Last updated 03/13/2023.
Despite our best efforts to create a book with zero defects, our vigilant
readers have pointed out some bugs. Please report any new errata to
Randy Bryant and Dave O'Hallaron.
Note that some of these errors have been corrected in more recent
printings.
We maintain errata for the North American edition.
Errata for editions in other languages can be found as follows:
Note on the Global Edition: Unfortunately, the publisher
arranged for the generation of a different set of practice and
homework problems in the global edition. The person doing this didn't
do a very good job, and so these problems and their solutions have
many errors. We have not created an errata for this edition.
North American Edition (ISBN-10: 0-13-409266-X)
- Preface
- Chapter 1: A Tour of Computer Systems
- Chapter 2: Representing and Manipulating Information
- p. 45, code for show_bytes.
Variable i should be declared to have type size_t.
Posted 07/11/2015. Randal E. Bryant
- p. 47, aside “New to C? Formatted printing with printf,” second paragraph.
The referenced data type should be int32_t, not int_32t.
Posted 05/23/2016. Yili Gong
- p. 48, aside “New to C? Pointer creation and dereferencing,” fourth line.
The third pointer data type should be void *, not void **.
Posted 02/09/2013. Yili Gong
- p. 59, Aside at top of page.
The second-to-last sentence should
state: “This behavior is not guaranteed for C programs, however,
and so shift amounts should be kept less than the number of bits in
the value being shifted.”
Posted 09/28/2015. Ruth Anderson
- p. 71, first full paragraph, line 6.
The range should be 0 ≤ x ≤ Umaxw.
Posted 10/25/2015. Shoeb Mohammed
- p. 82, second line.
′x should be
x′.
Posted 10/25/2015. Shoeb Mohammed
- p. 82, third line. Value x should be computed with function
B2Uw(x), not B2Tw(x).
Posted 11/13/2015. Parinya Suparit
- p. 82, first equation in DERIVATION. The left-hand side should be computed with function
B2Uw(x), not B2Tw(x).
Posted 11/13/2015. Parinya Suparit
- p. 93, Figure 2.26.
In the axis labels, the minus signs are incorrectly printed as the numeral ‘2’.
A correct version of the image can be found here.
Posted 09/23/2016. Matt Toups
- p. 111, Practice Problem 2.45.
The entry in the third row should be 25/16.
Posted 11/23/2015. Mathieu Borderé
- p. 125, Line before Practice Problem 2.54.
The number shown should be -2147483648 rather than -21483648.
Posted 09/05/2016. JiaSheng Chen
- p. 153, Solution to Problem 2.32.
The sentence starting on third line should state
“In fact, the opposite is true: tsub_ok(x, TMin) should yield 1 when
x is negative and 0 when it is nonnegative.”
Posted 10/25/2015. Shoeb Mohammed
- p. 154, Solution to Problem 2.35, second line of part 3.
It would be more precise to state
|r| < |x| ≤ 2w-1.
(Note that the argument only requires
|r| < 2w,
and so the original statement
and proof are valid.)
Posted 07/12/2018. Wang Lei
- Chapter 3: Machine-Level Representation of Programs
- p. 171, second-to-last line.
248 is 256 terabytes, not 64 terabytes.
Posted 09/09/2016. Don Bagert
- p. 174, code annotation at top of page.
It should read “Disassembly of function multstore in binary file mstore.o.”
Posted 10/25/2015. Shoeb Mohammed
- p. 175, code annotation in middle of page.
It should read “Disassembly of function multstore in binary file prog.”
Posted 10/25/2015. Shoeb Mohammed
- p. 175, Paragraph starting “This code is almost identical” in middle of page.
The end of the sentence should read “disassembly of mstore.o.”
Posted 09/03/2020. Weicheng Pei
- p. 177, aside “ATT versus Intel assembly-code
formats,” last sentence of first paragraph.
The assembly code shown is for the function multstore, not sum.
Posted 11/09/2015. Shoeb Mohammed
- p. 179, Section 3.4, first paragraph.
The ranges given for the eight named registers for the three different instruction sets
are incorrect. They should be:
- For 8086: %ax through %sp.
- For IA32: %eax through %esp.
- For x86-64: %rax through %rsp.
Posted 07/19/2017. Yili Gong
- p. 183, sample code near bottom of page, line 4. The code should be “movb $-17, (%rsp).”
Posted 10/24/2015. Jason Waterman
- p. 184, (Clarification, not an erratum) Figure 3.5.
Although there is an instruction movzbq, the GCC compiler
typically generates the instruction movzbl for this
purpose, relying on the property that an instruction generating a 4-byte with a register as destination
will fill the upper 4 bytes of the register with zeros.
Posted 04/27/2018. Randal Bryant
- p. 188, aside “Some examples of pointers,”
call to printf. The format string should be
"a = %ld, b = %ld\n".
Posted 11/09/2015. Shoeb Mohammed
- p. 191, second full paragraph. The first sentence should state:
“The third column of Figure 3.9 illustrates the effect of executing the instruction popq %rdx …”
Posted 09/28/2015. Max Ma
- p. 191, fourth line of paragraph that begins “The third column of Figure 3 …”
It should state that the value 0x123 remains at memory location 0x100.
Posted 11/09/2015. Carlos Galdino
- p. 198, first line of first full paragraph. There should be a period after the word “forms”.
Posted 09/19/2016. Anise Ghorbani
- p. 199, third full paragraph. The reference to instruction idivl should be to idivq instead.
Posted 02/21/2017. Changan Wang
- p. 199, fourth full paragraph. The paragraph should start with “For most applications of 64-bit division …”
Posted 09/19/2015. Xingda Zhai
- p. 200, first line of text.
It should state “In this code, argument qp must first be saved in a different register (line 2), …”
Posted 08/16/2015. Dmitry Neverov
- p. 201, Second paragraph of Section 3.6.1. The rule for setting the CF flag when subtraction is performed requires special consideration. If t=a-b, then the flag will be set when (unsigned) t > (unsigned) a.
Posted 10/13/2021. Tung Luu
- p. 201, 9 lines from bottom.
For the shift instruction, it should state: “The overflow flag is changed only when the shift amounts is 1,
following rules that depend on the shift type.”
Posted 10/30/2017. Jiaheng Wang
- p. 204, Practice Problem 3.13, second paragraph.
It should state “Suppose a is in some portion of %rdi while …”
Posted 10/15/2015. Ronald Greenberg
- p. 205, Practice Problem 3.13, code for D.
The second instruction should be “setne %al.”
Posted 11/09/2015. Carlos Galdino
- p. 214, third paragraph.
The second sentence should be “The function computes the
absolute value of the difference of its two arguments x
and y, …”
Posted 11/11/2015. Yili Gong
- p. 214, Figure 3.17(c).
The comment on line 8 of the code should state “Return rval”.
Posted 03/18/2017.
- p. 218, assembly code for cread.
In the annotations for lines 3 and 5, “x”
should be “xp”.
Posted 11/19/2015. Vlad Buslov
- p. 226, last sentence before Practice Problem 3.25. It should
state “The compiler has determined that the loop can only be
entered when n > 1, and that, as n is decremented,
it will reach value 1 before reaching a value less than 1.”
Posted 06/11/2017. Daniel O'Brien
- p. 234, first and second lines of text.
They should state “ … indexed by register %rsi, which holds … ”
Posted 09/02/2015. Dmitry Neverov
- p. 235, Figure 3.23, annotation for line 5 of code.
It should state “Goto *jt[index].”
Posted 12/25/2015. Xinzhen Chen
- p. 238, Problem 3.31, Comments in assembly code at top of page.
The register allocation is listed incorrectly.
It should state
“a in %rdi,
b in %rsi,
c in %rdx,
dest in %rcx.”
Posted 07/25/2015. Yu Zhong and Carlos Galdino
- p. 241, fourth and fifth lines.
They should state “Procedure P can pass up to six integral values (i.e., pointers and integers) in registers … ”
Posted 09/02/2015. Dmitry Neverov
- p. 243, Figure 3.27a, line 2.
The annotation should state “L1: y+2.”
Posted 02/15/2016. Xinzhen Chen
- p. 244, first full paragraph.
The value 0x400054e should be 0x40054e.
Posted 07/26/2017. Daniel O'Brien
- p. 244, second full paragraph.
The value 0x4000560 should be 0x400560.
Posted 07/26/2017. Daniel O'Brien
- p. 244, Comment after line 4.
It should state “Disassembly of first(long x)”
Posted 09/02/2015. Dmitry Neverov
- p. 249, Fourth line from bottom. Sentence ending “… and 17 (s3).” should be
“… and 17 (x4).”
Posted 11/29/2016. Hyun Chun
- p. 252, Practice Problem 3.34.
The first two sentences should state “Consider a
function P, which generates local values,
named a0–a7. It then calls
function Q, which has no arguments.”
Posted 11/23/2015. Vlad Buslov
- p. 257, Table near the bottom of the page.
The assembly code for the first entry should be “movq %rdx,%rax.”
Posted 10/03/2015. Zhiwei Xin
- p. 259, Figure 3.36
The fourth block from the bottom should have the label A[3][2].
Posted 04/04/2017. Dean Kajmakci
- p. 259, assembly code in middle of page, line 3.
The annotation should state: “Read from M[xA + 12i + 4j].”
Posted 10/25/2015. Karan Dwivedi
- p. 263, assembly code in middle of page, line 3.
There is a missing ‘)’ at the end of the annotation.
Posted 02/11/2016. Elizabeth White
- p. 264, Annotations of last block of assembly code.
It should state that j is in register %rdx.
Posted 07/26/2017. Daniel O'Brien
- p. 266, middle of page. The declaration and initialization of r should
be:
struct rect r = { 0, 0, 10, 20, 0xFF00FF };
Posted 08/04/2015. Yu Zhong
- p. 277, first bullet, second and third lines after declarations.
They should state “… if the object has type T, then the pointer has type T *.”
Posted 09/02/2015. Dmitry Neverov
- p. 281, Figure 3.40.
The first sentence of the caption should read “Character array buf is just below part of the saved state.”
Posted 12/01/2015. Yili Gong
- p. 282, Table with heading “Characters typed.”
The second entry should have the range 8–23.
Posted 11/6/2017. Weicheng Pei
- p. 282, Practice Problem 3.46, second paragraph.
It should state that get_line is called with return address 0x400076.
Posted 11/23/2015. Vlad Buslov
- p. 284, Practice Problem 3.46, part E.
The question for ask for three things wrong with the code.
Posted 07/26/2017. Daniel O'Brien
- p. 293, Practice Problem 3.49, first paragraph.
The third line should state “… let s1 denote …” (delete “let us”).
Posted 12/27/2020. Ashwin Najappa
- p. 293, Practice Problem 3.49, second paragraph.
The text should state: “… there may be an offset of e2 bytes between the values of s2 and p.”
Posted 11/09/2015. Shoeb Mohammed
- p. 298. Line 11 of paragraph starting with “The vunpcklps instruction …”.
It should read “converting x0 to double precision.”
Posted 02/24/2022. Jinbeom Hong
- p. 295, Figure 3.45. The lower 128 bits of register %ymm13 should be labeled %xmm13.
Posted 10/28/2015. David Hirschv
- p. 300, Problem 3.51, second line.
The return value has type “dest_t.”
Posted 10/13/2016. Anise Ghorbani
- p. 301, Problem 3.51. The first entry should be
“vcvtsi2sdq %rdi, %xmm0, %xmm0”
Posted 07/02/2022. Bartosz Zbiciak
- p. 305, Figure 3.50.
The two instructions for data type double should be
“vxorpd” and
“vandpd”.
Posted 07/02/2022. Bartosz Zbiciak
- p. 306, Table listing two instructions.
These should be vucomiss and vucomisd (AVX instructions), rather than
ucomiss and ucomisd (SSE instructions).
Posted 03/09/2018. Weicheng Pei
- p. 313, Problem 3.60, first line of code.
Argument n should be declared as being of type int.
Posted 07/21/2015. Lauren Cooper
- p. 326, Solution to Problem 3.3, third line of text.
These examples do not rely on the rules of integral promotion.
These rules concern operations performed on data values smaller than those of type int. See, for example
https://wiki.sei.cmu.edu/confluence/display/c/INT02-C.+Understand+integer+conversion+rules. This topic is not covered in the book.
Posted 04/26/2018. Randal Bryant
- p. 326, Solution to Problem 3.3, fifth line of code.
The code should be movq %rax, $0x123 to be consistent with the problem statement,
although both versions have the same error.
Posted 06/02/2015. Zhi Li
- p. 326, Solution to Problem 3.3, sixth line of code.
The code should be movl %eax, %rdx to be consistent with the problem statement,
although both versions have the same error.
Posted 09/02/2015. Elizabeth White
- (Clarification, not an erratum) p. 326, Solution to Problem 3.4, seventh line of code.
The GCC compiler generates the instruction movzbl for this case, even though the goal is to
extend the 1-byte value to 8 bytes. See the note on Figure 3.5 (p. 184).
Posted 04/27/2018. Randal Bryant
- p. 330, Solution to Problem 3.14D.
The comparison is for <=.
Posted 08/16/2015. Dmitry Neverov
- p. 331, Solutions to Problems 3.15B and 3.15D.
The values starting with “0x0x”
should start with “0x” instead.
Posted 12/04/2015. Carlos Galdino and Yili Gong
- p. 333, Solution to Problem 3.20.
The comment on the third line of code should read “Test x”.
Posted 06/02/2015. Curtis Gagliardi
- p. 338, Solution to Problem 3.30, third bullet.
The final sentence should say “Thus, case labels 3 and 6 are missing in the switch statement body.”
Posted 09/27/2015. Xinyun Zhao
- p. 339, Solution to Problem 3.32, line labeled “F4”.
The instruction should be “repz retq.”
Posted 10/13/2016. Wenjun Huang
- p. 339, Solution to Problem 3.33, line 4.
The expression should be *v += b
Posted 06/24/2019. Deepayan Patra
- p. 346, Solution to Problem 3.46C.
The program is attempting to return to address 0x400034, not 0x040034.
Posted 10/24/2016. Wenjun Huang
- p. 347, Solution to Problem 3.46E.
In addition, the program should test the value returned by gets to make sure it's not NULL.
Posted 07/26/2017. Daniel O'Brien
- p. 348, Solution to Problem 3.51.
In the third entry of the table, the conversion is from double to float, as is requested in the problem statement.
Posted 08/09/2015. Yu Zhong
- p. 349, Solution to Problem 3.53, last paragraph of text.
The ambiguity arises from the commutativity of addition, not multiplication.
Posted 11/09/2015. Shoeb Mohammed
- Chapter 4: Processor Architecture
- p. 360, Problem 4.2A.
There should be an additional byte with hexadecimal value 00 at the end of the sequence.
Posted 08/16/2016. James Timmins
- p. 360, Problem 4.2B.
There should be an additional byte with hexadecimal value 90 at the end of the sequence.
Posted 12/01/2015. Shoeb Mohammed
- p. 365, Figure 4.6, Line 9 of x86-64 code. The comment should say “Test count”
Posted 01/02/2017. Lee Wintervold
- p. 366, middle of page. The four words listed are incorrect. They should be
0x000d000d000d,
0x00c000c000c0,
0x0b000b000b00, and
0xa000a000a000.
Posted 06/23/2017. Yiling Gong
- p. 369, paragraph in middle of page, sixth line. The value 0xabcdabcdabcdabcd should be
0xabcdabcdabcd.
Posted 10/02/2020. Radoslaw Jurga
- p. 371, Practice Problem 4.7. The annotation for line 7 of the code should state “Return 0 or 8.”
Posted 12/02/2015. Alex Knaust
- p. 377, Figure 4.13(b):
The last selection in the HCL code should read “1 : B;” (The number 1 rather than the letter l).
Posted 06/02/2015. Paul Anagnostopoulos
- p. 404, Figure 4.26, 7th entry.
The name should be IOPQ, not IOPL.
Posted 10/24/2016. Wenjun Huang
- p. 404, Figure 4.26, 14th entry.
The name should be RRSP, not RESP.
Posted 08/26/2015. Yu Zhong
- p. 405, Caption of Figure 4.27.
It should state “Ten bytes are read from the instruction memory …”
Posted 07/25/2015. Randal E. Bryant
- p. 407, parenthetical remark before HCL code for srcA.
Should state “recall that RRSP is the register ID of %rsp.”
Posted 08/26/2015. Yu Zhong
- p. 420, second line from bottom.
Reference to Figure 4.38(c) should be to Figure 4.38(d).
Posted 07/23/2018. Jong-won Choi
- p. 439, fourth and fifth lines from bottom.
The sentence should state “The value for register %rbx is also forwarded from the memory to the decode stage.”
Posted 08/09/2015. Liemin Ma
- p. 434, Figure 4.47.
The instruction at address 0x016 should be addq.
Posted 11/23/2015. Warren Crasta
- p. 441, Figure 4.53.
The instruction at address 0x32 should be addq %rbx,%rax.
Posted 07/28/2015. Randal Bryant
- p. 443, Figure 4.55. Three errors:
- The program is prog6.
- The instruction at address 0x000 should be irmovq Stack,%rsp.
- The instruction at address 0x013 should be irmovq $10,%rdx.
Posted 07/28/2015. Randal Bryant + 12/21/2017 Dave Ohlsson
- p. 444, Figure 4.56.
The instuctions at addresses 0x016 and 0x020 should be irmovq.
Posted 10/24/2016. Wenjun Huang
- p. 445, first complete paragraph. As a general note, our
processor designs do not attempt to set the program counter
(PC) to a consistent value when an exception occurs. In a more
complete implementation, the PC would be set to the address of the
instruction that causes the exception, but this is not the case for
the pipelined implementations.
Posted 06/05/2018. Eadren King
- p. 454, last sentence on page.
It should read “In this figure, you can also see that many of the values in pipeline registers M and W ….”
Posted 12/25/2015. Jiwen He
- p. 473, Bibliographic Notes. In the last line, “Intel-compatible x86-64 processor” should
be “Intel-compatible IA32 processor.”
Posted 05/04/2015. Scott Wright
- p. 482, Solution to Problem 4.4. Second line from bottom.
The instruction should be jle return, and the comment should be
“If count <= 0, return 0”
Posted 01/11/2016. Xinzhen Chen
- p. 491, Solution to Problem 4.33. Last sentence should state
“ … register %rax will be set to the incremented stack pointer … ”
Posted 10/13/2021. Yili Gong
- p. 492, Solution to Problem 4.35. Second to last sentence should state:
“ … the conditional move source value 0x123
gets forwarded into ALU input valA, while input valB correctly gets operand value
0x321.”
Posted 07/21/2015. Liemin Ma
- p. 494, Solution to Problem 4.44. The answers to part A and part C are both off by 2 instructions:
- The inner loop of the code using the conditional jump contains 9
instructions, 8 of which are executed when the array element is positive.
- The inner loop of the code using the conditional move contains 8
instructions.
- The conditional jump code requires an average of 9.5 cycles.
- The conditional move code requires an average of 8.0 cycles.
Posted 12/15/2015. Nathaniel Green
- Chapter 5: Optimizing Program Performance
- p. 520, around 1/3 way down the page. The refererence to register “%eax”
should instead be to “%rax.”
Posted 12/01/2015. Shoeb Mohammed
- p. 533, second to last line in first paragraph.
It should state “ … up to a factor of 6.”
Posted 07/26/2017. Daniel O’Brien
- p. 533, code annotations for inner loop of combine5.
Variable limit is stored in register %rbp.
Posted 12/01/2015. Shoeb Mohammed
- p. 533–534, sentence spanning these two pages.
It should state: “The loop unrolling leads to
two vmulsd instructions—one to
multiply acc by data[i], and the second to multiply acc by data[i+1].
Posted 01/02/2016. Mark Morrissey
- p. 546. Web Aside OPT:SIMD.
The assembly code should reference register “%rcx,” not “%rcs.”
Posted 05/08/2016. Sheldon Guo
- p. 547. Table in description of Web Aside OPT:SIMD.
-
The headings under the category “Floating point” should read float and double, rather than
int and long.
Posted 07/25/2015. Randal E. Bryant
-
Several entries in the line “Scalar throughput bound” are incorrect. The eight bounds should be:
0.50, 1.00, 0.50, 1.00, 1.00, 0.50, 1.00, 0.50.
Posted 12/12/2022. Jonathan Phillips
- p. 551, Table in middle of page. The code for combine4b should be listed as being on page 551.
Posted 02/21/2022. Waqas Younas
- p. 561, Practice Problem 5.11, last paragraph. It should state “Perform an analysis similar to those
shown for combine4 … ”.
Posted 03/22/2022. Pedro Henrique
- p. 556, Figure 5.33 caption. dest should be dst
Posted 07/31/2018. Tanvir Alam
- p. 568, Last paragraph of Section 5.14. By Amdahl’s Law (Equation 1.2), the predicted speedup is 1/(1-α) = 39.4
Posted 10/02/2022. Yili Gong
- Chapter 6: The Memory Hierarchy
- p. 602, Section 6.1.4, paragraph 4, line 9. 1980 should be 1985.
Posted 1/11/2016. Yili Gong
- p. 612, [Clarification]: The last sentence of the Cache Hits section refers to Figure 6.22.
Posted 07/31/2018. Dave Ohlsson
- p. 640, Figure 6.40, line 22. “i++” in for loop should be “i += stride”.
Posted 03/01/2016. Shoeb Mohammed
- p. 640, Figure 6.40, line 35. sizeof(double) should be sizeof(long)
Posted 07/31/2018. Weicheng Pei
- p. 645, Figure 6.44. It should be mentioned somewher that the C
matrix is assumed to be initially all zeros. Also line 6 of
version ijk could eliminate an unncessary load by replacing
C[i][j] += sum with C[i][j] = sum. However, since this line
is not in the inner loop, it doesn't affect the analysis either way.
Posted 07/31/2018. Debbie Neft
- p. 651, Problem 6.29A,
The diagram should have 12 boxes instead of 13.
Posted 11/23/2015. Tj Gilbrough
- p. 661, The solution to Problem 6.4:
-
For parts A and B, the 1 MB file consists of 2,048 512-byte logical blocks, not 2,000 as stated in the solution.
Posted 07/30/2018. Jonatan Schroeder
-
To be precise, part B should include the total transfer time for the blocks, which
is 2 x Tmax rotation = 12 ms.
Posted 07/30/2018. Li Qiuhao
- p. 662, Solution to Part (c) Problem 6.5, the expected lifetime is ~17,535 years, not 140 years.
Posted 03/01/2016. David Hirsch
- Chapter 7: Linking
- p. 680, Section 7.6.1. Recent versions of gcc such
as gcc10 now default to -fno-common, so programs
with multiply-defined weak symbols will now elicit a linker error.
Posted 2/14/2022. Michaël Cadilhac
- p. 682. paragraph 2. in the output from foobar5, the values
of x and y are corrupted, but the precise values
are system-dependent.
Posted 12/16/2015. Yili Gong
- p. 683, [Clarification]: The variable x referred to in
the first paragraph is from foo5.c.
Posted 07/31/2018. Dave Ohlsson
- p. 690, paragraph above Figure 7.9 that reads “Recall from Section 3.6.3” should be “Recall from Section 3.6.4.”
Posted 7/30/2018. Yili Gong and Jianxun Xie.
- p. 698, Figure 7.15.
The address for the user stack (248-1) is wrong.
The correct address is much smaller, and varies from process to process.
Posted 07/09/2015. Dave O'Hallaron
- p. 698, Figure 7.15.
%esp should be %rsp.
Posted 10/21/2015. Liz White
- p. 715, Problem 7.8, Module 2. static int main=1[ should be static int main=1;
Posted 07/30/2018. Michael Hinton and Michael Ross
- Chapter 8: Exceptional Control Flow
- p. 735, Figure 8.13. Two errors:
- The address for the user stack (248-1) is wrong.
The correct address is much smaller, and varies from process to process.
Posted 07/14/2015. Dave O'Hallaron
- %esp should be %rsp.
Posted 07/31/2018. Dave Ohlsson
- p. 748. First paragraph. Remove the last sentence: “The only
correct assumption is that each possible outcome is equally likely” or
replace it with “The only correct assumption is that each possible
outcome can indeed occur.”
Posted 07/31/2018. Dave Ohlsson
- p. 751, definition of getenv, “Returns: pointer
to name if it exists” should be “Returns: pointer to value
associated with name if it exists”
Posted 7/30/2018. Yili Gong and Jianxun Xie.
- p. 752, Practice problem 8.6, output of ./myecho. “Command-ine” should be “Command-line.”
Posted 7/30/2018. Li Du.
- p. 756, Figure 8.25. parseline() should check that the capacity of
argv (= MAXARGS, defined in Figure 8.23) is not exceeded.
Posted 7/30/2018. Dave Ohlsson.
- p. 762. Definition of signal function. On error, this
function sets errno to indicate the cause.
r>Posted 07/31/2018. Dave Ohlsson
- p. 763, Figure 8.30. For consistency, pause() should be Pause().
Posted 02/26/2018. Bill Nace
- p. 768, Fig 8.34. Fixed sio_putl so that it correctly handles negative numbers.
Posted 10/19/2016. Randy Bryant
- p. 773, Fig 8.37, line 5. To avoid the signal handler potentially
blocking on long-running jobs, waitpid should be called
with the WNOHANG option.
Posted 12/11/2018. Fahim Faisal
- p. 777, Fig 8.39, line 9. To avoid the signal handler potentially
blocking on long-running jobs, waitpid should be called
with the WNOHANG option.
Posted 12/11/2018. Fahim Faisal
- p. 779, Fig 8.40, line 8. To avoid the signal handler potentially
blocking on long-running jobs, waitpid should be called
with the WNOHANG option.
Posted 12/11/2018. Fahim Faisal
- p. 781. In the description of sigsuspend in the second paragraph, line 1
should be sigprocmask(SIG_SETMASK, &mask, &prev);
Posted 10/28/2015. Dave O'Hallaron
- p. 796. Figure 8.47. On the top line of the process graph, p1: x=1
should be p2: x=1.
Posted 11/8/2015. Eric Adlam
- p. 797. Figure 8.49. In the process graph, the bottom exit should
be exit(2)
Posted 1/14/2019. Roslyn Cyrus
- Chapter 9: Virtual Memory
- p. 829, Figure 9.26. The kernel portion of the address space is
identical for each process. There is no part of the kernel virtual
memory that is different for each process.
Posted 05/08/2018. Godmar Bak
- p. 831, bottom of the page, “fvm_start” should be “vm_start”
Posted 03/01/2016. Ruth Anderson
- p. 842, first paragraph. The heap shown in Figure 9.34
consists of 18 words, not 16 words.
Posted 06/26/2016. Ruth Anderson
- p. 873, Section 9.11.6, paragraph 2. The explanation for why the
binheapDelete function is buggy is correct for the version
of ANSI C described in the K&R book, where the
unary --, ++, and * operators have
the same precedence. However, in more recent versions of C,
-- and ++ have higher precedence
than *. So while the given code is always incorrect, the
reason is different for different versions of C.
Posted 06/26/2016. Xinyang Zhang
- p. 881, Solution to Problem 9.1. 16,384 P could be expressed more
clearly as 16 E.
Posted 07/31/2018. Daniel O'Brien and Dave Ohlsson
- p. 884, solution code. The #endif on line 12 should be removed.
Posted 07/30/2018. Wenjun Huang
- Chapter 10: System-Level I/O
- p. 893, Second bullet point. The relative path name should be ../droh/hello.c
rather than ../home/droh/hello.c.
Posted 06/26/2016. Mathieu Bordere
- p. 895, read() returns 0 not only on EOF, but also if n is 0.
Posted 07/31/2016. Dave Ohlsson
- p. 904, Figure 10.10 and p. 906, Figure 10.11.
main() should check that argc is at least equal
to 2 before accessing argv[1].
Posted 07/31/2016. Dave Ohlsson
- Chapter 11: Network Programming
- p. 949, Aside. “Marc Andreesen” should be “Marc Andreessen”
Posted 07/31/2016. Dave Ohlsson
- p. 956, Figure 11.28. In the telnet trace, there should be
“Connection: close” header, which is generated by
the adder.c CGI program, sandwiched between lines 8 and 9.
Posted 06/26/2016. Laura M. Roberts
- p. 959, Figure 11.31. The use of multiple sprintf calls to
generate body violates the following rule: C99 and
POSIX.1-2001 specify that the results are undefined if a call to
sprintf(), snprintf(), vsprintf(), or vsnprintf() would cause copying to take place between
objects that overlap (e.g., if the target string array and one of
the supplied input arguments refer to the same buffer).
Posted 04/25/2019. Zeng Fan Pu and Urvi Agrawal
- p. 962, Figure 11.34. The use of multiple sprintf calls to
generate buf violates the same rule as p. 959.
Posted 04/25/2019. Zeng Fan Pu and Urvi Agrawal
- pp. 959 and 962. Updated the Tiny code on the CS:APP Web site to fix the previous two bugs.
Posted 11/14/2019. Dave O'Hallaron
- Chapter 12: Concurrent Programming
- p. 989, first sentence. Delete the entire sentence that begins
with “If the main thread calls pthread_exit, it waits for all other threads to terminate...” In fact,
Pthread_exit never waits for other threads to terminate.
Posted 07/30/2018. Yuan Tang
- p. 997, The text describing Figure 12.17 is correct, but the Figure itself has two typos:
- The Ui instruction should be addq $1, %rdx.
- The Si instruction should be movq %rdx, cnt(%rip).
Posted 02/22/2018. Dr. Joann J Ordille
- p. 1002, first box. In the declaration of sem_init, the
middle parameter should be int pshared.
Posted 07/31/2018. Dave Ohlsson
- p. 1015, Figure 12.31, code line 25. sem_init
should be Sem_init
Posted 07/31/2018. Dave Ohlsson
- p. 1023, ten lines from the bottom. “the static next variable”
should be “the static next_seed variable”
Posted 07/31/2018. Dave Ohlsson
- p. 1029. The definition of the mutex ordering rule is too
strong. It should read simply “a program is deadlock free if each thread
acquires its mutexes in the same order.” In fact, the order that mutexes are
released does not matter, since V operations never block.
Posted 07/30/2018. Einar Rasmussen
- Appendix A: Error Handling
- Index
- p. 1076. There should be no index entry for RESP.
Instead, it should be for RRSP with the same information.
Posted 08/26/2015. Yu Zhong
CS:APP3e Instructor's Manual
- Chapter 9: Virtual Memory
- p. 107, Problem 9.16. The answer for the first case should be 16
bytes, rather than 20 bytes. Smallest possible allocated block is 12 bytes
(|hdr|4-byte payload|ftr|). Smallest free block is 16 bytes
(|hdr|pred|succ|ftr|). Thus, a minimum block size of 16 bytes.
Posted 03/01/2016. Daniel Rushton
