Sun Nov 3 14:34:37 2013 fox

848* tools/readelf.pl: New/experimental script

847* driver/Makefile: Use -gdwarf-2 debug info so we work on

Ubuntu 13.10/gcc-4.8 (else mkctf.sh dies).

Sat Nov 2 18:14:37 2013 fox

846* ctfconvert/dwarf.c: Silly hack for DWARF-4 issue. I still cant

figure out libdwarf vs libdw and how they not only evolve but

the features dont allow one to be implemented from another.

This is a fricking mess.

Sun Oct 20 10:48:45 2013 fox

845* Archive: 1.0151

Sat Oct 19 17:47:23 2013 fox

844* driver/dtrace_linux.c: Fix for /proc/dtrace/trace on 3.11 kernel

which doesnt like big blobs of output.

Fri Oct 18 22:51:33 2013 fox

843* driver/systrace.c: Dont GPF on an amd64 kernel with no

32-bit syscalls support when catting /proc/dtrace/syscall.

Thu Oct 17 23:20:00 2013 fox

842* sdt_linux.c, prov_common.c, dtrace_linux.c: Support for

dynamic/private SDT probes in loadable modules.

Sat Oct 12 08:07:40 2013 fox

841* intr.c: Fix for 3.11 kernel (store_gdt)

Sat Aug 3 21:33:35 2013 fox

840* driver/proc_compat.h: create_proc_entry only if < 2.6.18 kernel.

(Centos 5.5 fix)

Thu Jul 11 23:00:12 2013 fox

839* driver/vminfo.c: Fix for 3.9 kernel.

Sun Jun 23 21:27:38 2013 fox

838* libproc/common/Psymtab.c: Avoid core dump on "dtrace -c ls".

Missing prototype. Cleaned up some other compiler warnings.

"dtrace -c ls" still doesnt work.

Sat Jun 22 18:47:16 2013 fox

837* libdtrace/makefile: Create drti.o with -fPIC to avoid a build

error with "ruby".

Thu May 30 23:45:21 2013 fox

836* tools/mkport.pl: Better detection for

HAVE_ELF_GETSHDRSTRNDX. Without this, Brendan Giggs

"struct sock" test broke.

http://sourceforge.net/apps/trac/elftoolchain/wiki/LibElfIncompatibilityRedHatElfUtils

has the gory details - they changed (fixed) the return code

for elf_getshstrndx which is annoying since we need to know the version

of elfutils to see which API it conforms to. I am seriously unimpressed

that sprawled through the releases - it was a broken API change and

it is a broken API change. WTF are two dyslexically impossible to

read function names used?

Tue May 28 23:10:14 2013 fox

835* driver/dtrace_linux.c: New fix for dtrace_gethrtime(). For kernels

which expose native_sched_clock(), use that for nsec timer accuracy.

Mon May 27 20:49:14 2013 fox

834* driver/intr_x86-64.S: Fix for when CONFIG_PARAVIRT is not turned on.

Sat May 25 17:05:25 2013 fox

833* dtrace_linux.c: Fix for dtrace_gethrtime() for later kernels

where global 'xtime'/'xtime_cache' isnt available.

Sun May 5 17:03:01 2013 fox

832* ctfconvert/dwarf.c: Allow DWARF-2 to DWARF-4. This is possibly

dangerous without code changes and proper validation.

Sat May 4 19:17:55 2013 fox

831* cmd/ctfconvert: Some partial changes to move to libdw.

830* driver/intr.c: Fix FBT probes for i386/x64 due to bug put in

for the __arm__ port.

829* driver/systrace.c: Fix for >=3.8 kernels, since they went and

hid sys_call_table (i386/amd64 only - not ARM, yet).

Fri May 3 23:09:37 2013 fox

828* cmd/convert/dwarf.c: Allow for

struct fred {} xx[10];

(struct lock_class_key)

Thu May 2 22:25:29 2013 fox

827* Archive: 1.0148

Wed May 1 21:06:11 2013 fox

826* driver/systrace.c: Fix stupid typo for x86_64 whereby

syscall:::return probes may not fire if i386 syscall

entry doesnt exist. (I said it was a stupid typo).

Sun Apr 28 08:59:18 2013 fox

825* Archive: 1.0147

824* driver/systrace.c: Re-enable syscall::: for x86 arch.

Tue Apr 23 22:32:39 2013 fox

823* tools/load.pl: Fix for missing ia32_sys_call_table on amd64

kernel.

Sun Mar 17 07:13:40 2013 fox

822* libdtrace/dt_subr.c: Fix for sysconf(_SC_CPUID_MAX) returning 1

more than the actual cpus (this is a 'max' not a counter).

821* driver/taskq.c: Fixes for ARM.

820* driver/dtrace_linux.c: Be careful when kernel not compiled

with CONFIG_PROFILE (ARM) as we would die on a driver reload.

819* driver/dtrace_linux.c: Fix xtime_cache_ptr so that

the timestamp function works on Ubuntu 12.04 (or any >= 3.4

kernel).

Wed Mar 6 22:49:14 2013 fox

818* dtrace.c: dtrace_probe: When checking for Xen probes with

interrupts disabled, be careful of deref of null ptr.

Fri Mar 1 21:53:13 2013 fox

817* linux/sys/regset.h: Fix for i386 build after rearranging for

ARM.

Thu Feb 21 23:25:12 2013 fox

816* Various: Changes to compile on 3.6.8/ARM/RaspberryPi kernel.

ARM is still no good to run. This build *should* compile

apart from the very last section on divmod64.c processing,

but am releasing to keep current. Theres still a lot

of execution time work to do.

Mon Feb 11 21:26:31 2013 fox

815* libdtrace/dt_link.c: Start to implement dt_modtext for ARM.

814* cmd/dtrace/dtrace.c: Sign extension fix for ARM/getopt.

813* driver/dtrace_isa.c: Fix use of previous_esp because it

disappears in later kernels.

812* driver/dcpc.c: Fix #elseif issue.

Sun Feb 10 00:43:12 2013 fox

811* Various: User-mode DTrace compiles on ARM (dtrace cmd runs

but does not currently get passed the USDT behavior); some

stuff stubbed out in libproc.

Kernel does not compile on ARM at present.

810* libdtrace/dt_link.c: Fix/placeholders for ARM.

809* linux/sys/elf_arm.h: New file.

808* tests/makefile: Build sys32 for ARM.

807* tools/libgcc.pl: Dont do any checks if on ARM.

Tue Feb 5 21:42:17 2013 fox

806* driver/dtrace_linux.c: Fix "pte" unref for <=2.6.24 kernels.

805* driver/xen.c: Avoid invoking Xen include file if we dont have Xen.

804* driver/vminfo.c: Conditionalise UNEVICTABLE_MLOCKFREED for 3.7

Mon Feb 4 21:19:52 2013 fox

803* driver/dtrace_linux.c: Fix for 2.6.18 kernel caused by

Xen call to set_pte_atomic.

802* linux/sys/regset.h: Fix for greg_t conflict with 3.7 kernel.

801* driver/xen.c: Fix compile issues on 2.6.32 which lack the Xen interface

we need.

Fri Jan 25 21:30:43 2013 fox

800* bug.sh: Remove the dtrace@crisp.demon.co.uk address

Thu Jan 17 22:13:57 2013 fox

799* README: Fix one of the github addresses.

Wed Dec 19 22:50:45 2012 fox

798* libdtrace/dt_open.c: Change /usr/ccs/lib/cpp to "cpp".

Mon Dec 10 22:23:10 2012 fox

797* driver/intr.c: 'Fix' for i386 to avoid use of undef structures in

Xen hypercall. Not verified. But Apurba reported the issue.

796* libdtrace/dt_module.c: Fix for core dump on i386 when doing usdt.

Fri Nov 30 19:45:05 2012 fox

795* intr.c: Fix for xen on i386 (not currently supporting this).

Thu Nov 29 20:53:05 2012 fox

794* Add 'safe' support - make it work, but turn off by default.

You can see the setting in /proc/dtrace/stats.

echo safe=1 >/proc/dtrace/stats

to turn it on - in which case, probes whilst interrupts are

disabled, are silently dropped.

Tue Nov 27 22:44:19 2012 fox

793* driver/dtrace.c, driver/dtrace_linux.c: Add safe mode support,

but doesnt seem to fire. (See /proc/dtrace/stats to watch

the counters for dropped/safe probes).

792* libdtrace/dt_module.c: Ensure we give kernel stack

traces even if we dont load linux.ctf file.

Mon Nov 26 23:18:25 2012 fox

791* driver/dtrace_isa.c: Avoid GPF when doing

dtrace -n syscall:::'{stack();}'

since we have no stack frame to play with. Also, add

GPF protection in case we walk off end of stack.

Sun Nov 25 20:13:07 2012 fox

790* driver/dtrace_isa.c: Fix for stack() - we could panic the

kernel, or for some kernels, not provide a stack.

789* driver/intr.c, driver/intr_x86-64.S, driver/xen.c, driver/x_call.c

Attempt to get Xen IPI to work, but it wont. This means

using FBT or fast firing probes can result in kernel panics.

Tue Nov 13 21:59:44 2012 fox

788* driver/intr.c, driver/intr_x86-64.S: Fix Xen interrupt

patching, and conditionalise/verify we can compile and

run on non-Xen kernels.

Fri Nov 9 22:55:23 2012 fox

787* driver/intr_x86-64.S: Fix for Xen interrupts; handle

kernel traps properly for passthru. (fbt -- int3 and int1

now appear to work properly).

Sun Nov 4 21:35:49 2012 fox

786* taskq.c, dtrace_linux.c: Avoid GPF if we directly load the

driver.

785* driver/intr.c: Updates for Xen compatibility.

Sat Oct 20 22:29:22 2012 fox

784* driver/intr.c: Fix for CONFIG_PARAVIRT (Amazon EC2 instance)

Dont inline sgdt/sidt if in paravirt mode - the instructions

are not valid.

783* driver/*: Add wp_disable/wp_enable calls and stop using

memory_set_rw to turn off write protected pages as this

doesnt work in virtualised kernels (Amazon). Thanks

to Eric Chaves for giving me access to an EC2 instance

to debug the issues here.

782* driver/dtrace.c, libdtrace/dt_open.c dt_names.c, etc/io.d

Add initial support for the fds[] array. The call of

d_path() function is not done in a mutex free environment.

If you use this on the read/write/... syscalls, all should be

well, but calling it from low level context could deadlock

the CPU. This is an initial experiment.

Comment out the io.d stuff because it may cause dtrace to

fail on hosts where we dont have the full linux.ctf file.

Sun Oct 14 21:54:24 2012 fox

781* libdtrace/dt_parser.c: Allow use of variables stored

in the linux-`uname -r`.ctf file (eg cur_thread) without

prefixing with "linux`..."

780* driver/ctf_struct.c: Add <linux/fdtable.h> for etc/io.d

Tue Oct 9 22:07:16 2012 fox

779* ctf_struct.c: Add support for <linux/migrate.h> which can

cause ctfconvert to fail due to a missing structure(?)

(Dan Mick)

778* tools/mkport.pl: Add support for HAVE_LINUX_MIGRATE_H

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