First complete draft.

diff --git a/ta-advice/HW3 b/ta-advice/HW3
index 6143c24bd7674465746be26aadadb19a9d7ad4f5..c0860c8e8d8211d57cc8d72a996647de094d14f7 100644 (file)
--- a/ta-advice/HW3
+++ b/ta-advice/HW3
@@ -61,8 +61,9 @@ A1:
 
 A2:
 
-    The answer should describe code, not data; the data structures
-    have already been explained in A1.
+    The answer generally boils down to traversing the data structure
+    from A1.  The answer should describe code, not data; the data
+    structures have already been explained in A1.
 
 A3:
 
@@ -113,8 +114,6 @@ A5:
     easier to find all the pages that belong to a given process (for
     cleanup when a process dies).
 
-Check the submitted code, by hand, for the following: 
-
                       PAGING TO AND FROM DISK
                       =======================
 
@@ -255,14 +254,16 @@ B8:
     Several possible answers:
 
        - Pre-lock all the pages that the system call will access,
-          then unlock them when the system call completes.
+          then unlock them when the system call completes.  There must
+          be a discussion of how to avoid deadlock among processes
+          that, among them have all of the user pool locked, and want
+          more.
 
        - Page fault in kernel.
 
-       - Acquire a global lock needed to evict pages.  (But there
-          should be some additional explanation of how the system call
-          makes sure that those pages are in main memory in the first
-          place.)
+       - Acquire a global lock needed to evict pages.  There should
+          be some additional explanation of how the system call makes
+          sure that those pages are in main memory in the first place.
 
 B9:
 
@@ -270,9 +271,8 @@ B9:
     to claim that the conditions for deadlock are "rare" or that the
     window in which deadlock can occur is "short" are not acceptable.
     In a proper design, deadlock is impossible, not just rare.  Deduct
-    points.
-
-Check the submitted code, by hand, for the following:
+    points.  (This deduction is listed under B5 because similar claims
+    are common there.)
 
                         MEMORY MAPPED FILES
                         ===================
@@ -293,7 +293,7 @@ C2:
     For some reason, students sometimes talk about permanently
     assigning or pre-reserving memory mapped files to swap slots or to
     frames in physical memory here.  That's incorrect.  The assignment
-    is explicit that mapping should be lazily loaded and written back
+    is explicit that mappings should be lazily loaded and written back
     to their files, not to swap.
 
 C3:

diff --git a/ta-advice/hw3.txt b/ta-advice/hw3.txt
new file mode 100644 (file)
index 0000000..48169d8
--- /dev/null
+++ b/ta-advice/hw3.txt
@@ -0,0 +1,113 @@
+HW3 DESIGN SUMMARY
+==================
+
+OVERALL
+
+  DOCUMENTATION
+    -20 Egregious violation of style versus design document example
+    -10 Numerous excessively long lines (> 79 characters)
+    -10 Numerous capitalization, punctuation, spelling, or grammar errors
+
+  DESIGN
+    -10 Failure to check return value of malloc()
+    -10 Use of ASSERT to check something that can actually fail, e.g. malloc()
+
+  CODING STYLE
+    -10 Inconsistent or bad coding style: no indentation, cramming
+        many statements into one line, other issues at TA's discretion
+    -10 Numerous very long source code lines (> 100 characters)
+    -10 Commented-out or #if'd out code makes real code hard to read
+    -10 Many missing comments on structure, structure members,
+        global or static variables, or function definitions
+    -10 Function(s) should be decomposed for clarity [indicate function]
+    -10 Cut-and-pasted code should be made into function [indicate where]
+    -10 Uninformative or deceptive identifiers
+
+  EXTRA CREDIT
+     +5 Sharing
+
+Subtotal: /10 (not capped)
+
+PROBLEM 1: PAGE TABLE MANAGEMENT
+
+  DOCUMENTATION
+    -25 Grossly inaccurate: documentation has no resemblance to code
+    -13 Important inaccuracies: documentation and code differ significantly
+        [indicate how]
+     -5 Minor inaccuracies: documentation and code differ [indicate how]
+     -5 A1: Missing entirely/missing comments or purpose/too long
+     -5 A2: Missing/non-responsive/too long/too short
+     -5 A3: Missing/non-responsive/too long/too short
+     -5 A3: Sharing implemented, but no explanation of how accessed
+        and dirty bits are coordinated
+     -5 A3: Sharing not implemented, but explanation states or implies
+        that a page can have multiple aliases
+     -5 A4: Missing/non-responsive/too long/too short
+     -5 A5: Missing/non-responsive/too long/too short
+
+  DESIGN
+    -30 Not implemented
+    -10 Page dirty status must be determined accurately, not approximately
+
+Subtotal: /25 (capped at 0)
+
+PROBLEM 2: PAGING TO AND FROM DISK
+
+  DOCUMENTATION
+    -45 Grossly inaccurate: documentation has no resemblance to code
+    -23 Important inaccuracies: documentation and code differ significantly
+        [indicate how]
+     -5 Minor inaccuracies: documentation and code differ [indicate how]
+     -5 B1: Missing entirely/missing comments or purpose/too long
+     -5 B2: Missing/non-responsive/too long/too short
+     -5 B3: Missing/non-responsive/too long/too short
+     -5 B4: Missing/non-responsive/too long/too short
+     -5 B4: Heuristic does not make sense
+     -2 B4: Claim that "struct intr_frame" is on user stack
+     -5 B5: Missing/non-responsive/too long/too short
+     -5 B5: Claim that deadlock is "rare" or has a "short" window
+     -5 B6: Missing/non-responsive/too long/too short
+     -3 B6: Didn't answer first part of question
+     -3 B6: Didn't answer second part of question
+     -3 B5: Disables interrupts
+     -5 B7: Missing/non-responsive/too long/too short
+     -5 B8: Missing/non-responsive/too long/too short
+     -3 B8: When pre-locking of pages is used, no discussion of
+        deadlock prevention
+     -3 B8: When global eviction lock is used, no discussion of
+        how pages are brought into main memory
+     -5 B9: Missing/non-responsive/too long/too short
+
+  DESIGN
+    -45 Not implemented
+     -5 Clock algorithm implementation literally uses a timer
+    -10 Clock algorithm implementation resets dirty bits
+     -5 Stack fault heuristic allows more than 32 bytes below stack pointer
+     -3 Stack limit is smaller than 1 MB
+
+Subtotal: /45 (capped at 0)
+
+PROBLEM 3: MEMORY-MAPPED FILES
+
+  DOCUMENTATION
+    -20 Grossly inaccurate: documentation has no resemblance to code
+    -10 Important inaccuracies: documentation and code differ significantly
+        [indicate how]
+     -5 Minor inaccuracies: documentation and code differ [indicate how]
+     -5 C1: Missing entirely/missing comments or purpose/too long
+     -5 C2: Missing/non-responsive/too long/too short
+     -5 C3: Missing/non-responsive/too long/too short
+     -2 C3: No mention of stack expansion
+     -2 C3: Claim that memory mappings in different processes can "overlap"
+     -5 C4: Missing/non-responsive/too long/too short
+
+  DESIGN
+    -20 Not implemented
+     -5 Pre-reserves or permanently assigning file pages to frames
+     -5 Pre-reserves or permanently assigning file pages to swap slots
+     -5 Global counter for mapid_t lacks synchronization
+     -5 Global table of file mappings lacks synchronization
+     -5 Global table allows multiple processes to access same file mapping
+     -5 Superfluous locking on per-thread data structure [which]
+
+Subtotal: /20 (capped at 0)