMBIOS is present on this system.
*/
#if defined(__xpv)
#else
/*
* Startup the memory scrubber.
* XXPV This should be running somewhere ..
*/
if (get_hwenv() != HW_XEN_HVM)
#endif
}
/*
* Complete CPU module initialization
*/
/*
*/
/*
* ON4.0: Force /proc module in until clock interrupt handle fixed
*/
(void) i_ddi_attach_hw_nodes("pit_beep");
#if defined(__i386)
/*
* Check for required functional Floating Point hardware,
* unless FP hardware explicitly disabled.
*/
halt("No working FP hardware found");
#endif
pg_init();
}
static int
{
}
void
release_bootstrap(void)
{
int root_is_ramdisk;
extern void kobj_boot_unmountroot(void);
#if !defined(__xpv)
#endif
/* unmount boot ramdisk and release kmem usage */
/*
* We're finished using the boot loader so free its pages.
*/
PRM_POINT("Unmapping lower boot pages");
/*
* If root isn't on ramdisk, destroy the hardcoded
* ramdisk node now and release the memory. Else,
* ramdisk memory is kept in rd_pages.
*/
if (!root_is_ramdisk) {
(void) ddi_remove_child(dip, 0);
}
PRM_POINT("Releasing boot pages");
while (bootpages) {
/* Keep pages for the lower 64K */
lower_pages = pp;
continue;
}
ramdisk_end)) {
continue;
}
}
PRM_POINT("Boot pages released");
#if !defined(__xpv)
/* XXPV -- note this following bunch of code needs to be revisited in Xen 3.0 */
/*
* Find 1 page below 1 MB so that other processors can boot up or
* so that any processor can resume.
* Make sure it has a kernel VA as well as a 1:1 mapping.
* We should have just free'd one up.
*/
/*
* 0x10 pages is 64K. Leave the bottom 64K alone
* for BIOS.
*/
continue;
break;
}
panic("No page below 1M available for starting "
"other processors or for resuming from system-suspend");
#endif /* !__xpv */
}
/*
* Initialize the platform-specific parts of a page_t.
*/
void
{
}
/*
* kphysm_init() initializes physical memory.
*/
static pgcnt_t
{
struct memseg *cur_memseg;
pgcnt_t pages_done = 0;
extern pfn_t ddiphysmin;
extern int mnode_xwa;
/*
* In a 32 bit kernel can't use higher memory if we're
* not booting in PAE mode. This check takes care of that.
*/
continue;
/*
* align addr and size - they may not be at page boundaries
*/
if ((addr & MMU_PAGEOFFSET) != 0) {
addr += MMU_PAGEOFFSET;
}
/* only process pages below or equal to physmax */
if (num == 0)
continue;
pages_done += num;
if (prom_debug)
" pgs=0x%lx pfn 0x%lx-0x%lx\n",
/*
* Ignore pages below ddiphysmin to simplify ddi memory
* allocation with non-zero addr_lo requests.
*/
if (base_pfn < ddiphysmin) {
continue;
}
/*
* mnode_xwa is greater than 1 when large pages regions can
* cross memory node boundaries. To prevent the formation
* of these large pages, configure the memsegs based on the
* memory node ranges which had been made non-contiguous.
*/
if (mnode_xwa > 1) {
/*
* current range spans more than 1 memory node.
* Set num to only the pfn range in the start
* memory node.
*/
+ 1;
}
}
for (;;) {
/*
* Build the memsegs entry
*/
/*
* Insert into memseg list in decreasing pfn range
* order. Low memory is typically more fragmented such
* that this ordering keeps the larger ranges at the
* front of the list for code that searches memseg.
* This ASSERTS that the memsegs coming in from boot
* are in increasing physical address order and not
* contiguous.
*/
}
/*
* add_physmem() initializes the PSM part of the page
* struct by calling the PSM back with add_physmem_cb().
* In addition it coalesces pages into larger pages as
* it initializes them.
*/
cur_memseg++;
availrmem_initial += num;
break;
/* process next memory node range */
ms++;
}
}
return (pages_done);
}
/*
* Kernel VM initialization.
*/
static void
kvm_init(void)
{
/*
* Put the kernel segments in kernel address space.
*/
as_avlinit(&kas);
(void) segkmem_create(&ktextseg);
(void) segkmem_create(&kvalloc);
(void) segkmem_create(&kvseg);
if (core_size > 0) {
PRM_POINT("attaching kvseg_core");
&kvseg_core);
(void) segkmem_create(&kvseg_core);
}
if (segziosize > 0) {
PRM_POINT("attaching segzio");
&kzioseg);
(void) segkmem_zio_create(&kzioseg);
/* create zio area covering new segment */
}
(void) segkmem_create(&kdebugseg);
/*
* Ensure that the red zone at kernelbase is never accessible.
*/
PRM_POINT("protecting redzone");
/*
* Make the text writable so that it can be hot patched by DTrace.
*/
/*
* Make data writable until end.
*/
}
#ifndef __xpv
/*
* Solaris adds an entry for Write Combining caching to the PAT
*/
void
pat_sync(void)
{
if (!(x86_feature & X86_PAT))
return;
/* disable caching and flush all caches and TLBs */
} else {
reload_cr3();
}
/* add our entry to the PAT */
/* flush TLBs and cache again, then reenable cr0 caching */
} else {
reload_cr3();
}
}
#endif /* !__xpv */
#if defined(_SOFT_HOSTID)
/*
* On platforms that do not have a hardware serial number, attempt
* not exist, assume that we are to generate a new hostid and set
* it in the kernel, for subsequent saving by a userland process
* once the system is up and the root filesystem is mounted r/w.
*
* In order to gracefully support upgrade on OpenSolaris, if
*
* In an attempt to make the hostid less prone to abuse
* in rot47 format.
*/
extern volatile unsigned long tenmicrodata;
static int atoi(char *);
static int32_t
set_soft_hostid(void)
{
char tokbuf[MAXNAMELEN];
int done = 0;
int i;
unsigned char *c;
/*
* random number to use at the hostid. A nice way to do this
* is to read the real time clock. To remain xen-compatible,
* we can't poke the real hardware, so we use tsc_read() to
* read the real time clock. However, there is an ominous
* warning in tsc_read that says it can return zero, so we
* deal with that possibility by falling back to using the
* (hopefully random enough) value in tenmicrodata.
*/
/*
* hostid file not found - try to load sysinit module
* and see if it has a nonzero hostid value...use that
* instead of generating a new hostid here if so.
*/
(void) modunload(i);
}
if (hostid == HW_INVALID_HOSTID) {
if (tsc == 0) /* tsc_read can return zero sometimes */
else
}
} else {
/* hostid file found */
while (!done) {
switch (token) {
case POUND:
/*
* skip comments
*/
break;
case STRING:
/*
* un-rot47 - obviously this
* nonsense is ascii-specific
*/
for (c = (unsigned char *)tokbuf;
*c != '\0'; c++) {
*c += 47;
if (*c > '~')
*c -= 94;
else if (*c < '!')
*c += 94;
}
/*
* now we should have a real number
*/
"Bad value %s for hostid",
tokbuf);
else
break;
case EOF:
done = 1;
/* FALLTHROUGH */
case NEWLINE:
break;
default:
break;
}
}
"hostid missing or corrupt");
}
/*
* new hostid we generated in this routine or HW_INVALID_HOSTID if not
* set.
*/
return (hostid);
}
static int
atoi(char *p)
{
int i = 0;
while (*p != '\0')
i = 10 * i + (*p++ - '0');
return (i);
}
#endif /* _SOFT_HOSTID */
void
get_system_configuration(void)
{
char prop[32];
} else {
}
else
else
segmapfreelists = 0; /* use segmap driver default */
else
segmapfreelists = (int)lvalue;
/* physmem used to be here, but moved much earlier to fakebop.c */
}
/*
* Add to a memory list.
* start = start of new memory segment
* len = length of new memory segment in bytes
* new = pointer to a new struct memlist
* memlistp = memory list to which to add segment.
*/
void
{
while (cur) {
return;
}
return;
}
}
}
void
{
}
{
}
/*ARGSUSED*/
{
panic("unexpected call to kobj_texthole_alloc()");
/*NOTREACHED*/
return (0);
}
/*ARGSUSED*/
void
{
panic("unexpected call to kobj_texthole_free()");
}
/*
* This is called just after configure() in startup().
*
* The ISALIST concept is a bit hopeless on Intel, because
* there's no guarantee of an ever-more-capable processor
* given that various parts of the instruction set may appear
* and disappear between different implementations.
*
* While it would be possible to correct it and even enhance
* it somewhat, the explicit hardware capability bitmask allows
* more flexibility.
*
* So, we just leave this alone.
*/
void
setx86isalist(void)
{
char *tp;
extern char *isa_list;
#define TBUFSIZE 1024
*tp = '\0';
#if defined(__amd64)
#endif
switch (x86_vendor) {
case X86_VENDOR_Intel:
case X86_VENDOR_AMD:
case X86_VENDOR_TM:
if (x86_feature & X86_CMOV) {
/*
* Pentium Pro or later
*/
"+mmx pentium_pro " : " ");
}
/*FALLTHROUGH*/
case X86_VENDOR_Cyrix:
/*
* The Cyrix 6x86 does not have any Pentium features
* accessible while not at privilege level 0.
*/
if (x86_feature & X86_CPUID) {
"+mmx pentium " : " ");
}
break;
default:
break;
}
}
#ifdef __amd64
void *
{
}
void
{
}
#else /* __i386 */
void *
{
uintptr_t v;
return (NULL);
ASSERT(v >= kernelbase);
++start;
}
return (vaddr);
}
void
{
ASSERT(v >= kernelbase);
++start;
}
}
/*
* returns 1st address in range that is in device arena, or NULL
* if len is not NULL it returns the length of the toxic range
*/
void *
{
/*
* if called very early by kmdb, just return NULL
*/
if (toxic_bit_map == NULL)
return (NULL);
/*
* First check if we're completely outside the bitmap range.
*/
return (NULL);
/*
* Trim ends of search to look at only what the bitmap covers.
*/
if (v < kernelbase)
v = kernelbase;
if (end >= toxic_bit_map_len)
return (NULL);
return ((void *)v);
}
#endif /* __i386 */