list_splice_init(list, head) does unneeded job if it is known that
list_empty(head) == 1. We can use list_replace_init() instead.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Under certain timing conditions, a race during boot occurs where timer
ticks are being processed on remote CPUs. The remote timer ticks can
increment jiffies, and if this happens during a window when a timeout is
very close to expiring but a local tick has not yet been delivered, you can
end up with
1) No softirq pending
2) A local timer wheel which is not synced to jiffies
3) No high resolution timer active
4) A local timer which is supposed to fire before the current jiffies value.
In this circumstance, the comparison in next_timer_interrupt overflows,
because the base of the comparison for high resolution timers is jiffies,
but for the softirq timer wheel, it is relative the the current base of the
wheel (jiffies_base).
Signed-off-by: Zachary Amsden <zach@vmware.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Few of the notifier_chain_register() callers use __init in the definition
of notifier_call. It is incorrect as the function definition should be
available after the initializations (they do not unregister them during
initializations).
This patch fixes all such usages to _not_ have the notifier_call __init
section.
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Few of the notifier_chain_register() callers use __devinitdata in the
definition of notifier_block data structure. It is incorrect as the
data structure should be available after the initializations (they do
not unregister them during initializations).
This was leading to an oops when notifier_chain_register() call is
invoked for those callback chains after initialization.
This patch fixes all such usages to _not_ have the notifier_block data
structure in the init data section.
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
We need the boot CPU's tvec_bases[] entry to be initialised super-early in
boot, for early_serial_setup(). That runs within setup_arch(), before even
per-cpu areas are initialised.
The patch changes tvec_bases to use compile-time initialisation, and adds a
separate array `tvec_base_done' to keep track of which CPU has had its
tvec_bases[] entry initialised (because we can no longer use the zeroness of
that tvec_bases[] entry to determine whether it has been initialised).
Thanks to Eugene Surovegin <ebs@ebshome.net> for diagnosing this.
Cc: Eugene Surovegin <ebs@ebshome.net>
Cc: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
If the HPET timer is enabled, the clock can drift by ~3 seconds a day.
This is due to the HPET timer not being initialized with the correct
setting (still using PIT count).
If HZ changes, this drift can become even more pronounced.
HPET patch initializes tick_nsec with correct tick_nsec settings for
HPET timer.
Vojtech comments:
"It's not entirely correct (it assumes the HPET ticks totally
exactly), but it's significantly better than assuming the PIT error
there."
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
this changes if() BUG(); constructs to BUG_ON() which is
cleaner, contains unlikely() and can better optimized away.
Signed-off-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Currently, count_active_tasks() calls both nr_running() &
nr_interruptible(). Each of these functions does a "for_each_cpu" & reads
values from the runqueue of each cpu. Although this is not a lot of
instructions, each runqueue may be located on different node. Depending on
the architecture, a unique TLB entry may be required to access each
runqueue.
Since there may be more runqueues than cpu TLB entries, a scan of all
runqueues can trash the TLB. Each memory reference incurs a TLB miss &
refill.
In addition, the runqueue cacheline that contains nr_running &
nr_uninterruptible may be evicted from the cache between the two passes.
This causes unnecessary cache misses.
Combining nr_running() & nr_interruptible() into a single function
substantially reduces the TLB & cache misses on large systems. This should
have no measureable effect on smaller systems.
On a 128p IA64 system running a memory stress workload, the new function
reduced the overhead of calc_load() from 605 usec/call to 324 usec/call.
Signed-off-by: Jack Steiner <steiner@sgi.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Since base and new_base are of the same type now, we can save one 'if'
branch and simplify the code a bit.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Commit a4a6198b80:
[PATCH] tvec_bases too large for per-cpu data
introduced "struct tvec_t_base_s boot_tvec_bases" which is visible at
compile time. This means we can kill __init_timer_base and move
timer_base_s's content into tvec_t_base_s.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This removes the support for pps. It's completely unused within the kernel
and is basically in the way for further cleanups. It should be easier to
readd proper support for it after the rest has been converted to NTP4
(where the pps mechanisms are quite different from NTP3 anyway).
Signed-off-by: Roman Zippel <zippel@linux-m68k.org>
Cc: Adrian Bunk <bunk@stusta.de>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
alarm() calls the kernel with an unsigend int timeout in seconds. The
value is stored in the tv_sec field of a struct timeval to setup the
itimer. The tv_sec field of struct timeval is of type long, which causes
the tv_sec value to be negative on 32 bit machines if seconds > INT_MAX.
Before the hrtimer merge (pre 2.6.16) such a negative value was converted
to the maximum jiffies timeout by the timeval_to_jiffies conversion. It's
not clear whether this was intended or just happened to be done by the
timeval_to_jiffies code.
hrtimers expect a timeval in canonical form and treat a negative timeout as
already expired. This breaks the legitimate usage of alarm() with a
timeout value > INT_MAX seconds.
For 32 bit machines it is therefor necessary to limit the internal seconds
value to avoid API breakage. Instead of doing this in all implementations
of sys_alarm the duplicated sys_alarm code is moved into a common function
in itimer.c
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Make the softlockup detector purely timer-interrupt driven, removing
softirq-context (timer) dependencies. This means that if the softlockup
watchdog triggers, it has truly observed a longer than 10 seconds
scheduling delay of a SCHED_FIFO prio 99 task.
(the patch also turns off the softlockup detector during the initial bootup
phase and does small style fixes)
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
With internal Xen-enabled kernels we see the kernel's static per-cpu data
area exceed the limit of 32k on x86-64, and even native x86-64 kernels get
fairly close to that limit. I generally question whether it is reasonable
to have data structures several kb in size allocated as per-cpu data when
the space there is rather limited.
The biggest arch-independent consumer is tvec_bases (over 4k on 32-bit
archs, over 8k on 64-bit ones), which now gets converted to use dynamically
allocated memory instead.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The pointer to the current time interpolator and the current list of time
interpolators are typically only changed during bootup. Adding
__read_mostly takes them away from possibly hot cachelines.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add a compiler barrier so that we don't read jiffies before updating
jiffies_64.
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Also from Thomas Gleixner <tglx@linutronix.de>
Function next_timer_interrupt() got broken with a recent patch
6ba1b91213 as sys_nanosleep() was moved to
hrtimer. This broke things as next_timer_interrupt() did not check hrtimer
tree for next event.
Function next_timer_interrupt() is needed with dyntick (CONFIG_NO_IDLE_HZ,
VST) implementations, as the system can be in idle when next hrtimer event
was supposed to happen. At least ARM and S390 currently use
next_timer_interrupt().
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
On some platforms readq performs additional work to make sure I/O is done
in a coherent way. This is not needed for time retrieval as done by the
time interpolator. So we can use readq_relaxed instead which will improve
performance.
It affects sparc64 and ia64 only. Apparently it makes a significant
difference on ia64.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Cc: john stultz <johnstul@us.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This provides an interface for arch code to find out how many
nanoseconds are going to be added on to xtime by the next call to
do_timer. The value returned is a fixed-point number in 52.12 format
in nanoseconds. The reason for this format is that it gives the
full precision that the timekeeping code is using internally.
The motivation for this is to fix a problem that has arisen on 32-bit
powerpc in that the value returned by do_gettimeofday drifts apart
from xtime if NTP is being used. PowerPC is now using a lockless
do_gettimeofday based on reading the timebase register and performing
some simple arithmetic. (This method of getting the time is also
exported to userspace via the VDSO.) However, the factor and offset
it uses were calculated based on the nominal tick length and weren't
being adjusted when NTP varied the tick length.
Note that 64-bit powerpc has had the lockless do_gettimeofday for a
long time now. It also had an extremely hairy routine that got called
from the 32-bit compat routine for adjtimex, which adjusted the
factor and offset according to what it thought the timekeeping code
was going to do. Not only was this only called if a 32-bit task did
adjtimex (i.e. not if a 64-bit task did adjtimex), it was also
duplicating computations from kernel/timer.c and it wasn't clear that
it was (still) correct.
The simple solution is to ask the timekeeping code how long the
current jiffy will be on each timer interrupt, after calling
do_timer. If this jiffy will be a different length from the last one,
we then need to compute new values for the factor and offset used in
the lockless do_gettimeofday. In this way we can keep xtime and
do_gettimeofday in sync, even when NTP is varying the tick length.
Note that when adjtimex varies the tick length, it almost always
introduces the variation from the next tick on. The only case I could
see where adjtimex would vary the length of the current tick is when
an old-style adjtime adjustment is being cancelled. (It's not clear
to me why the adjustment has to be cancelled immediately rather than
from the next tick on.) Thus I don't see any real need for a hook in
adjtimex; the rare case of an old-style adjustment being cancelled can
be fixed up at the next tick.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Acked-by: john stultz <johnstul@us.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
hrtimer subsystem core. It is initialized at bootup and expired by the timer
interrupt, but is otherwise not utilized by any other subsystem yet.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch contains the following cleanups:
- make needlessly global functions static
- every file should include the headers containing the prototypes for
it's global functions
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Acked-by: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Define jiffies_64 in kernel/timer.c rather than having 24 duplicated
defines in each architecture.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Remove timer_list.magic and associated debugging code.
I originally added this when a spinlock was added to timer_list - this meant
that an all-zeroes timer became illegal and init_timer() was required.
That spinlock isn't even there any more, although timer.base must now be
initialised.
I'll keep this debugging code in -mm.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Create a macro shift_right() that avoids the numerous ugly conditionals in the
NTP code that look like:
if(a < 0)
b = -(-a >> shift);
else
b = a >> shift;
Replacing it with:
b = shift_right(a, shift);
This should have zero effect on the logic, however it should probably have
a bit of testing just to be sure.
Also replace open-coded min/max with the macros.
Signed-off-by : John Stultz <johnstul@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Every user of init_timer() also needs to initialize ->function and ->data
fields. This patch adds a simple setup_timer() helper for that.
The schedule_timeout() is patched as an example of usage.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
These functions don't need schedule_timeout()'s barrier.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add schedule_timeout_{,un}interruptible() interfaces so that
schedule_timeout() callers don't have to worry about forgetting to add the
set_current_state() call beforehand.
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Christoph Lameter <clameter@engr.sgi.com>
When using a time interpolator that is susceptible to jitter there's
potentially contention over a cmpxchg used to prevent time from going
backwards. This is unnecessary when the caller holds the xtime write
seqlock as all readers will be blocked from returning until the write is
complete. We can therefore allow writers to insert a new value and exit
rather than fight with CPUs who only hold a reader lock.
Signed-off-by: Alex Williamson <alex.williamson@hp.com>
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch adds a new kernel debug feature: CONFIG_DETECT_SOFTLOCKUP.
When enabled then per-CPU watchdog threads are started, which try to run
once per second. If they get delayed for more than 10 seconds then a
callback from the timer interrupt detects this condition and prints out a
warning message and a stack dump (once per lockup incident). The feature
is otherwise non-intrusive, it doesnt try to unlock the box in any way, it
only gets the debug info out, automatically, and on all CPUs affected by
the lockup.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-Off-By: Matthias Urlichs <smurf@smurf.noris.de>
Signed-off-by: Richard Purdie <rpurdie@rpsys.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
With CONFIG_PREEMPT && !CONFIG_SMP, it's possible for sys_getppid to
return a bogus value if the parent's task_struct gets reallocated after
current->group_leader->real_parent is read:
asmlinkage long sys_getppid(void)
{
int pid;
struct task_struct *me = current;
struct task_struct *parent;
parent = me->group_leader->real_parent;
RACE HERE => for (;;) {
pid = parent->tgid;
#ifdef CONFIG_SMP
{
struct task_struct *old = parent;
/*
* Make sure we read the pid before re-reading the
* parent pointer:
*/
smp_rmb();
parent = me->group_leader->real_parent;
if (old != parent)
continue;
}
#endif
break;
}
return pid;
}
If the process gets preempted at the indicated point, the parent process
can go ahead and call exit() and then get wait()'d on to reap its
task_struct. When the preempted process gets resumed, it will not do any
further checks of the parent pointer on !CONFIG_SMP: it will read the
bad pid and return.
So, the same algorithm used when SMP is enabled should be used when
preempt is enabled, which will recheck ->real_parent in this case.
Signed-off-by: David Meybohm <dmeybohmlkml@bellsouth.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The comment for msleep_interruptible() is wrong, as it will ignore
wait-queue events, but will wake up early for signals.
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
In kernel/sched.c the return value from preempt_count() is cast to an int.
That made sense when preempt_count was defined as different types on is not
needed and should go away. The patch removes the cast.
In kernel/timer.c the return value from preempt_count() is assigned to a
variable of type u32 and then that unsigned value is later compared to
preempt_count(). Since preempt_count() returns an int, an int is what
should be used to store its return value. Storing the result in an
unsigned 32bit integer made a tiny bit of sense back when preempt_count was
different types on different archs, but no more - let's not play signed vs
unsigned comparison games when we don't have to. The patch modifies the
code to use an int to hold the value. While I was around that bit of code
I also made two changes to a nearby (related) printk() - I modified it to
specify the loglevel explicitly and also broke the line into a few pieces
to avoid it being longer than 80 chars and clarified the text a bit.
Signed-off-by: Jesper Juhl <juhl-lkml@dif.dk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch splits del_timer_sync() into 2 functions. The new one,
try_to_del_timer_sync(), returns -1 when it hits executing timer.
It can be used in interrupt context, or when the caller hold locks which
can prevent completion of the timer's handler.
NOTE. Currently it can't be used in interrupt context in UP case, because
->running_timer is used only with CONFIG_SMP.
Should the need arise, it is possible to kill #ifdef CONFIG_SMP in
set_running_timer(), it is cheap.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch tries to solve following problems:
1. del_timer_sync() is racy. The timer can be fired again after
del_timer_sync have checked all cpus and before it will recheck
timer_pending().
2. It has scalability problems. All cpus are scanned to determine
if the timer is running on that cpu.
With this patch del_timer_sync is O(1) and no slower than plain
del_timer(pending_timer), unless it has to actually wait for
completion of the currently running timer.
The only restriction is that the recurring timer should not use
add_timer_on().
3. The timers are not serialized wrt to itself.
If CPU_0 does mod_timer(jiffies+1) while the timer is currently
running on CPU 1, it is quite possible that local interrupt on
CPU_0 will start that timer before it finished on CPU_1.
4. The timers locking is suboptimal. __mod_timer() takes 3 locks
at once and still requires wmb() in del_timer/run_timers.
The new implementation takes 2 locks sequentially and does not
need memory barriers.
Currently ->base != NULL means that the timer is pending. In that case
->base.lock is used to lock the timer. __mod_timer also takes timer->lock
because ->base can be == NULL.
This patch uses timer->entry.next != NULL as indication that the timer is
pending. So it does __list_del(), entry->next = NULL instead of list_del()
when the timer is deleted.
The ->base field is used for hashed locking only, it is initialized
in init_timer() which sets ->base = per_cpu(tvec_bases). When the
tvec_bases.lock is locked, it means that all timers which are tied
to this base via timer->base are locked, and the base itself is locked
too.
So __run_timers/migrate_timers can safely modify all timers which could
be found on ->tvX lists (pending timers).
When the timer's base is locked, and the timer removed from ->entry list
(which means that _run_timers/migrate_timers can't see this timer), it is
possible to set timer->base = NULL and drop the lock: the timer remains
locked.
This patch adds lock_timer_base() helper, which waits for ->base != NULL,
locks the ->base, and checks it is still the same.
__mod_timer() schedules the timer on the local CPU and changes it's base.
However, it does not lock both old and new bases at once. It locks the
timer via lock_timer_base(), deletes the timer, sets ->base = NULL, and
unlocks old base. Then __mod_timer() locks new_base, sets ->base = new_base,
and adds this timer. This simplifies the code, because AB-BA deadlock is not
possible. __mod_timer() also ensures that the timer's base is not changed
while the timer's handler is running on the old base.
__run_timers(), del_timer() do not change ->base anymore, they only clear
pending flag.
So del_timer_sync() can test timer->base->running_timer == timer to detect
whether it is running or not.
We don't need timer_list->lock anymore, this patch kills it.
We also don't need barriers. del_timer() and __run_timers() used smp_wmb()
before clearing timer's pending flag. It was needed because __mod_timer()
did not lock old_base if the timer is not pending, so __mod_timer()->list_add()
could race with del_timer()->list_del(). With this patch these functions are
serialized through base->lock.
One problem. TIMER_INITIALIZER can't use per_cpu(tvec_bases). So this patch
adds global
struct timer_base_s {
spinlock_t lock;
struct timer_list *running_timer;
} __init_timer_base;
which is used by TIMER_INITIALIZER. The corresponding fields in tvec_t_base_s
struct are replaced by struct timer_base_s t_base.
It is indeed ugly. But this can't have scalability problems. The global
__init_timer_base.lock is used only when __mod_timer() is called for the first
time AND the timer was compile time initialized. After that the timer migrates
to the local CPU.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Renaud Lienhart <renaud.lienhart@free.fr>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Replace a number of memory barriers with smp_ variants. This means we won't
take the unnecessary hit on UP machines.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
