The advantages of letting the compiler drive preprocessing and
assemblying are
 * we do not have to look for cpp or the assembler
 * we can use COMPILE_OPTIONS to pass through options to the assembler
 * we can use Clang's assembler

Issues:
* Clang's preprocessor does not seem to preserve whitespaces as
  required
* Clang's assembler does not support flags we need (maybe there
  are equivalent flags we have to use)
* Clang does not show assembler help with -Wa,-help.

Fixes #1675

Change-Id: I8ae45c39f071853cb95744d551993aba38faec1c

This silences assembly deprecation warnings from some test cases. We might miss some valid assembly warnings, but the flag is only set when building tests. For builds without tests, we still get the warnings.

Fixes #1675

Adds support for several conflicts between aflags and xax on x86: failing
to reserve xax due to lazy aflags still residing in xax; failing to reserve
aflags if xax is taken; and failing to get the app aflags value if xax is
taken.  For the first one, we throw away the lazy aflags.  For the other
two, we reserve a temporary scratch register, xchg it with xax, and restore
it afterward.  We place aflags in TLS and do not try to keep it in a
register.

Issue: #511

Fixes a regression from f6e8e1c4 where drutil_insert_get_mem_addr() gets
the wrong address for a far memref that uses the destination register, due
to improper ordering of operations.

Also optimizes from two steps to one for a far memref that uses the
destination but not the scratch register and has only one of base or index.

Fixes #2913

Adds static TLS data to support drreg routines being called during process
init before thread init has happened.

Also adds support for drmgr_current_bb_phase(), which is called by drrreg,
being called during process init.

Adds a test case to drreg-test.

Fixes #2910

Fixes Mac build breakage from 8471d5cc.

Issue: #2037

This uses the encoding scheme introduced in #2811.

Everything expect the tests are auto-generated now.

I raised issue for the failing tests on ARM. Ignoring those failures in
ARM pre/postcommit testing makes it easier to spot new regressions.

Issues: #2416, #2892, #2893, #2894

2e9f20e4 made get_clean_call_switch_stack_size() 8-byte aligned for ARM,
which is required by the "Procedure Call Standard for the ARM Architecture
[AAPCS]" [1]. priv_mcontext_t on ARM only contains 31 8-byte slots,
which is why an extra adjustment is needed in insert_push_all_registers.

Fixes #2799

[1] http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.faqs/ka4127.html

Refactors find_executable_vm_areas() to share its map entry skipping with
the re-takeover re-walk from b06a7022, but not its module list or
executable area updates.  This entry skipping for vmheap turns out to make
a big performance difference when attaching.

Removes individual updates to memcache for entries inside vmheap which were
already bulk-added for the find_executable_vm_areas() walk.

Issue: #2037

Added dr_where_am_i() to better support client self-profiling via
sampling.  This also provides the fragment tag, by refactoring
the pcprofile code into a new helper fcache_refine_whereami().

Renamed the whereami types to better avoid name conflicts:
s/WHERE_xxx/DR_WHERE_xxx/ and s/where_am_i_t/dr_where_am_i_t/.
Exported the dr_where_am_i_t enum.

Fixed a bug where an uninitialized mcontext was passed to a
client timer callback: we only filled in the mcontext for a
DR-internal callback.

Added a test of client sampling to client.timer

Issue: #140

Adds new versions of drmgr's pre and post syscall events that take in user data parameters.
Adds tests of the new functions.

Issue: #2833

Adds new versions of drmgr's module load/unload events that take in user data parameters.
Adds tests of the new functions.
    
Issue: #2833

Adds a missing error capture in raw2trace_t::read_and_map_modules() where
an error in module parsing was accidentally ignored.

Allows for zero data or instruction refs on a core and thus no printed miss
rate, to reduce flakiness in the burst_threads test.

Issue: #2092

Adds a complete maps file walk to update the memcache on re-taking-over the
process for dr_api_start.  The memcache is cleared beforehand to avoid both
false positives and negatives in later queries.  This helps to solve issues
with a gap between dr_app_setup() and dr_app_start().

Does not update the executable areas or module list: they are more
difficult to re-walk, and existing lazy updates to those will suffice for
now, with a low risk of false positives.

Adds updating of the memcache on a query miss.  Previously we would just
continue to miss and walk the maps file every time.

Tested manually by disabling the i#2114 change so that a signal does a
query, adding signals to the burst_threads test, and calling dr_app_setup()
before creating the test's threads, causing queries to miss when delivering
signals.  It is difficult to create a regression test for this as the
consequences are performance degradations rather than correctness, and
these degradations only really show up at scale with hundreds of threads
whose missing stacks are queried at once with no caching.

Fixes #2037

Avoids name conflicts by renaming the LOG_xxx macros to have a DR_ prefix,
so DR_LOG_xxx.  Puts in place a compatibility CMake option
DynamoRIO_LOG_COMPATIBILITY that requests the old LOG_xxx macros and
automatically sets it if the client targets DR version 7.0.0 or older.

Updates all non-core uses of LOG_xxx.

Given the time since the RC1 we're retroactively considering it to be an
official release.  The base version was bumped to 7.0 already in ad850c4c.
Here we split the changelist.  Additionally I added the missing
page_drltrace page to match page_drstrace.

Removes the incorrect removal of all itimers in signal_thread_exit(), which
was disabling app itimers on detach.

Adds a test to api.static_signal.

Fixes #2871

As noted in the added code comment, there are points in time where it's
possible that the app_sigaction is NULL for a signal. This PR makes us
more defensive to avoid a crash on an internal unit test.

Issue: #2114

In environments where there are a lot of signals being delivered to
threads of a process, there is possibly a race where one thread
links/unlinks while another is waiting to do it themselves. With this
commit, we have threads skip linking/unlinking if they notice another
thread did this (after they have obtained the change_linking_lock).

This fixes an ASSERT I'm hitting with an app:
https://github.com/DynamoRIO/dynamorio/blob/9d9129a/core/link.c#L1896

Issue #2066

Previously, if we weren't able to take over all threads after a few
retries we would have a debug-level assert but otherwise continue on
with running the application. This can cause lots of correctness
problems in the application, and it would be better to cleanly crash
ahead of time rather than wait for an application crash to happen.

Issue #2723

Because relocating can invoke an ifunc which accesses TLS, for UNIX we move
loader_init() to after thread init.  The custom delayed init function
calling is moved back to privload finalization.  The ELF TLS block setup
for the primary thread is moved up to TLS block discovery time and is moved
to after relocation (but before init function calling).  Windows remains
unchanged as it has other ordering requirements (i#338).

Fixes #2751
Issue: #338

We don't want flaky tests to derail package deployment, so we disable test
running for such builds.  We've already run the tests for the same commit
via regular master-push triggers: these package builds are coming from a
cron trigger (Travis) or a tag addition (Appveyor), not a code change.

Issue: #1967

Relaxes the burst_threads output further to handle non-determinism in the
test which had been causing sporadic failures.

Yet another attempt to get Appveyor and Travis to upload files to the same
GitHub Release, by using not only the same tag but the same name.

Issue: #1967

Since Appveyor appears to clobber the GitHub Release name and description
set by Travis, we set similar values for Appveyor.

Sets the Appveyor deploy file to use the artifact name, as the path (with
wildcards) failed.

Issue: #1967

Adds the proper path for our package files to be "artifacts" that can then
be deployed to GitHub Releases.

Issue: #1967

Adds package building in each Appveyor build.

Adds force_update for Appveyor deployment to try and get it in the same
GitHub Release as Travis.

Issue: #1967

Fixes an error in ad850c4c where only the 64-bit package was deployed.
However, it is not clear that Travis supports deployments from multiple
parallel jobs, and it is also not clear how racy tagging and deploying is
going to work.  This is an attempt to see what happens: the 2nd tag may
fail and abort its deploy.

Sets the GitHub Release title and description.
Fixes a git diff error on Appveyor tagged builds.

Issue: #1967

Adds support for Travis builds to produce auto-published package files.
Each job can create its own package file, using existing support in
runsuite*.cmake and package.cmake with some additions here to enable and
tweak that code.  Longer-term we may want to use package.cmake instead and
even make official builds on Travis (i#2861).

Adds Travis deployment to Github Releases of produced packages files.  The
tag is "cronbuild-${VERSION_NUMBER}": e.g., "cronbuild-7.0.17592".

Adds Appveyor tag-triggered deployment.

Adds better support for 32-bit-only or 64-bit-only x86 package files by
naming them differently and removing the drrun warnings on incomplete
packages when missing the opposite bitwidth.

Adds better build order handling to ensure release build is the default
for tool files.

Issue: #1967

Eliminates the memory query loop in copy_frame_to_stack() which checks
whether each part of the app stack is writable before copying over the
signal frame.  This loop grabs the global all_memory_areas lock and if that
lookup misses it goes to a maps lookup.  This has shown to cause
noticeable contention.  Instead we perform an overlap check for
is_executable_area_writable followed by a TRY_EXCEPT safe write.

Adds a new test linux.bad-signal-stack which tests this by setting up an
unwritable alternate stack.  It also tests checking for SA_ONSTACK, and we
fix #2017 here to make that work, which is needed for a good test.  The
existing security-common.decode-bad-stack hits the
is_executable_area_writable() case.

Issue: #2214
Fixes #2017

Fixes a Mac build error introduced by 3bc33154.
Fixes a Mac test build error that shows up on newer toolchains.

Issue: #975

Adds a feature where a client library can request that the private loader
complain if malloc & co. are called at any time other than process init or
exit, to help clients that want to support being linked statically with the
app.  Because it has to be early, the feature is triggered by a variable
declaration DR_DISALLOW_UNSAFE_STATIC.  It can be overridden
dynamically by a new API routine dr_allow_unsafe_static_behavior().

Fixes drcachesim to use placement new for its offline custom module data
allocations.

Issue: #975, #2006

The trace header refactoring in aedf9be2 broke post-processing of legacy
trace headers where the first timestamp precedes the first thread id.  We
fix that here.

Issue: #2843

Adds a new option -cpu_scheduling which schedules trace execution on
simulated cores to match the new recorded cpu mappings.  The cpu's are
assigned to cores in a round-robin fashion, and each thread region is assigned
to the core that owns the cpu in the thread region's header markers.

Adds -cpu_scheduling to several tests with multiple threads.

Adds the documentation for the new option.

Fixes #2843

For tools that have many options, passing a long list of parameters is
cumbersome, error-prone, and difficult to extend with newly added options.
We change cache_simulator, tlb_simulator, and reuse_distance to use a
struct of knobs instead.  Tools with just a couple of options are left
using parameters.

Issue: #2006

Adds synchronization to avoid fewer than one thread per core in the
tool.drcacheoff.burst_threads test, to eliminate flakiness in output
matching.

Adds a timestamp marker and a cpu marker to the trace buffer header for
each thread's buffer unit output.  The timestamp was already in the raw
offline trace format, but now it is in the final trace for both offline and
online as a new marker type.  The cpu is completely new and specifies which
core that thread executed on at the granularity of the buffer unit.

Refactors the initial and per-output buffer headers to fix warts in the
tracer: now the initial buffer's timestamp is from output time rather than
thread init time; the initial header is more cleanly skipped for virt2phys;
header uses are more normalized and easier to understand.

Updates the basic_counts tool to separate these new scheduling marks from
kernel transfer and other markers.  Updates the corresponding
documentation.

A forthcoming change will update the cache simulator to schedule threads
based on executed cores rather than a thread round-robin scheme.

Issue: #2843

Resets callback functions at exit to better support re-attach.

Adds re-attach testing to the burst_threads and burst_thread*filter tests.

Adds "firstglob" support to runmulti to post-process only one of the 4
re-attach traces to keep the tests simpler and faster.

Adds testing of the thread filter callback being reset across re-attach by
passing in a specific user param.

Issue: #2175, #2820