Assignment 3 tips. OS/161 Kernel. Theoretical Typical Address Space Layout. Real R3000 Address Space Layout. Kernel Address Space Layout

Transcription

1 ssignment tips Kernel Stack Virtual ddress Shared Libraries SS (heap) Data Text (Code) 8 0 Theoretical Typical ddress Layout Stack region is at top, and can grow down Heap has free space to grow up Text is typically read-only Kernel is in a reserved, protected, shared region Real R000 ddress Layout : gigabytes TL translated (mapping loaded from page table) Cacheable (depending on N bit) user-mode and kernel mode accessible Page size is K 0xFFFFFFFF kseg Page table per-user address space Switching processes switches the translation (page table) for 0xFFFFFFFF kseg Proc Proc Proc Kernel ddress Layout kseg : megabytes Fixed translation window to physical memory - 0xfffffff virtual = - 0xfffffff physical TL not used Cacheable Only kernel-mode accessible Usually where the kernel is placed Placed in Kseg0 lower part of physical memory meg of physical RM busctl ramsize=, in sys.conf #define PDDR_TO_KVDDR(paddr) ((paddr)+mips_kseg0) void ram_getsize (paddr_t *lo, paddr_t *hi); Physical Memory

2 Implementing alloc_kpage()/free_kpage() The low-level functions that kmalloc()/kfree() use to allocate/free memory in its memory pool. You can assume page at a time ut should return NULL (no memory) if you get a call for more than a page. Consider using an assert() for debugging in case you get such a request ddresses must be in the address range of DUMVM firstpaddr RM/physical memory Memory access 8 Frame Table RM/physical memory How to place my frame table? Memory access struct frame_table_entry *frame_table; location = top_of_ram - (top_of_ram / K * ); /* assume byte entries */ frame_table = (struct frame_table_entry *) location; How can my my allocator work before and after it is intialised? struct frame_table_entry *frame_table = 0; alloc_kpages() { if (ft == 0) { /* use ram_stealmem */ } else { /* use my allocator as frame table is now initialised */ } } Stack Other s? Virtual ddress Data Text (Code) 8 0 KUseg layout Stack region is at top, and can grow down Other regions determined by ELF file see load_elf() number can vary permissions specified also cs-objdump -p testbin/huge

3 thresher% cs-objdump -h../bin/true thresher% cs-objdump -p../bin/true../bin/true: file format elf-tradbigmips../bin/true: file format elf-tradbigmips Sections: Idx Name Size VM LM File off lgn 0.reginfo ** CONTENTS, LLOC, LOD, REDONLY, DT, LINK_ONCE_SME_SIZE.text 00000d b b b0 ** CONTENTS, LLOC, LOD, REDONLY, CODE ** CONTENTS, LLOC, LOD, DT.sbss ** LLOC.bss ** LLOC.comment **0 CONTENTS, REDONLY.pdr 00000a ** CONTENTS, REDONLY.mdebug.abi e0 **0 CONTENTS, REDONLY thresher% Program Header: 0x off 0x vaddr 0x00000 paddr 0x00000 align ** filesz 0x memsz 0x flags r-- LOD off vaddr 0x paddr 0x align ** filesz 0x memsz 0x flags r-x LOD off 0x vaddr 0x paddr 0x align ** filesz memsz 0x flags rwprivate flags = 0: [abi=o] [mips] [not bitmode] thresher% Process Layout Process layout in KUseg regions specified by calls to as_define_() as_define_region() usually implemented as a linked list of region specifications as_prepare_load() make REDONLY regions REDWRITE for loading purposes as_complete_load() enforce REDONLY again 0x other? Process Layout Need to keep translation table for KUSEG Page Table 0x other? 0x x as_create() allocate a structure used to keep track of an address space i.e. regions PT Level proc_getas() used to get access to current address space struct as_destroy() deallocate book keeping and page tables. deallocating frames used as_copy() allocates a new (destination) address space adds all the same regions as source roughly, for each mapped page in source allocate a frame in dest copy contents from source frame to dest frame add PT entry for dest as_activate() flush TL (or set the hardware asid) as_deactivate() flush TL (or flush an asid) 8

4 VM Fault pproximate Flow Chart vm_fault Load TL VM_FULT REDONLY lookup PT Valid Translation Look up region llocate Frame, Zero-fill, Insert PTE Valid EFULT Newly allocated user-level pages are expected to be zero-filled Load TL EFULT kprintf() Do not use it in vm_fault() after the TL write!!!! 0 trace can help with debugging The following additional options control trace's tracing and are ignored by sys: -f tracefile Set the file trace information is logged to. y default, stderr is used. Specifying -f- sends output to stdout instead of stderr. -t traceflags Tell System/ what to trace. The following flags are available: d Trace disk I/O e Trace emufs I/O j Trace jumps and branches k Trace instructions in kernel mode n Trace network I/O t Trace TL/MMU activity u Trace instructions in user mode x Trace exceptions Caution: tracing instructions generates huge amounts of output that may overwhelm smaller host systems. wagner% trace -tt kernel sys: System/ release., compiled Jun 00 ::0 sys: Tracing enabled: tlb trace: tlbp: 800/000 -> : [0] trace: tlbp: 80/000 -> : [] trace: tlbp: 80/000 -> : [] trace: tlbp: 80/000 -> : [] trace: tlbp: 80/000 -> : [] trace: tlbp: 80/000 -> : [] trace: tlbp: 80/000 -> : [] trace: tlbp: 80/000 -> : [] trace: tlbp: 808/000 -> : [8] trace: tlbp: 80/000 -> : [] trace: tlbp: 80a/000 -> : [] trace: tlbp: 80b/000 -> : [] trace: tlbp: 80c/000 -> : [] trace: tlbp: 80d/000 -> : [] trace: tlbp: 80e/000 -> : [] trace: tlbp: 80f/000 -> : [] trace: tlbp: 8/000 -> : [] trace: tlbp: 8/000 -> : [] trace: tlbp: 8/000 -> : [8] trace: tlbp: 8/000 -> : [] trace: tlbp: 8/000 -> : [0] trace: tlbp: 8/000 -> : [].. trace: tlbp: 8c/000 -> : [0] trace: tlbp: 8d/000 -> : [] trace: tlbp: 8e/000 -> : [] trace: tlbp: 8f/000 -> : [] trace: tlbp: 80/000 -> NOT FOUND trace: tlbwi: [ 0] 800/000 -> ==> 80/000 -> trace: tlbp: 8/000 -> NOT FOUND trace: tlbwi: [ ] 80/000 -> ==> 8/000 -> trace: tlbp: 8/000 -> NOT FOUND trace: tlbwi: [ ] 80/000 -> ==> 8/000 -> trace: tlbp: 8/000 -> NOT FOUND trace: tlbwi: [ ] 80/000 -> ==> 8/000 -> trace: tlbp: 8/000 -> NOT FOUND trace: tlbwi: [ ] 80/000 -> ==> 8/000 -> trace: tlbp: 8/000 -> NOT FOUND trace: tlbwi: [ ] 80/000 -> ==> 8/000 -> trace: tlbp: 8/000 -> NOT FOUND trace: tlbwi: [ ] 80/000 -> ==> 8/000 -> trace: tlbp: 8/000 -> NOT FOUND trace: tlbwi: [ ] 80/000 -> ==> 8/000 -> trace: tlbp: 88/000 -> NOT FOUND trace: tlbp: 8/000 -> : [].. trace: tlbp: 8c/000 -> NOT FOUND trace: tlbwi: [0] 8c/000 -> ==> 8c/000 -> trace: tlbp: 8d/000 -> NOT FOUND trace: tlbwi: [] 8d/000 -> ==> 8d/000 -> trace: tlbp: 8e/000 -> NOT FOUND trace: tlbwi: [] 8e/000 -> ==> 8e/000 -> trace: tlbp: 8f/000 -> NOT FOUND trace: tlbwi: [] 8f/000 -> ==> 8f/000 -> OS/ base system version. Copyright (c) 000, 00, 00, 00 President and Fellows of Harvard College. ll rights reserved. Put-your-group-name-here's system version 0 (SST #) Cpu is MIPS r000/r000 k physical memory available Device probe... lamebus0 (system main bus) emu0 at lamebus0

5 dvance ssignment Proc ddress Proc ddress Shared pages and copy-on-write Sbrk() Demand loading and mmap Paging Two (or more) processes running the same program and sharing a section Page Table Physical ddress Page Table sbrk The "break" is the end address of a process's heap region. The sbrk call adjusts the "break" by the amount. It returns the old "break". Thus, to determine the current "break", call sbrk(0). The heap region is initially empty, so at process startup, the beginning of the heap region is the same as the end and may thus be retrieved using sbrk(0). 0x x heap Memorymapped files and paging Memory mapped file K Y X N M L F E D C K L E C X Y F L M N Disk Physical ddress D 8 mmap semantics void *mmap(size_t length, int prot, int fd, off_t offset); int munmap(void *addr);