Changes

Thu Aug 23 07:47:40 2012  fox

     777* tools/mkport.c, libproc/common/Pcontrol.c:
          Fix for PTRACE_O_TRACEEXEC which may not exist
	  in <sys/ptrace.h> (Centos 5.3).

     776* driver/x_call.c: Fix for IPI on archlinux-3.4/i386

Wed Aug 22 23:57:50 2012  fox

     775* driver/dtrace_linux.c, tools/load.pl: Remove some of the 
     	  legacy symbols as we dont need them anymore.

Mon Aug 20 22:03:03 2012  fox

     774* driver/fbt_linux.c: Better fix for skipping .init section
          of drivers loaded after dtrace - avoids paniccing if
	  we probe on them (eg 'dtrace -n fbt:isofs::').

     773* driver/signal.c: Put in the missing assembler code for i386.

Sun Aug 19 16:05:01 2012  fox

     772* scripts/dt.pl: Rename to dt.

     771* scripts/dt.pl: Add 'dt func <pid>'

     770* archive: 1.0145

     769* libproc/common/Pcontrol.c: Some changes to Pxcreate to handle
          parent/child sync with a forked child from 'dtrace -c cmd'.

Sat Aug 18 16:15:47 2012  fox

     768* driver/signal.c, fasttrap_isa.c, dtrace_subr.c, and others:
          Intercept process signal delivery. If we are inside the
	  scratch area for the PID provider, reset PC to the original
	  instruction. This stops user app core dumping on return from
	  a signal handler because we would return to the scratch area
	  but the scratch area will most likely be overwritten with
	  a later instruction.

Fri Aug 17 08:46:57 2012  fox

     767* driver/x_call.c: Fix for Debian-6 - find any apic outlet to
          invoke.

     766* tools/mksyscall.pl: Fix for Debian-6.

     765* systrace.c: Fixes for Debian-6.

Sat Aug 11 10:05:56 2012  fox

     764* fasttrap_isa.c: Fix the last issue properly - 2 styles
          of indexed instructions werent being handled
	  properly. Now they are.

     763* archive: 1.0144

Thu Aug  9 22:48:18 2012  fox

     762* fasttrap_isa.c, fasttrap.c: Fix handling of 
             41 ff 14 c4             callq  *(%r12,%rax,8)
	  instruction.

	  Add private page for the scratch buffer trampoline.

Mon Aug  6 20:56:11 2012  fox

     761* Archive: 1.0143

     760* driver/systrace.c: Fix for 32b binaries on 64b kernel for
          Centos/RedHat releases.

Sun Aug  5 11:50:41 2012  fox

     759* tools/mkport.h: Fix for FUNC_DUMP_TRACE_ARGS on centos5.6.

     758* dtrace_asm.c: dtrace_casptr: Add a check that target is writable
          and avoid paniccing kernel when it isnt (eg syscall tracing,
	  caused by mem_set_writable code not working). Hm. Undo
	  this change - we can get deadlock if we trigger a violation.
	  Or something else - not sure what it is.

     757* fasttrap_isa.c, dtrace_linux.c: When patching in PID provider
          instructions, ensure page is executable. 
	  Centos5.3 2.6.18-128.el5 fix

Sat Aug  4 15:16:00 2012  fox

     756* cmd/ctfconvert: Doesnt depend on -lbfd

     755* libdtrace/dt_link.c: Allow us to avoid deleting the temp file.
          Fix file perms issue.

     754* libdtrace/dt_subr.c: dt_alloc: Use calloc() and not malloc()
          as this can cause bad ELF file generation.

     753* linux/sys/regset.h: Reorder the amd64 registers to match
          the interrupt trap, so that single stepping pid probes work
	  properly.

     752* driver/fasttrap.c: Significant speed up for /proc/dtrace/fasttrap.
          Not ideal or linear, but better than before.

Fri Aug  3 11:40:18 2012  fox

     751* driver/intr_x86-64.S, intr_x86-32.S: Dont corrupt RAX register on 
          an INT3 trap. Was breaking pid provider/fasttrap.

Thu Aug  2 08:25:51 2012  fox

     750* driver/x_call.c: xcall_init() Ensure xcalls[] is setup even
          if no APIC (Centos 5.5 issue).

Wed Aug  1 15:24:01 2012  fox

     749* libproc/common/Psymtab.c: Fix for _dl_initial_error_catch_tsd on
          Centos 5.3.

     748* driver/taskq.c: Fix for centos 5.3 (2.6.18-129) where invoking
          workqueue callback derefs null ptr.

     747* cmd/ctfconvert/tdata.c: Avoid core dump in tdesc_layouthash
          if we have a struct/union with no members (fwd def/ref?)

Tue Jul 31 22:00:40 2012  fox

     746* driver/taskq.c: Fix for centos 5 GPL function issue.

     745* Psymtab.c: Fix for older kernels where AT_BASE auxv vector
          is not defined, so couldnt use 'dtrace -n pidXXX:::' on them.
	  (Centos 5).

     744* dtrace_linux.c: mem_set_writable: Fix (clflush) for centos
          kernel.

Mon Jul 30 08:28:16 2012  fox

     743* Archive: 1.0142

     742* driver/dtrace_linux.c: Fix issue with determining data model of 32b
          proc on 64b kernel.

     741* ctl.c, ctf_struct.h: Build fixes for older kernels.

     740* taskq.c: Add older kernel support.

Sun Jul 29 18:06:43 2012  fox

     739* libproc/common/Psymtab.c: Fix for reading /proc/pid/auxv for
          a 32b proc on a 64b kernel.

     738* driver/dtrace_linux.c: dtrace_data_model: Implement this
          for 64b systems running 32b binaries.

     737* driver/dtrace_linux.c: Fix issue affecting 32b binaries
          on 32b kernel (fasttrap miscomputing instr sizes).

Sat Jul 28 08:41:53 2012  fox

     736* All drivers: Add ".owner = THIS_MODULE", to protect from
          attempts to unload driver whilst still have open references.

     735* prov_common.c: Add proc:::fault probe.

     734* Psymtab.c: Fix last change; wasnt handling 64b binaries properly.

Fri Jul 27 13:27:25 2012  fox

     733* Archive: 1.0141

     732* libproc/common/Pcontrol.c: Fix for setting the model of
          a spawned process (32 vs 64 bit).

     731* driver/ctl.c: Temp comment out the rcu_read_lock calls.

     730* driver/ctl.c: Avoid direct reference to find_task_by_vpid.

     729* driver/ctl.c: Add permission checking and update comments.

     728* driver/systrace.c: Bug fix for /proc/dtrace/syscall - SIGSEGV
          and confusion on 32b kernel.

     727* libproc/common/Pcontrol.c: Use the /dev/dtrace_ctl driver instead
          of ptrace(). We can now use PID provider on stopped jobs or even
	  procs being debugged in a debugger.

Thu Jul 26 08:33:03 2012  fox

     726* archive: 1.0140

     725* x_call.c: fix the xcalls[][] array. Some draconian measures
          for dtrace_ack_apic() to avoid crashing VirtualBox.

     724* x_call.c: Dynamically allocate xcalls[][] array according
          to CPUs available, not max possible, to reduce memory footprint.

     723* x_call.c: Redo native_apic_write_mem, for maximum portability
          across releases. May not be brilliant if we have machines
	  without APICs (embedded?) but lets worry about them when
	  we find them. Fix #pragma warning.

     722* missing.c, dtrace_linux.c: Rename clflush() and call it
          dtrace_clflush() to avoid kernel version conflicts.

     721* liblinux.c/proc2.c: When getting proc status, handle suspended
          procs.

Wed Jul 25 23:19:13 2012  fox

     720* driver/taskq.c: Dont GPF if on older kernel and cannot find
          __alloc_workqueue_key_ptr. Works on Ubuntu 8 (2.6.28)

     719* driver/x_call.c: Fix up some portability issues. Getting difficult
          to ensure this works on all kernels, so having to be more
	  draconian.

Tue Jul 24 22:45:32 2012  fox

     718* driver/x_call.c: GPL fix for 2.6.28 (Ubuntu 8/32) kernel.

Mon Jul 23 23:32:39 2012  fox

     717* driver/x_call.c: Fix compile issue for 2.6.28 kernel (Ubuntu 8)

Sun Jul 22 17:30:38 2012  fox

     716* Archive: 1.0139

     715* Various: Tidied up some compiler warnings.

     714* taskq.c, taskq.h, dtrace_linux.c, dtrace.c, fasttrap.c, 
          Fix the PID provider. It may not be perfect and needs lots of 
	  testing, but some basic stuff works. Added the taskq_create()
	  and timeout functions to aid in probe garbage collection.

Fri Jul 20 00:15:04 2012  fox

     713* tools/get-deps-fedora.sh: Add in 'make'.

Sat Jul 14 08:32:08 2012  fox

     712* fasttrap_isa.c: FASTTRAP_T_COMMON: Setup addr for the
          trapped instruction. (Previously, we had a TODO() item)

     711* Pcontrol.c: Pread_live: do ptrace(PTRACE_ATTACH)/waitpid to
          read someone elses memory, even if we are root. Seems to
	  make pid provider work *much* better.

     710* driver/systrace.c: Fix typo for ia32_execve.

Wed Jul 11 23:02:16 2012  fox

     709* driver/fasttrap_linux.c: Oops. Forgot to implement the ioctl()
          interface, so PID provider couldnt get off the ground.

     708* tools/load.pl: Make /dev/fasttrap world read/writable (not sure if
          this is safe - but fasttrap will do its own perms checkings).

     707* liblinux/proc2.c: Stupid read() call for /proc/%d/exe wrong way around.

Tue Jul 10 23:28:21 2012  fox

     706* liblinux/gmatch.c: Result of match was reversed. Stopped pid
          provider from kicking in.


Sun Jul  8 16:53:33 2012  fox

     705* driver/tcp.c: Add udp:::send and udp:::receive (crude).

     704* dtrace_linux.c: Fix the pmd hacks.

     703* tcp.c: Add the accept and send/receive probes.

Thu Jul  5 22:42:36 2012  fox

     702* tcp.c: Add tcp::connect-established.

     701* prov_common.c: Add prov_add_callback, so we can start handling
          argument specific callbacks.

Wed Jul  4 16:03:51 2012  fox

     700* systrace.c: Cater for x32 syscalls not being enabled.

Sun Jul  1 16:22:42 2012  fox

     699* driver/dtrace_linux.c: Specialise last fix for 3.2/3.3 kernels.
          3.4 kernels dont have same glitch in the header files.

     698* driver/dtrace_linux.c: Fix for i386 3.2 and above kernels
          in page table manipulation.

Fri Jun 29 07:23:53 2012  fox

     697* tools/load.pl: ALlow insmod to be in /usr/bin

     696* vminfo.c: Fix for 3.4 kernels.

     695* systrace.c: Fix for 3.4 kernels.

Sat Jun  2 22:37:18 2012  fox

     694* README: Update install instructions to run tools/get-deps.pl

Thu May 24 18:57:48 2012  fox

     693* makefile: Dont fail on "make install" if build/linux.ctf not available.

     692* dtrace_linux.c: Make par_alloc/par_free do something - remove the
          stub, because without this, we dont get probes for modules.

     691* x_call.c: Put in a warning if NCPU is too large - can cause
          problems at driver load time (Centos 6.2 has NCPU==4096 which 
          equates to more than 40MB of memory for the xcall array).

Mon Apr  9 16:01:07 2012  fox

     690* driver/sdt_linux.c: Avoid compile errors on 3.3 kernels.

Thu Mar  1 21:52:36 2012  fox

     689* driver/x_call.c: Centos 4.7 change

     688* driver/dtrace.c: Avoid native_read_pmc for now - we dont need it
          anyway.

     687* driver/mutex.c: Put in workaround for Centos 4.7/Ubuntu 10.x

Fri Feb 24 19:21:39 2012  fox

     686* driver/vminfo.c: Use CONFIG_COMPACTION to avoid compile time errors.

Fri Feb 17 23:43:53 2012  fox

     685* driver/x_call.c, driver/dtrace_linux.c: Changes to workaround
          issues with accessing the GPL symbol apic.

Thu Feb 16 21:00:48 2012  fox

     684* x_call.c: Make it GPL compliant.

Sat Feb 11 10:25:20 2012  fox

     683* driver/dtrace_linux.h: Mark as CDDL driver.

     682* driver/prov_common.c: Mark as CDDL.

     681* Archive: 1.0138

     680* libdtrace/dt_cc.c, dt_impl.h, dt_open.c, 
          driver/dtrace.c, uts/common/sys/dtrace.h:
	  Add two experimental actions, so I can play (test1() and test2()).

     679* dtrace_linux.c: par_alloc: Remove temporary stub-out - we werent
          providing fbt module probes.

Mon Feb  6 21:47:58 2012  fox

     678* libdtrace/dt_pid.c: dt_pid_usdt_mapping: Dont call 
          pr_open/pr_ioctl (see comment in the code). This allows
	  us to dtrace against a USDT probe application.

     677* cmd/dtrace/dtrace.c: Add some env var defns to the usage.

Sun Feb  5 10:18:55 2012  fox

     676* driver/intr.c: Fix license.

     675* liblinux/proc2.c: Fix to avoid leaving target USDT proc in
          a stopped state. (We did this due to bugs in libproc handling
	  but we shouldnt even if we are buggy). Not detecting 't'
	  process state properly.

     674* Archive: 1.0137

     673* dtrace_linux.c: Fix clflush issue in <= 2.6.21 kernels

     672* tools/mkport.pl: Possible fix for SMP_CALL_FUNCTION_SINGLE_ARGS for
          2.6.18 kernel.

     671* driver/dtrace.c, driver/fasttrap_linux.c: Rip out all the fasstrap
          helpers when unloading the driver, so we dont panic on a reload
	  if a USDT proc is still running.

Wed Feb  1 20:14:09 2012  fox

     670* x_call.c: Fix for 2.6.24 kernel in cpumask_t

     669* intr.c, tools/mkport.pl: Remove private vmalloc_sync_all if kernel
          supports it.

     668* linux/sys/privregs.h, tools/mkport.pl: Add autodetect of
          BX vs EBX register for i386/2.6.24 type kernels.

     667* driver/cyclic_linux.c: Fix for 2.6.24/debian-4 kernel.

     666* driver/intr.c: Fix commenting error for older kernels.

Tue Jan 31 20:31:24 2012  fox

     665* driver/toxic.c: Allow "page_fault" on earlier kernels.

     664* driver/intr.c: Avoid problems when store_gdt/store_idt are
          not available.

     663* intr.c: Avoid problem when swapper_pg_dir is a #define on older
          kernels.

     662* mkport.pl, dtrace_linux.c: Dont compile against atomic_notifier_chain_register
          if its not available.

Mon Jan 30 20:33:20 2012  fox

     661* Archive: 1.0136

     660* driver/dtrace_linux.c: Move intr_init() so we do it last, so
          we can maximise the effect of ensuring all page tables are in sync.

     659* driver/intr.c: Solution for the big impossible problem. Call
          vmalloc_sync_all if possible, else a half-hearted emulation.
	  (We could just copy vmalloc_sync_all from the kernel but its
	  GPL, so we will think about a better solution based
	  on the kernel one; only a problem for older i386 kernels).

     658* driver/intr-x86-32.S: Ensure we setup %gs for kernels >= 2.6.31.
          Else user space apps die, because %gs can be corrupted when
	  returning back to app.

     657* driver/dtrace.c: Remove workaround for what I thought was a compiler
          bug. Avoid dtrace/memcpy_with_error - this was working around the
	  wrong issue. If the page fault handler works properly, then these
	  two things were bogus.

Wed Jan 25 21:49:03 2012  fox

     656* tools/tty.pl: Fix defunct process leakage.

Wed Jan 11 23:33:51 2012  fox

     655* instr_linux.c: Add -ltr instructions.

Fri Jan  6 22:11:05 2012  fox

     654* instr_linux.c: Add support for detecting rd/wr to GDT and writing
          to CR3 register. (Trying to find where the page table gets set
	  in the kernel to debug the page_fault problem for i386).

Mon Jan  2 23:16:35 2012  fox

     653* driver/instr_linux.c: Add support for SIDT instruction and fix
	  encoding for LIDT.

Sat Dec 31 00:01:52 2011  fox

     652* Archive: 1.0.135

     651* driver/intr.c: Move the interrupt/idt code to its own file.
          During driver init, swap out the old idt, make updates and
	  swap back in the new idt. Avoids funny crashes on driver
	  reload. 

	  Added /proc/dtrace/idt driver to see the IDT table.

     650* Archive: 1.0134

     649* dtrace_linux.c: set_idt_entry: Grab the existing IDT
          entry and just set the new address; dont lose/change the
	  dpl or type of the gate.

     648* printf.c, dtrace_linux.c: Move set_console() to printf.c

     647* makefile: Set LANG=C to avoid UTF-8 output from gcc.

Thu Dec 29 10:23:49 2011  fox

     646* driver/intr_x86-32.S: Kernels >= 2.6.31 save the %GS
          register on the stack, so we need to do this so that pt_regs
	  agrees. This fixes the horror of Ubuntu 11.10/i686 not
	  working properly.

Tue Dec 27 15:42:06 2011  fox

     645* tools/get-deps.pl: Fix for Ubuntu 11.10/i686 - very broken
          default install (missing /usr/include/sys).

     644* linux/sys/elftypes.h: Fix for Ubuntu 11.10/i686.

     643* tests/makefile: Handle 32-bit only system

Sun Dec 25 00:11:37 2011  fox

     642* prov_common.c: Fire for all matching probes, not just the
          first one, so that proc:::create and proc:::start will fire
	  together.

     641* prov_common.c: Add support for 
     	  proc:::create - Fires when a new process is created.
	  proc:::start - Fires when process is created (same as create)
	  proc:::end - Fires on process termination

     640* dtrace/dtrace_subr.c: When looking at the invop handlers, broadcast
          the breakpoint to all invop handlers. This allows two providers
	  to sit on the same instruction.

     639* fasttrap.c: Basic /proc/dtrace/fasttrap support.

Thu Dec 22 14:43:40 2011  fox

     638* driver/intr_x86-32.S/intr_x86-64.S: We werent allowing 
          user breakpoint traps to be detected by USDT. This works
	  again.

     637* Archive: 1.0133

     636* cmd/dtrace/dtrace.c: Add support for -arch [i386|x86_64], similar
          to Apple's dtrace.

     635* usdt/c/makefile: Fix so that build/simple-c can run from anywhere.

     634* libdtrace/dt_link.c: process_obj: For ELF32, write the SHT_REL entry
          back to the appropriate section, not the SHT_RELA section. Seems
	  like a Solaris symmetry bug.

	  We now get USDT probes, but nobody is garbage collection them. Strange....

Wed Dec 21 09:19:53 2011  fox

     633* dtrace.c: Temp remove the module-loaded chain because of fault
          on a reload.

     632* dtrace_linux.c: prfind(): Map pid to pid structure, and then
          lookup.

     631* ctl.c, driver_linux.c: Ensure fn_pid_task is setup, so prfind
          doesnt die on USDT probe.

     630* sys/dtrace.h: Fix for 32b build.

     629* cmd/instr.c: Fix size_t issue.

Mon Dec 19 21:48:57 2011  fox

     628* ctf_struct.c: Avoid referring to user32 struct so we can
          compile on RH4 kernel.

     627* dtrace_asm.c: Avoid compiler issue for old gcc/linux kernels
          for VMREAD instruction.

Sun Dec 18 00:17:43 2011  fox

     626* fbt_linux.c: Fix the arg0, .., arg4 args. We were relying
          on solaris stack layout and not linux. (Thanks Nigel for pointing
	  this out).

Sat Dec 17 13:53:47 2011  fox

     625* Archive: 1.0132

Fri Dec 16 20:27:53 2011  fox

     624* sys/systm.h: Use correct sysconf() mapping #defines.

Mon Dec 12 23:58:33 2011  fox

     623* dtrace.c, dtrace.h, dtrace_impl.h: Upgrade to Jolent 201111
          extras. (ECB delayed removal, vmreg support).

Sun Dec 11 21:23:47 2011  fox

     622* mutex.c: Call scheduler periodically to avoid deadlock when
          number of dtrace's exceeds number of physical cpus.

Fri Dec  9 20:01:25 2011  fox

     621* tools/tail.pl: Tool to do tail -f on any file, but specific
          char special device files. Defaults to /proc/dtrace/trace.

     620* Archive: 1.0131

     619* dtrace.c: Added some monitoring around the kmem_free in 
          dtrace_state_destroy. Not needed, but something to watch.

     618* x_call.c: Whilst priming the other cpus for action, attempt
          to drawin any pending call against ourselves. Reduce the
	  poll time (ack_wait) a little, to try and gain better performance.

     617* mutex.c: Dont disable interrupts for interrupt level mutexes.
          (And, by implication, we dont need to restore them). Parameterise
	  this change.
	  Every 500m iterations, dump a "taking a long time message". Havent
	  seen this get hit.

Thu Dec  8 21:44:49 2011  fox

     616* toxic.c: Re-enable more of the functions. We shouldnt have the old
          problems anymore.

Wed Dec  7 20:57:52 2011  fox

     615* tools/cpustuck.pl: Simple tool to look for stuck cpus.

Sun Dec  4 12:03:45 2011  fox

     614* dtrace.c: Tone down the teardown logic to avoid two
          dtraces fighting each other in a death match.

     613* mutex.c: Dont break out of mutex_enter, and dont dump
          stacks, as this can lead to inter-cpu corruption and assertion
	  failure.

     612* dtrace.c: When hitting an assertion/panic issue, avoid
          problems if another panic hits at the same time, to avoid
	  frantic console scrolling (we only allow the message once).

     611* Archive: 1.0130

     610* mutex.c: Dtrace is *NOT* expecting the mutexes to be recursive.
          By disabling recursion, we fix the "dtrace -l & dtrace -l &"
	  issue which would cause deadlock or assertions. We increased
	  the break-out check - we need to turn this off. Under very
	  heavy load (eg two copies of fbt::: running), the mutex
	  might believe somethings broke and panic dtrace, effectively
	  disabling it.

     609* dtrace_subr.c: Dump out the hdlr chain so we can diagnose
          if fbt is not at the front of the list.

Sat Dec  3 08:56:10 2011  fox

     608* dtrace_linux.c, fbt_linux.c: Arrange for fbt to be first in the
          invop chain (optimisation).

     607* x_call.c: Remove the per-interrupt xcall state. I think it was
          potentially wrong, as we were using our interrupt state instead
	  of the other cpu state. I am not sure it matters, because
	  we try to call xslave2 during fast-teardown to empty any
	  pending buffers, so the amount of contention should be reduced.

     606* instr_linux.c: Small parsing optimisation and opcode name fix
          for repnz.

     605* instr_linux.c: Add ud2 instructions to the INSTR provider. Would allow us to
          take a probe when a BUG_ON is hit in the kernel. (The
	  kernel will kill the calling process anyhow and dump a stack,
	  but at least we could do something useful if we were interested).

     604* prov_common.c: We were incorrectly assuming the file/line info
          for the BUG_ON/ud2 instruction followed the instruction. This
	  caused random disassembly failure leading to problems when that
	  code was executed. (FC16 flush_exec_old function would kill
	  new processes).
	  Avoid print mangling when debug enabled for tracing disassembler.

Wed Nov 30 22:12:40 2011  fox

     603* fbt_linux.c: Add an overrun indicator to /proc/dtrace/fbt ("*")

Tue Nov 29 09:00:40 2011  fox

     602* cmd/instr: New utility for getting instruction encodings.

     601* cyclic_linux.c: Avoid race conditions during timer closedown.
          We might have the timer active whilst trying to kill it, and
	  corrupt memory leading to a panic later on.

     600* dtrace_linux.c: Search fbt probes before instr, since we are likely
          to hit them first. (Minor optimisation).

Mon Nov 28 22:18:05 2011  fox

     599* sdt_linux.c, systrace.c: Avoid kernel memcpy function (cleanup
          rather than a bug fix).

     598* dtrace_linux.c: dtrace_printf/printk: Avoid calling vprintk,
          since we can get a recursion deadlock if we try and probe
	  fbt::form*:. Now we are detached from the kernel.

     597* dtrace_linux.c: dtrace_printf: Avoid deadlock if we interrupt
          our own cpu doing a printf.

Sun Nov 27 17:36:18 2011  fox

     596* dtrace.c, dtrace_linux.c: Add simple lock around dtrace_printf,
          so we dont get interleaved output. Add "stack" command to
	  /proc/dtrace/trace. Modify the fast-teardown to try and stop at
	  a red light, the other cpus. We use the original mechanism but
	  we try to prevent the flood of probes during teardown. 
	  Appox 20m probes could happen when fbt::: with 50,000+ probes
	  are enabled. This reduces it to may 1m. See /proc/dtrace/trace
	  for the results.

Sat Nov 26 21:40:32 2011  fox

     595* dtrace_linux.c: Add timestamp and cpu# to all /proc/dtrace/trace
          entries.

Fri Nov 25 17:26:03 2011  fox

     594* driver/dtrace.c: Rework the fast-teardown code since it didnt
          work reliably. This is potentially much better. We save
	  up the kmem_free's of the ECBs til the end and free them up
	  at the end.

     593* cpu_x86.c: Add missing emulation for POPF (0x9d) instruction.
          Fedora core does that in "not_same" label in this_cpu_cmpxchg16b_emu
	  subroutine.

Thu Nov 24 23:48:02 2011  fox

     592* cpu_x86.c: Remove curious code handler for iret.

     591* dtrace_linux.c: Map printk to dtrace_printf.

     590* vminfo.c: Got if..endif conditional wrong.

Wed Nov 23 21:04:52 2011  fox

     589* syscalls.c: Avoid hanging on sys64.

     588* systrace.c: Add "*" to show probed syscalls.

     587* vminfo.c: Avoid compile issue on older kernels.

Tue Nov 22 22:30:00 2011  fox

     586* driver/cpu_x86.c: When handling PUSHF instruction, dont leave
          interrupts disabled. Cause random crashes/kernel assertion issues.

Mon Nov 21 22:25:54 2011  fox

     585* driver/*: Remove some debug left behind. Some portability fixes.

     584* tests/syscalls.c: Add more syscalls. Fix potential hang on
          sigsuspend.

Sun Nov 20 10:46:20 2011  fox

     583* systrace.c: Add syscall counting stats to /proc/dtrace/stats.

     582* systrace.c: Add /proc/dtrace/syscall so we can see how many hits
          and do coverage analysis.

Fri Nov 18 20:21:52 2011  fox

     581* intr_x86-64.S: Disable NMIs whilst we are loaded to avoid calamity.

     580* x_call.c: Fix NMI (dont invoke it).

     579* x_call.c: Fix #elif.

Thu Nov 17 22:09:35 2011  fox

     578* dt_module.c: Fix unsigned isue leading to core dump.

     577* dtrace_linux.c, mutex.c: Add better diagnostics/error detection
          to mutex. Fix issue in dmutex_init which could lead to memory
	  corruption/kernel core dump.

Tue Nov 15 22:53:23 2011  fox

     576* driver/dtrace_linux.c: dtrace_printf() Add support for %ll
          specifier. Avoid core dump on printing null string.

	  Add magic number checking to vmem allocate/free functions.

Sun Nov 13 11:45:49 2011  fox

     575* Archive: 1.0128

     574* systrace.c: Fix for rt_sigsuspend in 2.6.26 and above kernels.
          Was causing build/sys64 to core dump or panic the kernel.

Fri Nov 11 23:03:36 2011  fox

     573* tests/instr.c: Instruction dumper, for diagnosing instruction
          single stepping issues. We dont build it, but theres a comment
	  to show how to build it.

Thu Nov 10 23:12:04 2011  fox

     572* fbt_linux.c: /proc/dtrace/fbt now include the probe hit count
          and an instruction dumper so we can verify if we are handling
	  all the fbt probes properly when single stepping.

Wed Nov  9 20:37:30 2011  fox

     571* dtrace_linux.c: Remove redundant mtx counters, add extra int3
          counters. Remove dtrace_printf() mutex since that cannot work.

     570* tests/syscalls.c: Remove non-determinism and random segfaults.

Tue Nov  8 21:53:44 2011  fox

     569* vminfo.c: Fix config issue for fc15/2.6.38 kernel.

Mon Nov  7 23:00:12 2011  fox

     568* mutex.c, linux_types.h, cyclic_linux.c: Revert cyclic_linux.c
          to using a spinlock - not a mutex or semaphore, since we
	  run from IRQ context/tasklet context.

	  Revert mutex.c to use pure semaphores, not hand crafted
	  mutexes. If the kernel is doing lock checking, it will
	  tell us if we might block in an interrupt routine, which is
	  bad. (So, why did I do mutex.c at all? Probably got confused
	  by some other issues).

	  linux_types.h: Wrap the semaphores with the dmutex call
	  wrappers. Dont really need this.

	  Anyway, I still kept some of the lock-tightening code which
	  was possibly upsetting the kernels.

Sun Nov  6 00:41:47 2011  fox

     567* tools/tty.pl: Commarise numbers when dumping stats. Dont
          keep printing the prompt message.

     566* mutex.c: Calling the wrong panic routine!

     565* dtrace_linux.h: Dont map dtrace_panic to panic.

     564* driver/vminfo.c: Compile warnings cleanup; avoid table overflow.

     563* driver/prov_common.c: Avoid table overflow.

     562* driver/dtrace_isa.c: Fix stack() which paniced us because of
          changes in Linux 2.6.39 and above.

     561* Archive: 1.0127

     560* driver/dtrace_linux.c: Turn off interrupt level dtrace_printf()
          calls. Might be affecting Liux 3.0 kernel (probably not). But
	  removes noise from /proc/dtrace/trace.

	  Add mutex protection to the par_alloc family of functions.

	  Add some extra stats to /proc/dtrace/stats.

	  Dont update a PTE entry if what we are doing is the same as it is
	  (performance optimisation). Add multipage support.

     559* driver/dtrace_linux.h: Add libc_ prototypes.

     558* driver/dtrace_subr.c: Hide dtrace_sync_func since we need the
          real one to be global for xcall.c to see it.

     557* driver/fbt_linux.c: Put some protection in, in case the par_alloc
          fails to allocate (normally, because I was testing/disabling par_alloc).

     556* driver/*: Replace mutex calls with dmutex calls (to avoid
          name clash).

     555* driver/cyclic_linux.c: Replace kernel spinlocks with our
          mutex calls.

     554* driver/ctf_struct.c: Add cpumask typedef. Trying to fix the
          ctfconvert issue, but this change doesnt fix it. Need to
	  dive into the dwarf code to see what gcc on Ubuntu 11.10
	  is tickling the "cant resolve type" bug.

     553* driver/libc.c: Some simple string functions. I really want
          "nm -u build/driver/dtracedrv.ko" to be clean. It isnt, but
	  some of these functions are only called during load time.
	  We'll defer proper libc cleanup til a later date.

     552* driver/mutex.c: Our simple spinlock mutexes.

     551* linux_types.h: Use our own mutex/dmutex implementation to
          avoid problems with Linux mutexes trying to preempt us.

     550* cmd/ctfconvert/dwarf.c: Avoid bailing out when hitting a DWARF
          issue. This may be caused by newer gcc's emitting something
	  we dont understand. For now, just work around it.
	  (cpumask is causing the problem).

Sat Nov  5 16:46:29 2011  fox

     549* bug.sh: Dont keep telling people to contact me - just on the
          first failed build.

Sun Oct 30 12:25:08 2011  fox

     548* dtrace_linux.c: Add a panic notifier (dont need it, just for
          debugging).

     547* prov_common.c: Dont let us reparse the common provider table
          multiple times.

     546* tools/tty.pl: New load causing tool.

Wed Oct 26 22:19:06 2011  fox

     545* vminfo.c: Make it work (or at least compile) for <2.6.39 kernels.

     544* prov_common.c: Changes so that we can add support for instruction
          level providers (ie. vminfo). Change the map[] table to
	  allow room for expansion (ideally a linked list), and provide
	  support for instruction level kernel disassembly callbacks.

     543* vminfo.c: New file. vminfo provider. Similar to Solaris, but
          annotating all the /proc/vmstats counters.

     542* dtrace_linux.c: Move some of the init calls to after the end of
          syms_write, so we can share the prcom code with the vminfo code.

     541* dtrace_linux.c: Fix PAGE_NX issue for cpu_core_exec[] structure
          on 3.x kernels.

Tue Oct 11 23:49:58 2011  fox

     540* cmd/ctfconvert/makefile: Fix bad linking for unused ctfdump
          command.

Mon Oct 10 21:50:42 2011  fox

     539* Fix last change to dtrace_isa.c to get the if/else/endif the
          right way around. Thanks Nigel.

     538* dtrace_isa.c: Fix for kernels >= 2.6.39, since the stacktrace_ops
          have lost the .warning/.warning_symbol members.

Fri Oct  7 19:53:05 2011  fox

     537* mkport.pl, ctfconvert/dwarf.c: Handle missing cplus_demangle()
          function. Reported on SLE11.

Mon Jul 18 23:15:08 2011  fox

     536* driver/systrace.c: Fix for execv on 2.6.38 kernels.

Wed Jul 13 21:30:01 2011  fox

     535* dt_lex.l: Handle // comments at outer scope.

     534* tests/tests.d: Better use of timestamps to exit test after 5s
          to avoid problems with dropped probes.

Tue Jul 12 07:07:55 2011  fox

     533* driver/prov_common.c: Avoid null ptr deref for some probes.

Fri Jul  8 09:19:49 2011  fox

     532* prov_common.c: Provide common framework, similar to notifier.c
          but more generic for more Solaris defined probes.

Sat Jun 25 23:44:06 2011  fox

     531* notifier.c: New provider to intercept the notifier chains.

Fri Jun 24 20:12:27 2011  fox

     530* tools/load.pl: Use uname -r to find the right build dir when
          we have multiplate build/ directories in the same build area.

     529* x_call.c: Fix the targetted cpu scenario.

Thu Jun 23 20:37:10 2011  fox

     528* driver/systrace.c: Fix for i386 sigreturn() syscall.

Wed Jun 22 21:18:17 2011  fox

     527* tests/syscalls.c: i386 now invokes sigreturn() syscall - which is
          panicing the 64b kernel...off to fix that.

Tue Jun 21 22:44:58 2011  fox

     526* tools/mksyscalls.pl: Better unistd.h detection and tbl file generation.

Mon Jun 20 20:17:31 2011  fox

     525* tools/mksyscalls.pl: Updates for 2.6.18 kernel.

     524* dtrace_linux.c, x_call.c: Fixes for 2.6.18 kernel and >8 cpus.

     523* driver/intr_x86*.S: Add cnt_nmi1/cnt_nmi2 so we can see if we are
          called.

     522* doc/security.txt: Update the security instructions.

Sun Jun 19 00:13:41 2011  fox

     521* dtrace_linux.c, x_call.c, intr_x86*.S: Add support for NMI interrupts
          when IPI is not responding.

     520* Archive: 1.0126

     519* tools/mkport.pl: Try the /lib/modules/ver/build dir for probing
          when theres no /lib/modules/ver/source directory (for Centos 4).

     518* driver/x_call.c: Lockless implementation. Use NCPU*NCPU*2
          xcalls array to avoid interference with parallel xcalls from
	  other cpus or our own interrupt routines.

     517* Archive: 1.0125

Fri Jun 17 22:15:28 2011  fox

     516* systrace.c: Fix execve() for 32b apps on 64b kernel.

Thu Jun 16 23:19:02 2011  fox

     515* Archive: 1.0124

     514* dtrace_linux.c: Provide multi cpu stack dumping via NMI code in
          kernel.

     513* x_call.c: Use cmpxchg instruction to ensure visibility between
          cpus.

Sun Jun 12 00:01:13 2011  fox

     512* mkport.pl, dtrace_isa.c: Handle dump_trace() specially with
          respect to number of args.

     511* driver/sdt_linux.c: Fix stupid issue in io::: probe on
          AS4 (2.6.9 kernels).

Sat Jun 11 14:52:06 2011  fox

     510* archive: 1.0123

     509* systrace.c: Implement the 64b/32b ptregs syscalls.

Wed Jun  8 22:42:07 2011  fox

     508* tests/syscall.c: dummy app to allow us to put in lots of syscalls.
          Am debugging whether "dtrace -n syscall:::" can crash system for
	  some syscalls, especially the pt_regs calls for a 32b binary
	  on a 64b kernel.

     507* dtrace_isa.c: Fix for 2.6.38 kernel.

Tue Jun  7 22:00:51 2011  fox

     506* systrace.c: Fix 32b syscalls on 64b kernels (forgot to finish
          this off last night).

Mon Jun  6 22:23:23 2011  fox

     505* Archive: 1.0122

     504* systrace.c: Fix for 64-bit. Looks like it works, but need to
          cleanup the ptregs calls - dont think we need the special
	  patching code.

	  Tested on Ubuntu 10.04 only so far.

     503* systrace.c: Revamp for 32-bit kernels, so that we dont
          get occasional ERESTART issues on syscalls.

	  This took a lot of effort to get going, but the code is
	  better for 32-bit kernels. Now off to refix the 64-bit
	  kernels.

Sat May 14 08:38:13 2011  fox

     502* systrace.c: Fix bug where we overwrote the saved underlyer
          when doing a syscall trace, and another /usr/sbin/dtrace does
	  something unrelated (like 'dtrace -l').

Tue May 10 23:05:18 2011  fox

     501* systrace.c: Avoid two user-mode dtraces causing syscall interception
          to point to itself and hang/panic/stack overflow the kernel.

Mon Apr 25 12:14:15 2011  fox

     500* dtrace_linux.c: dtrace_xcall: Take out the local_irq_enable causing
          load-time kernel warnings to be displayed.

     499* systrace.c: Fix 32-bit syscalls on 64-bit kernel causing possible
          coredumps (saw this on exit/exitgroup syscall).

     498* dtrace_linux.c: Disable the new xcall code - it wont/cannot work
          since if we fire a timer on another cpu, the timer may not
	  fire until we exit the current interrupt. The other
	  code seems better, (having enabled irq's), lets see how well.

Sun Apr 24 12:06:54 2011  fox

     497* Archive: 1.0121

     496* tests/tests.d: Some more random tests.

     495* driver/systrace.c: Rework the syscall trampoline code to
          avoid RAX/EAX corruptions, along with some other fixes.

     494* driver/dtrace_linux.c: Implement out own safe dtrace_xcall
          using cross-cpu timers. Not brilliant, but seems to avoid
	  deadlocks or kernel warnings. We have a deadman timeout to
	  avoid real deadlocks. (We occasionally see the impossible in a 
	  VM where timers dont fire or fire too late; I think we handle
	  this ok). dtrace_xcalls are not that frequent, so it should have
	  minimal performance decrease.

Tue Apr 19 21:04:57 2011  fox

     493* driver/systrace.c: Rewrite syscall wrappers to handle OpenSuse
          kernel - this method appears to be better/more generic for
	  all other platforms.

	  Code is looking messy - dups of functions and horrible 32/64 
	  ifdefs and will look to clean up at a later date.

Mon Apr 18 22:30:20 2011  fox

     492* tools/load.pl: Add in support for vmalloc_exec.
	  dtrace_linux.c: Add support for vmalloc_exec.

Fri Apr 15 22:06:53 2011  fox

     491* tools/mkport.pl: Updates for OpenSuse 11.

Thu Apr 14 21:51:25 2011  fox

     490* mksyscall.pl: Fix for opensuse syscall table finding.

     489* tools/mkport.pl, dt_impl.h: Add support for yytext[] array which
          some lex/flex's will generate for us.

Sun Apr 10 10:11:43 2011  fox

     488* Archive: 1.0120

     487* driver/dtrace.c: Fix gcc 4.2.4 compiler bug in dtrace_probe().

     486* dtrace_isa.c: dtrace_copyinstr: Fix the direction of copying.

Sat Apr  9 08:32:59 2011  fox

     485* dtrace.c: dtrace_probe(): Implement reentrancy avoidance mechanism.

     484* dtrace.c, dtrace_asm.c, dtrace_impl.h: Add support for the
          illval error/ptr for a faulted return, so that arg5 on an ERROR
	  probe has access to it.

Thu Apr  7 22:26:26 2011  fox

     483* intr_x86-64.S: Better detection in dtrace_memcpy_with_error
     	  that we had been aborted (%rcx != 0)

     482* driver/dtrace.c: dtrace_load##bits: More accurate
          transcription of fault handling when we have a bad address. Werent
	  quite copying what on/usr/src/uts/i86pc/ml/locore.s does -
	  we needed to set the CPU_DTRACE_NOFAULT flag because the
	  dtrace_memcpy_with_error() wont be processed by the Linux
	  exception table, so we need to avoid propagating the page faults.

Wed Apr  6 22:22:26 2011  fox

     481* intr_x86-32.S, intr_x86-64.S: Fix sillyness when taking a handled
          page fault where we forgot to remove the error code.

Mon Apr  4 20:48:25 2011  fox

     480* prov_common.c: Handle the BUG_ON instruction.

Sun Apr  3 13:36:39 2011  fox

     479* Archive: 1.0119

     478* sdt_linux.c: Avoid 'dtrace -n io:::' panicing kernel on the
          io:::return probes.

     477* etc/io.d, etc/sched.d: Avoid need to delete the files - we rely
          on pragma depends_on module linux to handle a missing linux.ctf
	  file.

Sat Apr  2 18:09:03 2011  fox

     476* intr_x86-32.S: Refactor to use same macro as intr_x86-64.S

     475* makefile, tools/tests.pl: Add actual testing for specific 
          probe scenarios and interrupt handlers.

     474* intr_x86-64.S, dtrace_linux.c: Fixes for page fault issues
          (cpu_table wasnt set up properly). Rewrote interrupt routines
	  to use a common macro.

Wed Mar 30 20:59:41 2011  fox

     473* intr_x86-32.S: Add missing dtrace_int11 code.

     472* dtrace_linux.c: Remove temp debug code which broke i386 build.

     471* Archive: 1.0118

Tue Mar 29 23:00:47 2011  fox

     470* driver/dtrace_isa.c: Add some safety checks around the stack
          probing so we cannot trigger a kernel pgfault (but a Dtrace intercepted
	  one is ok).

     469* driver/dtrace_linux.c: Add fixed dtrace_int13_handler and add
          dtrace_int11_handler (not sure we need this, for segment-not-present).
	  Add counters to /proc/dtrace/stats to allow us to see these firing.

     468* driver/instr_linux.c: Add support for LIDT instruction probes
          (either it doesnt work, or there arent any in 64-bit space which
	  the disassembler finds; will fix later - I wanted it to help debug
	  early version of the pgfault handler problem).

     467* driver/intr_x86-64.S: Fix pgfault interrupt to understand the
          fault code on the stack.

Tue Mar 22 19:29:23 2011  fox

     466* cmd/ctfconvert: Some changes to try and support this on RHEL4, but it
          wont work since RH4 doesnt have a good libdwarf.so on the system.

	  libdwarfext.h: Added to provide missing definitions. Not needed for
	  other platforms.

     465* tools/mkport.pl: Fix typo on elf_getshstrndx

Sun Mar 20 13:07:00 2011  fox

     464* tools/mkport.pl: Fix HAVE_ELF_GETSHDRSTRNDX detection for
          FC-14 (which fixes the etc/shed.d issue).

     463* mkctf.sh: Force ue of dwarf-2 so that Fedora-14 doesnt
          bomb out since its defaulting to dwarf-3.

     462* dtrace_isa.c: Fix for 2.6.38 kernels.

     461* cyclic_linux.c: Add support for 2.6.38 kernels.

     460* usdt/c/makefile: Add -x nolibs to avoid build problems on etc/sched.d

Fri Mar 18 23:20:47 2011  fox

     459* Archive: 1.0117

     458* driver/dtrace.c: Add some annotations for debugging illaddr
          traps.

     457* driver/ctf_create.c, driver/ctf_create.h: Fix the buf_t
          definition so that it works with etc/io.d. 

	  etc/io.d: Now works to trace elements of fileinfo_t, dev_t, and
	  bufinfo_t (io probe provider).

	  sdt_linux.c: Properly handle the buf_t type for io provider.

	  Note: some fields may not yet be populated or populated properly
	  but at least we now get device names and path names and a few
	  other key fields.

	  The following now "works":

	  dtrace -n io:::start'{printf("%x %x %x %s %s %p", 
	  	args[1]->dev_major, args[1]->dev_minor, 
		args[1]->dev_instance, args[2]->fi_pathname, 
		args[1]->dev_pathname, arg2);exit(0);}'

Tue Mar  8 22:02:52 2011  fox

     456* usdt/c/makefile: Fix circular dependency on linux.ctf file. We wont
          have built it yet, but we dont need it yet, either.

Mon Mar  7 23:12:26 2011  fox

     455* sdt_linux.c: Fix the io::start/done to populate bcount and size.
          Add buf_t.b_addr also.

	  I still dont like this emulation of io::start/done - its intercepting
	  the VFS read/write syscalls, and I need to see if Solaris is doing
	  this at a lower level. But the demo/io/whoio.d script now works

Sun Mar  6 12:43:06 2011  fox

     454* driver/sdt_linux.c: Correct the bcount for io::start/done.

Fri Mar  4 23:29:07 2011  fox

     453* driver/dtrace_linux.c, driver/ctf_struct.c: Expose some
          new structures so that etc/sched.d can work, and we can return
	  the "cpu" we hit a probe on.

Wed Mar  2 18:52:46 2011  fox

     452* Archive: 1.0116

     451* makefile: Allow DESTDIR prefix for installing into 
          non-system areas.

	  sebastian AT pipping DOT org

Tue Feb 22 23:36:43 2011  fox

     450* libdtrace/dt_module.c: Ensure the ctf file is loaded with a distinct
          name from /proc/kallsyms.

Thu Feb 17 23:36:52 2011  fox

     449* dt_module.c: Allow us to read linux-XXX-generic.ctf from the
          dtrace build dir when debugging. 

	  We have a catch-22 during build because when we build the USDT
	  stuff, we invoke dtrace without the linux.ctf file being ready.
	  But etc/sched.d currently needs this for dtrace_cpu_t.

	  Will need to change the order of building to resolve this, but
	  am unhappy that we cannot put good conditionals into sched.d
	  (like #if or "#pragma D depends_on type XXX").

Tue Feb 15 21:37:43 2011  fox

     448* libdtrace/dt_module.c: Fix for HAVE_ELF_C_READ_MMAP

     447* tools/mkport.pl: Fix for HAVE_ELF_C_READ_MMAP

     446* tools/mksyscall.pl: Picking up the 64-bit unistd.h for the 64/32 table.

Sun Feb 13 15:19:12 2011  fox

     445* libdtrace: Move references to /usr/lib into dt_get_libdir() function.

     444* driver/linux_pure.c, driver/dtrace.c: remove: linux_pure.c

     443* Archive: 1.0115

     442* dtrace_linux.c, tools.pl, systrace.c: Add in ia32_sys_call_table as
          an external symbol so we can intercept 32b syscalls on AS4/64b kernels.

     441* driver/systrace.c: Fix problems with not handlign 32b syscalls on 64b 
          kernels.

Sun Feb  6 09:05:12 2011  fox

     440* Archive: 1.0114

     439* systrace.c: When creating probes, give a module name (x64 or x32)
          so we can distinguish 64b syscalls from 32b ones. Avoids
	  rampant probe creation as we kept creating the probes
	  on every use of dtrace.

Sat Feb  5 10:24:19 2011  fox

     438* dtrace_linux.c: Init syscalls before fbt/instr provider. We can 
          reduce memory this way. Bizarre. But true.

     437* Archive: 1.0113

     436* dtrace_linux.c: Fix par_free. Now correctly handles modules being
          loaded/unloaded for fbt tracing.

Fri Feb  4 22:09:23 2011  fox

     435* fbt_linux.c, instr_linux.c: par_alloc/par_lookup: We had a collision
          on modules registered by the two providers because they seem the
	  same mem addresses. Add support or 'domains' to allow us to
	  distinguish the clashing entries. This should fix the issue where
	  the instr provider stops fbt from providing fbt probes for a module.

Thu Feb  3 22:10:03 2011  fox

     434* dtrace.c: Add support for module_loaded notifier.

Sun Jan 30 22:07:28 2011  fox

     433* systrace.c: Initial attempt at supporting 32b/64b mode. Not yet
          finished, but now have two syscall tables to look after.

     432* mksyscall.pl: Generated sorted and expanded syscall tables. i386 and
          amd64 have no commonality of syscalls, and we cannot rely on
	  #includes to access the symbol syscall names.

Wed Jan 26 23:36:19 2011  fox

     431* mksyscall.pl: Generate both 32 and 64 bit sys call table 
          irrespective of kernel.

Sun Jan 23 09:12:58 2011  fox

     430* cmd/dtrace/makefile: Always use -lelf and stop trying to be
          too clever. Try and fix the USDT link time problems.

     429* tools/mkport.pl: Fix silly in "my" definitions.

Sat Jan 22 09:33:14 2011  fox

     428* tools/yacc.pl: New file to handle bison/yacc discrepancies.

Thu Jan 20 22:18:52 2011  fox

     427* Archive: 1.0112

     426* various: fixes for Linux 2.6.37 - rejig some mutex/semaphore
          defines, fix ioctl/unlocked_ioctls, comment out unneeded ioctls,
	  fix some driver warnings.

Wed Jan 19 22:10:14 2011  fox

     425* dt_link.c: Avoid the '#undef linux' as this can cause a bad
          i386 simple.o being generated, and stopping the build process.
	  May want to investigate further if I break amd64 support.

Mon Jan 17 21:43:03 2011  fox

     424* cmd/dtrace/makefile: Fix invocation of "-lelf" so we pick up the
          right library.

     423* linux_types.h: Change MUTEX_DEFINE to be optionally DEFINE_MUTEX
          instead of DECLARE_MUTEX.

     422* instr_linux.c: Remove redundant ioctl/compile error.

     421* ctf_impl.h: Include linux/slab.h if we have it, for kmalloc.

     420* mkport.pl: Add support for linux/slab.h

     419* driver/ctf_mod.c: Remove redundant ioctl (compile issues).

     418* dt_module.c, tools/mkport.c: Handle not having ELF_C_READ_MMAP.

Sat Oct 16 21:36:01 2010  fox

     417* linux_types.h: Some cleanups due to Ubuntu 10.10

     416* libproc/common/pr_stat.c: Similar to the below - handle
          missing blksize_t typedef.

     415* libproc/common/Pcore.c: Horrible hack because the glibc
          in Ubuntu 10.10 is broken (or maybe correct) compared to Ubuntu
	  10.04. We cannot get things to compile on both systems without
	  the "extern size_t pread64();" hacks you will see in there.

Fri Oct 15 23:11:15 2010  fox

     414* dt_grammar.y: Add more aliases for token idents.

Thu Oct 14 23:59:33 2010  fox

     413* scripts/bkl.d: Simple big kernel log script.

Sun Oct  3 19:24:38 2010  fox

     412* mkport.pl, dt_module.c: Allow for elf_getshstrndx if 
          elf_getshdrstrndx is not available.

Sat Sep 18 21:24:49 2010  fox

     411* dt_grammar.y: Try and avoid enum names in bison error messages
          (such as DT_TOK_IDENT).

Wed Sep 15 23:20:31 2010  fox

     410* dt_module.c: Changes so that we can refer to "kernel`symbol" in
          D scripts. At present, we fake all symbols in /proc/kallsyms
	  to look like functions. (Not a good move, but sufficient for
	  now).

Sat Sep  4 21:29:24 2010  fox

     409* dt_module.c: Fix loading of /proc/kallsyms so that "stack()"
          prints out all symbols. (We werent sorting the data from
	  /proc/kallsyms so we lost some of the stack names).

     408* Archive: 1.0111

     407* libdtrace/dt_link.c: Dont use 'linux' to enable my code changes;
          usdt/c doesnt seem to want this anymore. Why? No idea. Could be
	  my build environment was impure before, but is sane now. Or maybe
	  vice-versa.

Tue Aug 31 21:18:28 2010  fox

     406* dt_module.c: Some performance optimisations.

     405* driver/ctf_struct.c, ctf_struct.h, sdt_linux.c: Refactor the public
          structures.

Mon Aug 30 19:52:32 2010  fox

     404* Archive: 1.0110

     403* driver/ctf_struct.c: Add more structs so that io.d will work
          properly.

     402* makefile: Add /usr/lib/dtrace/linux-`uname -r`.ctf and remove
          types.d; fix issues with io.d

     401* libdtrace/dt_module.c: Load /usr/lib/dtrace/linux-`uname -r`.ctf on
          startup to get the CTF symbols for the kernel.

Tue Aug 24 23:08:48 2010  fox

     400* cmd/ctfconvert/dwarf.c: Workaround for pathological GCC 
          structure members which would cause us to abort.
	  (Set GCC_WORKAROUND to a non-blank value to see the errors you 
	  might get). If you compile the driver with -g (edit driver/Makefile),
	  then you can do this:

	  build/ctfconvert -L lab1 -o /tmp/ctf build/driver/cpu_x86.o

	  to get a file to play with. (Use ctfdump to see whats in it).

	  Next up is to have a compile-only file with debug symbols
	  so we can generate a kernel CTF image, to store in /usr/lib/dtrace.

Mon Aug 23 21:52:56 2010  fox

     399* cmd/ctfconvert: Add 'ctfdump' utility to list out the CTF sections.

     398* Archive: 1.0109

     397* cmd/ctfconvert: New binary to support CTF/.SUNW_ctf sections.
          You need the libdwarf-dev and binutils-dev package for this to
	  work.

Mon Aug 16 21:48:52 2010  fox

     396* dtrace.c: Rework the dtrace_loadNNbits functions to avoid taking
          a segmentation violation if user is being nasty (or a bug
	  in a translator), by using the Linux validation framework.
	  This gives us a win over Solaris as we dont have to check toxic
	  ranges. Could optimise further for the 8/16/32 bit case to avoid
	  the call to dtrace_memcpy_with_error(), but probably
	  most of the time its a 64-bit move (on a 64b kernel) so not
	  a big loss. Maybe we will optimise one day when we can profile
	  if it is a real rather than an imaginary problem.

Sun Aug 15 09:31:16 2010  fox

     395* driver/dtrace_isa.c: Fix invocation of call to __range_not_ok 
          which was born in 2.6.31 kernels.

Tue Aug 10 21:58:37 2010  fox

     394* Psymtab.c: avoid core dump when ^C-ing dtrace after dumping ustack().

Mon Aug  9 23:27:04 2010  fox

     393* Archive: 1.0108

     392* dtrace_isa.c: Better ustack handling - try and walk the stack
          better to avoid false positives.

     391* Psymtab.c: Dont use a.out for the app name in stack dumps, use
          /proc/pid/exe. 

Sun Aug  8 10:46:01 2010  fox

     390* mkport.pl: Wasnt picking up stacktrace.h properly from later kernels.

Sat Aug  7 10:38:52 2010  fox

     389* archive: 1.0107

     388* intr_x86-32.S: Fix assembler sillyisms for dtrace_memcpy_with_error.

     387* dtrace_isa.c: Use same logic for 32-bit kernels as 64-bit when
          doing dtrace_getupcstack.

     386* libdtrace/dt_module.c: Fix the way we load /proc/kallsyms to 
          get the correct symbol sizes, to help ustack().

     385* dtrace_isa.c: Remove the dwarf test code and use a different algorithm
     	  for 64-bit stacks - more in line with what the kernel does on stack dumps.
	  (stack() function).

Fri Aug  6 21:55:59 2010  fox

     384* intr_x86-32.S/intr_x86-64.S: Add dtrace_memcpy_with_error to
          detect GPF in D programs.

     383* etc/linux.d: Add uid_t type so we can access "uid" function.

Thu Aug  5 23:53:55 2010  fox

     382* dtrace_isa.c: Change the copyin/out functions to handle
          user level segmentation violations and report back to the user.

Wed Aug  4 23:10:26 2010  fox

     381* libproc/Pcontrol.c: Dont ptrace(ATTACH) to a proc if we want readonly
          mode. (Avoid ustack() hanging procs).

     380* dtrace.h, dt_consume.c: Add stacksymbols option from Apple to turn
          off symbolic stack traces. (Debugging or speed up).

     379* dtrace_asm.c: Avoid kernel GPF if given a bad user addr to copy.
          Not perfect as we should generate a DTRACE fault - but cannot
	  simulate this just yet.

Tue Aug  3 20:54:35 2010  fox

     378* scripts/dt.pl: Add timestamps, support for signals and show
          filenames in "files" subsystem.

     377* dtrace.c: Fix handling of DIF_VAR_PPID to avoid GPF.

Mon Aug  2 20:13:02 2010  fox

     376* dtrace_linux.c, dtrace_linux.h: Create our own read_cr2_register()
          function to avoid OS dependencies.

Sun Aug  1 22:23:22 2010  fox

     375* cmd/dtrace/dtrace.c: Avoid core dump if doing "dtrace -S -v -s script"
          and no CTF data for the internal SDT probe.

Fri Jul 30 09:07:10 2010  fox

     374* privregs.h: Fix register names for sdt_linux.c (32-bit kernels).

Thu Jul 29 00:10:39 2010  fox

     373* dtrace_isa.c: Fix dtrace_copyin: Had args reversed in copy_from_user.
          Cause copyinstr() to panic the kernel.

     372* archive: 1.0106

     371* privregs.h, sdt_linux.c: Get the proper args to the calling function
          for the probe from user space. Now io::start has the right
	  arg0, arg1, ..., but we still havent made available a working
	  io.d and no fileinfo_t (etc) args.

Wed Jul 28 08:42:55 2010  fox

     370* libdtrace/dt_lex.l: Better // handling.

     369* libdtrace/dt_lex.l: #define ECHO as nothing to avoid writing to
          stdout.

     368* scripts/mkinstall.pl: Install /usr/lib/dtrace/ regs.d, signal.d,
          unistd.d

     367* instr_linux.c: Same as #366 - avoid aliasing issues.
          These changes fixed an issue where a fbt::do_sync_read:return
	  would never fire, because the SDT provider was stealing the probe
	  despite it not having been enabled.

     366* sdt_implh, sdt_linux.c: Mark a probe as enabled or not so
          we avoid aliasing issues if fbt is probing the same instruction.

     365* archive: 1.0104 (build.pl was missing)

Tue Jul 27 19:47:26 2010  fox

     364* cpu_x86.c: Add a NOP instruction after the instruction we are
          going to single step, since some instructions may execute
	  that even if the single-step trap flag is set. E.g.
	  "MOV %CR3,%EAX".

Mon Jul 26 15:38:00 2010  fox

     363* uncompress.c, mkport.pl: Allow us to not have zlib functions in 
          kernel. Dtrace mostly works without this, but may affect the 
	  ctf functions.

     362* tools/mkport.pl: Fix for Debian 4-i386 kernel (smp_call_function_single)

     361* dtrace_isa.c: Fix for Debian 4-i386 kernel (2.6.24)

Sun Jul 25 20:18:41 2010  fox

     360* archive: 1.0103

     359* sdt_linux.c: Add io::start and io::done exits.

     358* prov_common.c: Add dtrace_function_size() utility.

     357* driver/dis_tables.c: Grab this from Darwin-10.6.4. It has the
          fixes needed to support multibyte nop instructions and a silly
	  coding error in handling mode operations.

Fri Jul 16 23:56:16 2010  fox

     356* fbt_linux.c, prov_common.c: Migrate the function parser code
          to a reusable function so we can share it with other providers.

Thu Jul 15 21:00:35 2010  fox

     355* archive: 1.0102

     354* Better parsing for SMP_CALL_FUNCTION_SINGLE_ARGS and SMP_CALL_FUNCTION_ARGS

     353* toxic.c: Add ia32_sysenter_target for x86-64 kernels when 32 bit
          apps invoke a i386 SYSENTER syscall.

Wed Jul 14 20:51:58 2010  fox

     352* dtrace_linux.c: dtrace_int13_handler: Annotate a dtrace induced
          segmentation violation so that the error can be propagated to
	  the user.

     351* dtrace_linux.h, dtrace_linux.c: Try and handle smp_call_function()
          with 3 or 4 args depending on kernel version.

Tue Jul 13 20:57:56 2010  fox

     350* driver/systrace.c: Fix sigaltstack for 2.6.32/64 bit kernel.

     349* driver/systrace.c: Fix fork/vfork() for 2.6.32/64 bit kernel.
          scripts/fork.d is crashing the kernel on the copyinstr.

     348* driver/systrace.c: Fix iopl() syscall for 2.6.32/64 bit kernel.

     347* driver/systrace.c: Fix rt_sigreturn() for 2.6.32/64 bit kernel.

     346* driver/systrace.c: Fix execve() for 2.6.32/64 bit kernel.

Mon Jul 12 22:24:58 2010  fox

     345* systrace.c: Fix clone() syscall for 2.6.32/64 bit kernel.

Sun Jul 11 21:40:15 2010  fox

     344* dtrace_linux.c: dtrace_xcall: Dont call on_each_cpu as its not
          interrupt safe.

Sat Jul 10 21:40:00 2010  fox

     343* Archive: 1.0101

     342* Release: 20100710

     341* ctl.c, dtrace_linux.c: Changes for Centos 5.5 (
          (Nathan Slingerland at RotateRight, LLC)

     340* dt_lex.l: Add support for C++ // comments.

     339* dtrace_linux.c: When doing an ioctl, pass down CRED() so we can
          probe memory if we are root.

Tue Jul  6 11:56:02 2010  fox

     338* libdtrace/dt_dis.c: Fix typo where we get a double "0x" in the output
          for "-S -s" switches.

Mon Jul  5 21:07:32 2010  fox

     337* dtrace_isa.c: Uncomment dtrace_getarg() - but its not good for
          now. This allows tests/dt021 to "run" (but not properly).

Sun Jul  4 21:30:37 2010  fox

     336* dtrace_linux.c: __validate_ptr, using "=a" instead of "=&a"
          when calling this inline asm. Seems to cure the problem on
	  line 1259 (instr_in_text_seg) where we GPF'd if we didnt
	  have a print statement.

Sat Jul  3 22:03:17 2010  fox

     335* common/ctf/ctf_create.c: Update from OpenSolaris. This fixes the
          tests/dt020 bug.

     334* load.pl, makefile: Add "/sbin" to insmod and rmmod. 
          (slingn@rotateright.com)
	   
Wed Jun 30 23:16:29 2010  fox

     333* driver/dis_tables.h: Set OPLEN to a teeny number if not disassembling
          since we might break the interrupt stack frame.

Fri Feb 12 22:10:46 2010  fox

     332* libdtrace/i386/dt_isadep.c: Fix references to text[] array as
          pointed our by Sanjay. (sanjay3000-at-gmail.com)

Tue Jan  5 22:09:05 2010  fox

     331* tools/load.pl: Allow i386die_chain as a synonym for die_chain for 2.6.16
          kernels. <aacosta@gmv.com>

Wed Dec 23 10:26:39 2009  fox

     330* dtrace_linux.c: mem_set_writable: Patch from Soumendu to fix problem on
          2.6.27.39.x86_64 kernel. satapathy.soumendu&at;gmail.com.

Sat Oct 10 15:51:22 2009  fox

     329* libproc/*: Rename dprintf to p_dprintf to avoid glibc issues.

     328* dtrace_linux.c, dtrace_isa.c: Fixes for 2.6.31 (hopefully 2.6.32)
          kernels. May have broken older kernels.

Mon Aug  3 23:18:25 2009  fox

     327* dtrace_isa.c: Fix for oldrsp in 2.6.30.4 kernel.

Fri Jul 24 20:13:15 2009  fox

     326* Archive: 1.0100

     325* Below two changes for RH 2.6.18-92.1.13.el5xen kernel
          (Oracle platform)

     324* intr_x86-32.S: Add conditional if __KERNEL_PERCPU does not
          exist in this kernel.

	  missing.c: Add checks for strcmp/strlen to avoid asm/string.h
	  conflicts.

	  profile.c: Add missing strcasecmp() prototype.

     323* missing.c: smp_call_function_single: Only defined if SMP, not
          a kernel version specific function.

	  tools/mkport.pl: Add special detection for smp_call_function_single
	  since theres a variety of scenarios to support.

Thu Jul 23 20:07:18 2009  fox

     322* Pcontrol.c, Putil.c: Rename dprintf to p_dprintf to avoid glibc
          clashes.

Mon Jul 20 21:10:50 2009  fox

     321* driver/dwarf.c: Subroutine code for stack walking and for the
          build/dwarf userland binary.

     320* Archive: 1.0099

Tue Jul 14 21:15:07 2009  fox

     319* toxic.c: Add level3_kernel_pgt and level2_kernel_pgt because
          these are data areas, not code areas. (Nicolos.Willams@sun.com)

Fri Jul 10 22:50:35 2009  fox

     318* dtrace_linux.c: Patch to handle smp_call_function_single() when
          invoking current cpu by <Nicolas.Williams@sun.com>

	  strcasecmp: Fix const correctness.

	  dtrace_isa.c: fix call to dump_trace on 2.6.18 kernel.

Sun Jun 28 18:08:39 2009  fox

     317* dtrace_linux.c: instr_in_text_seg: Use strncmp instead of strcmp
          to avoid potential GPF. Dont know why this may help, but its
	  a useful addition.

	  <lahti@student.chalmers.se>

Sat Jun 27 20:23:53 2009  fox

     316* dtrace_isa.c, dtrace_linux.c: Make stack() work - using the
          kernels dump_trace code - if its present, else a cheaper
	  stack walker.

	  We have problems getting a stack on syscall::: provider.

Tue Jun 23 23:52:03 2009  fox

     315* dtrace_linux.c: Disable the GPF interrupt hook for now -- we
          dont need it, and on a 32b kernel, reloading dtrace lots
	  of times can crash the kernel.

     314* tools/build.pl: Detect if we have bison/yacc/flex.

Mon Jun 22 21:03:36 2009  fox

     313* Archive: 1.0098

     312* dt_proc.c: Handle -p <pid> command line switch.

Sun Jun 21 21:27:25 2009  fox

     311* Archive: 1.0097

     310* dt_proc.c, Pcontrol.c: Make dtrace -c <cmd> work.

Wed Jun 17 23:29:46 2009  fox

     309* Archive: 1.0094

     308* libdtrace/makefile: Put dt_grammar.h in the build dir so that
          it is not part of the release, to avoid polluting peoples
	  systems with my yacc or bison build.

Tue Jun 16 22:00:24 2009  fox

     307* dtrace_linux.c: lx_get_curthread_id: Use current->pid, rather
          than zero. <mauritz.sundell@viem.se>

Sun Jun 14 09:58:36 2009  fox

     306* Archive: 1.0093

     305* dtrace_linux.c: We need to preallocate and change the
          process shadowing code, so that when we use predicates such
	  as

	  dtrace -n 'syscall::: /pid == NNN/ {}'

	  that we can actually intercept the targetted process and not
	  confuse dtrace_probe when we share the shadow structure.

     304* Various: change dprintf() to be dprintf1() because GLIBC
          likes to break working apps and defines a POSIX dprintf().
	  POSIX like to break working apps too, and should be ashamed
	  of themselves.

     303* systrace.c, dtrace_linux.c: Make the pt_regs available to 
          dtrace_probe() so we can access register specific state.
	  Handle nesting of interrupts (so we dont destroy the
	  pt_regs pointer in the event of an interrupt on a syscall).

     302* dtrace.c, dtrace_linux.c: Add 'ipl' probe function support.

Fri Jun 12 19:51:37 2009  fox

     301* dtrace_linux.c: Avoid proc shadowing - dont think we need
          this and can do it with per-cpu allocation instead.
	  (par_setup_thread1).

     300* cpu_x86.c: Fixed the issue with CALLR and the other call
          instructions which manipulate the stack as we step over them.
	  32 and 64 bit kernels should be safe - til we hit another
	  instruction we have never seen before.

     299* Archive: 1.0092

Wed Jun 10 18:34:03 2009  fox

     298* dtrace_linux.c: Remove dependency on init_mm for 2.6.30 kernel.

Tue Jun  9 21:32:20 2009  fox

     297* cpu_x86.c: Add support for stepping over 0x7n Jump relative
          instructions for the instruction provider.

Mon Jun  8 22:39:17 2009  fox

     296* instr_linux.c: Instruction provider. Not fully tested/debugged.

     295* dtrace_linux.c: Fix for 2.6.29 task_struct/cred structure compile.

Sun Jun  7 11:12:40 2009  fox

     294* Archive: 1.0091

     293* dtrace.h, dtrace.c, cpu_var.h: Add some dtrace fixes from
          opensolaris-20090607.

     292* fbt_linux.c: Added /proc/dtrace/fbt so we can get info about
          probes. This is CPU intensive and O(N^2) in performance, so treat
	  with care. This is to help debug or add perf stats.

Thu Jun  4 22:08:45 2009  fox

     291* strings.h: Dont undef __USE_BSD. Dont know why Sun insist on using
          the legacy strings.h and not ISO string.h. Oh well.

Sun May 31 10:39:20 2009  fox

     290* dtrace_linux.c, intr_x86*.S: Add support for INT13 (GPF), so
          we can disable ourselves when we run into issues.

     289* toxic.c: Add system_call: This is a very special kernel
          entry point and if we probe it, we may get a kernel double-fault
	  because the stack wont be ready for kernel space.

Fri May 29 19:50:52 2009  fox

     288* driver/cpu_x86.c: New file - put the single step cpu specific code
          in here, from dtrace_linux.c.

	  Add support for 0xEn opcodes (call/jump/loop instructions).

Thu May 28 23:55:45 2009  fox

     287* dtrace_linux.c: Handle the FF .. opcodes (indirect jump/calls).
          fbt::__delay: now works on my Ubuntu 32bit kernel.

	  (Not yet proven for 64bit kernels).

Wed May 27 20:25:44 2009  fox

     286* instr_size.c: Workaround for LOCK/REPZ/REPNZ prefix instructions.

Mon May 25 21:44:26 2009  fox

     285* fbt_linux.c, dtrace_linux.c: Enable more 32-bit instructions
          for :entry traces. Not all have been validated yet, but
	  the code has been left in a safe state.

     284* dtrace_linux.c: Dont do RIP addressing mode for 32-bit cpu.

Sat May 23 19:18:09 2009  fox

     283* intr_x86-32.S, intr_x86-64.S: Add a dummy mcount function
          in case kernel is compiled with -pg. We mustnt call into
	  the kernels mcount else we hit a recursion error.

     282* fbt_linux.c: Re-enable more x86-64 opcodes which were disabled
          whilst debugging the interrupt handlers.

     281* fbt_linux.c, dtrace_linux.c, cpuvar.h: Compute modrm byte
          during analysis phase, not during an interrupt, to lower overhead
	  of RIP addressing mode handler.

     280* Archive: 1.0090

Thu May 21 23:10:10 2009  fox

     279* systrace.c: Fix for write-protected interrupt handlers.
	  (patch_enable).

Wed May 20 21:55:34 2009  fox

     278* dtrace_linux.c: Move /dev/dtrace into /proc/dtrace and segregate
          the debug streams into separate entries (debug, security, stats, trace).
	  This is easier to handle >4K reads for the trace buffer.

Sun May 17 20:51:17 2009  fox

     277* intr_x86-64.S: Allow us to probe an IRET instruction.
          (irq_return)

Sat May 16 22:19:52 2009  fox

     276* dtrace_linux.c: Add support for RIP relative addressing mode
          single step.

     275* dtrace_linux.c: Add dtrace_memcpy to avoid reentrancy issues when
          we probe memcpy(). This happens because we use this inside our
	  single step handler.

     274* fasttrap.c: Avoid kernel panic when invoking a USDT since
          we are tuned to the number of cpus, not the max of cpus we
	  could have.

Fri May 15 21:56:07 2009  fox

     273* Archive: 1.0089

     272* tools/load.pl, makefile: Load /etc/dtrace.conf into the kernel 
          if present. SUDO when appropriate because now we need to be 
	  root to talk to the kernel.

	  Please edit /etc/dtrace.conf - it contains my uid in there which
	  isnt appropriate for your system, but change according to taste.

	  It wont get overwritten in a new 'make install', so its
	  persistent.

     271* dtrace.c, dtrace_linux.c: Fix security policy issues.

Thu May 14 21:30:46 2009  fox

     270* fbt_linux.c: Allow opcode 0x48 for 64-bit. More than doubles 
     	  Ubuntu 8.10 / 2.6.27.8 kernel probes. We can allow them all 
	  through but just taking it slowly.

Wed May 13 21:27:06 2009  fox

     269* dtrace_linux.c: Add security parsing for /dev/dtrace.

Tue May 12 20:34:40 2009  fox

     268* Archive: 1.0088

     267* driver/*: Fix compilation issues with earlier kernels regarding
          placement of semaphores and atomic_t and related issues with
	  mutex macros.

Mon May 11 20:42:02 2009  fox

     266* Archive: 1.0087

     265* dtrace_linux.c: par_alloc: We had some static initialisation
          in here which was breaking predicates for processes.
	  The process shadowing was broken - we kept cleaning out
	  the predicate hints. Now, when doing

	  dd if=/dev/zero of=/dev/null

	  dtrace -n 'syscall::read: /pid == 7331/ {printf("pid=%d", pid);}'

	  we get consistent read:entry + read:return pairs, and we
	  dont just stop probe collecting.

	  Thanks to <mauritz.sundell@viem.se> for pointing out this
	  obvious silliness on my behalf.

     264* dtrace_linux.c, fbt_linux.c, sdt_linux.c: If we have a recursive
          probe, then when we disable the probe, dont let it fire !
	  Else we get multiple recursions leading to a cpu triple fault.

	  This seems to fix the 32-bit instability issue, and let me see
	  where the recursion issue has happened.

Sun May 10 21:47:15 2009  fox

     263* linux_types.h, driver/*: Use semaphores instead of mutexes
          since mutexes arent available in interrupt context.

Sat May  9 09:57:35 2009  fox

     262* libdtrace/makefile: Put dt_lex.c in the build dir since it 
          depends on the version of flex we are using.

     261* Archive: 1.0086

     260* dtrace_linux.c: Some portability cleanups - so we can handle
          the differing kernels.

     259* utils/make_kernel.pl, Makefile: Allow us to cross build/validate
          i386 kernels on a 64 bit machine. (For my benefit only).

     258* dtrace_linux.c: Fixes for 2.6.22/i386 set_idt_entry.

Thu May  7 22:29:11 2009  fox

     257* dtrace_linux.c: Fix compile issue for 2.6.22/i386 on desc_defs.h
          include.

Wed May  6 20:29:31 2009  fox

     256* dtrace_linux.c: dtrace_gethrtime: Never call do_gettimeofday, since
          we will block if that is interrupted.

	  dtrace_linux.c: Comment out the register_die_notifier code since
	  we no longer need it. We are directly attached.

	  Allow /dev/dtrace to be read to print out some debug stats.
	  Nothing useful.

Sat May  2 14:22:15 2009  fox

     255* Archive: 1.0085

     254* dtrace_linux.c: Use direct INT3 vector attachments to avoid
          long toxic lists.

	  intr_x86-64.S: New file for the INT3 interrupt handler.

Mon Apr 27 20:07:36 2009  fox

     253* Archive: 1.0084

Sun Apr 26 11:22:44 2009  fox

     252* cpu_64bit.c, fbt_linux.c: Add support for XOR REG,REG

Sat Apr 25 23:00:24 2009  fox

     251* fbt_linux.c: Fix handling of module sections for pre-2.6.19
          kernels. We couldnt properly validate if symbols in .text section.

	  Add support for PUSH %RCX entry point; change amd64 
	  entry point so we only accept first instruction in function,
	  like we do for i386.

Fri Apr 24 21:45:05 2009  fox

     250* fbt_linux.c: Fix issue with 64-bit entry probe searching.
          Should never have worked for funcs which start PUSH %RBP, but
	  did.

Thu Apr 23 21:15:58 2009  fox

     249* scripts/dt.pl -- make it more usable. This is to replace
          the various scripts/*.d test examples.

     248* dtrace_isa.c: Use copy_from_user to avoid kernel panic if
          copyinstr() called on an argument which isnt a pointer type
	  arg to a syscall.

Wed Apr 22 21:00:26 2009  fox

     247* driver/cyclic_linux.c: Use old timers for non-hrtimer
          aware kernels.

Tue Apr 21 08:13:00 2009  fox

     246* dtrace_linux.c: Allow us to read/write from /dev/dtracedrv for
          debugging.

     245* Archive: 1.0083

     244* driver/systrace.c: More surgery to ensure we pick up
          the right syscall for all syscalls, and avoid dependency
	  on GCC local variable allocations for finding the syscall.
	  Also get rid of a few magic numbers in dtrace_systrace_syscall
	  which were causing some of the troubles.

     243* driver/systrace.c: Put in the handler for 2.6.9 special
          case for execve requiring pt_regs passing by value.

Mon Apr 20 23:41:08 2009  fox

     242* driver/systrace.c: Replace the assembler sequence for
          execve() and rt_sigreturn() with a patch sequence to
	  allow for more generic kernel version handling.

Sun Apr 19 00:02:12 2009  fox

     241* dt_subr.c: dt_sysconf: Was returning wrong number of cpus.

     240* Archive: 1.0082

Sat Apr 18 12:45:26 2009  fox

     239* driver/cyclic_linux.c: Handle API/struct changes in 2.6.28 
          kernels.

     238* driver/cyclic_linux.c: Fix the hrtimer restart issue
          which was crashing the kernel (not increasing the time
	  locks the kernel since timer keeps refiring).

Thu Apr 16 20:20:57 2009  fox

     237* dtrace_linux.c: Tidy up the syms[] array which is filled in
          from userspace to bootstrap us.

     236* Archive: 1.0080

     235* dtrace_linux.c: Fix typo in on_notifier_list, and avoid
          repeated calls to get_proc_addr when entries dont exist.

     234* tools/load.pl: Try /boot/System.map-`uname -r` for symbols
          which may be missing (AS4). Remove sys_syscall which isnt
	  valid at all.

Tue Apr 14 20:17:10 2009  fox

     233* cyclic_linux.c: Emergency patch to disable cyclic timer
          til i fix the crashing...

     232* Remove all compiler warnings for the driver (still working on
          init_mm), so I can see wood-for-trees on 2.6.9;
	  kernel 2.6.28+ is clean, but 2.6.9 has a few noisy differences.

     231* driver/cyclic_linux.c: Remove warnings and fix way
          hrtimer callbacks work (based on 2.6.28)

     230* Various: cleanup the exit functions for the drivers, since
          Linux 2.6.9 doesnt like calls to misc_deregister() if we
	  never registered the structure.

	  dtrace_linux.c:
	  Remove redundant calls to remove_proc_entry() which can panic
	  2.6.9.

Mon Apr 13 11:03:10 2009  fox

     229* dtrace_linux.c: par_alloc. Allocate memory with GFP_ATOMIC since
          we may be in interrupt context and AS4 doesnt like us.

     228* cpuvar.h, fasttrap_impl.h: Avoid struct padding for Linux
          2.6.9 where we end up with a negative array size since
	  sizeof(mutex_t) is > 64 bytes.

     227* fbt_linux.c: Avoid possible GPF when loading driver and trawling
          the available modules.

     226* Archive: 1.0079

     225* driver/dtrace_linux.c: Fix re-entrancy problem for fbt:::
          due to kprobes notifier wanting to be first.

Tue Apr  7 23:55:48 2009  fox

     224* systrace.c: Fix calling sequence for fork/vfork/iopl/sigaltstack.

Mon Apr  6 21:54:17 2009  fox

     223* Archive: 1.0078

     222* systrace.c: We broke 32-bit syscalls. Now fixed.

     221* makefile: Move to tools/build.pl, so we can start to extend
          for other cpu types.

     220* systrace.c: Fix silly assemblerism.

     219* dtrace_linux.c, fasttrap.c, missing.c: Add support for AS4
          2.6.9+patches kernel.

Sun Apr  5 18:05:55 2009  fox

     218* systrace.c: Fix fork/vfork/sigaltstack - forgot to put the
          switch statement into get_interposer.

     217* usdt/c/makefile: Keep the src dir clean - all objects now in the build dir.

Sat Apr  4 10:44:41 2009  fox

     216* Archive: 1.0077

     215* systrace.c: Add handlers for fork/vfork/iopl/sigaltstack.

     214* systrace.c: Fix amd64 clone() syscall handler.

     213* Archive: 1.0076

Sun Mar 29 15:27:08 2009  fox

     212* Archive: 1.0074

     211* fbt_linux.c: Fix for 2.6.29

     210* dtrace.c: Fixes for 2.6.29

     209* linux_types.h: Fix for including <sys/cred.h> on 2.6.29
          kernel.

Sun Mar 22 09:28:26 2009  fox

     208* Archive: 1.0073

     207* dtrace_linux.c: Undo last change - it wasnt what the problem was.
          Problem is that an SMP kernel may say we have 64 cpus, when we
	  only have 1 or 2, so buffer allocation went thru the roof -- 
	  64 * 4MB of RAM, and we OOM'ed. Now we utilise number of online
	  cpus at driver load time. (We need to do work if we online
	  cpus whilst system is up, since the arrays will be too small).

     206* dtrace.c: Do some rough memory calcs when allocating the state 
          buffers to avoid starvation on 512MB systems.

     205* ctl.c: Fix to compile on 2.6.18.

     204* dtrace_linux.c: Fix PTE modifications for x86-64 which uses
          a 4-level page table and for which later kernels try hard
	  to stop us from patching the syscall table.

     203* fbt_linux.c: Re-put-back the sanity check on kernel functions
          so we dont crash when we hit code in non-standard .text sections.

Wed Mar 11 23:30:11 2009  fox

     202* tools/load.pl: Add more diagnostics if things go wrong.

     201* tools/load.pl: Add missing -mem_alloc switch.

     200* dtrace_asm.c: Fix the 64-bit assembler to avoid call frame
          issues with differing compilers. 

	  "dtrace -n fbt:::" works again for 64-bit.

Mon Mar  9 20:44:28 2009  fox

     199* dtrace_linux.c: Support for non ATOM_NOTIFIER_HEAD kernels.
          (AS4)

Sun Mar  8 00:39:01 2009  fox

     198* fbt_linux.c: Fix 64-bit fbt.

     197* cpu_32bit.c, fbt_linux.c: Emulate PUSH EAX instruction.

     196* cpu_32bit.c, fbt_linux.c: Emulate PUSH $nn instruction.

     195* cpu_32bit.c, fbt_linux.c: Emulate CLI instruction.

     194* Archive: 1.0070

     193* toxic.c: Add __mutex_* functions to the list since we
          invoke mutex_lock/mutex_unlock and these are part of
	  the implementation.

	  dtrace -n fbt:::

	  seems to work now - at least under lighter load.

     192* fbt_linux.c: Dont include functions from the .exit section
          of a module.

     191* fbt_linux.c, dtrace_linux.c: Add on_notifier_list() to 
          detect if a target probe point is on one of our notifier
	  chains, and will cause kernel mayhem if we probe it.

Sat Mar  7 09:02:09 2009  fox

     190* missing.c: Add strlen(), and strncmp() for older kernels/gcc's.

     189* fbt_linux.c: When examining module symbols, ignore any in
          the ".init" sections, else we can be trying to touch reclaimed
	  pages.

     188* fbt_linux.c: Dont allow us to create probes for module "init"
          functions which are likely sitting in unmapped pages. This
	  caused problems in dtrace_close() as we tried to unpatch them,
	  but they may not be mapped or contain data since the pages would
	  have been freed.

     187* tools/load.pl, dtrace_linux.c: Add a -mem_alloc option
          to turn on memory tracing. (Its too voluminous if enabled
	  with -here).

     186* fbt_linux.c:fbt_disable: Dont unpatch a probe we never
          enabled in the first place (eg, because kernel unmapped
	  a .init section).

     185* Archive: 1.0069

     184* toxic.c: Add do_int3, since this is the core of the
          kernel for invoking us when a probe fires.

     183* dtrace_linux.c: par_setup_thread: Hack for the moment to
          avoid reentrancy issues in dtrace_probe.

Fri Mar  6 14:09:04 2009  fox

     182* toxic.c: Remove non-toxic functions (not all, but many).
          Add notifier_call_chain.

     181* cpu_32bit.c: Fix "SUBL $0,%ESP".

     180* cpu_32bit.c: Fix "MOV %ESP,%EAX" emulation which was
          causing kernel to die when we trace local_bh_disable.

Thu Mar  5 23:31:02 2009  fox

     179* cpu_32bit.c: Fix "TEST %EAX,%EAX" emulation.

Tue Mar  3 10:33:15 2009  fox

     178* dtrace_linux.c: dtrace_gethrtime() Fix bug stopping us probe
          nr_active and friends and other timer related functions, since
	  we deadlocked inside our probe interrupt. Need to fix dtrace_gethrtime
	  since we do a read from a multiword item, and this can change
	  whilst we read it, leading to potentially non-monotonic
	  clock.

     177* Archive: 1.0068

     176* systrace.c, fbt_linux.c, dtrace_linux.c: For fbt probes,
          the i386 kernel is write protected; make pages writable,
	  to avoid paniccing the driver.

Sun Mar  1 00:10:21 2009  fox

     175* Archive: 1.0067

     174* toxic.c: Added to allow filtering of bad probe funcs in the
          kernel. Not finished yet (for my kernel).

     173* dtrace.c, fbt_linux.c: Add support for fbt:kernel:... so
          we can start tracing real kernel functions. This is dangerous
	  at this time (wait for next update?) since you can trace
	  anything - including the bits of the kernel dtrace depends on.
	  Next step is to hide those symbols we depend on so you can
	  do:

	  dtrace -n fbt:kernel::

     172* cpu_32bit.c: Add support for 89 c0..c7 (mov %eax,reg)

     171* Archive: 1.0066

     170* dtrace_asm.c: Fix dtrace_interrupt_disable - we broke
          the C/asm interface semantics - would crash a real cpu, but
	  not vmware for some reason.

Sat Feb 28 22:45:59 2009  fox

     169* makefile: recreate build symlink each time in case of kernel
          upgrade.

Fri Feb 27 20:16:31 2009  fox

     168* makefile: Add a 'build' number when we create a release in case
          we release multiple times per day.

     167* fbt_linux.c, cpu_32bit.c: Add support for 

     	XOR reg,reg
	JMP nnnn

	instructions

     166* driver/cpu_emulate.c: Split into cpu_32bit.c and cpu_64bit.c

     165* Archive: 1.0065

     164* dtrace_asm.c: Fix nesting of interrupt disable for i386.

Wed Feb 25 19:11:10 2009  fox

     163* load.pl: If we have an error, tell them a little more about
          why the driver load failed.

     162* makefile: Add a dependency checker before people build.

Mon Feb 23 21:45:13 2009  fox

     161* cpu_emulate.c: Emulate SUB $4,ESP

     160* cpu_emulate.c: Emulate TEST EAX,EAX

     159* cpu_emulate.c: Fix SUB $nn,ESP where nn == 0x18

Sun Feb 22 11:31:20 2009  fox

     158* cpu_emulate.c: Add support for SUB $8,ESP

     157* cpu_emulate.c: Fix problem with SUB $12,ESP instruction

     156* Archive: 1.0064

     155* systrace.c: Fix for sys_clone - now we dont get the core dumps.

     154* systrace.c: Fix for sigreturn needing a proper call frame so
          it can patch the registers. [i386 only]

	  This appears to work

	  dtrace -n syscall:::

Sat Feb 21 10:25:07 2009  fox

     153* cpu_emulate.c: Emulate PUSH ESI.
          We need to avoid having a zillion case/defines where we
	  can use the CPU opcode decoding to get big-bang-for-buck.
	  We need to pass in the missing opcode so we can decide
	  which is better to do.

     152* dtrace_linux.c: Use vmalloc for large buffer allocations to avoid
          smacking the kernel badly with our large probe buffer.

     151* cpu_emulate.c: Emulate PUSH EDI

Fri Feb 20 20:12:05 2009  fox

     150* cpu_emulate.c: Emulate PUSH EBX

     149* makefile+tools/bug.sh: Add a .release file, so we know
          what people are building on when reporting a bug.

     148* drivers/missing.c: Fix for 2.6.23.1-42.fc8 missing
          find_task_by_vpid.

Thu Feb 19 21:55:51 2009  fox

     147* cpu_emulate.c: Add support for "SUB $12,ESP" - where the
          reti stack frame overlapped our saved regs. Special case.
	  We can merge this with the generic (>12) case later.
	  Still need to handle "SUB $8,ESP" and "SUB $4,ESP"...which
	  should be easier for we have the $12 code.

     146* Few cleanups and fixes for AS4 release.

Wed Feb 18 21:58:50 2009  fox

     145* dt_link.c: Fix properly for 32 + 64 bit architectures.
          Same reason as below. Also create a temp file to contain
	  the intermediate linker object rather than rely on /dev/fd/NNN.

Tue Feb 17 22:07:40 2009  fox

     144* dt_link.c: When writing the USDT output object file, use
          a section type of SHT_PROGBITS and not SHT_SUNW_dof, since
	  some older binutils/ld will not like what we did (too
	  over zealous validation).

Mon Feb 16 22:59:10 2009  fox

     143* Archive: 1.0061

     142* cpu_emulate.c: Add support for "SUB $nnn,%ESP"

Sun Feb 15 10:47:15 2009  fox

     141* fbt_linux.c, dtrace_linux.c: Use INT3 traps for entry/return
          Need to handle more than just LEAVE/RET for FBT on i386.
	  Fix INT3 handler to emulate the blotted-out instruction.

     140* Archive: 1.0060

Sun Feb  8 10:02:02 2009  fox

     139* Archive: v1.0059

     138* cpu_emulate.c: The fbt for i386 wont work on Linux since gcc
          generates lots of differing instructions for a function
	  (also, potentially inline assembler). We need to do more
	  of a job for i386. So start to handle the new instructions
	  we are seeing in the various kernel functions.

Sat Feb  7 22:08:31 2009  fox

     137* dtrace_linux.c: Fix FBT so it works for x86-64
          Not working properly for 32bit kernel yet.

Fri Feb  6 11:09:54 2009  fox

     136* Archive: 1.0057

     135* drivers/include -> ./include due to new build structure.
          build/ is a symlink to build-`uname -r`. 
	  drivers/dtrace/* -> driver

     134* makefiles: Allow us to label the build dir with the appropriate
          kernel version, so we can do multiple kernel builds.

Thu Feb  5 10:35:22 2009  fox

     133* Archive: 1.055

     132* fasttrap.c: Werent initting the fasttrap_proc_t mutex, so
          we died if a USDT process exited.

     131* Fix dtrace_casptr for i386 which would cause copyinstr
          to fail.

Wed Feb  4 21:17:32 2009  fox

     130* Various: changes to better manage multikernels. Compiled
          against 2.6.9, 2.6.24, 2.6.26, 2.6.28

Tue Feb  3 13:03:37 2009  fox

     129* Archive: 1.0054

     128* systrace.c: Re-engineered to the syscall table patcher --
          now works inside vmware and a real 32bit cpu.

	  Kernel can still panic if you do some syscall tracing
	  and try and do copyinstr(arg0) - need to find out why.

Mon Feb  2 17:27:39 2009  fox

     127* ctl.c: Disable the hunt_init/hunt_cleanup code for the moment
          as its crashing the x86 cpu (32b).

     126* dtrace_asm.c: Found an issue with the membar functions which
          dont work on a real x86(32b) cpu, but do on a VMware
	  session on a 64b processor, and in 64bit mode. Use SFENCE
	  instruction.

     125* Archive: 1.0053

     124* Now works on Ubuntu 2.6.24-16 (32bit) kernel

     123* sys/privregs.h: Move the horrors of register and regs/pt_regs
          into here. Need to cater for older kernels and i386/x86-64 too.

Sun Feb  1 15:28:56 2009  fox

     122* Various: tidy up some kernel compile warnings; remove some
          unneeded zero length headers. 

	  Arrange for proc_exit notification to wipe out fasttrap
	  probes.

     121* Various: refactor datamodel computation because we may treat
          a 64b proc as a 32b one and access the wrong values for 
	  copyinstr().

	  We still have some USDT issues: dtrace doesnt know the target
	  proc exited, so leaves the probes behind.

     120* Archive: 1/0052

Sat Jan 31 10:42:32 2009  fox

     119* fasttrap_isa.c: Changes to compile on 32bit kernel (tested on
          2.6.28 kernel).

Fri Jan 30 22:48:51 2009  fox

     118* dt_lex.l: Compat support for AS4 version of flex.

Tue Jan 27 23:22:35 2009  fox

     117* Archive: 1.0051

Mon Jan 26 22:42:20 2009  fox

     116* dtrace_linux.c: Fix uwrite() interface.

Sun Jan 18 18:51:13 2009  fox

     115* dtrace_linux.c: Put in a trap handler for INT3 and pass onto
          to dtrace. Not there yet....

Thu Jan 15 20:56:19 2009  fox

     114* Archive: 1.0050

     113* tests/dt019: Simple usdt example we are trying to make work.

     112* fasttrap.c: Avoid kernel panic when we find the target to intercept.
          Still not write because we commented out uread/uwrite.

Tue Jan 13 23:52:50 2009  fox

     111* Archive: 1.0049

     110* fasttrap.c: Define fasttrap_max so that USDT starts to work.
          libdtrace/: Lots of hackery to allow us to launch a process.
	  (Not yet working to completion yet)

Sun Dec 28 00:02:27 2008  fox

     109* dtrace_linux.c: Remove old no-longer needed kernel dependencies
          so we can compile on 2.6.28

     108* drivers/dtrace/systrace.c: Fixes to build on 2.6.27.10 kernel
          for 32-bit.

     107* dt_link.c: Generate correct ELF patchups for 32-bit libc6 code.
          (usdt now builds the user space simple executable)

Sat Dec 27 04:35:47 2008  fox

     106* liblinux/getopt.c: Add this because some getopt()s are doing
          weird stuff compared to GNU getopt.

Sat Dec 20 22:34:13 2008  fox

     105* makefile, tools/build.pl: Handle i686 (32-bit) platform.

Sat Dec 13 08:45:52 2008  fox

     104* systrace.c: Avoid panicing 2.6.27.8 kernel due to bad
          syscall name.

     103* ctl.c: Fix crash on unloading driver.
          <ppluzhnikov@google.com>

Fri Dec 12 21:03:00 2008  fox

     102* ctl.c: Fix find_task_by_pid, and improve make-me script.

     101* Archive: 1.0048

     100* dtrace_linux.c: Suggestion to use find_task_by_vpid vs
          find_task_by_pid to avoid kernel-ifdefs.
          <ppluzhnikov@google.com>

      99* drivers/dtrace/dtrace_asm.c, sdt_linux.c, dtrace_isa.c
          Some more OpenSolaris 20081211 alignment changes. Nothing
	  functional fixed, since many of the ON source changes are
	  more in the Sun kernel code than the common stuff here.

      98* drivers/dtrace/dtrace.c: Merge in changes from OpenSolaris 20081211.

Wed Dec 10 10:46:25 2008  fox

      97* tools/mksyscalls.pl: [PATCH] Add verbosity flag; print KGDB module load address.
          <ppluzhnikov@google.com>

      96* tools/mksyscalls.pl: Change to find unistd.h which keeps moving.
          <ppluzhnikov@google.com>

Mon Dec  8 22:07:48 2008  fox

      95* dtrace_linux.c: Patch for on_each_cpu, smp_call_function, and
          asm/stacktrace.h
          <ppluzhnikov@google.com>

Thu Dec  4 20:53:33 2008  fox

      94* drivers/dtrace/dtrace_isa.c: Fix to handle kernel 2.6.26 and
          above.
          (ppluzhnikov@google.com).

      93* drivers/dtrace/dtrace.c: Fix dtrace_errlock bad mutex defn.
          (ppluzhnikov@google.com).

Wed Dec  3 23:16:40 2008  fox

      92* Archive: 1.0046

      91* drivers/dtrace/ctl.c: We now have a /proc/$$/ctl driver visible.
          It doesnt do anything yet, but we have now patched a running
	  kernel.

Wed Nov 12 22:10:51 2008  fox

      90* usdt/*/makefile: Use our compiled dtrace, rather than anything in
          the environment.

Sun Oct 26 16:54:54 2008  fox

      89* dtrace_linux.c: Enable /dev/sdt

Sat Oct 25 10:18:22 2008  fox

      88* fasttrap.c: Can now run a USDT enabled app, but not showing up
          on 'dtrace -l'. No longer hangs/crashes the kernel.

Sun Oct 19 09:58:28 2008  fox

      87* dt_link.c: Now generates correct binary, but drti.o isnt
          correct yet.

      86* Archive: 1.0045

Sun Oct 12 09:47:14 2008  fox

      85* drti.c: Allow us to specify $DTRACE_DRTI_O so we can find
          the drti.o object file;

	  linux_types.h: Fix include issues for dlfcn.h.

	  dtrace_linux.c: Now invoke the dtrace_ioctl_helper function.

Sat Oct 11 10:00:48 2008  fox

      84* drivers/dtrace.c, dtrace_linux.c: Start to enable the 
          /dev/dtrace_helper driver, needed for USDT.

Thu Oct  9 22:36:16 2008  fox

      83* linux_types.h: Add _LITTLE_ENDIAN definition, else dt_link.c
          generates bad ELF .o files.

Wed Oct  8 22:38:54 2008  fox

      82* libdtrace/dt_link.c: Some changes to allow dtrace -G to work.
          Not there yet.

      81* usdt/ New directory to create a simple USDT so we can
          get things sorted.

Sat Aug  2 11:58:07 2008  fox

      80* Archive: 1.0042

      79* dt_module.c: Fixes for 64b kernels - we can now see kernel stack
          traces.

Tue Jul 29 21:27:11 2008  fox

      78* makefile: Build libelf/Gelf structures properly for 32b machines.

Sun Jul 27 08:56:23 2008  fox

      77* dtrace_isa.c: Add in support for getstack to handle stack()
          calls. Not brilliant with -fomit-frame-pointer in the kernel,
	  but that will teach people to use this option :-(

      76* dt_module.c: Add support for reading /proc/kallsyms so we
          can get kernel level stack traces.

Thu Jul 24 08:54:01 2008  fox

      75* dtrace.c: dtrace_probe: Fix issue with 32-bit arguments
          not being sign extended when stored in return struct. Without
	  this, arg0 is never < 0 for failed syscalls.

      74* load.pl: rewrite load shell script as perl as we want to do some
          extra things for some kernels.

      73* fbt_linux.c: Add support for i386/modules[] on 2.6.23.1

Wed Jul 23 07:32:30 2008  fox

      72* fbt_linux.c: Implement work around for 2.6.23.1 kernel
          without public sys_call_table.

      71* dtrace_linux.c: par_alloc should memset the memory
          to zero to avoid firing a signal when it wasnt asked for.

      70* systrace.c, dtrace.c: Implement support for raise/stop actions
          which can send a signal to a process.

      69* load: Detect missing symbols and try to warn user something
          may fail.

Wed Jul 16 22:55:01 2008  fox

      68* systrace.c: Avoid bogus syscalls which can index
          off the end of the array. Likely due to bad stack reference.
	  Avoids SIGKILL near kernel panic when it happens.
	  Trying to bootstrap ustack() call which involves 
	  examining to the proc in user space to grab the symbol
	  table. Race condition (which is catered for) -- the proc
	  may no longer exist by the time dtrace is running.

Sun Jul 13 10:53:44 2008  fox

      67* Archive: 1.0037

      66* Pisadep.c: Got rid of most of our stubs and use proper libproc
          code. This code wont work since its depending on Solaris /procfs
	  but we can start to work our way through this.

Sat Jul 12 10:59:36 2008  fox

      65* Archive: 1.0039

      64* libproc: Upgrade complete.

      63* libproc: Start upgrade to Solaris 20080623, since we cannot
          call ustack() without starting to emulate /procfs on Solaris.

      62* dtrace_isa.c: Fix issue with coying to/from user space.
          arg0/arg1/... now works.

Sun Jul  6 10:29:42 2008  fox

      61* Archive: 1.0035

      60* fasttrap_isa.c: Merged/compiled in - we have some TODOxxx items
          to fix before we can use this.

      59* fasttrap.c: Now compiles, but need to implement timeout()
          function. Next, get fasttrap_isa.c to compile.

Sat Jul  5 14:53:52 2008  fox

      58* dtrace.c, dtrace_linux.c: Implement a parallel shadow to curthread
          so we dont die when doing thread local variables (dt9 test).
	  We now need a garbage collector - otherwise we may leak memory,
	  or have an issue when a new thread comes along on the same address.

	  dt9: demo/vars/rtime.d no longer panics the kernel.

Fri Jul  4 20:35:34 2008  fox

      57* dt_lex.l: Another attempt at fixing the lexing issues with
          flex.

      56* dtrace.c, fasttrap.c: Upgrade to latest ON Solaris release
          20080623.

Thu Jul  3 23:16:19 2008  fox

      55* Move cmd/dtrace/demo to toplevel dir so its more accessible.
          Create tools/runtests.sh so we can start proving the quality
	  of the yacc/lex port.

      54* Archive: 1.0032

      53* profile.c: Now shows up in dtrace -l; allow us to talk to
          /dev/dtrace_profile driver to change max profile probes.

Wed Jul  2 20:46:02 2008  fox

      52* profile.c: Add missing profile driver. Not yet proven and the
          omni support is missing...

      51* utils/libgcc.pl: Script to locate your libgcc.a file rather than
          me supply one.

      50* Mojmir Svoboda: Fix for raw_local_irq_disable() /
          raw_local_irq_enable() to be more backward kernel compliant.
	  Mojmir suggests issues with kmem_cache_create - I think kernel
	  code changed function calling so need to incorporate a kernel
	  specific fix.

Tue Jul  1 21:53:15 2008  fox

      49* dt_lex.l: Lets hope this works because I am mightily sick
         of flex - getting it to work on all platforms. Its an
	 embarrassment.

      48* Archive: 1.0031

      47* put dtrace binary in build/ directory.

      46* dt_lex.l, makefile, dt_impl.h: Get lex to build us properly so
          we can parse the scripts.

Mon Jun 30 20:50:07 2008  fox

      45* cyclic_linux.c: Enable linux cyclic timer code.

Sun Jun 29 11:50:58 2008  fox

      44* Archive: 1.0030

      43* dtrace_linux.c: syscall tracing finally working on 32-bit kernel.

      42* Archive: 1.0029

      41* liblinux/proc2.c: Fix gethrtime and pthread_cond_reltimedwait_np()
          to avoid dtrace eating lots of cpu and filling /var/log/messages.

Thu Jun 26 22:51:51 2008  fox

      40* i386 kernel sys_call_table patching now works, just need to get the
          syscall collector to work.

Mon Jun 23 23:13:08 2008  fox

      39* divmod64.c: Fix the 64-bit divide issue.

Wed Jun 18 23:19:11 2008  fox

      38* Archive: 1.0027

      37* More 32-bit portability changes.

Tue Jun 17 23:02:33 2008  fox

      36* Various: changes to build 32-bit dtrace + driver

Mon Jun 16 22:57:02 2008  fox

      35* cyclic.c: Need the full whack from Solaris. Without this, the
          D Engine will kill us after a while because it thinks we are hanging
	  and may endanger the system stability.

	  We could just write our own, but its doing a lot of complex
	  and well thought out issues to do with accuracy and 
	  cpu alignment, and its not worth hacking this. FreeBSD has
	  its own (slightly simplified) version of the Solaris code.

Sun Jun 15 09:39:19 2008  fox

      34* dtrace_linux.c, fbt_linux.c: Add a (slow/horrid) parallel 
          allocator. Avoids reprobing modules.

      33* Archive: 1.0026

      32* syscall tracing now working !

Sat Jun 14 00:20:36 2008  fox

      31* Archive: 1.0025

      30* dtrace.c: Fix kernel panic when creating an ecb event.
          dt4 now works for BEGIN/END functionality but syscalls not
	  being triggered.

Mon Jun  9 22:05:12 2008  fox

      29* Archive: 1.0024

      28* fbt_linux.c, dtrace_linux.c: Implement a validate_ptr() function
          since we were dying due to bad VM inside the kernel.

Mon Jun  2 21:42:23 2008  fox

      27* fbt_linux.c: Avoid issue with call to __symbol_get() cause
          incremented mod ref count, which stops us unloading.

Sun Jun  1 00:17:23 2008  fox

      26* Archive: 1.0021

Thu May 29 23:07:39 2008  fox

      25* dtrace_linux.c: Create the cpu_list[] structure which dtrace wants.
          Avoids kernel panic when we hit DTRACEIOC_GO.

      24* drivers/dtrace/dtrace.c: More merge-in from ON solaris.

      23* Archive: 1.0020

Wed May 28 22:31:37 2008  fox

      22* drivers/dtrace/dtrace_ioctl.c: Remove this - integrate back in
          with dtrace.c since we are now more aligned and nearly working.

Mon May 26 14:31:54 2008  fox

      21* drivers/dtrace/*: Fix issue where 'dtrace -l' would keep
          adding the same probes. Needed to mimic the init/attach/open
	  cycle of a proper solaris driver. Fixed various crashes due
	  to this, and allows driver to unload properly - no issues in simple
	  testing, but basic probe placement is not working yet - cannot 
	  plant the probes. Not sure why..yet.

Sun May 25 00:35:12 2008  fox

      20* drivers/dtrace.c: Update to latest opensolaris code to help
          find bugs in driver.

      19* Archive: 1.0018

Tue May 20 23:50:24 2008  fox

      18* cmd/dtrace/dtrace.c: Incorporate getopt() fixes from Apple/Darwin.
          Fixes 'dtrace -n BEGIN', but note this will presently crash your
	  kernel if you run it !

Tue May 13 21:56:09 2008  fox

      17* cmd/dtrace/dtrace.c: Fix issue with getopt() not working when
          presented with this command line:

	  cmd/dtrace/dtrace cmd/dtrace/demo/agg/denorm.d /bin/ls

      16* Update the cmd/dtrace demo to the latest version. They broke
          backwards compatibility ! Why ?! The #pragma changed syntax.
	  Thats very very naughty of them. I do hate lex+flex.

Sun May 11 20:54:47 2008  fox

      15* Update libdtrace to latest opensolaris version.
          Update cmd/dtrace/dtrace.c to latest verison.

Mon May  5 09:48:18 2008  fox

      14* /dev/ctf driver now loads but theres lots of inter-driver
          dependencies which are a pain to manage, so may try
	  and create a consolidated driver.

      13* Archive: 1.0010

Mon Apr 28 20:36:54 2008  fox

      12* Archive: 1.0008

      11* Fix dt_lex.l and dt_grammar.y compile fixed so we dont copy
          files from my install tree to the remote install tree.
	  Fixes problem with dtrace plainly not working at all.

      10* dt_grammar.y: Some minor tweaks to handle compiling with
          bison. Sun have an error in their file (missing ';').

Sun Apr 27 00:03:52 2008  fox

       9* Archive: 1.0007

       8* Archive: 1.0006

       7* Put all binaries into the build/ dir rather than littering the
          source. Not done for drivers yet, since Linux kernel build
	  pollutes us. Also made code do a clean build from start to finish.
	  Ignore warnings - will address these over the course of time.

       6* Archive: 1.0005

       5* Various changes for clean build on FC8/2.6.24.4 kernel. Slightly
          broken Ubuntu...to fix next. First part of the driver works, and
	  making some changes to allow cmd/dtrace to talk to the driver.

Mon Apr 21 22:10:40 2008  fox

       4* Surprise: my first kernel panic. /dev/dtrace is the interface
          and if you cat it ... surprise !

Sun Apr 20 20:51:52 2008  fox

       3* Now builds under ubuntu 2.6.22-14-generic kernel and
          RedHat FC* 2.6.24.4-64.fc8. More portability.

	  If you load the driver, you will have 

	  /proc/dtrace/helper
	  /proc/dtrace/dtrace

	  entries. (Non functional since no plumbing at present).

Sat Apr 19 23:08:09 2008  fox

       2* drivers/dtracedrv.ko finally links correctly. However theres still
          quite a few stubs and broken definitions, so be careful - will
	  likely crash your kernel.

Wed Apr  9 21:56:09 2008  fox

       1* dt_lex.l: Fix flex/lex issues - now parses strings properly and
           dtrace -x nolibs -n BEGIN -e -S
	   seems to work - or at least it doesnt core dump or generate
	   a diagnostic.