From: Ben Gras Date: Mon, 12 Apr 2010 11:25:24 +0000 (+0000) Subject: let vm use physically fragmented memory for allocations. X-Git-Tag: v3.1.7~154 X-Git-Url: http://zhaoyanbai.com/repos/%22/xml/v3/zones/static/dnssec-keyfromlabel.html?a=commitdiff_plain;h=c78250332d821120f6e832644ffa51eed86cab95;p=minix.git let vm use physically fragmented memory for allocations. map_copy_ph_block is replaced by map_clone_ph_block, which can replace a single physical block by multiple physical blocks. also, . merge map_mem.c with region.c, as they manipulate the same data structures . NOTRUNNABLE removed as sanity check . use direct functions for ALLOC_MEM and FREE_MEM again . add some checks to shared memory mapping code . fix for data structure integrity when using shared memory . fix sanity checks --- diff --git a/servers/vm/Makefile b/servers/vm/Makefile index e6e435b80..c6d91bad9 100644 --- a/servers/vm/Makefile +++ b/servers/vm/Makefile @@ -4,7 +4,7 @@ PROG= vm SRCS= main.c alloc.c utility.c exec.c exit.c fork.c break.c \ signal.c mmap.c slaballoc.c region.c pagefaults.c addravl.c \ - physravl.c rs.c queryexit.c map_mem.c + physravl.c rs.c queryexit.c DPADD+= ${LIBSYS} LDADD+= -lsys diff --git a/servers/vm/alloc.c b/servers/vm/alloc.c index 9c259a000..8a903ae86 100644 --- a/servers/vm/alloc.c +++ b/servers/vm/alloc.c @@ -41,13 +41,14 @@ #include "pagerange.h" #include "addravl.h" #include "sanitycheck.h" +#include "memlist.h" /* AVL tree of free pages. */ addr_avl addravl; /* Used for sanity check. */ PRIVATE phys_bytes mem_low, mem_high; -#define vm_assert_range(addr, len) \ +#define assert_range(addr, len) \ vm_assert((addr) >= mem_low); \ vm_assert((addr) + (len) - 1 <= mem_high); @@ -73,13 +74,15 @@ PRIVATE struct hole *free_slots;/* ptr to list of unused table slots */ FORWARD _PROTOTYPE( void del_slot, (struct hole *prev_ptr, struct hole *hp) ); FORWARD _PROTOTYPE( void merge, (struct hole *hp) ); FORWARD _PROTOTYPE( void free_pages, (phys_bytes addr, int pages) ); -FORWARD _PROTOTYPE( phys_bytes alloc_pages, (int pages, int flags) ); +FORWARD _PROTOTYPE( phys_bytes alloc_pages, (int pages, int flags, + phys_bytes *ret)); #if SANITYCHECKS FORWARD _PROTOTYPE( void holes_sanity_f, (char *fn, int line) ); #define CHECKHOLES holes_sanity_f(__FILE__, __LINE__) -#define MAXPAGES (1024*1024*1024/VM_PAGE_SIZE) /* 1GB of memory */ +#define PAGESPERGB (1024*1024*1024/VM_PAGE_SIZE) /* 1GB of memory */ +#define MAXPAGES (2*PAGESPERGB) #define CHUNKS BITMAP_CHUNKS(MAXPAGES) PRIVATE bitchunk_t pagemap[CHUNKS]; @@ -87,32 +90,6 @@ PRIVATE bitchunk_t pagemap[CHUNKS]; #define CHECKHOLES #endif -/* Sanity check for parameters of node p. */ -#define vm_assert_params(p, bytes, next) { \ - vm_assert((p) != NO_MEM); \ - vm_assert(!((bytes) % VM_PAGE_SIZE)); \ - vm_assert(!((next) % VM_PAGE_SIZE)); \ - vm_assert((bytes) > 0); \ - vm_assert((p) + (bytes) > (p)); \ - vm_assert((next) == NO_MEM || ((p) + (bytes) <= (next))); \ - vm_assert_range((p), (bytes)); \ - vm_assert_range((next), 1); \ -} - -/* Retrieve size of free block and pointer to next block from physical - * address (page) p. - */ -#define GET_PARAMS(p, bytes, next) { \ - phys_readaddr((p), &(bytes), &(next)); \ - vm_assert_params((p), (bytes), (next)); \ -} - -/* Write parameters to physical page p. */ -#define SET_PARAMS(p, bytes, next) { \ - vm_assert_params((p), (bytes), (next)); \ - phys_writeaddr((p), (bytes), (next)); \ -} - #if SANITYCHECKS @@ -127,7 +104,7 @@ int line; if(!(c)) { \ printf("holes_sanity_f:%s:%d: %s failed\n", file, line, #c); \ util_stacktrace(); \ - panic("assert failed"); } \ + panic("vm_assert failed"); } \ } int h, c = 0, n = 0; @@ -187,9 +164,9 @@ int line; #endif /*===========================================================================* - * alloc_mem_f * + * alloc_mem * *===========================================================================*/ -PUBLIC phys_clicks alloc_mem_f(phys_clicks clicks, u32_t memflags) +PUBLIC phys_clicks alloc_mem(phys_clicks clicks, u32_t memflags) { /* Allocate a block of memory from the free list using first fit. The block * consists of a sequence of contiguous bytes, whose length in clicks is @@ -208,7 +185,7 @@ PUBLIC phys_clicks alloc_mem_f(phys_clicks clicks, u32_t memflags) if(vm_paged) { vm_assert(CLICK_SIZE == VM_PAGE_SIZE); - mem = alloc_pages(clicks, memflags); + mem = alloc_pages(clicks, memflags, NULL); } else { CHECKHOLES; prev_ptr = NIL_HOLE; @@ -253,7 +230,7 @@ CHECKHOLES; if(o > 0) { phys_clicks e; e = align_clicks - o; - FREE_MEM(mem, e); + free_mem(mem, e); mem += e; } } @@ -263,9 +240,9 @@ CHECKHOLES; } /*===========================================================================* - * free_mem_f * + * free_mem * *===========================================================================*/ -PUBLIC void free_mem_f(phys_clicks base, phys_clicks clicks) +PUBLIC void free_mem(phys_clicks base, phys_clicks clicks) { /* Return a block of free memory to the hole list. The parameters tell where * the block starts in physical memory and how big it is. The block is added @@ -414,7 +391,7 @@ struct memory *chunks; /* list of free memory chunks */ to = CLICK2ABS(chunks[i].base+chunks[i].size)-1; if(first || from < mem_low) mem_low = from; if(first || to > mem_high) mem_high = to; - FREE_MEM(chunks[i].base, chunks[i].size); + free_mem(chunks[i].base, chunks[i].size); total_pages += chunks[i].size; first = 0; } @@ -465,7 +442,7 @@ PUBLIC void memstats(int *nodes, int *pages, int *largest) /*===========================================================================* * alloc_pages * *===========================================================================*/ -PRIVATE PUBLIC phys_bytes alloc_pages(int pages, int memflags) +PRIVATE PUBLIC phys_bytes alloc_pages(int pages, int memflags, phys_bytes *len) { addr_iter iter; pagerange_t *pr; @@ -494,7 +471,8 @@ PRIVATE PUBLIC phys_bytes alloc_pages(int pages, int memflags) while((pr = addr_get_iter(&iter))) { SLABSANE(pr); - if(pr->size >= pages) { + vm_assert(pr->size > 0); + if(pr->size >= pages || (memflags & PAF_FIRSTBLOCK)) { if(memflags & PAF_LOWER16MB) { if(pr->addr + pages > boundary16) return NO_MEM; @@ -518,6 +496,8 @@ PRIVATE PUBLIC phys_bytes alloc_pages(int pages, int memflags) printf("VM: alloc_pages: alloc failed of %d pages\n", pages); util_stacktrace(); printmemstats(); + if(len) + *len = 0; #if SANITYCHECKS if(largest >= pages) { panic("no memory but largest was enough"); @@ -528,6 +508,22 @@ PRIVATE PUBLIC phys_bytes alloc_pages(int pages, int memflags) SLABSANE(pr); + if(memflags & PAF_FIRSTBLOCK) { + vm_assert(len); + /* block doesn't have to as big as requested; + * return its size though. + */ + if(pr->size < pages) { + pages = pr->size; +#if SANITYCHECKS + wantpages = firstpages - pages; +#endif + } + } + + if(len) + *len = pages; + /* Allocated chunk is off the end. */ mem = pr->addr + pr->size - pages; @@ -556,6 +552,10 @@ PRIVATE PUBLIC phys_bytes alloc_pages(int pages, int memflags) memstats(&finalnodes, &finalpages, &largest); sanitycheck(); + if(finalpages != wantpages) { + printf("pages start: %d req: %d final: %d\n", + firstpages, pages, finalpages); + } vm_assert(finalnodes == wantnodes); vm_assert(finalpages == wantpages); #endif @@ -754,7 +754,7 @@ PUBLIC int do_deldma(message *msg) if (j >= NR_DMA) { /* Last segment */ - FREE_MEM(dmatab[i].dt_seg_base, + free_mem(dmatab[i].dt_seg_base, dmatab[i].dt_seg_size); } } @@ -822,7 +822,7 @@ PUBLIC void release_dma(struct vmproc *vmp) } if (!found_one) - FREE_MEM(base, size); + free_mem(base, size); msg->VMRD_FOUND = found_one; #endif @@ -867,7 +867,7 @@ int usedpages_add_f(phys_bytes addr, phys_bytes len, char *file, int line) vm_assert(!(addr % VM_PAGE_SIZE)); vm_assert(!(len % VM_PAGE_SIZE)); vm_assert(len > 0); - vm_assert_range(addr, len); + assert_range(addr, len); pagestart = addr / VM_PAGE_SIZE; pages = len / VM_PAGE_SIZE; @@ -892,3 +892,91 @@ int usedpages_add_f(phys_bytes addr, phys_bytes len, char *file, int line) } #endif + +/*===========================================================================* + * alloc_mem_in_list * + *===========================================================================*/ +struct memlist *alloc_mem_in_list(phys_bytes bytes, u32_t flags) +{ + phys_bytes rempages; + struct memlist *head = NULL, *ml; + + vm_assert(bytes > 0); + vm_assert(!(bytes % VM_PAGE_SIZE)); + + rempages = bytes / VM_PAGE_SIZE; + + /* unless we are told to allocate all memory + * contiguously, tell alloc function to grab whatever + * block it can find. + */ + if(!(flags & PAF_CONTIG)) + flags |= PAF_FIRSTBLOCK; + + do { + struct memlist *ml; + phys_bytes mem, gotpages; + mem = alloc_pages(rempages, flags, &gotpages); + + if(mem == NO_MEM) { + free_mem_list(head, 1); + return NULL; + } + + vm_assert(gotpages <= rempages); + vm_assert(gotpages > 0); + + if(!(SLABALLOC(ml))) { + free_mem_list(head, 1); + free_pages(mem, gotpages); + return NULL; + } + + USE(ml, + ml->phys = CLICK2ABS(mem); + ml->length = CLICK2ABS(gotpages); + ml->next = head;); + head = ml; + rempages -= gotpages; + } while(rempages > 0); + + for(ml = head; ml; ml = ml->next) { + vm_assert(ml->phys); + vm_assert(ml->length); + } + + return head; +} + +/*===========================================================================* + * free_mem_list * + *===========================================================================*/ +void free_mem_list(struct memlist *list, int all) +{ + while(list) { + struct memlist *next; + next = list->next; + vm_assert(!(list->phys % VM_PAGE_SIZE)); + vm_assert(!(list->length % VM_PAGE_SIZE)); + if(all) + free_pages(list->phys / VM_PAGE_SIZE, + list->length / VM_PAGE_SIZE); + SLABFREE(list); + list = next; + } +} + +/*===========================================================================* + * print_mem_list * + *===========================================================================*/ +void print_mem_list(struct memlist *list) +{ + while(list) { + vm_assert(list->length > 0); + printf("0x%lx-0x%lx", list->phys, list->phys+list->length-1); + printf(" "); + list = list->next; + } + printf("\n"); +} + diff --git a/servers/vm/arch/i386/pagetable.c b/servers/vm/arch/i386/pagetable.c index 66e28c520..5946b8986 100644 --- a/servers/vm/arch/i386/pagetable.c +++ b/servers/vm/arch/i386/pagetable.c @@ -17,6 +17,7 @@ #include #include #include +#include #include #include @@ -45,7 +46,7 @@ PRIVATE int proc_pde = 0; PRIVATE int bigpage_ok = 0; /* Our process table entry. */ -struct vmproc *vmp = &vmproc[VM_PROC_NR]; +struct vmproc *vmprocess = &vmproc[VM_PROC_NR]; /* Spare memory, ready to go after initialization, to avoid a * circular dependency on allocating memory and writing it into VM's @@ -114,6 +115,8 @@ PUBLIC void pt_sanitycheck(pt_t *pt, char *file, int line) for(i = proc_pde; i < I386_VM_DIR_ENTRIES; i++) { if(pt->pt_pt[i]) { + int pte; + MYASSERT(vm_addrok(pt->pt_pt[i], 1)); if(!(pt->pt_dir[i] & I386_VM_PRESENT)) { printf("slot %d: pt->pt_pt[%d] = 0x%lx, but pt_dir entry 0x%lx\n", slot, i, pt->pt_pt[i], pt->pt_dir[i]); @@ -208,17 +211,26 @@ PRIVATE u32_t findhole(pt_t *pt, u32_t vmin, u32_t vmax) PRIVATE void vm_freepages(vir_bytes vir, vir_bytes phys, int pages, int reason) { vm_assert(reason >= 0 && reason < VMP_CATEGORIES); - if(vir >= vmp->vm_stacktop) { + if(vir >= vmprocess->vm_stacktop) { vm_assert(!(vir % I386_PAGE_SIZE)); vm_assert(!(phys % I386_PAGE_SIZE)); - FREE_MEM(ABS2CLICK(phys), pages); - if(pt_writemap(&vmp->vm_pt, arch_vir2map(vmp, vir), + free_mem(ABS2CLICK(phys), pages); + if(pt_writemap(&vmprocess->vm_pt, arch_vir2map(vmprocess, vir), MAP_NONE, pages*I386_PAGE_SIZE, 0, WMF_OVERWRITE) != OK) panic("vm_freepages: pt_writemap failed"); } else { printf("VM: vm_freepages not freeing VM heap pages (%d)\n", pages); } + +#if SANITYCHECKS + /* If SANITYCHECKS are on, flush tlb so accessing freed pages is + * always trapped, also if not in tlb. + */ + if((sys_vmctl(SELF, VMCTL_FLUSHTLB, 0)) != OK) { + panic("VMCTL_FLUSHTLB failed"); + } +#endif } /*===========================================================================* @@ -281,7 +293,7 @@ PUBLIC void *vm_allocpage(phys_bytes *phys, int reason) static int level = 0; void *ret; - pt = &vmp->vm_pt; + pt = &vmprocess->vm_pt; vm_assert(reason >= 0 && reason < VMP_CATEGORIES); level++; @@ -289,7 +301,7 @@ PUBLIC void *vm_allocpage(phys_bytes *phys, int reason) vm_assert(level >= 1); vm_assert(level <= 2); - if(level > 1 || !(vmp->vm_flags & VMF_HASPT) || !meminit_done) { + if(level > 1 || !(vmprocess->vm_flags & VMF_HASPT) || !meminit_done) { int r; void *s; s=vm_getsparepage(phys); @@ -304,8 +316,8 @@ PUBLIC void *vm_allocpage(phys_bytes *phys, int reason) /* VM does have a pagetable, so get a page and map it in there. * Where in our virtual address space can we put it? */ - loc = findhole(pt, arch_vir2map(vmp, vmp->vm_stacktop), - vmp->vm_arch.vm_data_top); + loc = findhole(pt, arch_vir2map(vmprocess, vmprocess->vm_stacktop), + vmprocess->vm_arch.vm_data_top); if(loc == NO_MEM) { level--; printf("VM: vm_allocpage: findhole failed\n"); @@ -315,9 +327,9 @@ PUBLIC void *vm_allocpage(phys_bytes *phys, int reason) /* Allocate page of memory for use by VM. As VM * is trusted, we don't have to pre-clear it. */ - if((newpage = ALLOC_MEM(CLICKSPERPAGE, 0)) == NO_MEM) { + if((newpage = alloc_mem(CLICKSPERPAGE, 0)) == NO_MEM) { level--; - printf("VM: vm_allocpage: ALLOC_MEM failed\n"); + printf("VM: vm_allocpage: alloc_mem failed\n"); return NULL; } @@ -326,7 +338,7 @@ PUBLIC void *vm_allocpage(phys_bytes *phys, int reason) /* Map this page into our address space. */ if((r=pt_writemap(pt, loc, *phys, I386_PAGE_SIZE, I386_VM_PRESENT | I386_VM_USER | I386_VM_WRITE, 0)) != OK) { - FREE_MEM(newpage, CLICKSPERPAGE); + free_mem(newpage, CLICKSPERPAGE); printf("vm_allocpage writemap failed\n"); level--; return NULL; @@ -339,7 +351,7 @@ PUBLIC void *vm_allocpage(phys_bytes *phys, int reason) level--; /* Return user-space-ready pointer to it. */ - ret = (void *) arch_map2vir(vmp, loc); + ret = (void *) arch_map2vir(vmprocess, loc); return ret; } @@ -355,8 +367,8 @@ PUBLIC void vm_pagelock(void *vir, int lockflag) u32_t flags = I386_VM_PRESENT | I386_VM_USER; pt_t *pt; - pt = &vmp->vm_pt; - m = arch_vir2map(vmp, (vir_bytes) vir); + pt = &vmprocess->vm_pt; + m = arch_vir2map(vmprocess, (vir_bytes) vir); vm_assert(!(m % I386_PAGE_SIZE)); @@ -376,6 +388,51 @@ PUBLIC void vm_pagelock(void *vir, int lockflag) return; } +/*===========================================================================* + * vm_addrok * + *===========================================================================*/ +PUBLIC int vm_addrok(void *vir, int writeflag) +{ +/* Mark a page allocated by vm_allocpage() unwritable, i.e. only for VM. */ + pt_t *pt = &vmprocess->vm_pt; + int pde, pte; + vir_bytes v = arch_vir2map(vmprocess, (vir_bytes) vir); + + /* No PT yet? Don't bother looking. */ + if(!(vmprocess->vm_flags & VMF_HASPT)) { + return 1; + } + + pde = I386_VM_PDE(v); + pte = I386_VM_PTE(v); + + if(!(pt->pt_dir[pde] & I386_VM_PRESENT)) { + printf("addr not ok: missing pde %d\n", pde); + return 0; + } + + if(writeflag && + !(pt->pt_dir[pde] & I386_VM_WRITE)) { + printf("addr not ok: pde %d present but pde unwritable\n", pde); + return 0; + } + + if(!(pt->pt_pt[pde][pte] & I386_VM_PRESENT)) { + printf("addr not ok: missing pde %d / pte %d\n", + pde, pte); + return 0; + } + + if(writeflag && + !(pt->pt_pt[pde][pte] & I386_VM_WRITE)) { + printf("addr not ok: pde %d / pte %d present but unwritable\n", + pde, pte); + return 0; + } + + return 1; +} + /*===========================================================================* * pt_ptalloc * *===========================================================================*/ @@ -412,6 +469,32 @@ PRIVATE int pt_ptalloc(pt_t *pt, int pde, u32_t flags) return OK; } +PRIVATE char *ptestr(u32_t pte) +{ +#define FLAG(constant, name) { \ + if(pte & (constant)) { strcat(str, name); strcat(str, " "); } \ +} + + static char str[30]; + if(!(pte & I386_VM_PRESENT)) { + return "not present"; + } + str[0] = '\0'; + FLAG(I386_VM_WRITE, "W"); + FLAG(I386_VM_USER, "U"); + FLAG(I386_VM_PWT, "PWT"); + FLAG(I386_VM_PCD, "PCD"); + FLAG(I386_VM_ACC, "ACC"); + FLAG(I386_VM_DIRTY, "DIRTY"); + FLAG(I386_VM_PS, "PS"); + FLAG(I386_VM_GLOBAL, "G"); + FLAG(I386_VM_PTAVAIL1, "AV1"); + FLAG(I386_VM_PTAVAIL2, "AV2"); + FLAG(I386_VM_PTAVAIL3, "AV3"); + + return str; +} + /*===========================================================================* * pt_writemap * *===========================================================================*/ @@ -436,14 +519,8 @@ PUBLIC int pt_writemap(pt_t *pt, vir_bytes v, phys_bytes physaddr, * what's actually written into the PTE if I386_VM_PRESENT * isn't on, so we can just write MAP_NONE into it. */ -#if SANITYCHECKS - if(physaddr != MAP_NONE && !(flags & I386_VM_PRESENT)) { - panic("pt_writemap: writing dir with !P"); - } - if(physaddr == MAP_NONE && flags) { - panic("pt_writemap: writing 0 with flags"); - } -#endif + vm_assert(physaddr == MAP_NONE || (flags & I386_VM_PRESENT)); + vm_assert(physaddr != MAP_NONE || !flags); finalpde = I386_VM_PDE(v + I386_PAGE_SIZE * pages); @@ -454,11 +531,7 @@ PUBLIC int pt_writemap(pt_t *pt, vir_bytes v, phys_bytes physaddr, */ for(pdecheck = I386_VM_PDE(v); pdecheck <= finalpde; pdecheck++) { vm_assert(pdecheck >= 0 && pdecheck < I386_VM_DIR_ENTRIES); - if(pt->pt_dir[pdecheck] & I386_VM_BIGPAGE) { - printf("pt_writemap: trying to write 0x%lx into 0x%lx\n", - physaddr, v); - panic("pt_writemap: BIGPAGE found"); - } + vm_assert(!(pt->pt_dir[pdecheck] & I386_VM_BIGPAGE)); if(!(pt->pt_dir[pdecheck] & I386_VM_PRESENT)) { int r; if(verify) { @@ -497,7 +570,8 @@ PUBLIC int pt_writemap(pt_t *pt, vir_bytes v, phys_bytes physaddr, /* Make sure page directory entry for this page table * is marked present and page table entry is available. */ - vm_assert((pt->pt_dir[pde] & I386_VM_PRESENT) && pt->pt_pt[pde]); + vm_assert((pt->pt_dir[pde] & I386_VM_PRESENT)); + vm_assert(pt->pt_pt[pde]); #if SANITYCHECKS /* We don't expect to overwrite a page. */ @@ -509,7 +583,7 @@ PUBLIC int pt_writemap(pt_t *pt, vir_bytes v, phys_bytes physaddr, } if(writemapflags & WMF_FREE) { - FREE_MEM(ABS2CLICK(physaddr), 1); + free_mem(ABS2CLICK(physaddr), 1); } /* Entry we will write. */ @@ -521,12 +595,23 @@ PUBLIC int pt_writemap(pt_t *pt, vir_bytes v, phys_bytes physaddr, maskedentry &= ~(I386_VM_ACC|I386_VM_DIRTY); /* Verify pagetable entry. */ if(maskedentry != entry) { - printf("pt_writemap: 0x%lx found, masked 0x%lx, 0x%lx expected\n", - pt->pt_pt[pde][pte], maskedentry, entry); + printf("pt_writemap: mismatch: "); + if((entry & I386_VM_ADDR_MASK) != + (maskedentry & I386_VM_ADDR_MASK)) { + printf("pt_writemap: physaddr mismatch (0x%lx, 0x%lx); ", entry, maskedentry); + } else printf("phys ok; "); + printf(" flags: found %s; ", + ptestr(pt->pt_pt[pde][pte])); + printf(" masked %s; ", + ptestr(maskedentry)); + printf(" expected %s\n", ptestr(entry)); return EFAULT; } } else { /* Write pagetable entry. */ +#if SANITYCHECKS + vm_assert(vm_addrok(pt->pt_pt[pde], 1)); +#endif pt->pt_pt[pde][pte] = entry; } @@ -644,8 +729,7 @@ PUBLIC void pt_init(phys_bytes usedlimit) phys_bytes sparepages_ph; /* Shorthand. */ - newpt = &vmp->vm_pt; - + newpt = &vmprocess->vm_pt; /* Get ourselves spare pages. */ if(!(sparepages_mem = (vir_bytes) aalloc(I386_PAGE_SIZE*SPAREPAGES))) @@ -679,9 +763,9 @@ PUBLIC void pt_init(phys_bytes usedlimit) free_pde = id_map_high_pde+1; /* Initial (current) range of our virtual address space. */ - lo = CLICK2ABS(vmp->vm_arch.vm_seg[T].mem_phys); - hi = CLICK2ABS(vmp->vm_arch.vm_seg[S].mem_phys + - vmp->vm_arch.vm_seg[S].mem_len); + lo = CLICK2ABS(vmprocess->vm_arch.vm_seg[T].mem_phys); + hi = CLICK2ABS(vmprocess->vm_arch.vm_seg[S].mem_phys + + vmprocess->vm_arch.vm_seg[S].mem_len); vm_assert(!(lo % I386_PAGE_SIZE)); vm_assert(!(hi % I386_PAGE_SIZE)); @@ -713,9 +797,9 @@ PUBLIC void pt_init(phys_bytes usedlimit) } /* Move segments up too. */ - vmp->vm_arch.vm_seg[T].mem_phys += ABS2CLICK(moveup); - vmp->vm_arch.vm_seg[D].mem_phys += ABS2CLICK(moveup); - vmp->vm_arch.vm_seg[S].mem_phys += ABS2CLICK(moveup); + vmprocess->vm_arch.vm_seg[T].mem_phys += ABS2CLICK(moveup); + vmprocess->vm_arch.vm_seg[D].mem_phys += ABS2CLICK(moveup); + vmprocess->vm_arch.vm_seg[S].mem_phys += ABS2CLICK(moveup); /* Allocate us a page table in which to remember page directory * pointers. @@ -731,14 +815,14 @@ PUBLIC void pt_init(phys_bytes usedlimit) * like regular processes have. */ extra_clicks = ABS2CLICK(VM_DATATOP - hi); - vmp->vm_arch.vm_seg[S].mem_len += extra_clicks; + vmprocess->vm_arch.vm_seg[S].mem_len += extra_clicks; /* We pretend to the kernel we have a huge stack segment to * increase our data segment. */ - vmp->vm_arch.vm_data_top = - (vmp->vm_arch.vm_seg[S].mem_vir + - vmp->vm_arch.vm_seg[S].mem_len) << CLICK_SHIFT; + vmprocess->vm_arch.vm_data_top = + (vmprocess->vm_arch.vm_seg[S].mem_vir + + vmprocess->vm_arch.vm_seg[S].mem_len) << CLICK_SHIFT; /* Where our free virtual address space starts. * This is only a hint to the VM system. @@ -746,7 +830,7 @@ PUBLIC void pt_init(phys_bytes usedlimit) newpt->pt_virtop = 0; /* Let other functions know VM now has a private page table. */ - vmp->vm_flags |= VMF_HASPT; + vmprocess->vm_flags |= VMF_HASPT; /* Now reserve another pde for kernel's own mappings. */ { @@ -813,19 +897,19 @@ PUBLIC void pt_init(phys_bytes usedlimit) /* Give our process the new, copied, private page table. */ pt_mapkernel(newpt); /* didn't know about vm_dir pages earlier */ - pt_bind(newpt, vmp); + pt_bind(newpt, vmprocess); /* new segment limit for the kernel after paging is enabled */ ep_data.data_seg_limit = free_pde*I386_BIG_PAGE_SIZE; /* the memory map which must be installed after paging is enabled */ - ep_data.mem_map = vmp->vm_arch.vm_seg; + ep_data.mem_map = vmprocess->vm_arch.vm_seg; /* Now actually enable paging. */ if(sys_vmctl_enable_paging(&ep_data) != OK) panic("pt_init: enable paging failed"); /* Back to reality - this is where the stack actually is. */ - vmp->vm_arch.vm_seg[S].mem_len -= extra_clicks; + vmprocess->vm_arch.vm_seg[S].mem_len -= extra_clicks; /* All OK. */ return; diff --git a/servers/vm/arch/i386/vm.c b/servers/vm/arch/i386/vm.c index 3df08eb28..804ea334b 100644 --- a/servers/vm/arch/i386/vm.c +++ b/servers/vm/arch/i386/vm.c @@ -18,6 +18,7 @@ #include #include +#include #include #include "proto.h" diff --git a/servers/vm/exec.c b/servers/vm/exec.c index 754e6a596..5bcac2c08 100644 --- a/servers/vm/exec.c +++ b/servers/vm/exec.c @@ -88,8 +88,6 @@ PUBLIC int do_exec_newmem(message *msg) vmp= &vmproc[proc_n]; ptr= msg->VMEN_ARGSPTR; - NOTRUNNABLE(vmp->vm_endpoint); - if(msg->VMEN_ARGSSIZE != sizeof(args)) { printf("VM: exec_newmem: args size %d != %ld\n", msg->VMEN_ARGSSIZE, sizeof(args)); @@ -160,8 +158,6 @@ SANITYCHECK(SCL_DETAIL); if (!sh_mp) /* Load text if sh_mp = NULL */ msg->VMEN_FLAGS |= EXC_NM_RF_LOAD_TEXT; - NOTRUNNABLE(vmp->vm_endpoint); - return OK; } @@ -225,8 +221,8 @@ vir_bytes *stack_top; /* top of process stack */ */ SANITYCHECK(SCL_DETAIL); if(hadpt) { - pt_free(&rmp->vm_pt); rmp->vm_flags &= ~VMF_HASPT; + pt_free(&rmp->vm_pt); } vm_assert(!(vmpold->vm_flags & VMF_INUSE)); *vmpold = *rmp; /* copy current state. */ @@ -236,11 +232,11 @@ SANITYCHECK(SCL_DETAIL); if(!hadpt) { if (find_share(rmp, rmp->vm_ino, rmp->vm_dev, rmp->vm_ctime) == NULL) { /* No other process shares the text segment, so free it. */ - FREE_MEM(rmp->vm_arch.vm_seg[T].mem_phys, rmp->vm_arch.vm_seg[T].mem_len); + free_mem(rmp->vm_arch.vm_seg[T].mem_phys, rmp->vm_arch.vm_seg[T].mem_len); } /* Free the data and stack segments. */ - FREE_MEM(rmp->vm_arch.vm_seg[D].mem_phys, + free_mem(rmp->vm_arch.vm_seg[D].mem_phys, rmp->vm_arch.vm_seg[S].mem_vir + rmp->vm_arch.vm_seg[S].mem_len - rmp->vm_arch.vm_seg[D].mem_vir); @@ -271,6 +267,7 @@ SANITYCHECK(SCL_DETAIL); SANITYCHECK(SCL_DETAIL); printf("VM: new_mem: failed\n"); if(ptok) { + rmp->vm_flags &= ~VMF_HASPT; pt_free(&rmp->vm_pt); } *rmp = *vmpold; /* undo. */ @@ -304,9 +301,9 @@ SANITYCHECK(SCL_DETAIL); } else { phys_clicks new_base; - new_base = ALLOC_MEM(text_clicks + tot_clicks, 0); + new_base = alloc_mem(text_clicks + tot_clicks, 0); if (new_base == NO_MEM) { - printf("VM: new_mem: ALLOC_MEM failed\n"); + printf("VM: new_mem: alloc_mem failed\n"); return(ENOMEM); } diff --git a/servers/vm/exit.c b/servers/vm/exit.c index 3af26c56f..7c339a5ab 100644 --- a/servers/vm/exit.c +++ b/servers/vm/exit.c @@ -16,6 +16,7 @@ #include #include +#include #include #include "glo.h" @@ -25,11 +26,11 @@ PUBLIC void free_proc(struct vmproc *vmp) { + map_free_proc(vmp); if(vmp->vm_flags & VMF_HASPT) { vmp->vm_flags &= ~VMF_HASPT; pt_free(&vmp->vm_pt); } - map_free_proc(vmp); vmp->vm_regions = NULL; #if VMSTATS vmp->vm_bytecopies = 0; @@ -77,7 +78,7 @@ SANITYCHECK(SCL_DETAIL); SANITYCHECK(SCL_DETAIL); } else { /* Free the data and stack segments. */ - FREE_MEM(vmp->vm_arch.vm_seg[D].mem_phys, + free_mem(vmp->vm_arch.vm_seg[D].mem_phys, vmp->vm_arch.vm_seg[S].mem_vir + vmp->vm_arch.vm_seg[S].mem_len - vmp->vm_arch.vm_seg[D].mem_vir); @@ -86,7 +87,7 @@ SANITYCHECK(SCL_DETAIL); /* No other process shares the text segment, * so free it. */ - FREE_MEM(vmp->vm_arch.vm_seg[T].mem_phys, + free_mem(vmp->vm_arch.vm_seg[T].mem_phys, vmp->vm_arch.vm_seg[T].mem_len); } } diff --git a/servers/vm/fork.c b/servers/vm/fork.c index b75935ebb..2ab6dac95 100644 --- a/servers/vm/fork.c +++ b/servers/vm/fork.c @@ -58,8 +58,6 @@ PUBLIC int do_fork(message *msg) vmc = &vmproc[childproc]; /* child */ vm_assert(vmc->vm_slot == childproc); - NOTRUNNABLE(vmp->vm_endpoint); - if(vmp->vm_flags & VMF_HAS_DMA) { printf("VM: %d has DMA memory and may not fork\n", msg->VMF_ENDPOINT); return EINVAL; @@ -74,7 +72,7 @@ PUBLIC int do_fork(message *msg) vmc->vm_regions = NULL; vmc->vm_endpoint = NONE; /* In case someone tries to use it. */ vmc->vm_pt = origpt; - vmc->vm_flags |= VMF_HASPT; + vmc->vm_flags &= ~VMF_HASPT; #if VMSTATS vmc->vm_bytecopies = 0; @@ -85,6 +83,8 @@ PUBLIC int do_fork(message *msg) return ENOMEM; } + vmc->vm_flags |= VMF_HASPT; + if(fullvm) { SANITYCHECK(SCL_DETAIL); @@ -102,7 +102,7 @@ PUBLIC int do_fork(message *msg) SANITYCHECK(SCL_DETAIL); } else { vir_bytes sp; - phys_bytes d_abs, s_abs; + struct vir_region *heap, *stack; vir_bytes text_bytes, data_bytes, stack_bytes, parent_gap_bytes, child_gap_bytes; @@ -147,30 +147,32 @@ PUBLIC int do_fork(message *msg) return r; } - if((d_abs = map_lookup_phys(vmc, VRT_HEAP)) == MAP_NONE) - panic("couldn't lookup data"); - if((s_abs = map_lookup_phys(vmc, VRT_STACK)) == MAP_NONE) + if(!(heap = map_region_lookup_tag(vmc, VRT_HEAP))) + panic("couldn't lookup heap"); + vm_assert(heap->phys); + if(!(stack = map_region_lookup_tag(vmc, VRT_STACK))) panic("couldn't lookup stack"); + vm_assert(stack->phys); /* Now copy the memory regions. */ if(vmc->vm_arch.vm_seg[T].mem_len > 0) { - phys_bytes t_abs; - if((t_abs = map_lookup_phys(vmc, VRT_TEXT)) == MAP_NONE) + struct vir_region *text; + if(!(text = map_region_lookup_tag(vmc, VRT_TEXT))) panic("couldn't lookup text"); - if(sys_abscopy(CLICK2ABS(vmp->vm_arch.vm_seg[T].mem_phys), - t_abs, text_bytes) != OK) + vm_assert(text->phys); + if(copy_abs2region(CLICK2ABS(vmp->vm_arch.vm_seg[T].mem_phys), + text, 0, text_bytes) != OK) panic("couldn't copy text"); } - if(sys_abscopy(CLICK2ABS(vmp->vm_arch.vm_seg[D].mem_phys), - d_abs, data_bytes) != OK) - panic("couldn't copy data"); + if(copy_abs2region(CLICK2ABS(vmp->vm_arch.vm_seg[D].mem_phys), + heap, 0, data_bytes) != OK) + panic("couldn't copy heap"); - if(sys_abscopy( - CLICK2ABS(vmp->vm_arch.vm_seg[D].mem_phys + + if(copy_abs2region(CLICK2ABS(vmp->vm_arch.vm_seg[D].mem_phys + vmc->vm_arch.vm_seg[D].mem_len) + parent_gap_bytes, - s_abs + child_gap_bytes, stack_bytes) != OK) + stack, child_gap_bytes, stack_bytes) != OK) panic("couldn't copy stack"); } @@ -187,9 +189,6 @@ PUBLIC int do_fork(message *msg) panic("do_fork can't sys_fork: %d", r); } - NOTRUNNABLE(vmp->vm_endpoint); - NOTRUNNABLE(vmc->vm_endpoint); - if(fullvm) { vir_bytes vir; /* making these messages writable is an optimisation diff --git a/servers/vm/glo.h b/servers/vm/glo.h index 49f82ec82..bd9f5bbfe 100644 --- a/servers/vm/glo.h +++ b/servers/vm/glo.h @@ -31,3 +31,4 @@ EXTERN int total_pages; EXTERN long vm_paged; EXTERN int meminit_done; + diff --git a/servers/vm/main.c b/servers/vm/main.c index e3c312af5..3f23c0fd1 100644 --- a/servers/vm/main.c +++ b/servers/vm/main.c @@ -24,6 +24,7 @@ #include #include #include +#include #include @@ -79,6 +80,8 @@ PUBLIC int main(void) /* SEF local startup. */ sef_local_startup(); + SANITYCHECK(SCL_TOP); + /* This is VM's main loop. */ while (TRUE) { int r, c; @@ -166,7 +169,6 @@ PRIVATE int sef_cb_init_fresh(int type, sef_init_info_t *info) #if SANITYCHECKS incheck = nocheck = 0; - FIXME("VM SANITYCHECKS are on"); #endif vm_paged = 1; @@ -278,7 +280,7 @@ PRIVATE int sef_cb_init_fresh(int type, sef_init_info_t *info) panic("VM: vmctl for new stack failed"); } - FREE_MEM(vmp->vm_arch.vm_seg[D].mem_phys + + free_mem(vmp->vm_arch.vm_seg[D].mem_phys + vmp->vm_arch.vm_seg[D].mem_len, old_stack); diff --git a/servers/vm/map_mem.c b/servers/vm/map_mem.c deleted file mode 100644 index 5364585db..000000000 --- a/servers/vm/map_mem.c +++ /dev/null @@ -1,308 +0,0 @@ - -#define _SYSTEM 1 - -#include -#include -#include -#include -#include - -#include -#include -#include -#include -#include - -#include "vm.h" -#include "proto.h" -#include "util.h" -#include "glo.h" -#include "region.h" -#include "sanitycheck.h" - -/*===========================================================================* - * split_phys * - *===========================================================================*/ -PRIVATE int split_phys(struct phys_region *pr, vir_bytes point) -{ - struct phys_region *newpr, *q, *prev; - struct phys_block *newpb; - struct phys_block *pb = pr->ph; -/* Split the phys region into 2 parts by @point. */ - - if(pr->offset >= point || pr->offset + pb->length <= point) - return OK; - if(!SLABALLOC(newpb)) - return ENOMEM; - - /* Split phys block. */ - *newpb = *pb; - pb->length = point - pr->offset; - newpb->length -= pb->length; - newpb->phys += pb->length; - - /* Split phys regions in a list. */ - for(q = pb->firstregion; q; q = q->next_ph_list) { - if(!SLABALLOC(newpr)) - return ENOMEM; - - *newpr = *q; - newpr->ph = newpb; - newpr->offset += pb->length; - - /* Link to the vir region's phys region list. */ - physr_insert(newpr->parent->phys, newpr); - - /* Link to the next_ph_list. */ - if(q == pb->firstregion) { - newpb->firstregion = newpr; - prev = newpr; - } else { - prev->next_ph_list = newpr; - prev = newpr; - } - } - prev->next_ph_list = NULL; - - return OK; -} - -/*===========================================================================* - * rm_phys_regions * - *===========================================================================*/ -PRIVATE void rm_phys_regions(struct vir_region *region, - vir_bytes begin, vir_bytes length) -{ -/* Remove all phys regions between @begin and @begin+length. - * - * Don't update the page table, because we will update it at map_memory() - * later. - */ - struct phys_region *pr; - physr_iter iter; - - physr_start_iter(region->phys, &iter, begin, AVL_GREATER_EQUAL); - while((pr = physr_get_iter(&iter)) && pr->offset < begin + length) { - pb_unreferenced(region, pr); - physr_remove(region->phys, pr->offset); - physr_start_iter(region->phys, &iter, begin, - AVL_GREATER_EQUAL); - SLABFREE(pr); - } -} - -/*===========================================================================* - * clean_phys_regions * - *===========================================================================*/ -PRIVATE void clean_phys_regions(struct vir_region *region, - vir_bytes offset, vir_bytes length) -{ -/* Consider @offset as the start address and @offset+length as the end address. - * If there are phys regions crossing the start address or the end address, - * split them into 2 parts. - * - * We assume that the phys regions are listed in order and don't overlap. - */ - struct phys_region *pr; - physr_iter iter; - - physr_start_iter_least(region->phys, &iter); - while((pr = physr_get_iter(&iter))) { - /* If this phys region crosses the start address, split it. */ - if(pr->offset < offset - && pr->offset + pr->ph->length > offset) { - split_phys(pr, offset); - physr_start_iter_least(region->phys, &iter); - } - /* If this phys region crosses the end address, split it. */ - else if(pr->offset < offset + length - && pr->offset + pr->ph->length > offset + length) { - split_phys(pr, offset + length); - physr_start_iter_least(region->phys, &iter); - } - else { - physr_incr_iter(&iter); - } - } -} - -/*===========================================================================* - * do_map_memory * - *===========================================================================*/ -PRIVATE int do_map_memory(struct vmproc *vms, struct vmproc *vmd, - struct vir_region *vrs, struct vir_region *vrd, - vir_bytes offset_s, vir_bytes offset_d, - vir_bytes length, int flag) -{ - struct phys_region *prs; - struct phys_region *newphysr; - struct phys_block *pb; - physr_iter iter; - u32_t pt_flag = PTF_PRESENT | PTF_USER; - vir_bytes end; - - /* Search for the first phys region in the source process. */ - physr_start_iter(vrs->phys, &iter, offset_s, AVL_EQUAL); - prs = physr_get_iter(&iter); - if(!prs) - panic("do_map_memory: no aligned phys region: %d", 0); - - /* flag: 0 -> read-only - * 1 -> writable - * -1 -> share as COW, so read-only - */ - if(flag > 0) - pt_flag |= PTF_WRITE; - - /* Map phys blocks in the source process to the destination process. */ - end = offset_d + length; - while((prs = physr_get_iter(&iter)) && offset_d < end) { - /* If a SMAP share was requested but the phys block has already - * been shared as COW, copy the block for the source phys region - * first. - */ - pb = prs->ph; - if(flag >= 0 && pb->refcount > 1 - && pb->share_flag == PBSH_COW) { - map_copy_ph_block(vms, vrs, prs); - pb = prs->ph; - } - - /* Allocate a new phys region. */ - if(!SLABALLOC(newphysr)) - return ENOMEM; - - /* Set and link the new phys region to the block. */ - newphysr->ph = pb; - newphysr->offset = offset_d; - newphysr->parent = vrd; - newphysr->next_ph_list = pb->firstregion; - pb->firstregion = newphysr; - physr_insert(newphysr->parent->phys, newphysr); - pb->refcount++; - - /* If a COW share was requested but the phys block has already - * been shared as SMAP, give up on COW and copy the block for - * the destination phys region now. - */ - if(flag < 0 && pb->refcount > 1 - && pb->share_flag == PBSH_SMAP) { - map_copy_ph_block(vmd, vrd, newphysr); - } - else { - /* See if this is a COW share or SMAP share. */ - if(flag < 0) { /* COW share */ - pb->share_flag = PBSH_COW; - /* Update the page table for the src process. */ - pt_writemap(&vms->vm_pt, offset_s + vrs->vaddr, - pb->phys, pb->length, - pt_flag, WMF_OVERWRITE); - } - else { /* SMAP share */ - pb->share_flag = PBSH_SMAP; - } - /* Update the page table for the destination process. */ - pt_writemap(&vmd->vm_pt, offset_d + vrd->vaddr, - pb->phys, pb->length, pt_flag, WMF_OVERWRITE); - } - - physr_incr_iter(&iter); - offset_d += pb->length; - offset_s += pb->length; - } - return OK; -} - -/*===========================================================================* - * map_memory * - *===========================================================================*/ -PUBLIC int map_memory(endpoint_t sour, endpoint_t dest, - vir_bytes virt_s, vir_bytes virt_d, vir_bytes length, int flag) -{ -/* This is the entry point. This function will be called by handle_memory() when - * VM recieves a map-memory request. - */ - struct vmproc *vms, *vmd; - struct vir_region *vrs, *vrd; - physr_iter iterd; - vir_bytes offset_s, offset_d; - int p; - int r; - - if(vm_isokendpt(sour, &p) != OK) - panic("map_memory: bad endpoint: %d", sour); - vms = &vmproc[p]; - if(vm_isokendpt(dest, &p) != OK) - panic("map_memory: bad endpoint: %d", dest); - vmd = &vmproc[p]; - - vrs = map_lookup(vms, virt_s); - vm_assert(vrs); - vrd = map_lookup(vmd, virt_d); - vm_assert(vrd); - - /* Linear address -> offset from start of vir region. */ - offset_s = virt_s - vrs->vaddr; - offset_d = virt_d - vrd->vaddr; - - /* Make sure that the range in the source process has been mapped - * to physical memory. - */ - map_handle_memory(vms, vrs, offset_s, length, 0); - - /* Prepare work. */ - clean_phys_regions(vrs, offset_s, length); - clean_phys_regions(vrd, offset_d, length); - rm_phys_regions(vrd, offset_d, length); - - /* Map memory. */ - r = do_map_memory(vms, vmd, vrs, vrd, offset_s, offset_d, length, flag); - - return r; -} - -/*===========================================================================* - * unmap_memory * - *===========================================================================*/ -PUBLIC int unmap_memory(endpoint_t sour, endpoint_t dest, - vir_bytes virt_s, vir_bytes virt_d, vir_bytes length, int flag) -{ - struct vmproc *vmd; - struct vir_region *vrd; - struct phys_region *pr; - struct phys_block *pb; - physr_iter iter; - vir_bytes off, end; - int p; - - /* Use information on the destination process to unmap. */ - if(vm_isokendpt(dest, &p) != OK) - panic("unmap_memory: bad endpoint: %d", dest); - vmd = &vmproc[p]; - - vrd = map_lookup(vmd, virt_d); - vm_assert(vrd); - - /* Search for the first phys region in the destination process. */ - off = virt_d - vrd->vaddr; - physr_start_iter(vrd->phys, &iter, off, AVL_EQUAL); - pr = physr_get_iter(&iter); - if(!pr) - panic("unmap_memory: no aligned phys region: %d", 0); - - /* Copy the phys block now rather than doing COW. */ - end = off + length; - while((pr = physr_get_iter(&iter)) && off < end) { - pb = pr->ph; - vm_assert(pb->refcount > 1); - vm_assert(pb->share_flag == PBSH_SMAP); - - map_copy_ph_block(vmd, vrd, pr); - - physr_incr_iter(&iter); - off += pb->length; - } - - return OK; -} - diff --git a/servers/vm/memlist.h b/servers/vm/memlist.h new file mode 100644 index 000000000..594fc07dc --- /dev/null +++ b/servers/vm/memlist.h @@ -0,0 +1,11 @@ + +#ifndef _MEMLIST_H +#define _MEMLIST_H 1 + +struct memlist { + struct memlist *next; + phys_bytes phys; /* physical address in bytes */ + phys_bytes length; /* length in bytes */ +}; + +#endif diff --git a/servers/vm/mmap.c b/servers/vm/mmap.c index 975ad1e11..8a5507ca2 100644 --- a/servers/vm/mmap.c +++ b/servers/vm/mmap.c @@ -15,10 +15,12 @@ #include #include #include +#include #include #include +#include #include #include #include @@ -59,12 +61,12 @@ PUBLIC int do_mmap(message *m) return EINVAL; } - if(m->VMM_FLAGS & MAP_CONTIG) mfflags |= MF_CONTIG; if(m->VMM_FLAGS & MAP_PREALLOC) mfflags |= MF_PREALLOC; if(m->VMM_FLAGS & MAP_LOWER16M) vrflags |= VR_LOWER16MB; if(m->VMM_FLAGS & MAP_LOWER1M) vrflags |= VR_LOWER1MB; if(m->VMM_FLAGS & MAP_ALIGN64K) vrflags |= VR_PHYS64K; if(m->VMM_FLAGS & MAP_SHARED) vrflags |= VR_SHARED; + if(m->VMM_FLAGS & MAP_CONTIG) vrflags |= VR_CONTIG; if(len % VM_PAGE_SIZE) len += VM_PAGE_SIZE - (len % VM_PAGE_SIZE); @@ -238,6 +240,24 @@ PUBLIC int do_remap(message *m) if (!(region = map_lookup(svmp, sa))) return EINVAL; + if(region->vaddr != sa) { + printf("VM: do_remap: not start of region.\n"); + return EFAULT; + } + + if(!(region->flags & VR_SHARED)) { + printf("VM: do_remap: not shared.\n"); + return EFAULT; + } + + if (size % VM_PAGE_SIZE) + size += VM_PAGE_SIZE - size % VM_PAGE_SIZE; + + if(size != region->length) { + printf("VM: do_remap: not size of region.\n"); + return EFAULT; + } + if ((r = map_remap(dvmp, da, size, region, &startv)) != OK) return r; diff --git a/servers/vm/pagefaults.c b/servers/vm/pagefaults.c index 080f28159..4cac55aa4 100644 --- a/servers/vm/pagefaults.c +++ b/servers/vm/pagefaults.c @@ -22,6 +22,7 @@ #include #include #include +#include #include diff --git a/servers/vm/proto.h b/servers/vm/proto.h index 477aef61b..1b3535664 100644 --- a/servers/vm/proto.h +++ b/servers/vm/proto.h @@ -18,7 +18,8 @@ struct phys_region; #include "vm.h" /* alloc.c */ -_PROTOTYPE( phys_clicks alloc_mem_f, (phys_clicks clicks, u32_t flags) ); +_PROTOTYPE( phys_clicks alloc_mem, (phys_clicks clicks, u32_t flags) ); +_PROTOTYPE( struct memlist *alloc_mem_in_list, (phys_bytes bytes, u32_t flags)); _PROTOTYPE( int do_adddma, (message *msg) ); _PROTOTYPE( int do_deldma, (message *msg) ); _PROTOTYPE( int do_getdma, (message *msg) ); @@ -28,12 +29,11 @@ _PROTOTYPE( void printmemstats, (void) ); _PROTOTYPE( void usedpages_reset, (void) ); _PROTOTYPE( int usedpages_add_f, (phys_bytes phys, phys_bytes len, char *file, int line) ); -_PROTOTYPE( void free_mem_f, (phys_clicks base, phys_clicks clicks) ); +_PROTOTYPE( void free_mem, (phys_clicks base, phys_clicks clicks) ); +_PROTOTYPE( void free_mem_list, (struct memlist *list, int all)); +_PROTOTYPE( void print_mem_list, (struct memlist *ml)); #define usedpages_add(a, l) usedpages_add_f(a, l, __FILE__, __LINE__) -#define ALLOC_MEM(clicks, flags) alloc_mem_f(clicks, flags) -#define FREE_MEM(base, clicks) free_mem_f(base, clicks) - _PROTOTYPE( void mem_init, (struct memory *chunks) ); /* utility.c */ @@ -109,6 +109,7 @@ _PROTOTYPE( void *vm_allocpage, (phys_bytes *p, int cat)); _PROTOTYPE( void pt_cycle, (void)); _PROTOTYPE( int pt_mapkernel, (pt_t *pt)); _PROTOTYPE( void vm_pagelock, (void *vir, int lockflag) ); +_PROTOTYPE( int vm_addrok, (void *vir, int write) ); #if SANITYCHECKS _PROTOTYPE( void pt_sanitycheck, (pt_t *pt, char *file, int line) ); @@ -123,7 +124,7 @@ _PROTOTYPE(void slabfree,(void *mem, int bytes)); _PROTOTYPE(void slabstats,(void)); _PROTOTYPE(void slab_sanitycheck, (char *file, int line)); #define SLABALLOC(var) (var = slaballoc(sizeof(*var))) -#define SLABFREE(ptr) slabfree(ptr, sizeof(*(ptr))) +#define SLABFREE(ptr) do { slabfree(ptr, sizeof(*(ptr))); (ptr) = NULL; } while(0) #if SANITYCHECKS _PROTOTYPE(void slabunlock,(void *mem, int bytes)); @@ -159,14 +160,14 @@ _PROTOTYPE(int map_remap, (struct vmproc *dvmp, vir_bytes da, size_t size, _PROTOTYPE(int map_get_phys, (struct vmproc *vmp, vir_bytes addr, phys_bytes *r)); _PROTOTYPE(int map_get_ref, (struct vmproc *vmp, vir_bytes addr, u8_t *cnt)); -_PROTOTYPE(int map_copy_ph_block, (struct vmproc *vmp, - struct vir_region *region, struct phys_region *ph)); _PROTOTYPE(void pb_unreferenced, (struct vir_region *region, struct phys_region *pr)); _PROTOTYPE(void get_usage_info, (struct vmproc *vmp, struct vm_usage_info *vui)); _PROTOTYPE(int get_region_info, (struct vmproc *vmp, struct vm_region_info *vri, int count, vir_bytes *nextp)); +_PROTOTYPE(int copy_abs2region, (phys_bytes abs, + struct vir_region *destregion, phys_bytes offset, phys_bytes len)); #if SANITYCHECKS _PROTOTYPE(void map_sanitycheck,(char *file, int line)); #endif diff --git a/servers/vm/region.c b/servers/vm/region.c index efa2bdc3c..4dda6609e 100644 --- a/servers/vm/region.c +++ b/servers/vm/region.c @@ -27,21 +27,25 @@ #include "region.h" #include "sanitycheck.h" #include "physravl.h" +#include "memlist.h" /* Should a physblock be mapped writable? */ #define WRITABLE(r, pb) \ (((r)->flags & (VR_DIRECT | VR_SHARED)) || \ (((r)->flags & VR_WRITABLE) && (pb)->refcount == 1)) -FORWARD _PROTOTYPE(struct phys_region *map_new_physblock, (struct vmproc *vmp, +FORWARD _PROTOTYPE(int map_new_physblock, (struct vmproc *vmp, struct vir_region *region, vir_bytes offset, vir_bytes length, - phys_bytes what)); + phys_bytes what, u32_t allocflags, int written)); FORWARD _PROTOTYPE(int map_ph_writept, (struct vmproc *vmp, struct vir_region *vr, struct phys_region *pr)); FORWARD _PROTOTYPE(struct vir_region *map_copy_region, (struct vmproc *vmp, struct vir_region *vr)); +FORWARD _PROTOTYPE(struct phys_region *map_clone_ph_block, (struct vmproc *vmp, + struct vir_region *region, struct phys_region *ph, physr_iter *iter)); + PRIVATE char *map_name(struct vir_region *vr) { static char name[100]; @@ -125,6 +129,7 @@ PRIVATE int map_sanitycheck_pt(struct vmproc *vmp, { struct phys_block *pb = pr->ph; int rw; + int r; if(!(vmp->vm_flags & VMF_HASPT)) return OK; @@ -134,8 +139,16 @@ PRIVATE int map_sanitycheck_pt(struct vmproc *vmp, else rw = 0; - return pt_writemap(&vmp->vm_pt, vr->vaddr + pr->offset, + r = pt_writemap(&vmp->vm_pt, vr->vaddr + pr->offset, pb->phys, pb->length, PTF_PRESENT | PTF_USER | rw, WMF_VERIFY); + + if(r != OK) { + printf("proc %d phys_region 0x%lx sanity check failed\n", + vmp->vm_endpoint, pr->offset); + map_printregion(vmp, vr); + } + + return r; } /*===========================================================================* @@ -360,6 +373,8 @@ int mapflags; struct phys_region *ph; physr_avl *phavl; + vm_assert(!(length % VM_PAGE_SIZE)); + SANITYCHECK(SCL_FUNCTIONS); startv = region_find_slot(vmp, minv, maxv, length, &prevregion); @@ -392,19 +407,15 @@ USE(newregion, /* If we know what we're going to map to, map it right away. */ if(what != MAP_NONE) { - struct phys_region *pr; + vm_assert(what); /* mapping in 0 is unlikely to be right */ vm_assert(!(what % VM_PAGE_SIZE)); - vm_assert(!(length % VM_PAGE_SIZE)); vm_assert(!(startv % VM_PAGE_SIZE)); vm_assert(!(mapflags & MF_PREALLOC)); - if(!(pr=map_new_physblock(vmp, newregion, 0, length, what))) { + if(map_new_physblock(vmp, newregion, 0, length, + what, PAF_CLEAR, 0) != OK) { printf("VM: map_new_physblock failed\n"); - SLABFREE(newregion->phys); - SLABFREE(newregion); - return NULL; - } - if(map_ph_writept(vmp, newregion, pr) != OK) { - printf("VM: map_region_writept failed\n"); + USE(newregion, + SLABFREE(newregion->phys);); SLABFREE(newregion); return NULL; } @@ -413,7 +424,8 @@ USE(newregion, if((flags & VR_ANON) && (mapflags & MF_PREALLOC)) { if(map_handle_memory(vmp, newregion, 0, length, 1) != OK) { printf("VM: map_page_region: prealloc failed\n"); - SLABFREE(newregion->phys); + USE(newregion, + SLABFREE(newregion->phys);); SLABFREE(newregion); return NULL; } @@ -474,7 +486,7 @@ PUBLIC void pb_unreferenced(struct vir_region *region, struct phys_region *pr) if(pb->refcount == 0) { vm_assert(!pb->firstregion); if(region->flags & VR_ANON) { - FREE_MEM(ABS2CLICK(pb->phys), + free_mem(ABS2CLICK(pb->phys), ABS2CLICK(pb->length)); } else if(region->flags & VR_DIRECT) { ; /* No action required. */ @@ -482,9 +494,37 @@ PUBLIC void pb_unreferenced(struct vir_region *region, struct phys_region *pr) panic("strange phys flags"); } SLABFREE(pb); + } else { + struct phys_region *others; + int n = 0; + + for(others = pb->firstregion; others; + others = others->next_ph_list) { + if(WRITABLE(region, others->ph)) { + if(map_ph_writept(others->parent->parent, + others->parent, others) != OK) { + printf("VM: map_ph_writept failed unexpectedly\n"); + } + } + n++; + } + vm_assert(n == pb->refcount); } } +PRIVATE struct phys_region *reset_physr_iter(struct vir_region *region, + physr_iter *iter, vir_bytes offset) +{ + struct phys_region *ph; + + physr_start_iter(region->phys, iter, offset, AVL_EQUAL); + ph = physr_get_iter(iter); + vm_assert(ph); + vm_assert(ph->offset == offset); + + return ph; +} + /*===========================================================================* * map_subfree * *===========================================================================*/ @@ -494,6 +534,7 @@ PRIVATE int map_subfree(struct vmproc *vmp, struct phys_region *pr, *nextpr; physr_iter iter; + #if SANITYCHECKS { physr_start_iter_least(region->phys, &iter); @@ -532,13 +573,13 @@ PRIVATE int map_subfree(struct vmproc *vmp, vm_assert(sublen < pr->ph->length); if(pr->ph->refcount > 1) { int r; - r = map_copy_ph_block(vmp, region, pr); - if(r != OK) - return r; + if(!(pr = map_clone_ph_block(vmp, region, + pr, &iter))) + return ENOMEM; } vm_assert(pr->ph->refcount == 1); if(!(region->flags & VR_DIRECT)) { - FREE_MEM(ABS2CLICK(pr->ph->phys), ABS2CLICK(sublen)); + free_mem(ABS2CLICK(pr->ph->phys), ABS2CLICK(sublen)); } USE(pr, pr->offset += sublen;); USE(pr->ph, @@ -560,10 +601,13 @@ PRIVATE int map_free(struct vmproc *vmp, struct vir_region *region) { int r; - if((r=map_subfree(vmp, region, region->length)) != OK) + if((r=map_subfree(vmp, region, region->length)) != OK) { + printf("%d\n", __LINE__); return r; + } - SLABFREE(region->phys); + USE(region, + SLABFREE(region->phys);); SLABFREE(region); return OK; @@ -624,169 +668,243 @@ vir_bytes offset; return NULL; } +PRIVATE u32_t vrallocflags(u32_t flags) +{ + u32_t allocflags = 0; + + if(flags & VR_PHYS64K) + allocflags |= PAF_ALIGN64K; + if(flags & VR_LOWER16MB) + allocflags |= PAF_LOWER16MB; + if(flags & VR_LOWER1MB) + allocflags |= PAF_LOWER1MB; + if(flags & VR_CONTIG) + allocflags |= PAF_CONTIG; + + return allocflags; +} /*===========================================================================* * map_new_physblock * *===========================================================================*/ -PRIVATE struct phys_region *map_new_physblock(vmp, region, offset, length, what_mem) +PRIVATE int map_new_physblock(vmp, region, start_offset, length, + what_mem, allocflags, written) struct vmproc *vmp; struct vir_region *region; -vir_bytes offset; +vir_bytes start_offset; vir_bytes length; phys_bytes what_mem; +u32_t allocflags; +int written; { - struct phys_region *newphysr; - struct phys_block *newpb; - phys_bytes mem_clicks, clicks; - vir_bytes mem; + struct memlist *memlist, given, *ml; + int used_memlist, r; + vir_bytes mapped = 0; + vir_bytes offset = start_offset; SANITYCHECK(SCL_FUNCTIONS); vm_assert(!(length % VM_PAGE_SIZE)); - NOTRUNNABLE(vmp->vm_endpoint); - - /* Allocate things necessary for this chunk of memory. */ - if(!SLABALLOC(newphysr)) - return NULL; - if(!SLABALLOC(newpb)) { - SLABFREE(newphysr); - return NULL; + if((region->flags & VR_CONTIG) && + (start_offset > 0 || length < region->length)) { + printf("VM: map_new_physblock: non-full allocation requested\n"); + return EFAULT; } /* Memory for new physical block. */ - clicks = CLICKSPERPAGE * length / VM_PAGE_SIZE; if(what_mem == MAP_NONE) { - u32_t af = PAF_CLEAR; - if(region->flags & VR_PHYS64K) - af |= PAF_ALIGN64K; - if(region->flags & VR_LOWER16MB) - af |= PAF_LOWER16MB; - if(region->flags & VR_LOWER1MB) - af |= PAF_LOWER1MB; - if((mem_clicks = ALLOC_MEM(clicks, af)) == NO_MEM) { - SLABFREE(newpb); - SLABFREE(newphysr); + allocflags |= vrallocflags(region->flags); + + if(!(memlist = alloc_mem_in_list(length, allocflags))) { printf("map_new_physblock: couldn't allocate\n"); - return NULL; + return ENOMEM; } - mem = CLICK2ABS(mem_clicks); + used_memlist = 1; } else { - mem = what_mem; + given.phys = what_mem; + given.length = length; + given.next = NULL; + memlist = &given; + used_memlist = 0; + vm_assert(given.phys); + vm_assert(given.length); } - SANITYCHECK(SCL_DETAIL); - /* New physical block. */ - USE(newpb, - newpb->phys = mem; - newpb->refcount = 1; - newpb->length = length; - newpb->firstregion = newphysr;); - - /* New physical region. */ - USE(newphysr, - newphysr->offset = offset; - newphysr->ph = newpb; - newphysr->parent = region; - newphysr->next_ph_list = NULL; /* No other references to this block. */); + r = OK; + + for(ml = memlist; ml; ml = ml->next) { + vm_assert(ml->phys); + vm_assert(ml->length); + } + + for(ml = memlist; ml; ml = ml->next) { + struct phys_region *newphysr = NULL; + struct phys_block *newpb = NULL; + + /* Allocate things necessary for this chunk of memory. */ + if(!SLABALLOC(newphysr) || !SLABALLOC(newpb)) { + printf("map_new_physblock: no memory for the ph slabs\n"); + if(newphysr) SLABFREE(newphysr); + if(newpb) SLABFREE(newpb); + r = ENOMEM; + break; + } + + vm_assert(ml->phys); + vm_assert(ml->length); + + /* New physical block. */ + vm_assert(!(ml->phys % VM_PAGE_SIZE)); + + USE(newpb, + newpb->phys = ml->phys; + newpb->refcount = 1; + newpb->length = ml->length; + newpb->firstregion = newphysr;); + + /* New physical region. */ + USE(newphysr, + newphysr->offset = offset; + newphysr->ph = newpb; + newphysr->parent = region; + /* No other references to this block. */ + newphysr->next_ph_list = NULL;); #if SANITYCHECKS - USE(newphysr, newphysr->written = 0;); + USE(newphysr, newphysr->written = written;); #endif - /* Update pagetable. */ - vm_assert(!(length % VM_PAGE_SIZE)); - vm_assert(!(newpb->length % VM_PAGE_SIZE)); - SANITYCHECK(SCL_DETAIL); - if(map_ph_writept(vmp, region, newphysr) != OK) { - if(what_mem == MAP_NONE) - FREE_MEM(mem_clicks, clicks); - SLABFREE(newpb); - SLABFREE(newphysr); - printf("map_new_physblock: map_ph_writept failed\n"); - return NULL; + /* Update pagetable. */ + if(map_ph_writept(vmp, region, newphysr) != OK) { + printf("map_new_physblock: map_ph_writept failed\n"); + r = ENOMEM; + break; + } + + physr_insert(region->phys, newphysr); + + offset += ml->length; + mapped += ml->length; } - physr_insert(region->phys, newphysr); + if(used_memlist) { + if(r != OK) { + offset = start_offset; + /* Things did not go well. Undo everything. */ + for(ml = memlist; ml; ml = ml->next) { + struct phys_region *physr; + offset += ml->length; + if((physr = physr_search(region->phys, offset, + AVL_EQUAL))) { + vm_assert(physr->ph->refcount == 1); + pb_unreferenced(region, physr); + physr_remove(region->phys, physr->offset); + SLABFREE(physr); + } + } + } else vm_assert(mapped == length); + + /* Always clean up the memlist itself, even if everything + * worked we're not using the memlist nodes any more. And + * the memory they reference is either freed above or in use. + */ + free_mem_list(memlist, 0); + } SANITYCHECK(SCL_FUNCTIONS); - return newphysr; + return r; } - /*===========================================================================* - * map_copy_ph_block * + * map_clone_ph_block * *===========================================================================*/ -PUBLIC int map_copy_ph_block(vmp, region, ph) +PRIVATE struct phys_region *map_clone_ph_block(vmp, region, ph, iter) struct vmproc *vmp; struct vir_region *region; struct phys_region *ph; +physr_iter *iter; { - int r; - phys_bytes newmem, newmem_cl, clicks; - struct phys_block *newpb; - u32_t af = 0; - - /* This is only to be done if there is more than one copy. */ - vm_assert(ph->ph->refcount > 1); + vir_bytes offset, length; + struct memlist *ml; + u32_t allocflags; + phys_bytes physaddr; + struct phys_region *newpr; + int written = 0; +#if SANITYCHECKS + written = ph->written; +#endif + SANITYCHECK(SCL_FUNCTIONS); - /* Do actual copy on write; allocate new physblock. */ - if(!SLABALLOC(newpb)) { - printf("VM: map_copy_ph_block: couldn't allocate newpb\n"); - return ENOMEM; - } + /* Warning: this function will free the passed + * phys_region *ph and replace it (in the same offset) + * with one or more others! So both the pointer to it + * and any iterators over the phys_regions in the vir_region + * will be invalid on successful return. (Iterators over + * the vir_region could be invalid on unsuccessful return too.) + */ - clicks = CLICKSPERPAGE * ph->ph->length / VM_PAGE_SIZE; - vm_assert(CLICK2ABS(clicks) == ph->ph->length); - if(region->flags & VR_PHYS64K) - af |= PAF_ALIGN64K; + /* This function takes a physical block, copies its contents + * into newly allocated memory, and replaces the single physical + * block by one or more physical blocks with refcount 1 with the + * same contents as the original. In other words, a fragmentable + * version of map_copy_ph_block(). + */ - NOTRUNNABLE(vmp->vm_endpoint); + /* Remember where and how much. */ + offset = ph->offset; + length = ph->ph->length; + physaddr = ph->ph->phys; - if((newmem_cl = ALLOC_MEM(clicks, af)) == NO_MEM) { - printf("VM: map_copy_ph_block: couldn't allocate new block\n"); - SLABFREE(newpb); - return ENOMEM; - } - newmem = CLICK2ABS(newmem_cl); - vm_assert(ABS2CLICK(newmem) == newmem_cl); + /* Now unlink the original physical block so we can replace + * it with new ones. + */ + SANITYCHECK(SCL_DETAIL); + SLABSANE(ph); + SLABSANE(ph->ph); + vm_assert(ph->ph->refcount > 1); pb_unreferenced(region, ph); - vm_assert(ph->ph->refcount > 0); - -USE(newpb, - newpb->length = ph->ph->length; - newpb->refcount = 1; - newpb->phys = newmem; - newpb->firstregion = ph;); + vm_assert(ph->ph->refcount >= 1); + physr_remove(region->phys, offset); + SLABFREE(ph); - USE(ph, ph->next_ph_list = NULL;); + SANITYCHECK(SCL_DETAIL); - NOTRUNNABLE(vmp->vm_endpoint); + /* Put new free memory in. */ + allocflags = vrallocflags(region->flags); + vm_assert(!(allocflags & PAF_CONTIG)); + vm_assert(!(allocflags & PAF_CLEAR)); - /* Copy old memory to new memory. */ - if((r=sys_abscopy(ph->ph->phys, newpb->phys, newpb->length)) != OK) { - printf("VM: map_copy_ph_block: sys_abscopy failed\n"); - return r; + if(map_new_physblock(vmp, region, offset, length, + MAP_NONE, allocflags, written) != OK) { + /* XXX original range now gone. */ + free_mem_list(ml, 0); + printf("VM: map_clone_ph_block: map_new_physblock failed.\n"); + return NULL; } -#if VMSTATS - vmp->vm_bytecopies += newpb->length; -#endif + /* Copy the block to the new memory. + * Can only fail if map_new_physblock didn't do what we asked. + */ + if(copy_abs2region(physaddr, region, offset, length) != OK) + panic("copy_abs2region failed, no good reason for that"); - /* Reference new block. */ - USE(ph, ph->ph = newpb;); + newpr = physr_search(region->phys, offset, AVL_EQUAL); + vm_assert(newpr); + vm_assert(newpr->offset == offset); - /* Update pagetable with new address. - * This will also make it writable. - */ - r = map_ph_writept(vmp, region, ph); - if(r != OK) - panic("map_copy_ph_block: map_ph_writept failed: %d", r); + if(iter) { + physr_start_iter(region->phys, iter, offset, AVL_EQUAL); + vm_assert(physr_get_iter(iter) == newpr); + } - return OK; + SANITYCHECK(SCL_FUNCTIONS); + + return newpr; } + /*===========================================================================* * map_pf * *===========================================================================*/ @@ -810,10 +928,9 @@ int write; SANITYCHECK(SCL_FUNCTIONS); - NOTRUNNABLE(vmp->vm_endpoint); - if((ph = physr_search(region->phys, offset, AVL_LESS_EQUAL)) && (ph->offset <= offset && offset < ph->offset + ph->ph->length)) { + phys_bytes blockoffset = ph->offset; /* Pagefault in existing block. Do copy-on-write. */ vm_assert(write); vm_assert(region->flags & VR_WRITABLE); @@ -829,12 +946,14 @@ int write; printf("VM: write RO mapped pages.\n"); return EFAULT; } else { - r = map_copy_ph_block(vmp, region, ph); + if(!map_clone_ph_block(vmp, region, ph, NULL)) + r = ENOMEM; } } } else { /* Pagefault in non-existing block. Map in new block. */ - if(!map_new_physblock(vmp, region, virpage, VM_PAGE_SIZE, MAP_NONE)) { + if(map_new_physblock(vmp, region, virpage, + VM_PAGE_SIZE, MAP_NONE, PAF_CLEAR, 0) != OK) { printf("map_new_physblock failed\n"); r = ENOMEM; } @@ -848,7 +967,8 @@ int write; } #if SANITYCHECKS - if(OK != pt_checkrange(&vmp->vm_pt, region->vaddr+offset, VM_PAGE_SIZE, write)) { + if(OK != pt_checkrange(&vmp->vm_pt, region->vaddr+virpage, + VM_PAGE_SIZE, write)) { panic("map_pf: pt_checkrange failed: %d", r); } #endif @@ -869,8 +989,6 @@ int write; int changes = 0; physr_iter iter; - NOTRUNNABLE(vmp->vm_endpoint); - #define FREE_RANGE_HERE(er1, er2) { \ struct phys_region *r1 = (er1), *r2 = (er2); \ vir_bytes start = offset, end = offset + length; \ @@ -881,8 +999,8 @@ int write; if(start < end) { \ int r; \ SANITYCHECK(SCL_DETAIL); \ - if(!map_new_physblock(vmp, region, start, \ - end-start, MAP_NONE) != OK) { \ + if(map_new_physblock(vmp, region, start, \ + end-start, MAP_NONE, PAF_CLEAR, 0) != OK) { \ SANITYCHECK(SCL_DETAIL); \ return ENOMEM; \ } \ @@ -906,24 +1024,18 @@ int write; physr = physr_get_iter(&iter); } -#define RESET_ITER(it, where, what) { \ - physr_start_iter(region->phys, &it, where, AVL_EQUAL); \ - what = physr_get_iter(&it); \ - if(!what) panic("thing missing"); \ - if(what->offset != where) panic("thing wrong"); \ -} - FREE_RANGE_HERE(NULL, physr); if(physr) { - RESET_ITER(iter, physr->offset, physr); + physr = reset_physr_iter(region, &iter, physr->offset); if(physr->offset + physr->ph->length <= offset) { physr_incr_iter(&iter); physr = physr_get_iter(&iter); FREE_RANGE_HERE(NULL, physr); if(physr) { - RESET_ITER(iter, physr->offset, physr); + physr = reset_physr_iter(region, &iter, + physr->offset); } } } @@ -936,14 +1048,12 @@ int write; if(write) { vm_assert(physr->ph->refcount > 0); if(!WRITABLE(region, physr->ph)) { - SANITYCHECK(SCL_DETAIL); - r = map_copy_ph_block(vmp, region, physr); - if(r != OK) { + if(!(physr = map_clone_ph_block(vmp, region, + physr, &iter))) { printf("VM: map_handle_memory: no copy\n"); - return r; + return ENOMEM; } changes++; - SANITYCHECK(SCL_DETAIL); } else { SANITYCHECK(SCL_DETAIL); if((r=map_ph_writept(vmp, region, physr)) != OK) { @@ -963,7 +1073,8 @@ int write; if(nextphysr) { if(nextphysr->offset >= offset + length) break; - RESET_ITER(iter, nextphysr->offset, nextphysr); + nextphysr = reset_physr_iter(region, &iter, + nextphysr->offset); } physr = nextphysr; } @@ -1058,15 +1169,65 @@ PRIVATE struct vir_region *map_copy_region(struct vmproc *vmp, struct vir_region USE(newph, newph->written = 0;); #endif physr_insert(newvr->phys, newph); +#if SANITYCHECKS vm_assert(countregions(vr) == cr); +#endif physr_incr_iter(&iter); } +#if SANITYCHECKS vm_assert(countregions(vr) == countregions(newvr)); +#endif return newvr; } +/*===========================================================================* + * copy_abs2region * + *===========================================================================*/ +PUBLIC int copy_abs2region(phys_bytes abs, struct vir_region *destregion, + phys_bytes offset, phys_bytes len) + +{ + vm_assert(destregion); + vm_assert(destregion->phys); + while(len > 0) { + phys_bytes sublen, suboffset; + struct phys_region *ph; + vm_assert(destregion); + vm_assert(destregion->phys); + if(!(ph = physr_search(destregion->phys, offset, AVL_LESS_EQUAL))) { + printf("VM: copy_abs2region: no phys region found (1).\n"); + return EFAULT; + } + vm_assert(ph->offset <= offset); + if(ph->offset+ph->ph->length <= offset) { + printf("VM: copy_abs2region: no phys region found (2).\n"); + return EFAULT; + } + suboffset = offset - ph->offset; + vm_assert(suboffset < ph->ph->length); + sublen = len; + if(sublen > ph->ph->length - suboffset) + sublen = ph->ph->length - suboffset; + vm_assert(suboffset + sublen <= ph->ph->length); + if(ph->ph->refcount != 1) { + printf("VM: copy_abs2region: no phys region found (3).\n"); + return EFAULT; + } + + if(sys_abscopy(abs, ph->ph->phys + suboffset, sublen) != OK) { + printf("VM: copy_abs2region: abscopy failed.\n"); + return EFAULT; + } + abs += sublen; + offset += sublen; + len -= sublen; + } + + return OK; +} + /*=========================================================================* * map_writept * *=========================================================================*/ @@ -1111,7 +1272,6 @@ struct vmproc *src; SANITYCHECK(SCL_FUNCTIONS); - PT_SANE(&src->vm_pt); for(vr = src->vm_regions; vr; vr = vr->next) { physr_iter iter_orig, iter_new; @@ -1151,9 +1311,10 @@ struct vmproc *src; /* If the phys block has been shared as SMAP, * do the regular copy. */ if(pb->refcount > 2 && pb->share_flag == PBSH_SMAP) { - map_copy_ph_block(dst, newvr, new_ph); + map_clone_ph_block(dst, newvr,new_ph, + &iter_new); } else { - pb->share_flag = PBSH_COW; + USE(pb, pb->share_flag = PBSH_COW;); } /* Get next new physregion */ @@ -1364,6 +1525,8 @@ PUBLIC int map_remap(struct vmproc *dvmp, vir_bytes da, size_t size, SANITYCHECK(SCL_FUNCTIONS); + vm_assert(region->flags & VR_SHARED); + /* da is handled differently */ if (!da) dst_addr = dvmp->vm_stacktop; @@ -1373,11 +1536,10 @@ PUBLIC int map_remap(struct vmproc *dvmp, vir_bytes da, size_t size, prev = NULL; /* round up to page size */ - if (size % I386_PAGE_SIZE) - size += I386_PAGE_SIZE - size % I386_PAGE_SIZE; + vm_assert(!(size % VM_PAGE_SIZE)); startv = region_find_slot(dvmp, dst_addr, VM_DATATOP, size, &prev); if (startv == (vir_bytes) -1) { - printf("map_remap: search %x...\n", dst_addr); + printf("map_remap: search 0x%x...\n", dst_addr); map_printmap(dvmp); return ENOMEM; } @@ -1410,11 +1572,13 @@ PUBLIC int map_remap(struct vmproc *dvmp, vir_bytes da, size_t size, physr_start_iter_least(vr->phys, &iter); while((ph = physr_get_iter(&iter))) { struct phys_block *pb = ph->ph; + vm_assert(!ph->next_ph_list); + USE(ph, ph->next_ph_list = pb->firstregion;); + USE(pb, pb->firstregion = ph;); USE(pb, pb->refcount++;); if(map_ph_writept(dvmp, vr, ph) != OK) { panic("map_remap: map_ph_writept failed"); } - physr_incr_iter(&iter); } @@ -1576,6 +1740,302 @@ PUBLIC void printregionstats(struct vmproc *vmp) return; } +/*===========================================================================* + * do_map_memory * + *===========================================================================*/ +PRIVATE int do_map_memory(struct vmproc *vms, struct vmproc *vmd, + struct vir_region *vrs, struct vir_region *vrd, + vir_bytes offset_s, vir_bytes offset_d, + vir_bytes length, int flag) +{ + struct phys_region *prs; + struct phys_region *newphysr; + struct phys_block *pb; + physr_iter iter; + u32_t pt_flag = PTF_PRESENT | PTF_USER; + vir_bytes end; + + SANITYCHECK(SCL_FUNCTIONS); + + /* Search for the first phys region in the source process. */ + physr_start_iter(vrs->phys, &iter, offset_s, AVL_EQUAL); + prs = physr_get_iter(&iter); + if(!prs) + panic("do_map_memory: no aligned phys region: %d", 0); + + /* flag: 0 -> read-only + * 1 -> writable + * -1 -> share as COW, so read-only + */ + if(flag > 0) + pt_flag |= PTF_WRITE; + + /* Map phys blocks in the source process to the destination process. */ + end = offset_d + length; + while((prs = physr_get_iter(&iter)) && offset_d < end) { + /* If a SMAP share was requested but the phys block has already + * been shared as COW, copy the block for the source phys region + * first. + */ + pb = prs->ph; + if(flag >= 0 && pb->refcount > 1 + && pb->share_flag == PBSH_COW) { + if(!(prs = map_clone_ph_block(vms, vrs, prs, &iter))) + return ENOMEM; + pb = prs->ph; + } + + /* Allocate a new phys region. */ + if(!SLABALLOC(newphysr)) + return ENOMEM; + + /* Set and link the new phys region to the block. */ + newphysr->ph = pb; + newphysr->offset = offset_d; + newphysr->parent = vrd; + newphysr->next_ph_list = pb->firstregion; + pb->firstregion = newphysr; + physr_insert(newphysr->parent->phys, newphysr); + pb->refcount++; + + /* If a COW share was requested but the phys block has already + * been shared as SMAP, give up on COW and copy the block for + * the destination phys region now. + */ + if(flag < 0 && pb->refcount > 1 + && pb->share_flag == PBSH_SMAP) { + if(!(newphysr = map_clone_ph_block(vmd, vrd, + newphysr, NULL))) { + return ENOMEM; + } + } + else { + /* See if this is a COW share or SMAP share. */ + if(flag < 0) { /* COW share */ + pb->share_flag = PBSH_COW; + /* Update the page table for the src process. */ + pt_writemap(&vms->vm_pt, offset_s + vrs->vaddr, + pb->phys, pb->length, + pt_flag, WMF_OVERWRITE); + } + else { /* SMAP share */ + pb->share_flag = PBSH_SMAP; + } + /* Update the page table for the destination process. */ + pt_writemap(&vmd->vm_pt, offset_d + vrd->vaddr, + pb->phys, pb->length, pt_flag, WMF_OVERWRITE); + } + + physr_incr_iter(&iter); + offset_d += pb->length; + offset_s += pb->length; + } + + SANITYCHECK(SCL_FUNCTIONS); + + return OK; +} + +/*===========================================================================* + * unmap_memory * + *===========================================================================*/ +PUBLIC int unmap_memory(endpoint_t sour, endpoint_t dest, + vir_bytes virt_s, vir_bytes virt_d, vir_bytes length, int flag) +{ + struct vmproc *vmd; + struct vir_region *vrd; + struct phys_region *pr; + struct phys_block *pb; + physr_iter iter; + vir_bytes off, end; + int p; + + /* Use information on the destination process to unmap. */ + if(vm_isokendpt(dest, &p) != OK) + panic("unmap_memory: bad endpoint: %d", dest); + vmd = &vmproc[p]; + + vrd = map_lookup(vmd, virt_d); + vm_assert(vrd); + + /* Search for the first phys region in the destination process. */ + off = virt_d - vrd->vaddr; + physr_start_iter(vrd->phys, &iter, off, AVL_EQUAL); + pr = physr_get_iter(&iter); + if(!pr) + panic("unmap_memory: no aligned phys region: %d", 0); + + /* Copy the phys block now rather than doing COW. */ + end = off + length; + while((pr = physr_get_iter(&iter)) && off < end) { + pb = pr->ph; + vm_assert(pb->refcount > 1); + vm_assert(pb->share_flag == PBSH_SMAP); + + if(!(pr = map_clone_ph_block(vmd, vrd, pr, &iter))) + return ENOMEM; + + physr_incr_iter(&iter); + off += pb->length; + } + + return OK; +} + +/*===========================================================================* + * split_phys * + *===========================================================================*/ +PRIVATE int split_phys(struct phys_region *pr, vir_bytes point) +{ + struct phys_region *newpr, *q, *prev; + struct phys_block *newpb; + struct phys_block *pb = pr->ph; +/* Split the phys region into 2 parts by @point. */ + + if(pr->offset >= point || pr->offset + pb->length <= point) + return OK; + if(!SLABALLOC(newpb)) + return ENOMEM; + + /* Split phys block. */ + *newpb = *pb; + pb->length = point - pr->offset; + newpb->length -= pb->length; + newpb->phys += pb->length; + + /* Split phys regions in a list. */ + for(q = pb->firstregion; q; q = q->next_ph_list) { + if(!SLABALLOC(newpr)) + return ENOMEM; + + *newpr = *q; + newpr->ph = newpb; + newpr->offset += pb->length; + + /* Link to the vir region's phys region list. */ + physr_insert(newpr->parent->phys, newpr); + + /* Link to the next_ph_list. */ + if(q == pb->firstregion) { + newpb->firstregion = newpr; + prev = newpr; + } else { + prev->next_ph_list = newpr; + prev = newpr; + } + } + prev->next_ph_list = NULL; + + return OK; +} + +/*===========================================================================* + * clean_phys_regions * + *===========================================================================*/ +PRIVATE void clean_phys_regions(struct vir_region *region, + vir_bytes offset, vir_bytes length) +{ +/* Consider @offset as the start address and @offset+length as the end address. + * If there are phys regions crossing the start address or the end address, + * split them into 2 parts. + * + * We assume that the phys regions are listed in order and don't overlap. + */ + struct phys_region *pr; + physr_iter iter; + + physr_start_iter_least(region->phys, &iter); + while((pr = physr_get_iter(&iter))) { + /* If this phys region crosses the start address, split it. */ + if(pr->offset < offset + && pr->offset + pr->ph->length > offset) { + split_phys(pr, offset); + physr_start_iter_least(region->phys, &iter); + } + /* If this phys region crosses the end address, split it. */ + else if(pr->offset < offset + length + && pr->offset + pr->ph->length > offset + length) { + split_phys(pr, offset + length); + physr_start_iter_least(region->phys, &iter); + } + else { + physr_incr_iter(&iter); + } + } +} + +/*===========================================================================* + * rm_phys_regions * + *===========================================================================*/ +PRIVATE void rm_phys_regions(struct vir_region *region, + vir_bytes begin, vir_bytes length) +{ +/* Remove all phys regions between @begin and @begin+length. + * + * Don't update the page table, because we will update it at map_memory() + * later. + */ + struct phys_region *pr; + physr_iter iter; + + physr_start_iter(region->phys, &iter, begin, AVL_GREATER_EQUAL); + while((pr = physr_get_iter(&iter)) && pr->offset < begin + length) { + pb_unreferenced(region, pr); + physr_remove(region->phys, pr->offset); + physr_start_iter(region->phys, &iter, begin, + AVL_GREATER_EQUAL); + SLABFREE(pr); + } +} + +/*===========================================================================* + * map_memory * + *===========================================================================*/ +PUBLIC int map_memory(endpoint_t sour, endpoint_t dest, + vir_bytes virt_s, vir_bytes virt_d, vir_bytes length, int flag) +{ +/* This is the entry point. This function will be called by handle_memory() when + * VM recieves a map-memory request. + */ + struct vmproc *vms, *vmd; + struct vir_region *vrs, *vrd; + physr_iter iterd; + vir_bytes offset_s, offset_d; + int p; + int r; + + if(vm_isokendpt(sour, &p) != OK) + panic("map_memory: bad endpoint: %d", sour); + vms = &vmproc[p]; + if(vm_isokendpt(dest, &p) != OK) + panic("map_memory: bad endpoint: %d", dest); + vmd = &vmproc[p]; + + vrs = map_lookup(vms, virt_s); + vm_assert(vrs); + vrd = map_lookup(vmd, virt_d); + vm_assert(vrd); + + /* Linear address -> offset from start of vir region. */ + offset_s = virt_s - vrs->vaddr; + offset_d = virt_d - vrd->vaddr; + + /* Make sure that the range in the source process has been mapped + * to physical memory. + */ + map_handle_memory(vms, vrs, offset_s, length, 0); + + /* Prepare work. */ + clean_phys_regions(vrs, offset_s, length); + clean_phys_regions(vrd, offset_d, length); + rm_phys_regions(vrd, offset_d, length); + + /* Map memory. */ + r = do_map_memory(vms, vmd, vrs, vrd, offset_s, offset_d, length, flag); + + return r; +} + /*========================================================================* * map_lookup_phys * *========================================================================*/ @@ -1606,3 +2066,6 @@ map_lookup_phys(struct vmproc *vmp, u32_t tag) return pr->ph->phys; } + + + diff --git a/servers/vm/region.h b/servers/vm/region.h index 5b2c980a8..b0aa9def6 100644 --- a/servers/vm/region.h +++ b/servers/vm/region.h @@ -65,6 +65,7 @@ struct vir_region { #define VR_PHYS64K 0x004 /* Physical memory must be 64k aligned. */ #define VR_LOWER16MB 0x008 #define VR_LOWER1MB 0x010 +#define VR_CONTIG 0x020 /* Must be physically contiguous. */ /* Mapping type: */ #define VR_ANON 0x100 /* Memory to be cleared and allocated */ @@ -79,7 +80,6 @@ struct vir_region { /* map_page_region flags */ #define MF_PREALLOC 0x01 -#define MF_CONTIG 0x02 #endif diff --git a/servers/vm/rs.c b/servers/vm/rs.c index 35337ec5d..623240eca 100644 --- a/servers/vm/rs.c +++ b/servers/vm/rs.c @@ -20,6 +20,7 @@ #include #include #include +#include #include "glo.h" #include "proto.h" diff --git a/servers/vm/sanitycheck.h b/servers/vm/sanitycheck.h index ef98da8ce..f341ff531 100644 --- a/servers/vm/sanitycheck.h +++ b/servers/vm/sanitycheck.h @@ -1,6 +1,8 @@ #ifndef _SANITYCHECK_H #define _SANITYCHECK_H 1 +#include + #include "vm.h" #include "glo.h" @@ -17,15 +19,15 @@ slab_sanitycheck(__FILE__, __LINE__); } #define SANITYCHECK(l) if(!nocheck && ((l) <= vm_sanitychecklevel)) { \ - struct vmproc *vmp; \ + struct vmproc *vmpr; \ vm_assert(incheck == 0); \ incheck = 1; \ usedpages_reset(); \ slab_sanitycheck(__FILE__, __LINE__); \ - for(vmp = vmproc; vmp < &vmproc[VMP_NR]; vmp++) { \ - if((vmp->vm_flags & (VMF_INUSE | VMF_HASPT)) == \ + for(vmpr = vmproc; vmpr < &vmproc[VMP_NR]; vmpr++) { \ + if((vmpr->vm_flags & (VMF_INUSE | VMF_HASPT)) == \ (VMF_INUSE | VMF_HASPT)) { \ - PT_SANE(&vmp->vm_pt); \ + PT_SANE(&vmpr->vm_pt); \ } \ } \ map_sanitycheck(__FILE__, __LINE__); \ @@ -50,22 +52,11 @@ } \ } -#define NOTRUNNABLE(ep) { \ - struct proc pr; \ - if(sys_getproc(&pr, ep) != OK) { \ - panic("VM: sys_getproc failed: %d", ep); \ - } \ - if(!pr.p_rts_flags) { \ - panic("VM: runnable: %d", ep); \ - } \ -} - #else #define SANITYCHECK #define SLABSANITYCHECK(l) #define USE(obj, code) do { code } while(0) #define SLABSANE(ptr) -#define NOTRUNNABLE(ep) #endif #endif diff --git a/servers/vm/slaballoc.c b/servers/vm/slaballoc.c index 4ec4c8182..7072cc372 100644 --- a/servers/vm/slaballoc.c +++ b/servers/vm/slaballoc.c @@ -16,6 +16,7 @@ #include #include +#include #include #include #include @@ -316,8 +317,10 @@ PUBLIC void *slaballoc(int bytes) vm_assert(s); firstused = LH(s, LIST_USED); vm_assert(firstused); +#if SANITYCHECKS vm_assert(firstused->sdh.magic1 == MAGIC1); vm_assert(firstused->sdh.magic2 == MAGIC2); +#endif vm_assert(firstused->sdh.nused < ITEMSPERPAGE(bytes)); for(i = firstused->sdh.freeguess; @@ -502,8 +505,6 @@ PUBLIC void slablock(void *mem, int bytes) SLABDATAUNWRITABLE(f); - FIXME("verify new contents"); - return; } diff --git a/servers/vm/util.h b/servers/vm/util.h index 00b3baf8c..d93ec296d 100644 --- a/servers/vm/util.h +++ b/servers/vm/util.h @@ -10,9 +10,9 @@ #if SANITYCHECKS #define vm_assert(cond) { \ if(vm_sanitychecklevel > 0 && !(cond)) { \ - printf("VM:%s:%d: assert failed: %s\n", \ + printf("VM:%s:%d: vm_assert failed: %s\n", \ __FILE__, __LINE__, #cond); \ - panic("assert failed"); \ + panic("vm_assert failed"); \ } \ } #else diff --git a/servers/vm/vm.h b/servers/vm/vm.h index 71298e389..c711dd617 100644 --- a/servers/vm/vm.h +++ b/servers/vm/vm.h @@ -7,6 +7,9 @@ #define PAF_ALIGN64K 0x04 /* Aligned to 64k boundary. */ #define PAF_LOWER16MB 0x08 #define PAF_LOWER1MB 0x10 +#define PAF_FIRSTBLOCK 0x20 /* alloc_mem: return first block */ + +#define MARK do { if(mark) { printf("%d\n", __LINE__); } } while(0) /* special value for v in pt_allocmap */ #define AM_AUTO ((u32_t) -1) diff --git a/servers/vm/vmproc.h b/servers/vm/vmproc.h index dc840956b..e05147382 100644 --- a/servers/vm/vmproc.h +++ b/servers/vm/vmproc.h @@ -5,6 +5,7 @@ #include #include #include +#include #include "vm.h"