Home | History | Annotate | Download | only in vm
      1 /*
      2  * CDDL HEADER START
      3  *
      4  * The contents of this file are subject to the terms of the
      5  * Common Development and Distribution License (the "License").
      6  * You may not use this file except in compliance with the License.
      7  *
      8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
      9  * or http://www.opensolaris.org/os/licensing.
     10  * See the License for the specific language governing permissions
     11  * and limitations under the License.
     12  *
     13  * When distributing Covered Code, include this CDDL HEADER in each
     14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
     15  * If applicable, add the following below this CDDL HEADER, with the
     16  * fields enclosed by brackets "[]" replaced with your own identifying
     17  * information: Portions Copyright [yyyy] [name of copyright owner]
     18  *
     19  * CDDL HEADER END
     20  */
     21 /*
     22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
     23  * Use is subject to license terms.
     24  */
     25 
     26 #pragma ident	"@(#)seg_kmem.c	1.53	07/12/10 SMI"
     27 
     28 #include <sys/types.h>
     29 #include <sys/t_lock.h>
     30 #include <sys/param.h>
     31 #include <sys/sysmacros.h>
     32 #include <sys/tuneable.h>
     33 #include <sys/systm.h>
     34 #include <sys/vm.h>
     35 #include <sys/kmem.h>
     36 #include <sys/vmem.h>
     37 #include <sys/mman.h>
     38 #include <sys/cmn_err.h>
     39 #include <sys/debug.h>
     40 #include <sys/dumphdr.h>
     41 #include <sys/bootconf.h>
     42 #include <sys/lgrp.h>
     43 #include <vm/seg_kmem.h>
     44 #include <vm/hat.h>
     45 #include <vm/page.h>
     46 #include <vm/vm_dep.h>
     47 #include <vm/faultcode.h>
     48 #include <sys/promif.h>
     49 #include <vm/seg_kp.h>
     50 #include <sys/bitmap.h>
     51 #include <sys/mem_cage.h>
     52 
     53 /*
     54  * seg_kmem is the primary kernel memory segment driver.  It
     55  * maps the kernel heap [kernelheap, ekernelheap), module text,
     56  * and all memory which was allocated before the VM was initialized
     57  * into kas.
     58  *
     59  * Pages which belong to seg_kmem are hashed into &kvp vnode at
     60  * an offset equal to (u_offset_t)virt_addr, and have p_lckcnt >= 1.
     61  * They must never be paged out since segkmem_fault() is a no-op to
     62  * prevent recursive faults.
     63  *
     64  * Currently, seg_kmem pages are sharelocked (p_sharelock == 1) on
     65  * __x86 and are unlocked (p_sharelock == 0) on __sparc.  Once __x86
     66  * supports relocation the #ifdef kludges can be removed.
     67  *
     68  * seg_kmem pages may be subject to relocation by page_relocate(),
     69  * provided that the HAT supports it; if this is so, segkmem_reloc
     70  * will be set to a nonzero value. All boot time allocated memory as
     71  * well as static memory is considered off limits to relocation.
     72  * Pages are "relocatable" if p_state does not have P_NORELOC set, so
     73  * we request P_NORELOC pages for memory that isn't safe to relocate.
     74  *
     75  * The kernel heap is logically divided up into four pieces:
     76  *
     77  *   heap32_arena is for allocations that require 32-bit absolute
     78  *   virtual addresses (e.g. code that uses 32-bit pointers/offsets).
     79  *
     80  *   heap_core is for allocations that require 2GB *relative*
     81  *   offsets; in other words all memory from heap_core is within
     82  *   2GB of all other memory from the same arena. This is a requirement
     83  *   of the addressing modes of some processors in supervisor code.
     84  *
     85  *   heap_arena is the general heap arena.
     86  *
     87  *   static_arena is the static memory arena.  Allocations from it
     88  *   are not subject to relocation so it is safe to use the memory
     89  *   physical address as well as the virtual address (e.g. the VA to
     90  *   PA translations are static).  Caches may import from static_arena;
     91  *   all other static memory allocations should use static_alloc_arena.
     92  *
     93  * On some platforms which have limited virtual address space, seg_kmem
     94  * may share [kernelheap, ekernelheap) with seg_kp; if this is so,
     95  * segkp_bitmap is non-NULL, and each bit represents a page of virtual
     96  * address space which is actually seg_kp mapped.
     97  */
     98 
     99 extern ulong_t *segkp_bitmap;   /* Is set if segkp is from the kernel heap */
    100 
    101 char *kernelheap;		/* start of primary kernel heap */
    102 char *ekernelheap;		/* end of primary kernel heap */
    103 struct seg kvseg;		/* primary kernel heap segment */
    104 struct seg kvseg_core;		/* "core" kernel heap segment */
    105 struct seg kzioseg;		/* Segment for zio mappings */
    106 vmem_t *heap_arena;		/* primary kernel heap arena */
    107 vmem_t *heap_core_arena;	/* core kernel heap arena */
    108 char *heap_core_base;		/* start of core kernel heap arena */
    109 char *heap_lp_base;		/* start of kernel large page heap arena */
    110 char *heap_lp_end;		/* end of kernel large page heap arena */
    111 vmem_t *hat_memload_arena;	/* HAT translation data */
    112 struct seg kvseg32;		/* 32-bit kernel heap segment */
    113 vmem_t *heap32_arena;		/* 32-bit kernel heap arena */
    114 vmem_t *heaptext_arena;		/* heaptext arena */
    115 struct as kas;			/* kernel address space */
    116 struct vnode kvp;		/* vnode for all segkmem pages */
    117 struct vnode zvp;		/* vnode for zfs pages */
    118 int segkmem_reloc;		/* enable/disable relocatable segkmem pages */
    119 vmem_t *static_arena;		/* arena for caches to import static memory */
    120 vmem_t *static_alloc_arena;	/* arena for allocating static memory */
    121 vmem_t *zio_arena = NULL;	/* arena for allocating zio memory */
    122 vmem_t *zio_alloc_arena = NULL;	/* arena for allocating zio memory */
    123 
    124 /*
    125  * seg_kmem driver can map part of the kernel heap with large pages.
    126  * Currently this functionality is implemented for sparc platforms only.
    127  *
    128  * The large page size "segkmem_lpsize" for kernel heap is selected in the
    129  * platform specific code. It can also be modified via /etc/system file.
    130  * Setting segkmem_lpsize to PAGESIZE in /etc/system disables usage of large
    131  * pages for kernel heap. "segkmem_lpshift" is adjusted appropriately to
    132  * match segkmem_lpsize.
    133  *
    134  * At boot time we carve from kernel heap arena a range of virtual addresses
    135  * that will be used for large page mappings. This range [heap_lp_base,
    136  * heap_lp_end) is set up as a separate vmem arena - "heap_lp_arena". We also
    137  * create "kmem_lp_arena" that caches memory already backed up by large
    138  * pages. kmem_lp_arena imports virtual segments from heap_lp_arena.
    139  */
    140 
    141 size_t	segkmem_lpsize;
    142 static  uint_t	segkmem_lpshift = PAGESHIFT;
    143 int	segkmem_lpszc = 0;
    144 
    145 size_t  segkmem_kmemlp_quantum = 0x400000;	/* 4MB */
    146 size_t  segkmem_heaplp_quantum;
    147 vmem_t *heap_lp_arena;
    148 static  vmem_t *kmem_lp_arena;
    149 static  vmem_t *segkmem_ppa_arena;
    150 static	segkmem_lpcb_t segkmem_lpcb;
    151 
    152 /*
    153  * We use "segkmem_kmemlp_max" to limit the total amount of physical memory
    154  * consumed by the large page heap. By default this parameter is set to 1/8 of
    155  * physmem but can be adjusted through /etc/system either directly or
    156  * indirectly by setting "segkmem_kmemlp_pcnt" to the percent of physmem
    157  * we allow for large page heap.
    158  */
    159 size_t  segkmem_kmemlp_max;
    160 static  uint_t  segkmem_kmemlp_pcnt;
    161 
    162 /*
    163  * Getting large pages for kernel heap could be problematic due to
    164  * physical memory fragmentation. That's why we allow to preallocate
    165  * "segkmem_kmemlp_min" bytes at boot time.
    166  */
    167 static  size_t	segkmem_kmemlp_min;
    168 
    169 /*
    170  * Throttling is used to avoid expensive tries to allocate large pages
    171  * for kernel heap when a lot of succesive attempts to do so fail.
    172  */
    173 static  ulong_t segkmem_lpthrottle_max = 0x400000;
    174 static  ulong_t segkmem_lpthrottle_start = 0x40;
    175 static  ulong_t segkmem_use_lpthrottle = 1;
    176 
    177 /*
    178  * Freed pages accumulate on a garbage list until segkmem is ready,
    179  * at which point we call segkmem_gc() to free it all.
    180  */
    181 typedef struct segkmem_gc_list {
    182 	struct segkmem_gc_list	*gc_next;
    183 	vmem_t			*gc_arena;
    184 	size_t			gc_size;
    185 } segkmem_gc_list_t;
    186 
    187 static segkmem_gc_list_t *segkmem_gc_list;
    188 
    189 /*
    190  * Allocations from the hat_memload arena add VM_MEMLOAD to their
    191  * vmflags so that segkmem_xalloc() can inform the hat layer that it needs
    192  * to take steps to prevent infinite recursion.  HAT allocations also
    193  * must be non-relocatable to prevent recursive page faults.
    194  */
    195 static void *
    196 hat_memload_alloc(vmem_t *vmp, size_t size, int flags)
    197 {
    198 	flags |= (VM_MEMLOAD | VM_NORELOC);
    199 	return (segkmem_alloc(vmp, size, flags));
    200 }
    201 
    202 /*
    203  * Allocations from static_arena arena (or any other arena that uses
    204  * segkmem_alloc_permanent()) require non-relocatable (permanently
    205  * wired) memory pages, since these pages are referenced by physical
    206  * as well as virtual address.
    207  */
    208 void *
    209 segkmem_alloc_permanent(vmem_t *vmp, size_t size, int flags)
    210 {
    211 	return (segkmem_alloc(vmp, size, flags | VM_NORELOC));
    212 }
    213 
    214 /*
    215  * Initialize kernel heap boundaries.
    216  */
    217 void
    218 kernelheap_init(
    219 	void *heap_start,
    220 	void *heap_end,
    221 	char *first_avail,
    222 	void *core_start,
    223 	void *core_end)
    224 {
    225 	uintptr_t textbase;
    226 	size_t core_size;
    227 	size_t heap_size;
    228 	vmem_t *heaptext_parent;
    229 	size_t	heap_lp_size = 0;
    230 #ifdef __sparc
    231 	size_t kmem64_sz = kmem64_aligned_end - kmem64_base;
    232 #endif	/* __sparc */
    233 
    234 	kernelheap = heap_start;
    235 	ekernelheap = heap_end;
    236 
    237 #ifdef __sparc
    238 	heap_lp_size = (((uintptr_t)heap_end - (uintptr_t)heap_start) / 4);
    239 	/*
    240 	 * Bias heap_lp start address by kmem64_sz to reduce collisions
    241 	 * in 4M kernel TSB between kmem64 area and heap_lp
    242 	 */
    243 	kmem64_sz = P2ROUNDUP(kmem64_sz, MMU_PAGESIZE256M);
    244 	if (kmem64_sz <= heap_lp_size / 2)
    245 		heap_lp_size -= kmem64_sz;
    246 	heap_lp_base = ekernelheap - heap_lp_size;
    247 	heap_lp_end = heap_lp_base + heap_lp_size;
    248 #endif	/* __sparc */
    249 
    250 	/*
    251 	 * If this platform has a 'core' heap area, then the space for
    252 	 * overflow module text should be carved out of the end of that
    253 	 * heap.  Otherwise, it gets carved out of the general purpose
    254 	 * heap.
    255 	 */
    256 	core_size = (uintptr_t)core_end - (uintptr_t)core_start;
    257 	if (core_size > 0) {
    258 		ASSERT(core_size >= HEAPTEXT_SIZE);
    259 		textbase = (uintptr_t)core_end - HEAPTEXT_SIZE;
    260 		core_size -= HEAPTEXT_SIZE;
    261 	}
    262 #ifndef __sparc
    263 	else {
    264 		ekernelheap -= HEAPTEXT_SIZE;
    265 		textbase = (uintptr_t)ekernelheap;
    266 	}
    267 #endif
    268 
    269 	heap_size = (uintptr_t)ekernelheap - (uintptr_t)kernelheap;
    270 	heap_arena = vmem_init("heap", kernelheap, heap_size, PAGESIZE,
    271 	    segkmem_alloc, segkmem_free);
    272 
    273 	if (core_size > 0) {
    274 		heap_core_arena = vmem_create("heap_core", core_start,
    275 		    core_size, PAGESIZE, NULL, NULL, NULL, 0, VM_SLEEP);
    276 		heap_core_base = core_start;
    277 	} else {
    278 		heap_core_arena = heap_arena;
    279 		heap_core_base = kernelheap;
    280 	}
    281 
    282 	/*
    283 	 * reserve space for the large page heap. If large pages for kernel
    284 	 * heap is enabled large page heap arean will be created later in the
    285 	 * boot sequence in segkmem_heap_lp_init(). Otherwise the allocated
    286 	 * range will be returned back to the heap_arena.
    287 	 */
    288 	if (heap_lp_size) {
    289 		(void) vmem_xalloc(heap_arena, heap_lp_size, PAGESIZE, 0, 0,
    290 		    heap_lp_base, heap_lp_end,
    291 		    VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
    292 	}
    293 
    294 	/*
    295 	 * Remove the already-spoken-for memory range [kernelheap, first_avail).
    296 	 */
    297 	(void) vmem_xalloc(heap_arena, first_avail - kernelheap, PAGESIZE,
    298 	    0, 0, kernelheap, first_avail, VM_NOSLEEP | VM_BESTFIT | VM_PANIC);
    299 
    300 #ifdef __sparc
    301 	heap32_arena = vmem_create("heap32", (void *)SYSBASE32,
    302 	    SYSLIMIT32 - SYSBASE32 - HEAPTEXT_SIZE, PAGESIZE, NULL,
    303 	    NULL, NULL, 0, VM_SLEEP);
    304 
    305 	textbase = SYSLIMIT32 - HEAPTEXT_SIZE;
    306 	heaptext_parent = NULL;
    307 #else	/* __sparc */
    308 	heap32_arena = heap_core_arena;
    309 	heaptext_parent = heap_core_arena;
    310 #endif	/* __sparc */
    311 
    312 	heaptext_arena = vmem_create("heaptext", (void *)textbase,
    313 	    HEAPTEXT_SIZE, PAGESIZE, NULL, NULL, heaptext_parent, 0, VM_SLEEP);
    314 
    315 	/*
    316 	 * Create a set of arenas for memory with static translations
    317 	 * (e.g. VA -> PA translations cannot change).  Since using
    318 	 * kernel pages by physical address implies it isn't safe to
    319 	 * walk across page boundaries, the static_arena quantum must
    320 	 * be PAGESIZE.  Any kmem caches that require static memory
    321 	 * should source from static_arena, while direct allocations
    322 	 * should only use static_alloc_arena.
    323 	 */
    324 	static_arena = vmem_create("static", NULL, 0, PAGESIZE,
    325 	    segkmem_alloc_permanent, segkmem_free, heap_arena, 0, VM_SLEEP);
    326 	static_alloc_arena = vmem_create("static_alloc", NULL, 0,
    327 	    sizeof (uint64_t), vmem_alloc, vmem_free, static_arena,
    328 	    0, VM_SLEEP);
    329 
    330 	/*
    331 	 * Create an arena for translation data (ptes, hmes, or hblks).
    332 	 * We need an arena for this because hat_memload() is essential
    333 	 * to vmem_populate() (see comments in common/os/vmem.c).
    334 	 *
    335 	 * Note: any kmem cache that allocates from hat_memload_arena
    336 	 * must be created as a KMC_NOHASH cache (i.e. no external slab
    337 	 * and bufctl structures to allocate) so that slab creation doesn't
    338 	 * require anything more than a single vmem_alloc().
    339 	 */
    340 	hat_memload_arena = vmem_create("hat_memload", NULL, 0, PAGESIZE,
    341 	    hat_memload_alloc, segkmem_free, heap_arena, 0,
    342 	    VM_SLEEP | VMC_POPULATOR);
    343 }
    344 
    345 void
    346 boot_mapin(caddr_t addr, size_t size)
    347 {
    348 	caddr_t	 eaddr;
    349 	page_t	*pp;
    350 	pfn_t	 pfnum;
    351 
    352 	if (page_resv(btop(size), KM_NOSLEEP) == 0)
    353 		panic("boot_mapin: page_resv failed");
    354 
    355 	for (eaddr = addr + size; addr < eaddr; addr += PAGESIZE) {
    356 		pfnum = va_to_pfn(addr);
    357 		if (pfnum == PFN_INVALID)
    358 			continue;
    359 		if ((pp = page_numtopp_nolock(pfnum)) == NULL)
    360 			panic("boot_mapin(): No pp for pfnum = %lx", pfnum);
    361 
    362 		/*
    363 		 * must break up any large pages that may have constituent
    364 		 * pages being utilized for BOP_ALLOC()'s before calling
    365 		 * page_numtopp().The locking code (ie. page_reclaim())
    366 		 * can't handle them
    367 		 */
    368 		if (pp->p_szc != 0)
    369 			page_boot_demote(pp);
    370 
    371 		pp = page_numtopp(pfnum, SE_EXCL);
    372 		if (pp == NULL || PP_ISFREE(pp))
    373 			panic("boot_alloc: pp is NULL or free");
    374 
    375 		/*
    376 		 * If the cage is on but doesn't yet contain this page,
    377 		 * mark it as non-relocatable.
    378 		 */
    379 		if (kcage_on && !PP_ISNORELOC(pp))
    380 			PP_SETNORELOC(pp);
    381 
    382 		(void) page_hashin(pp, &kvp, (u_offset_t)(uintptr_t)addr, NULL);
    383 		pp->p_lckcnt = 1;
    384 #if defined(__x86)
    385 		page_downgrade(pp);
    386 #else
    387 		page_unlock(pp);
    388 #endif
    389 	}
    390 }
    391 
    392 /*
    393  * Get pages from boot and hash them into the kernel's vp.
    394  * Used after page structs have been allocated, but before segkmem is ready.
    395  */
    396 void *
    397 boot_alloc(void *inaddr, size_t size, uint_t align)
    398 {
    399 	caddr_t addr = inaddr;
    400 
    401 	if (bootops == NULL)
    402 		prom_panic("boot_alloc: attempt to allocate memory after "
    403 		    "BOP_GONE");
    404 
    405 	size = ptob(btopr(size));
    406 	if (BOP_ALLOC(bootops, addr, size, align) != addr)
    407 		panic("boot_alloc: BOP_ALLOC failed");
    408 	boot_mapin((caddr_t)addr, size);
    409 	return (addr);
    410 }
    411 
    412 static void
    413 segkmem_badop()
    414 {
    415 	panic("segkmem_badop");
    416 }
    417 
    418 #define	SEGKMEM_BADOP(t)	(t(*)())segkmem_badop
    419 
    420 /*ARGSUSED*/
    421 static faultcode_t
    422 segkmem_fault(struct hat *hat, struct seg *seg, caddr_t addr, size_t size,
    423 	enum fault_type type, enum seg_rw rw)
    424 {
    425 	pgcnt_t npages;
    426 	spgcnt_t pg;
    427 	page_t *pp;
    428 	struct vnode *vp = seg->s_data;
    429 
    430 	ASSERT(RW_READ_HELD(&seg->s_as->a_lock));
    431 
    432 	if (seg->s_as != &kas || size > seg->s_size ||
    433 	    addr < seg->s_base || addr + size > seg->s_base + seg->s_size)
    434 		panic("segkmem_fault: bad args");
    435 
    436 	/*
    437 	 * If it is one of segkp pages, call segkp_fault.
    438 	 */
    439 	if (segkp_bitmap && seg == &kvseg &&
    440 	    BT_TEST(segkp_bitmap, btop((uintptr_t)(addr - seg->s_base))))
    441 		return (SEGOP_FAULT(hat, segkp, addr, size, type, rw));
    442 
    443 	if (rw != S_READ && rw != S_WRITE && rw != S_OTHER)
    444 		return (FC_NOSUPPORT);
    445 
    446 	npages = btopr(size);
    447 
    448 	switch (type) {
    449 	case F_SOFTLOCK:	/* lock down already-loaded translations */
    450 		for (pg = 0; pg < npages; pg++) {
    451 			pp = page_lookup(vp, (u_offset_t)(uintptr_t)addr,
    452 			    SE_SHARED);
    453 			if (pp == NULL) {
    454 				/*
    455 				 * Hmm, no page. Does a kernel mapping
    456 				 * exist for it?
    457 				 */
    458 				if (!hat_probe(kas.a_hat, addr)) {
    459 					addr -= PAGESIZE;
    460 					while (--pg >= 0) {
    461 						pp = page_find(vp, (u_offset_t)
    462 						    (uintptr_t)addr);
    463 						if (pp)
    464 							page_unlock(pp);
    465 						addr -= PAGESIZE;
    466 					}
    467 					return (FC_NOMAP);
    468 				}
    469 			}
    470 			addr += PAGESIZE;
    471 		}
    472 		if (rw == S_OTHER)
    473 			hat_reserve(seg->s_as, addr, size);
    474 		return (0);
    475 	case F_SOFTUNLOCK:
    476 		while (npages--) {
    477 			pp = page_find(vp, (u_offset_t)(uintptr_t)addr);
    478 			if (pp)
    479 				page_unlock(pp);
    480 			addr += PAGESIZE;
    481 		}
    482 		return (0);
    483 	default:
    484 		return (FC_NOSUPPORT);
    485 	}
    486 	/*NOTREACHED*/
    487 }
    488 
    489 static int
    490 segkmem_setprot(struct seg *seg, caddr_t addr, size_t size, uint_t prot)
    491 {
    492 	ASSERT(RW_LOCK_HELD(&seg->s_as->a_lock));
    493 
    494 	if (seg->s_as != &kas || size > seg->s_size ||
    495 	    addr < seg->s_base || addr + size > seg->s_base + seg->s_size)
    496 		panic("segkmem_setprot: bad args");
    497 
    498 	/*
    499 	 * If it is one of segkp pages, call segkp.
    500 	 */
    501 	if (segkp_bitmap && seg == &kvseg &&
    502 	    BT_TEST(segkp_bitmap, btop((uintptr_t)(addr - seg->s_base))))
    503 		return (SEGOP_SETPROT(segkp, addr, size, prot));
    504 
    505 	if (prot == 0)
    506 		hat_unload(kas.a_hat, addr, size, HAT_UNLOAD);
    507 	else
    508 		hat_chgprot(kas.a_hat, addr, size, prot);
    509 	return (0);
    510 }
    511 
    512 /*
    513  * This is a dummy segkmem function overloaded to call segkp
    514  * when segkp is under the heap.
    515  */
    516 /* ARGSUSED */
    517 static int
    518 segkmem_checkprot(struct seg *seg, caddr_t addr, size_t size, uint_t prot)
    519 {
    520 	ASSERT(RW_LOCK_HELD(&seg->s_as->a_lock));
    521 
    522 	if (seg->s_as != &kas)
    523 		segkmem_badop();
    524 
    525 	/*
    526 	 * If it is one of segkp pages, call into segkp.
    527 	 */
    528 	if (segkp_bitmap && seg == &kvseg &&
    529 	    BT_TEST(segkp_bitmap, btop((uintptr_t)(addr - seg->s_base))))
    530 		return (SEGOP_CHECKPROT(segkp, addr, size, prot));
    531 
    532 	segkmem_badop();
    533 	return (0);
    534 }
    535 
    536 /*
    537  * This is a dummy segkmem function overloaded to call segkp
    538  * when segkp is under the heap.
    539  */
    540 /* ARGSUSED */
    541 static int
    542 segkmem_kluster(struct seg *seg, caddr_t addr, ssize_t delta)
    543 {
    544 	ASSERT(RW_LOCK_HELD(&seg->s_as->a_lock));
    545 
    546 	if (seg->s_as != &kas)
    547 		segkmem_badop();
    548 
    549 	/*
    550 	 * If it is one of segkp pages, call into segkp.
    551 	 */
    552 	if (segkp_bitmap && seg == &kvseg &&
    553 	    BT_TEST(segkp_bitmap, btop((uintptr_t)(addr - seg->s_base))))
    554 		return (SEGOP_KLUSTER(segkp, addr, delta));
    555 
    556 	segkmem_badop();
    557 	return (0);
    558 }
    559 
    560 static void
    561 segkmem_xdump_range(void *arg, void *start, size_t size)
    562 {
    563 	struct as *as = arg;
    564 	caddr_t addr = start;
    565 	caddr_t addr_end = addr + size;
    566 
    567 	while (addr < addr_end) {
    568 		pfn_t pfn = hat_getpfnum(kas.a_hat, addr);
    569 		if (pfn != PFN_INVALID && pfn <= physmax && pf_is_memory(pfn))
    570 			dump_addpage(as, addr, pfn);
    571 		addr += PAGESIZE;
    572 		dump_timeleft = dump_timeout;
    573 	}
    574 }
    575 
    576 static void
    577 segkmem_dump_range(void *arg, void *start, size_t size)
    578 {
    579 	caddr_t addr = start;
    580 	caddr_t addr_end = addr + size;
    581 
    582 	/*
    583 	 * If we are about to start dumping the range of addresses we
    584 	 * carved out of the kernel heap for the large page heap walk
    585 	 * heap_lp_arena to find what segments are actually populated
    586 	 */
    587 	if (SEGKMEM_USE_LARGEPAGES &&
    588 	    addr == heap_lp_base && addr_end == heap_lp_end &&
    589 	    vmem_size(heap_lp_arena, VMEM_ALLOC) < size) {
    590 		vmem_walk(heap_lp_arena, VMEM_ALLOC | VMEM_REENTRANT,
    591 		    segkmem_xdump_range, arg);
    592 	} else {
    593 		segkmem_xdump_range(arg, start, size);
    594 	}
    595 }
    596 
    597 static void
    598 segkmem_dump(struct seg *seg)
    599 {
    600 	/*
    601 	 * The kernel's heap_arena (represented by kvseg) is a very large
    602 	 * VA space, most of which is typically unused.  To speed up dumping
    603 	 * we use vmem_walk() to quickly find the pieces of heap_arena that
    604 	 * are actually in use.  We do the same for heap32_arena and
    605 	 * heap_core.
    606 	 *
    607 	 * We specify VMEM_REENTRANT to vmem_walk() because dump_addpage()
    608 	 * may ultimately need to allocate memory.  Reentrant walks are
    609 	 * necessarily imperfect snapshots.  The kernel heap continues
    610 	 * to change during a live crash dump, for example.  For a normal
    611 	 * crash dump, however, we know that there won't be any other threads
    612 	 * messing with the heap.  Therefore, at worst, we may fail to dump
    613 	 * the pages that get allocated by the act of dumping; but we will
    614 	 * always dump every page that was allocated when the walk began.
    615 	 *
    616 	 * The other segkmem segments are dense (fully populated), so there's
    617 	 * no need to use this technique when dumping them.
    618 	 *
    619 	 * Note: when adding special dump handling for any new sparsely-
    620 	 * populated segments, be sure to add similar handling to the ::kgrep
    621 	 * code in mdb.
    622 	 */
    623 	if (seg == &kvseg) {
    624 		vmem_walk(heap_arena, VMEM_ALLOC | VMEM_REENTRANT,
    625 		    segkmem_dump_range, seg->s_as);
    626 #ifndef __sparc
    627 		vmem_walk(heaptext_arena, VMEM_ALLOC | VMEM_REENTRANT,
    628 		    segkmem_dump_range, seg->s_as);
    629 #endif
    630 	} else if (seg == &kvseg_core) {
    631 		vmem_walk(heap_core_arena, VMEM_ALLOC | VMEM_REENTRANT,
    632 		    segkmem_dump_range, seg->s_as);
    633 	} else if (seg == &kvseg32) {
    634 		vmem_walk(heap32_arena, VMEM_ALLOC | VMEM_REENTRANT,
    635 		    segkmem_dump_range, seg->s_as);
    636 		vmem_walk(heaptext_arena, VMEM_ALLOC | VMEM_REENTRANT,
    637 		    segkmem_dump_range, seg->s_as);
    638 	} else if (seg == &kzioseg) {
    639 		/*
    640 		 * We don't want to dump pages attached to kzioseg since they
    641 		 * contain file data from ZFS.  If this page's segment is
    642 		 * kzioseg return instead of writing it to the dump device.
    643 		 */
    644 		return;
    645 	} else {
    646 		segkmem_dump_range(seg->s_as, seg->s_base, seg->s_size);
    647 	}
    648 }
    649 
    650 /*
    651  * lock/unlock kmem pages over a given range [addr, addr+len).
    652  * Returns a shadow list of pages in ppp. If there are holes
    653  * in the range (e.g. some of the kernel mappings do not have
    654  * underlying page_ts) returns ENOTSUP so that as_pagelock()
    655  * will handle the range via as_fault(F_SOFTLOCK).
    656  */
    657 /*ARGSUSED*/
    658 static int
    659 segkmem_pagelock(struct seg *seg, caddr_t addr, size_t len,
    660 	page_t ***ppp, enum lock_type type, enum seg_rw rw)
    661 {
    662 	page_t **pplist, *pp;
    663 	pgcnt_t npages;
    664 	spgcnt_t pg;
    665 	size_t nb;
    666 	struct vnode *vp = seg->s_data;
    667 
    668 	ASSERT(ppp != NULL);
    669 
    670 	/*
    671 	 * If it is one of segkp pages, call into segkp.
    672 	 */
    673 	if (segkp_bitmap && seg == &kvseg &&
    674 	    BT_TEST(segkp_bitmap, btop((uintptr_t)(addr - seg->s_base))))
    675 		return (SEGOP_PAGELOCK(segkp, addr, len, ppp, type, rw));
    676 
    677 	if (type == L_PAGERECLAIM)
    678 		return (ENOTSUP);
    679 
    680 	npages = btopr(len);
    681 	nb = sizeof (page_t *) * npages;
    682 
    683 	if (type == L_PAGEUNLOCK) {
    684 		pplist = *ppp;
    685 		ASSERT(pplist != NULL);
    686 
    687 		for (pg = 0; pg < npages; pg++) {
    688 			pp = pplist[pg];
    689 			page_unlock(pp);
    690 		}
    691 		kmem_free(pplist, nb);
    692 		return (0);
    693 	}
    694 
    695 	ASSERT(type == L_PAGELOCK);
    696 
    697 	pplist = kmem_alloc(nb, KM_NOSLEEP);
    698 	if (pplist == NULL) {
    699 		*ppp = NULL;
    700 		return (ENOTSUP);	/* take the slow path */
    701 	}
    702 
    703 	for (pg = 0; pg < npages; pg++) {
    704 		pp = page_lookup(vp, (u_offset_t)(uintptr_t)addr, SE_SHARED);
    705 		if (pp == NULL) {
    706 			while (--pg >= 0)
    707 				page_unlock(pplist[pg]);
    708 			kmem_free(pplist, nb);
    709 			*ppp = NULL;
    710 			return (ENOTSUP);
    711 		}
    712 		pplist[pg] = pp;
    713 		addr += PAGESIZE;
    714 	}
    715 
    716 	*ppp = pplist;
    717 	return (0);
    718 }
    719 
    720 /*
    721  * This is a dummy segkmem function overloaded to call segkp
    722  * when segkp is under the heap.
    723  */
    724 /* ARGSUSED */
    725 static int
    726 segkmem_getmemid(struct seg *seg, caddr_t addr, memid_t *memidp)
    727 {
    728 	ASSERT(RW_LOCK_HELD(&seg->s_as->a_lock));
    729 
    730 	if (seg->s_as != &kas)
    731 		segkmem_badop();
    732 
    733 	/*
    734 	 * If it is one of segkp pages, call into segkp.
    735 	 */
    736 	if (segkp_bitmap && seg == &kvseg &&
    737 	    BT_TEST(segkp_bitmap, btop((uintptr_t)(addr - seg->s_base))))
    738 		return (SEGOP_GETMEMID(segkp, addr, memidp));
    739 
    740 	segkmem_badop();
    741 	return (0);
    742 }
    743 
    744 /*ARGSUSED*/
    745 static lgrp_mem_policy_info_t *
    746 segkmem_getpolicy(struct seg *seg, caddr_t addr)
    747 {
    748 	return (NULL);
    749 }
    750 
    751 /*ARGSUSED*/
    752 static int
    753 segkmem_capable(struct seg *seg, segcapability_t capability)
    754 {
    755 	if (capability == S_CAPABILITY_NOMINFLT)
    756 		return (1);
    757 	return (0);
    758 }
    759 
    760 static struct seg_ops segkmem_ops = {
    761 	SEGKMEM_BADOP(int),		/* dup */
    762 	SEGKMEM_BADOP(int),		/* unmap */
    763 	SEGKMEM_BADOP(void),		/* free */
    764 	segkmem_fault,
    765 	SEGKMEM_BADOP(faultcode_t),	/* faulta */
    766 	segkmem_setprot,
    767 	segkmem_checkprot,
    768 	segkmem_kluster,
    769 	SEGKMEM_BADOP(size_t),		/* swapout */
    770 	SEGKMEM_BADOP(int),		/* sync */
    771 	SEGKMEM_BADOP(size_t),		/* incore */
    772 	SEGKMEM_BADOP(int),		/* lockop */
    773 	SEGKMEM_BADOP(int),		/* getprot */
    774 	SEGKMEM_BADOP(u_offset_t),	/* getoffset */
    775 	SEGKMEM_BADOP(int),		/* gettype */
    776 	SEGKMEM_BADOP(int),		/* getvp */
    777 	SEGKMEM_BADOP(int),		/* advise */
    778 	segkmem_dump,
    779 	segkmem_pagelock,
    780 	SEGKMEM_BADOP(int),		/* setpgsz */
    781 	segkmem_getmemid,
    782 	segkmem_getpolicy,		/* getpolicy */
    783 	segkmem_capable,		/* capable */
    784 };
    785 
    786 int
    787 segkmem_zio_create(struct seg *seg)
    788 {
    789 	ASSERT(seg->s_as == &kas && RW_WRITE_HELD(&kas.a_lock));
    790 	seg->s_ops = &segkmem_ops;
    791 	seg->s_data = &zvp;
    792 	kas.a_size += seg->s_size;
    793 	return (0);
    794 }
    795 
    796 int
    797 segkmem_create(struct seg *seg)
    798 {
    799 	ASSERT(seg->s_as == &kas && RW_WRITE_HELD(&kas.a_lock));
    800 	seg->s_ops = &segkmem_ops;
    801 	seg->s_data = &kvp;
    802 	kas.a_size += seg->s_size;
    803 	return (0);
    804 }
    805 
    806 /*ARGSUSED*/
    807 page_t *
    808 segkmem_page_create(void *addr, size_t size, int vmflag, void *arg)
    809 {
    810 	struct seg kseg;
    811 	int pgflags;
    812 	struct vnode *vp = arg;
    813 
    814 	if (vp == NULL)
    815 		vp = &kvp;
    816 
    817 	kseg.s_as = &kas;
    818 	pgflags = PG_EXCL;
    819 
    820 	if (segkmem_reloc == 0 || (vmflag & VM_NORELOC))
    821 		pgflags |= PG_NORELOC;
    822 	if ((vmflag & VM_NOSLEEP) == 0)
    823 		pgflags |= PG_WAIT;
    824 	if (vmflag & VM_PANIC)
    825 		pgflags |= PG_PANIC;
    826 	if (vmflag & VM_PUSHPAGE)
    827 		pgflags |= PG_PUSHPAGE;
    828 
    829 	return (page_create_va(vp, (u_offset_t)(uintptr_t)addr, size,
    830 	    pgflags, &kseg, addr));
    831 }
    832 
    833 /*
    834  * Allocate pages to back the virtual address range [addr, addr + size).
    835  * If addr is NULL, allocate the virtual address space as well.
    836  */
    837 void *
    838 segkmem_xalloc(vmem_t *vmp, void *inaddr, size_t size, int vmflag, uint_t attr,
    839 	page_t *(*page_create_func)(void *, size_t, int, void *), void *pcarg)
    840 {
    841 	page_t *ppl;
    842 	caddr_t addr = inaddr;
    843 	pgcnt_t npages = btopr(size);
    844 	int allocflag;
    845 
    846 	if (inaddr == NULL && (addr = vmem_alloc(vmp, size, vmflag)) == NULL)
    847 		return (NULL);
    848 
    849 	ASSERT(((uintptr_t)addr & PAGEOFFSET) == 0);
    850 
    851 	if (page_resv(npages, vmflag & VM_KMFLAGS) == 0) {
    852 		if (inaddr == NULL)
    853 			vmem_free(vmp, addr, size);
    854 		return (NULL);
    855 	}
    856 
    857 	ppl = page_create_func(addr, size, vmflag, pcarg);
    858 	if (ppl == NULL) {
    859 		if (inaddr == NULL)
    860 			vmem_free(vmp, addr, size);
    861 		page_unresv(npages);
    862 		return (NULL);
    863 	}
    864 
    865 	/*
    866 	 * Under certain conditions, we need to let the HAT layer know
    867 	 * that it cannot safely allocate memory.  Allocations from
    868 	 * the hat_memload vmem arena always need this, to prevent
    869 	 * infinite recursion.
    870 	 *
    871 	 * In addition, the x86 hat cannot safely do memory
    872 	 * allocations while in vmem_populate(), because there
    873 	 * is no simple bound on its usage.
    874 	 */
    875 	if (vmflag & VM_MEMLOAD)
    876 		allocflag = HAT_NO_KALLOC;
    877 #if defined(__x86)
    878 	else if (vmem_is_populator())
    879 		allocflag = HAT_NO_KALLOC;
    880 #endif
    881 	else
    882 		allocflag = 0;
    883 
    884 	while (ppl != NULL) {
    885 		page_t *pp = ppl;
    886 		page_sub(&ppl, pp);
    887 		ASSERT(page_iolock_assert(pp));
    888 		ASSERT(PAGE_EXCL(pp));
    889 		page_io_unlock(pp);
    890 		hat_memload(kas.a_hat, (caddr_t)(uintptr_t)pp->p_offset, pp,
    891 		    (PROT_ALL & ~PROT_USER) | HAT_NOSYNC | attr,
    892 		    HAT_LOAD_LOCK | allocflag);
    893 		pp->p_lckcnt = 1;
    894 #if defined(__x86)
    895 		page_downgrade(pp);
    896 #else
    897 		if (vmflag & SEGKMEM_SHARELOCKED)
    898 			page_downgrade(pp);
    899 		else
    900 			page_unlock(pp);
    901 #endif
    902 	}
    903 
    904 	return (addr);
    905 }
    906 
    907 static void *
    908 segkmem_alloc_vn(vmem_t *vmp, size_t size, int vmflag, struct vnode *vp)
    909 {
    910 	void *addr;
    911 	segkmem_gc_list_t *gcp, **prev_gcpp;
    912 
    913 	ASSERT(vp != NULL);
    914 
    915 	if (kvseg.s_base == NULL) {
    916 #ifndef __sparc
    917 		if (bootops->bsys_alloc == NULL)
    918 			halt("Memory allocation between bop_alloc() and "
    919 			    "kmem_alloc().\n");
    920 #endif
    921 
    922 		/*
    923 		 * There's not a lot of memory to go around during boot,
    924 		 * so recycle it if we can.
    925 		 */
    926 		for (prev_gcpp = &segkmem_gc_list; (gcp = *prev_gcpp) != NULL;
    927 		    prev_gcpp = &gcp->gc_next) {
    928 			if (gcp->gc_arena == vmp && gcp->gc_size == size) {
    929 				*prev_gcpp = gcp->gc_next;
    930 				return (gcp);
    931 			}
    932 		}
    933 
    934 		addr = vmem_alloc(vmp, size, vmflag | VM_PANIC);
    935 		if (boot_alloc(addr, size, BO_NO_ALIGN) != addr)
    936 			panic("segkmem_alloc: boot_alloc failed");
    937 		return (addr);
    938 	}
    939 	return (segkmem_xalloc(vmp, NULL, size, vmflag, 0,
    940 	    segkmem_page_create, vp));
    941 }
    942 
    943 void *
    944 segkmem_alloc(vmem_t *vmp, size_t size, int vmflag)
    945 {
    946 	return (segkmem_alloc_vn(vmp, size, vmflag, &kvp));
    947 }
    948 
    949 void *
    950 segkmem_zio_alloc(vmem_t *vmp, size_t size, int vmflag)
    951 {
    952 	return (segkmem_alloc_vn(vmp, size, vmflag, &zvp));
    953 }
    954 
    955 /*
    956  * Any changes to this routine must also be carried over to
    957  * devmap_free_pages() in the seg_dev driver. This is because
    958  * we currently don't have a special kernel segment for non-paged
    959  * kernel memory that is exported by drivers to user space.
    960  */
    961 static void
    962 segkmem_free_vn(vmem_t *vmp, void *inaddr, size_t size, struct vnode *vp,
    963     void (*func)(page_t *))
    964 {
    965 	page_t *pp;
    966 	caddr_t addr = inaddr;
    967 	caddr_t eaddr;
    968 	pgcnt_t npages = btopr(size);
    969 
    970 	ASSERT(((uintptr_t)addr & PAGEOFFSET) == 0);
    971 	ASSERT(vp != NULL);
    972 
    973 	if (kvseg.s_base == NULL) {
    974 		segkmem_gc_list_t *gc = inaddr;
    975 		gc->gc_arena = vmp;
    976 		gc->gc_size = size;
    977 		gc->gc_next = segkmem_gc_list;
    978 		segkmem_gc_list = gc;
    979 		return;
    980 	}
    981 
    982 	hat_unload(kas.a_hat, addr, size, HAT_UNLOAD_UNLOCK);
    983 
    984 	for (eaddr = addr + size; addr < eaddr; addr += PAGESIZE) {
    985 #if defined(__x86)
    986 		pp = page_find(vp, (u_offset_t)(uintptr_t)addr);
    987 		if (pp == NULL)
    988 			panic("segkmem_free: page not found");
    989 		if (!page_tryupgrade(pp)) {
    990 			/*
    991 			 * Some other thread has a sharelock. Wait for
    992 			 * it to drop the lock so we can free this page.
    993 			 */
    994 			page_unlock(pp);
    995 			pp = page_lookup(vp, (u_offset_t)(uintptr_t)addr,
    996 			    SE_EXCL);
    997 		}
    998 #else
    999 		pp = page_lookup(vp, (u_offset_t)(uintptr_t)addr, SE_EXCL);
   1000 #endif
   1001 		if (pp == NULL)
   1002 			panic("segkmem_free: page not found");
   1003 		/* Clear p_lckcnt so page_destroy() doesn't update availrmem */
   1004 		pp->p_lckcnt = 0;
   1005 		if (func)
   1006 			func(pp);
   1007 		else
   1008 			page_destroy(pp, 0);
   1009 	}
   1010 	if (func == NULL)
   1011 		page_unresv(npages);
   1012 
   1013 	if (vmp != NULL)
   1014 		vmem_free(vmp, inaddr, size);
   1015 
   1016 }
   1017 
   1018 void
   1019 segkmem_xfree(vmem_t *vmp, void *inaddr, size_t size, void (*func)(page_t *))
   1020 {
   1021 	segkmem_free_vn(vmp, inaddr, size, &kvp, func);
   1022 }
   1023 
   1024 void
   1025 segkmem_free(vmem_t *vmp, void *inaddr, size_t size)
   1026 {
   1027 	segkmem_free_vn(vmp, inaddr, size, &kvp, NULL);
   1028 }
   1029 
   1030 void
   1031 segkmem_zio_free(vmem_t *vmp, void *inaddr, size_t size)
   1032 {
   1033 	segkmem_free_vn(vmp, inaddr, size, &zvp, NULL);
   1034 }
   1035 
   1036 void
   1037 segkmem_gc(void)
   1038 {
   1039 	ASSERT(kvseg.s_base != NULL);
   1040 	while (segkmem_gc_list != NULL) {
   1041 		segkmem_gc_list_t *gc = segkmem_gc_list;
   1042 		segkmem_gc_list = gc->gc_next;
   1043 		segkmem_free(gc->gc_arena, gc, gc->gc_size);