Lecture: Virtual memory

Preparation

• read the xv6 book: §2, Page tables

Questions

• Exercise: system calls
  • “RFC 3339 format, 2016-9-30T0:23:17Z” - not strictly correct
  • “Oct 6 1:57:11 UTC 2016”
  • “the day of week was not directly provided in rtcdate” - see “Appendix B. Day of the Week” in RFC 3339
  • CMOS
  • network time
  • can we directly call cmostime from date.c? if not, what will prevent us from doing that?
• Lab 2
  • see the tools guide
  • enter/exit QEMU monitor using Ctrl-a c; try info pg

Paging

• pointers: virtual addresses
  • kernel: set up a page table
  • MMU: VA → PA translation & permission checking
  • TLB: cache translation results
  • now you understand what happens when you dereference a pointer, access an object field, etc.
  • basis for process isolation - see xv6
    • per-process page table: pgdir field in struct proc in proc.h
    • switch: switchuvm in vm.c
• x86 page table example: 4KB page size, two-level tree
  • see the xv6 book, Figure 2-1, x86 page table hardware
  • top-level: page directory
    • 1024 entries (PDE)
    • each PDE is 32-bit: address of page table page (20 bits) & flags (12 bits)
  • second-level: page table pages
    • each has 1024 entries (PTE)
    • each PTE is 32-bit: physical page address (20 bits) & flags (12 bits)
  • questions
    • would the physical RAM size affect the size of the page directory
    • how would you implement our QEMU’s info pg in JOS kernel monitor
• x86 programming interface
  • control registers
    • CR0: enable/disable paging
    • CR2: page fault address
    • CR3: physical address of the page directory
  • invlpg: invalidate TLB
    • TLB shootdown (multiprocessor)

Demos in JOS

volatile char *p = (volatile char *)0xcafebeef;
printf("XXX: %x\n", *p);

• invalid read/write
  • add two lines of code in JOS to read an invalid address
  • with unpatched QEMU: reboot
  • with patched QEMU: stop & print registers
    • useful for debugging
    • what are the values of EIP, CR2, CR3
• read address 0 vs. read KERNBASE (0xf0000000)
  • the same value - why?
  • how about write? what makes the difference?
• kernel runs in high virtual addresses
  • why
  • how does the kernel set this up