GitLab

The latter only show up because of implicit non-null attributes that gcc assumes
for the string functions, leading to warnings for the assertions in those
functions. The latter relates to a known-safe use of builtin_frame_address.

Newer gcc versions generate warnings because of the noreturn attribute on it and
different type signature.

Use new isa-debug-exit device for shutdown, switch name to
qemu-system-i386.

Modern versions of Perl prefer a caret in variable names over literal
control characters and issue a warning if the control character is used.
This fixes the warning.

When sysread encounters an error, it returns undef, which yields a warning
when compared against 0.  This fixes the problem.

(patches provided by David Mazieres).

involving invalid pages (e.g., position system call opcode with first
bytes in a valid page and remaining bytes in a nonexistent page).

PINTOSPATH should always point to the pintos src folder

fixed qemu path and added some kind of autodetection as multiple platform emulators are available on ubuntu 14.04

Recent clang version, detect attemps to dereference or call NULL, and
eliminate the code.

../../tests/userprog/bad-read.c:11:9: warning: indirection of
non-volatile null pointer will be deleted, not trap [-Wnull-dereference]
        *(int *) NULL);
        ^~~~~~~~~~~~~

This patchs adds the volatile modifier as necessary.
It allows to use clang to compile the tests.

On QEMU emulator version 1.7.0, the previous shutdown sequence (writing
`shutdown` to 0x8900), is not working properly.

This patch adds support for the ACPI shutdown code which works on qemu
1.7.0 and preserves the old sequence for compatibility.

Matthew Pirocchi <mrpirocchi@email.wm.edu>, via John Ousterhout
<ouster@cs.stanford.edu> reports that this fixes a problem on his system.

Fix gcc inline assembler operand constraints for syscall -- as pushl modifies %esp, variable operands must not reference the stack pointer.

Without this patch, gcc-4.6.1-9ubuntu3 miscompiles the read syscall to:
  000000b6 <read>:
    b6:   ff 74 24 0c             pushl  0xc(%esp)
    ba:   ff 74 24 08             pushl  0x8(%esp)  ; [!]
    be:   ff 74 24 04             pushl  0x4(%esp)  ; [!]
    ...

When compiled according to the instructions, bochs fails to load
the project 1 kernel.  This seems to be due to the latest gcc using an
instruction that bochs is not compiled to support.  Adding the
'--enable-cpu-level=6' configure option to Bochs, as this commit does,
seems to work.

bochs-2.2.6 failed to compile with the latest version of gcc.  This commit
fixes the problem

Reported by a student of Borja Sotomayor <borja@cs.uchicago.edu>.

This allows make to pass the proper command line arguments to the sub-make.

Reported by Francis Russell <fpr02@doc.ic.ac.uk>.

Reported by Francis Russell <fpr02@doc.ic.ac.uk>.

When the slave side of the pty is closed, the "read" system call returns
EIO.  In this situation, squish-pty failed to flush any output that it
already had in its buffer (up to 16 bytes) to stdout.  This could cause
the "Powering off..." message printed by Pintos just before exiting to
be missing from the output.

This usually didn't happen, because there are two other exit paths from
the relay() function that does most of squish-pty's work.  The most common
exit path was (evidently) the one which detected that the child process
had died, which did properly flush the buffer.  (The third path was wrong
too, and this patch fixes that one too.)

This patch fixes the problem by using a single exit path from relay() that
always reads any remaining input from the pty (if it is open) and flushes
it to stdout.

Reported by Borja Sotomayor <borja@cs.uchicago.edu>.

Reported by Borja Sotomayor <borja@cs.uchicago.edu>.

Some versions of ld accept /dev/null as a valid input file.  Others will
not.  The latter versions will give false negatives for --buil-id support,
which breaks linking of Pintos userspace programs.  So use a different
method, that I hope will be more reliable, to detect --build-id support.

Reported by Joseph Huang <jdhuang@stanford.edu>.

Here, $buf might have some data in it that has not yet been printed, but
we were discarding it (and trying to read more) without printing it.

(In Perl, "do" ensures that the inner block runs at least once; without
"do" the condition is evaluated first.)