330 likes | 483 Views
Virtual Memory. Case 1. In a 32-bit machine we subdivide the virtual address into 4 pieces as follows: We use a 3-level page table, such that the first 7 bits are for the first level and so on. Physical addresses are 44 bits and there are 4 protection bits per page. (Byte Alignment). Case 1.
E N D
Case 1 • In a 32-bit machine we subdivide the virtual address into 4 pieces as follows: • We use a 3-level page table, such that the first 7 bits are for the first level and so on. • Physical addresses are 44 bits and there are 4 protection bits per page. (Byte Alignment)
Case 1 • What is the page size in such system? • 4K (212) • How much memory is consumed by the page table and wasted by internal fragmentation for a process that has 128K of memory starting at address 0? • #L3-entry: 28 = 256, entry size: 5byte (44-12+4bits) • #L2-entry: 24 = 16, entry size: 5byte (44-8+4bits) • #L1-entry: 28 = 256 , entry size: 6byte (44-4+4bits)
Case 1 • How much memory is consumed by the page table and wasted by internal fragmentation for a process that has 128K of memory starting at address 0? • 128K = 4K * 32 1*L1 + 1*L2 + 1*L3 • L3: 256 * 5 = 1280 Bytes • L2: 16 * 5 = 80 Bytes • L1: 256 * 6 = 1536 Bytes • Memory: 1280 + 80 + 1536 = 2896 Bytes • Internal: 1*(5*8-36) + 1*(5*8-40) + 32*(6*8-44) = 132bits
Case 2 Part 1 • Assume program “runarun” contains • 2 pages of code and • 2 pages of data, and • libc library contains 1 page of code and 1 page of data, • as shown in figure 1. • Assume we have a machine with • 16 pages of physical memory, and • 64 pages of virtual memory. • User space takes 48 pages • kernel space takes 16 pages.
Case 2 Figure 1
Case 2 • The figure 2 shows • the runtime virtual memory layout of the program • the virtual memory areas • Each shadowed rectangle represents one page. • Fill in the flags of virtual memory areas with “read-only” or “read/write”
r/w r/w r r/w r/w r/w r
Case 2 Part 2 • There are two types of page fault: • major and minor faults. • major page faults occur • when data has to be read from disk • which is an expensive operation; • minor page fault concerns • no disk operation, • the fault could be handled by • just memory read/write/copy operations.
A sample from exam papers • During the lifetime of the program “runarun”, • every of the code and data pages are visited. • The process takes • one bss page, • one heap page • one stack page • During the lifetime of the program, how many pages faults occurred? • Number of major page faults: ________ • Number of minor page fault: ________ text data bss heap stack
A sample from exam papers Part 3 • Assume the system performs no page swapping. • Before the program starts, physical memory looks like the table in Figure 2. • Blacked boxes (PPN:2,9) are already allocated. • Parent process blocks itself and waits after forking, and continue after child exit
Case 2 Part 3 • Program binary and shared library are touched in the sequence of: • Parent process: <page4, page5, stack, page0, page2, bss, heap, fork, page1 > • Child Process: <stack, page1, page3, bss, page0, page2, heap> • Operations for data/bss/stack/heap are write • Don’t consider reap
Case 2 • A simple first-hit memory allocation policy is used. slot_t *find_a_mem_slot() { for(i = 0; i < 16; i++) if (slot_is_free) return slot[i]; if (i == 16) return fail; }
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork PPN:3 stack PPN:3 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:8 heap heap PPN:7 PPN:7 bss bss page3 page3 PPN:6 PPN:6 page2 page2 page1 page1 PPN:4 PPN:4 page0 page0
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork Child: Stack PPN:3 stack PPN:10 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:8 heap heap PPN:7 PPN:7 bss bss page3 page3 PPN:6 PPN:6 page2 page2 page1 page1 PPN:4 PPN:4 page0 page0
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork Child: Stack Page1 PPN:3 stack PPN:10 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:8 heap heap PPN:7 PPN:7 bss bss page3 page3 PPN:6 PPN:6 page2 page2 PPN:11 page1 page1 PPN:4 PPN:4 page0 page0
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork Child: Stack Page1 Page3 PPN:3 stack PPN:10 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:8 heap heap PPN:7 PPN:7 bss bss PPN:12 page3 page3 PPN:6 PPN:6 page2 page2 PPN:11 page1 page1 PPN:4 PPN:4 page0 page0
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork Child: Stack Page1 Page3 Bss PPN:3 stack PPN:10 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:8 heap heap PPN:7 PPN:13 bss bss PPN:12 page3 page3 PPN:6 PPN:6 page2 page2 PPN:11 page1 page1 PPN:4 PPN:4 page0 page0
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork Child: Stack Page1 Page3 Bss Page0 PPN:3 stack PPN:10 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:8 heap heap PPN:7 PPN:13 bss bss PPN:12 page3 page3 PPN:6 PPN:6 page2 page2 PPN:11 page1 page1 PPN:4 PPN:4 page0 page0
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork Child: Stack Page1 Page3 Bss Page0 Page2 PPN:3 stack PPN:10 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:8 heap heap PPN:7 PPN:13 bss bss PPN:12 page3 page3 PPN:6 PPN:14 page2 page2 PPN:11 page1 page1 PPN:4 PPN:4 page0 page0
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork Child: Stack Page1 Page3 Bss Page0 Page2 Heap PPN:3 stack PPN:10 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:15 heap heap PPN:7 PPN:13 bss bss PPN:12 page3 page3 PPN:6 PPN:14 page2 page2 PPN:11 page1 page1 PPN:4 PPN:4 page0 page0
Case 2 • Please fill in the flat page tables with PPN below for both the parent process and the child process just before they finished executing. Parent: Page4 Page5 Stack Page0 Page2 Bss Heap Fork Page1 Child: Stack Page1 Page3 Bss Page0 Page2 Heap PPN:3 stack PPN:10 stack PPN:1 PPN:1 page5 page5 page4 page4 PPN:0 PPN:0 PPN:8 PPN:15 heap heap PPN:7 PPN:13 bss bss PPN:12 page3 page3 PPN:6 PPN:14 page2 page2 PPN:11 PPN:11 page1 page1 PPN:4 PPN:4 page0 page0
Case 3 do_page_fault() { //page fault virtual address address = get_page_fault_addr(); //the reason of the page fault type = get_page_fault_type(); //find the vm area which the address falls in vma = find_vma(address); if (!vma) return fail; if (type == WRITE_PROTECT_FAULT) do_copy_on_write(); //not detailed here
Case 3 /* page not present */ if (vma->type == FILE_PAGE) { //file-backed vm area //load file page into memory and //update the page table load_file(); return success; } else if (vma->type == ANONYMOUS_PAGE) {// anonymous vm area //allocate a zero page and update the page table alloc_zero_page(); return success; } /* unknown type */ return fail; }
Case 3 find_vma(address) { //get the list head of the vm area vma = get_head_vma(); while(vma) { if (address > vma->vm_end) return NULL; if (address >= vma->vm_start && address <= vma->vm_end) return vma; vma = vma->vm_next; } return NULL; }
Case 3 • The page fault handler is buggy. • If the stack grows down one page, the page fault handler is likely to fail. • Why?
Case 4 • The following program executes on Pentium/Linux #define PAGE_SIZE 4 * 1024 #define BUF_SIZE 128*PAGE_SIZE int main(void) { char *p = NULL; int i; A: p = mmap(0, 128 * PAGE_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, 0, 0); printf("buffer start: %p\n", p); for(i = 0; i < BUF_SIZE; i += PAGE_SIZE) p[i] = 1; B: munmap(p, BUF_SIZE); return; }
Case 4 • When the program arrives at label A, the page table is like below. The number within block is the offset of PDE/PTE. The output of the function printf is “buffer start: 0xb7bdf000”. • Please write the page table like above, when the program arrives at label B? • NOTE: the white block without number means empty PDE/PTE, the white block with number means a filled PDE/PTE, the black block means multiple filled PDE/PTEs, and the grey block means a page with some data/code
Case 4 Start 0xb7bdf000 L1: 2DE L2: 3DF +128K End 0xb7c5f000 L1: 2DF L2: 05F