Assignment 3: Virtual Memory
- Due Dates and Mark Distribution
- Setting Up
- Tutorial Exercises
- Coding Assignment
- Advanced Assignment
1. Due Dates and Mark Distribution
Due Date: 23:59, Friday 6th June
Extension Date: 23:59, Monday 9th June
Marks: The base assignment is worth 30 marks (of the 100 available for the class mark component of the course)
The 10% bonus for one week early applies.
The advanced assignment is assessable for COMP3891/COMP9283 students. Only a subset of the advanced assignment is required. Marks are capped at 5 marks of the 100 class marks available. The paging component is for extra bonus marks only.
COMP3231/COMP9201 students can do the advanced part with the permission of the lecturer, and only if basic assignment is completed a week prior to the deadline. Marks obtained are added to any shortfall in the participation mark component. The bonus marks are also available.
Any bonus awarded for the advanced part can be used to make up for lost marks in any class mark component of the course.
The familiarisation questions contained herein are the subject of your week 11 tutorial. Please answer the questions and bring them to your tutorial.
Note: As noted in previous email and announcements, there is a video outline of how to approach the assignment in the week 11 extended lecture.
2. IntroductionIn this assignment you will implement the virtual memory sub-system of OS/161. The existing VM implementation in OS/161, dumbvm, is a minimal implementation with a number of shortcomings. In this assignment you will adapt OS/161 to take full advantage of the simulated hardware by implementing management of the MIPS software-managed Translation Lookaside Buffer (TLB). You will write the code to manage this TLB. You will also write code to manage system memory.
If you attempt the advanced portion of this assignment, you will implement paging, the mechanism by which memory pages of an active process can be sent to disk when memory is needed, and restored to memory when required by the program.
The System/161 TLBIn the System/161 machine, each TLB entry includes a 20-bit virtual page number and a 20-bit physical page number as well as the following five fields:
- global: 1 bit; if set, ignore the PID bits in the TLB.
- valid: 1 bit; set if the TLB entry contains a valid translation.
- dirty: 1 bit; enables writing to the page referenced by the entry; if this bit is 0, the page is only accessible for reading.
- nocache: 1 bit; unused in System/161. In a real processor, indicates that the hardware cache will be disabled when accessing this page.
- pid: 6 bits; a context or address space ID that can be used to allow entries to remain in the TLB after a context switch.
For this assignment, you may ignore the pid field. Note, however, that you must then flush the TLB on a context switch (why?).
The System/161 Virtual Address Space MapThe MIPS divides its address space into several regions that have hardwired properties. These are:
- kseg2, TLB-mapped cacheable kernel space
- kseg1, direct-mapped uncached kernel space
- kseg0, direct-mapped cached kernel space
- kuseg, TLB-mapped cacheable user space
The top of kuseg is 0x80000000. The top of kseg0 is 0xa0000000, and the top of kseg1 is 0xc0000000.
The memory map thus looks like this:
Address Segment Special properties 0xffffffff kseg2 0xc0000000 0xbfffffff kseg1 0xbfc00180 Exception address if BEV set. 0xbfc00100 UTLB exception address if BEV set. 0xbfc00000 Execution begins here after processor reset. 0xa0000000 0x9fffffff kseg0 0x80000080 Exception address if BEV not set. 0x80000000 UTLB exception address if BEV not set. 0x7fffffff kuseg 0x00000000
3. Setting Up Assignment 3
You will (again) be setting up the SVN repository that will contain your code. Remember to use a 3231 subshell (or continue using your modified PATH) for this assignment, as outlined in ASST0.
Only one of you needs to do the following. We suggest your partner sit in on this part of the assignment.
Check your umask is still set appropriately.
% umask 0007If not, you have done something wrong with your umask setup and need to fix it.
Now import the assignment sources.
% cd /home/cs3231/assigns % svn import asst3/src file:///home/osprjXXX/repo/asst3/trunk -m "Initial import"
Make an immediate branch of this import for easy reference when you generate your diff:
% svn copy -m "Tag initial import" file:///home/osprjXXX/repo/asst3/trunk file:///home/osprjXXX/repo/asst3/initial
You have now completed setting up a shared repository for both partners.
Checking out a working copyThe following instructions are now for both partners.
We'll assume from your cs3231 directory is intact from the previous assignment. Change to your directory.
% cd ~/cs3231
Now checkout a copy of the os161 sources to work on from your shared repository.
% svn checkout file:///home/osprjXXX/repo/asst3/trunk asst3-src
You should now have a asst3-src directory to work on.
UPDATED: You should now apply a patch that fixes a synchronisation bug in the original release. Download the following file into your home directory (or some location): asst3_fix.diff
% cd ~/cs3231/asst3-src % patch -p2 < ~/asst3_fix.diff
You will also need to increase the amount of physical memory to run some of the provided tests. Update ~/cs3231/root/sys161.conf so that the ramsize is as follows
31 mainboard ramsize=16777216 cpus=1Or, download a fresh, appropriately configured, version from here, and install it.
Configure OS/161 for Assignment 3
Remember to set your PATH environment variable as in previous assignments (e.g. run the 3231 command).
Before proceeding further, configure your new sources, and build and install the user-level libraries and binaries.
% cd ~/cs3231/asst3-src % ./configure % bmake % bmake install
You have to reconfigure your kernel before you can use the framework provided to do this assignment. The procedure for configuring a kernel is the same as before, except you will use the ASST3 configuration file:
% cd ~/cs3231/asst3-src/kern/conf % ./config ASST3
You should now see an ASST3 directory in the compile directory.
Building for ASST3
When you built OS/161 for ASST0, you ran bmake from compile/ASST0. When you built for ASST1, you ran bmake from compile/ASST1 ... you can probably see where this is heading:
% cd ../compile/ASST3 % bmake depend % bmake % bmake install
If you now run the kernel as you did for previous assignments, you should get to the menu prompt. If you try and run a program, it will fail with a message about an unimplemented feature (specifically refering to the as_* functions you must write). For example, run p /bin/true at the OS/161 prompt to run the program /bin/true in ~/cs3231/root.
You are now ready to start the assignment.
4. Tutorial ExercisesPlease answer the following questions and bring them to your tutorial in week 11. You should be familiar enough with navigating the kernel source that you can find the answers to the below questions by yourself (Hint: use the grep utility). You may also find the MIPS r3000 reference useful.
What is the difference between the different MIPS address space segments? What is the use of each segment?
What functions exist to help you manage the TLB? Describe their use. (Hint: look in kern/arch/mips/include/tlb.h)
What macros are used to convert from a physical address to a kernel virtual address?
What address should the initial user stack pointer be?
What are the entryhi and entrylo co-processor registers? Describe their contents.
What do the as_* functions do? Why do we need as_prepare_load() and as_complete_load()?
What does vm_fault() do? When is it called?
Assuming a 2-level hierarchical page table (4k pages), show for the following virtual addresses:
- The page number and offset;
- the translated address (after any page allocation); and
- the contents of the page table after the TLB miss.
The page table is initially empty, with no L2 pages. You may assume that the allocator returns frames in order, so that the first frame allocated is frame 0, then frames 1, 2, 3, etc.
5. Coding AssignmentThis assignment involves designing and implementing a number of data-structures. Before you start, you should work out what data you need to keep track of, and what operations are required.
Like previous assignments, you are required to submit a small design doc that identifies the major issues you tackled in this assignment, and also describes your solutions to these issues.
The document will be used to guide our markers in their evaluation of your solution to the assignment. In the case of a poor results in the functional testing combined with a poor design document, we will base our assessment on these components alone. If you can't describe your own solution clearly, you can't expect us to reverse engineer the code to a poor and complex solution to the assignment.
Place your design document in design.txt (which we have created for you) at the top of the source tree to OS/161 (i.e. in ~/cs3231/asst3-src/design.txt).
When you later commit your changes into your repository, your design doc will be included in the commit, and later in your submission.
Also, please word wrap you design doc if your have not already done so. You can use the unix fmt command to achieve this if your editor cannot.
Memory ManagementThis assignment requires you to keep track of physical memory. The current memory management implementation in dumbvm never frees memory; your implementation should handle both allocation and freeing of frames.
You will need a frametable containing information about the memory available in the system. In the basic assignment you will need to keep track of whether a frame is used or not. In the advanced part, you will need to keep track of reference statistics and other information about the frames.
The functions that deal with memory are described in kern/include/vm.h. You may assume that only one page will be allocated at a time—designing a page allocator that can allocate multiple pages at a time is surprisingly tricky. However, make sure that you never allocate memory (though kmalloc) that is larger than a page!
Note that alloc_kpages() should return the virtual address of the page, i.e., an address in kseg0.
Warning: alloc_kpages() can be called before vm_bootstrap(). The means that your implementation of alloc_kpages() must work before your frametable is initialised. You should just call ram_stealmem() if the frametable hasn't been initialised.
Address Space ManagementOS/161 has an address space abstraction, the struct addrspace. To enable OS/161 to interact with your VM implementation, you will need to fill in the functions in kern/vm/addrspace.c. The semantics of these functions is documented in kern/include/addrspace.h.
You may use a fixed-size stack region (say 16 pages) for each process.
Address TranslationThe main goal for this assignment is to provide virtual memory translation for user programs. To do this, you will need to implement a TLB refill handler. You will also need to implement a page table. For this assignment, you will implement a 2-level hierarchical page table.
Note that a hierarchical page table is a lazy data-structure. This means that the contents of the page table, including the second level pages, are only allocated when they are needed. You may find allocating the required pages at load time helps you start your assignment, however, your final solution should allocate pages only when a page-fault occurs.
The following questions may assist you in designing the contents of your page table
- What information do you need to store for each page?
- How does the page table get populated?
Testing and Debugging Your AssignmentTo test this assignment, you should run a process that requires more virtual memory than the TLB can map at any one time. You should also ensure that touching memory not in a valid region will raise an exception. The huge and faulter tests in testbin may be useful.
Apart from GDB, you may also find the trace161 command useful. trace161 will run the simulator with tracing, for example
% trace161 -t t -f outfile kernelwill record all TLB accesses in outfile.
HintsHave a close look at the dumbvm implementation, especially vm_fault(). Although it is simple, you should get an idea on how to approach the rest of the assignment.
One approach to implementing the assignment is in the following order:
- Understand how a page table works, and its relationship with the TLB.
- Understand the specification and the supplied code.
- Work out a basic design for you implementation.
You might start simply by using a different data structure initially to keep your code and data structures simple.
- Implement the TLB exception handlers in vm.c using this simplified page table. Note, your final solution should use a 2-level page table!
- Implement the functions in kern/vm/addrspace.c that are required for basic functionality (e.g. as_create(), as_prepare_load(), etc.). Allocating user pages in as_define_region() may also simplify your assignment.
- Test and debug this. Use the debugger!
If you really get stuck, submit at least this much of the solution, and you should get some marks for it.
- Understand how the page table works.
- Decide exactly what data structures you need.
- Work out the design for the proper solution, using the 2-level page table.
- Modify your implementation to include the 2-level page table.
- Write routines in kern/vm/frametable.c to manage free frames and copy pages. Modify kern/vm/vm.c to create and delete page table entries, and keep the TLB consistent with the page table.
- Use these routines to finish the functions in kern/vm/addrspace.c.
- Test and debug.
Alternatively, one could start on the frame table, and alloc_kpage() and free_kpage(). Note that alloc_kpage() and free_kpage() are the underlying routines used kmalloc() and kfree(), so ensure your implementation is correct. A broken alloc_kpage() will result in a broken kmalloc() and an unpredicably broken OS/161!
6. Basic Assignment SubmissionAs with the previous assignments, you again will be submitting a .diff file containing your solution.
You should first commit your changes back to the repository using the following command. Note: You will have to supply a comment on your changes. You also need to coordinate with your partner that the changes you have (or potentially both have) made are committed consistently by you and your partner, such that the repository contains the work you want from both partners.
% cd ~/cs3231/asst3-src % svn commit
The above command will commit any changes you have made to your shared repository.
Now generate a .diff
% svn diff file:///home/osprjXXX/repo/asst3/initial file:///home/osprjXXX/repo/asst3/trunk >~/asst3.diff
Beware! If you have created new files for this assignment, they will not be included in your submission unless you add them, using svn add:
% svn add filename.cIf you add files after running svn commit, you will need to run svn commit again.
Testing Your SubmissionLook here for information on testing and resubmitting your assignment.
Submitting Your AssignmentNow submit the diff as your assignment.
% cd ~ % give cs3231 asst3 asst3.diffYou're now done.
Even though the generated .diff file should represent all the changes you have made to the supplied code, occasionally students do something "ingenious" and generate non representative output.
We strongly suggest keeping your checkout intact to allow for recovery of your work if need be.
7. Advanced AssignmentThe advanced assignment is for COMP3891/COMP9283 students and consist of a student-chosen subset of the problem below, and capped at 5 marks.
COMP3231/COMP9201 students can do the advanced part with the permission of the lecturer, and only if basic assignment is completed a week prior to the deadline. Marks obtained are added to any shortfall in the participation mark component. The limited bonus marks are also available.
- (easy) 3 marks Shared pages and copy-on-write.
- (easy) 2 marks Implement sbrk() to enable user-level malloc() to function.
(hard) 5 marks.
- Implement a simplified mmap()
and munmap(). Note: you only need to support the
simplified case of mapping a file, and munmap() the entire
region that was mapped.
The prototypes are expected to be
void *mmap(size_t length, int prot, int fd, off_t offset); int munmap(void *addr);Where prot can be PROT_READ and/or PROT_WRITE. Compared to traditional mmap, there are no flags, and the OS chooses the virtual address to locate the region. You must ensure that applications can open a file, updated it, and have the updated file propagate to the filesystem.
- And, implement demand-loading. You should load pages only when they are referenced by the user process, as opposed to at process creation.
- Implement a simplified mmap() and munmap(). Note: you only need to support the simplified case of mapping a file, and munmap() the entire region that was mapped. The prototypes are expected to be
- Bonus section. (harder) 3 marks Implement paging. You should implement some page replacement algorithm and demonstrate your solution running under memory pressure.
Advanced Assignment Submission
Submission for the advanced assignment is similar to the basic assignment, except the advance component is given to a distinguished assignment name: asst3_adv Again, you need to generate a diff based on the original source tree.
% cd ~/cs3231/asst3-src % svn commit
Now generate a file containing the diff. (NOTE: How this works depends on how you have set up branches in svn.)
% cd ~ % svn diff file:///home/osprjXXX/repo/asst3/initial file:///home/osprjXXX/repo/asst3/adv >~/asst3_adv.diffSubmit your solution
% cd ~ % give cs3231 asst3_adv asst3_adv.diff
You're now done.
FAQ and Gotchas
See https://wiki.cse.unsw.edu.au/cs3231cgi/2014s1/Asst3 for an up to date list of potential issues you might encounter.