2005-10-10 16:36:14 +04:00
/ *
* PowerPC v e r s i o n
* Copyright ( C ) 1 9 9 5 - 1 9 9 6 G a r y T h o m a s ( g d t @linuxppc.org)
* Rewritten b y C o r t D o u g a n ( c o r t @cs.nmt.edu) for PReP
* Copyright ( C ) 1 9 9 6 C o r t D o u g a n < c o r t @cs.nmt.edu>
* Adapted f o r P o w e r M a c i n t o s h b y P a u l M a c k e r r a s .
* Low- l e v e l e x c e p t i o n h a n d l e r s a n d M M U s u p p o r t
* rewritten b y P a u l M a c k e r r a s .
* Copyright ( C ) 1 9 9 6 P a u l M a c k e r r a s .
* MPC8 x x m o d i f i c a t i o n s C o p y r i g h t ( C ) 1 9 9 7 D a n M a l e k ( d m a l e k @jlc.net).
*
* This f i l e c o n t a i n s t h e s y s t e m c a l l e n t r y c o d e , c o n t e x t s w i t c h
* code, a n d e x c e p t i o n / i n t e r r u p t r e t u r n c o d e f o r P o w e r P C .
*
* This p r o g r a m i s f r e e s o f t w a r e ; you can redistribute it and/or
* modify i t u n d e r t h e t e r m s o f t h e G N U G e n e r a l P u b l i c L i c e n s e
* as p u b l i s h e d b y t h e F r e e S o f t w a r e F o u n d a t i o n ; either version
* 2 of t h e L i c e n s e , o r ( a t y o u r o p t i o n ) a n y l a t e r v e r s i o n .
* /
# include < l i n u x / e r r n o . h >
# include < a s m / u n i s t d . h >
# include < a s m / p r o c e s s o r . h >
# include < a s m / p a g e . h >
# include < a s m / m m u . h >
# include < a s m / t h r e a d _ i n f o . h >
# include < a s m / p p c _ a s m . h >
# include < a s m / a s m - o f f s e t s . h >
# include < a s m / c p u t a b l e . h >
2006-09-25 12:19:00 +04:00
# include < a s m / f i r m w a r e . h >
2007-01-01 21:45:34 +03:00
# include < a s m / b u g . h >
2008-04-17 08:34:59 +04:00
# include < a s m / p t r a c e . h >
2008-04-17 08:35:01 +04:00
# include < a s m / i r q f l a g s . h >
2008-06-21 22:17:27 +04:00
# include < a s m / f t r a c e . h >
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
# include < a s m / h w _ i r q . h >
2013-05-13 20:16:43 +04:00
# include < a s m / c o n t e x t _ t r a c k i n g . h >
2005-10-10 16:36:14 +04:00
/ *
* System c a l l s .
* /
.section " .toc " , " aw"
.SYS_CALL_TABLE :
.tc .sys_call_table [ TC] ,. s y s _ c a l l _ t a b l e
/* This value is used to mark exception frames on the stack. */
exception_marker :
2008-04-17 08:34:59 +04:00
.tc _ E X C _ M A R K E R [ T C ] ,S T A C K _ F R A M E _ R E G S _ M A R K E R
2005-10-10 16:36:14 +04:00
.section " .text "
.align 7
# undef S H O W _ S Y S C A L L S
.globl system_call_common
system_call_common :
andi. r10 ,r12 ,M S R _ P R
mr r10 ,r1
addi r1 ,r1 ,- I N T _ F R A M E _ S I Z E
beq- 1 f
ld r1 ,P A C A K S A V E ( r13 )
1 : std r10 ,0 ( r1 )
std r11 ,_ N I P ( r1 )
std r12 ,_ M S R ( r1 )
std r0 ,G P R 0 ( r1 )
std r10 ,G P R 1 ( r1 )
2012-12-07 01:46:37 +04:00
beq 2 f / * i f f r o m k e r n e l m o d e * /
powerpc: Implement accurate task and CPU time accounting
This implements accurate task and cpu time accounting for 64-bit
powerpc kernels. Instead of accounting a whole jiffy of time to a
task on a timer interrupt because that task happened to be running at
the time, we now account time in units of timebase ticks according to
the actual time spent by the task in user mode and kernel mode. We
also count the time spent processing hardware and software interrupts
accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If
that is not set, we do tick-based approximate accounting as before.
To get this accurate information, we read either the PURR (processor
utilization of resources register) on POWER5 machines, or the timebase
on other machines on
* each entry to the kernel from usermode
* each exit to usermode
* transitions between process context, hard irq context and soft irq
context in kernel mode
* context switches.
On POWER5 systems with shared-processor logical partitioning we also
read both the PURR and the timebase at each timer interrupt and
context switch in order to determine how much time has been taken by
the hypervisor to run other partitions ("steal" time). Unfortunately,
since we need values of the PURR on both threads at the same time to
accurately calculate the steal time, and since we can only calculate
steal time on a per-core basis, the apportioning of the steal time
between idle time (time which we ceded to the hypervisor in the idle
loop) and actual stolen time is somewhat approximate at the moment.
This is all based quite heavily on what s390 does, and it uses the
generic interfaces that were added by the s390 developers,
i.e. account_system_time(), account_user_time(), etc.
This patch doesn't add any new interfaces between the kernel and
userspace, and doesn't change the units in which time is reported to
userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(),
times(), etc. Internally the various task and cpu times are stored in
timebase units, but they are converted to USER_HZ units (1/100th of a
second) when reported to userspace. Some precision is therefore lost
but there should not be any accumulating error, since the internal
accumulation is at full precision.
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-24 02:06:59 +03:00
ACCOUNT_ C P U _ U S E R _ E N T R Y ( r10 , r11 )
2012-12-07 01:46:37 +04:00
2 : std r2 ,G P R 2 ( r1 )
2005-10-10 16:36:14 +04:00
std r3 ,G P R 3 ( r1 )
2012-04-05 07:44:48 +04:00
mfcr r2
2005-10-10 16:36:14 +04:00
std r4 ,G P R 4 ( r1 )
std r5 ,G P R 5 ( r1 )
std r6 ,G P R 6 ( r1 )
std r7 ,G P R 7 ( r1 )
std r8 ,G P R 8 ( r1 )
li r11 ,0
std r11 ,G P R 9 ( r1 )
std r11 ,G P R 1 0 ( r1 )
std r11 ,G P R 1 1 ( r1 )
std r11 ,G P R 1 2 ( r1 )
2012-04-04 22:24:29 +04:00
std r11 ,_ X E R ( r1 )
2012-04-04 22:26:39 +04:00
std r11 ,_ C T R ( r1 )
2005-10-10 16:36:14 +04:00
std r9 ,G P R 1 3 ( r1 )
mflr r10
2012-04-05 07:44:48 +04:00
/ *
* This c l e a r s C R 0 . S O ( b i t 2 8 ) , w h i c h i s t h e e r r o r i n d i c a t i o n o n
* return f r o m t h i s s y s t e m c a l l .
* /
rldimi r2 ,r11 ,2 8 ,( 6 3 - 2 8 )
2005-10-10 16:36:14 +04:00
li r11 ,0 x c01
std r10 ,_ L I N K ( r1 )
std r11 ,_ T R A P ( r1 )
std r3 ,O R I G _ G P R 3 ( r1 )
2012-04-05 07:44:48 +04:00
std r2 ,_ C C R ( r1 )
2005-10-10 16:36:14 +04:00
ld r2 ,P A C A T O C ( r13 )
addi r9 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D
ld r11 ,e x c e p t i o n _ m a r k e r @toc(r2)
std r11 ,- 1 6 ( r9 ) / * " r e g s h e r e " m a r k e r * /
2012-07-25 09:56:04 +04:00
# if d e f i n e d ( C O N F I G _ V I R T _ C P U _ A C C O U N T I N G _ N A T I V E ) & & d e f i n e d ( C O N F I G _ P P C _ S P L P A R )
powerpc: Account time using timebase rather than PURR
Currently, when CONFIG_VIRT_CPU_ACCOUNTING is enabled, we use the
PURR register for measuring the user and system time used by
processes, as well as other related times such as hardirq and
softirq times. This turns out to be quite confusing for users
because it means that a program will often be measured as taking
less time when run on a multi-threaded processor (SMT2 or SMT4 mode)
than it does when run on a single-threaded processor (ST mode), even
though the program takes longer to finish. The discrepancy is
accounted for as stolen time, which is also confusing, particularly
when there are no other partitions running.
This changes the accounting to use the timebase instead, meaning that
the reported user and system times are the actual number of real-time
seconds that the program was executing on the processor thread,
regardless of which SMT mode the processor is in. Thus a program will
generally show greater user and system times when run on a
multi-threaded processor than on a single-threaded processor.
On pSeries systems on POWER5 or later processors, we measure the
stolen time (time when this partition wasn't running) using the
hypervisor dispatch trace log. We check for new entries in the
log on every entry from user mode and on every transition from
kernel process context to soft or hard IRQ context (i.e. when
account_system_vtime() gets called). So that we can correctly
distinguish time stolen from user time and time stolen from system
time, without having to check the log on every exit to user mode,
we store separate timestamps for exit to user mode and entry from
user mode.
On systems that have a SPURR (POWER6 and POWER7), we read the SPURR
in account_system_vtime() (as before), and then apportion the SPURR
ticks since the last time we read it between scaled user time and
scaled system time according to the relative proportions of user
time and system time over the same interval. This avoids having to
read the SPURR on every kernel entry and exit. On systems that have
PURR but not SPURR (i.e., POWER5), we do the same using the PURR
rather than the SPURR.
This disables the DTL user interface in /sys/debug/kernel/powerpc/dtl
for now since it conflicts with the use of the dispatch trace log
by the time accounting code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-08-26 23:56:43 +04:00
BEGIN_ F W _ F T R _ S E C T I O N
beq 3 3 f
/* if from user, see if there are any DTL entries to process */
ld r10 ,P A C A L P P A C A P T R ( r13 ) / * g e t p t r t o V P A * /
ld r11 ,P A C A _ D T L _ R I D X ( r13 ) / * g e t l o g r e a d i n d e x * /
2013-08-06 20:01:46 +04:00
addi r10 ,r10 ,L P P A C A _ D T L I D X
LDX_ B E r10 ,0 ,r10 / * g e t l o g w r i t e i n d e x * /
powerpc: Account time using timebase rather than PURR
Currently, when CONFIG_VIRT_CPU_ACCOUNTING is enabled, we use the
PURR register for measuring the user and system time used by
processes, as well as other related times such as hardirq and
softirq times. This turns out to be quite confusing for users
because it means that a program will often be measured as taking
less time when run on a multi-threaded processor (SMT2 or SMT4 mode)
than it does when run on a single-threaded processor (ST mode), even
though the program takes longer to finish. The discrepancy is
accounted for as stolen time, which is also confusing, particularly
when there are no other partitions running.
This changes the accounting to use the timebase instead, meaning that
the reported user and system times are the actual number of real-time
seconds that the program was executing on the processor thread,
regardless of which SMT mode the processor is in. Thus a program will
generally show greater user and system times when run on a
multi-threaded processor than on a single-threaded processor.
On pSeries systems on POWER5 or later processors, we measure the
stolen time (time when this partition wasn't running) using the
hypervisor dispatch trace log. We check for new entries in the
log on every entry from user mode and on every transition from
kernel process context to soft or hard IRQ context (i.e. when
account_system_vtime() gets called). So that we can correctly
distinguish time stolen from user time and time stolen from system
time, without having to check the log on every exit to user mode,
we store separate timestamps for exit to user mode and entry from
user mode.
On systems that have a SPURR (POWER6 and POWER7), we read the SPURR
in account_system_vtime() (as before), and then apportion the SPURR
ticks since the last time we read it between scaled user time and
scaled system time according to the relative proportions of user
time and system time over the same interval. This avoids having to
read the SPURR on every kernel entry and exit. On systems that have
PURR but not SPURR (i.e., POWER5), we do the same using the PURR
rather than the SPURR.
This disables the DTL user interface in /sys/debug/kernel/powerpc/dtl
for now since it conflicts with the use of the dispatch trace log
by the time accounting code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-08-26 23:56:43 +04:00
cmpd c r1 ,r11 ,r10
beq+ c r1 ,3 3 f
bl . a c c u m u l a t e _ s t o l e n _ t i m e
REST_ G P R ( 0 ,r1 )
REST_ 4 G P R S ( 3 ,r1 )
REST_ 2 G P R S ( 7 ,r1 )
addi r9 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D
33 :
END_ F W _ F T R _ S E C T I O N _ I F S E T ( F W _ F E A T U R E _ S P L P A R )
2012-07-25 09:56:04 +04:00
# endif / * C O N F I G _ V I R T _ C P U _ A C C O U N T I N G _ N A T I V E & & C O N F I G _ P P C _ S P L P A R * /
powerpc: Account time using timebase rather than PURR
Currently, when CONFIG_VIRT_CPU_ACCOUNTING is enabled, we use the
PURR register for measuring the user and system time used by
processes, as well as other related times such as hardirq and
softirq times. This turns out to be quite confusing for users
because it means that a program will often be measured as taking
less time when run on a multi-threaded processor (SMT2 or SMT4 mode)
than it does when run on a single-threaded processor (ST mode), even
though the program takes longer to finish. The discrepancy is
accounted for as stolen time, which is also confusing, particularly
when there are no other partitions running.
This changes the accounting to use the timebase instead, meaning that
the reported user and system times are the actual number of real-time
seconds that the program was executing on the processor thread,
regardless of which SMT mode the processor is in. Thus a program will
generally show greater user and system times when run on a
multi-threaded processor than on a single-threaded processor.
On pSeries systems on POWER5 or later processors, we measure the
stolen time (time when this partition wasn't running) using the
hypervisor dispatch trace log. We check for new entries in the
log on every entry from user mode and on every transition from
kernel process context to soft or hard IRQ context (i.e. when
account_system_vtime() gets called). So that we can correctly
distinguish time stolen from user time and time stolen from system
time, without having to check the log on every exit to user mode,
we store separate timestamps for exit to user mode and entry from
user mode.
On systems that have a SPURR (POWER6 and POWER7), we read the SPURR
in account_system_vtime() (as before), and then apportion the SPURR
ticks since the last time we read it between scaled user time and
scaled system time according to the relative proportions of user
time and system time over the same interval. This avoids having to
read the SPURR on every kernel entry and exit. On systems that have
PURR but not SPURR (i.e., POWER5), we do the same using the PURR
rather than the SPURR.
This disables the DTL user interface in /sys/debug/kernel/powerpc/dtl
for now since it conflicts with the use of the dispatch trace log
by the time accounting code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-08-26 23:56:43 +04:00
2012-03-01 08:40:23 +04:00
/ *
* A s y s c a l l s h o u l d a l w a y s b e c a l l e d w i t h i n t e r r u p t s e n a b l e d
* so w e j u s t u n c o n d i t i o n a l l y h a r d - e n a b l e h e r e . W h e n s o m e k i n d
* of i r q t r a c i n g i s u s e d , w e a d d i t i o n a l l y c h e c k t h a t c o n d i t i o n
* is c o r r e c t
* /
# if d e f i n e d ( C O N F I G _ T R A C E _ I R Q F L A G S ) & & d e f i n e d ( C O N F I G _ B U G )
lbz r10 ,P A C A S O F T I R Q E N ( r13 )
xori r10 ,r10 ,1
1 : tdnei r10 ,0
EMIT_ B U G _ E N T R Y 1 b ,_ _ F I L E _ _ ,_ _ L I N E _ _ ,B U G F L A G _ W A R N I N G
# endif
2009-07-24 03:15:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
wrteei 1
# else
2012-03-01 08:40:23 +04:00
ld r11 ,P A C A K M S R ( r13 )
2005-10-10 16:36:14 +04:00
ori r11 ,r11 ,M S R _ E E
mtmsrd r11 ,1
2009-07-24 03:15:59 +04:00
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
2005-10-10 16:36:14 +04:00
2012-03-01 08:40:23 +04:00
/ * We d o n e e d t o s e t S O F T E i n t h e s t a c k f r a m e o r t h e r e t u r n
* from i n t e r r u p t w i l l b e p a i n f u l
* /
li r10 ,1
std r10 ,S O F T E ( r1 )
2005-10-10 16:36:14 +04:00
# ifdef S H O W _ S Y S C A L L S
bl . d o _ s h o w _ s y s c a l l
REST_ G P R ( 0 ,r1 )
REST_ 4 G P R S ( 3 ,r1 )
REST_ 2 G P R S ( 7 ,r1 )
addi r9 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D
# endif
2012-07-05 08:41:35 +04:00
CURRENT_ T H R E A D _ I N F O ( r11 , r1 )
2005-10-10 16:36:14 +04:00
ld r10 ,T I _ F L A G S ( r11 )
andi. r11 ,r10 ,_ T I F _ S Y S C A L L _ T _ O R _ A
2013-01-09 03:47:36 +04:00
bne s y s c a l l _ d o t r a c e
2012-04-04 22:23:27 +04:00
.Lsyscall_dotrace_cont :
2005-10-10 16:36:14 +04:00
cmpldi 0 ,r0 ,N R _ s y s c a l l s
bge- s y s c a l l _ e n o s y s
system_call : /* label this so stack traces look sane */
/ *
* Need t o v e c t o r t o 3 2 B i t o r d e f a u l t s y s _ c a l l _ t a b l e h e r e ,
* based o n c a l l e r ' s r u n - m o d e / p e r s o n a l i t y .
* /
ld r11 ,. S Y S _ C A L L _ T A B L E @toc(2)
andi. r10 ,r10 ,_ T I F _ 3 2 B I T
beq 1 5 f
addi r11 ,r11 ,8 / * u s e 3 2 - b i t s y s c a l l e n t r i e s * /
clrldi r3 ,r3 ,3 2
clrldi r4 ,r4 ,3 2
clrldi r5 ,r5 ,3 2
clrldi r6 ,r6 ,3 2
clrldi r7 ,r7 ,3 2
clrldi r8 ,r8 ,3 2
15 :
slwi r0 ,r0 ,4
ldx r10 ,r11 ,r0 / * F e t c h s y s t e m c a l l h a n d l e r [ p t r ] * /
mtctr r10
bctrl / * C a l l h a n d l e r * /
syscall_exit :
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
std r3 ,R E S U L T ( r1 )
2005-10-10 16:36:14 +04:00
# ifdef S H O W _ S Y S C A L L S
bl . d o _ s h o w _ s y s c a l l _ e x i t
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
ld r3 ,R E S U L T ( r1 )
2005-10-10 16:36:14 +04:00
# endif
2012-07-05 08:41:35 +04:00
CURRENT_ T H R E A D _ I N F O ( r12 , r1 )
2005-10-10 16:36:14 +04:00
ld r8 ,_ M S R ( r1 )
2009-07-24 03:15:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 S
/* No MSR:RI on BookE */
2005-10-10 16:36:14 +04:00
andi. r10 ,r8 ,M S R _ R I
beq- u n r e c o v _ r e s t o r e
2009-07-24 03:15:59 +04:00
# endif
2012-03-01 08:40:23 +04:00
/ *
* Disable i n t e r r u p t s s o c u r r e n t _ t h r e a d _ i n f o ( ) - > f l a g s c a n ' t c h a n g e ,
2009-07-24 03:15:59 +04:00
* and s o t h a t w e d o n ' t g e t i n t e r r u p t e d a f t e r l o a d i n g S R R 0 / 1 .
* /
# ifdef C O N F I G _ P P C _ B O O K 3 E
wrteei 0
# else
2012-03-01 08:40:23 +04:00
ld r10 ,P A C A K M S R ( r13 )
2012-05-29 16:22:00 +04:00
/ *
* For p e r f o r m a n c e r e a s o n s w e c l e a r R I t h e s a m e t i m e t h a t w e
* clear E E . W e o n l y n e e d t o c l e a r R I j u s t b e f o r e w e r e s t o r e r13
* below, b u t b a t c h i n g i t w i t h E E s a v e s u s o n e e x p e n s i v e m t m s r d c a l l .
* We h a v e t o b e c a r e f u l t o r e s t o r e R I i f w e b r a n c h a n y w h e r e f r o m
* here ( e g s y s c a l l _ e x i t _ w o r k ) .
* /
li r9 ,M S R _ R I
andc r11 ,r10 ,r9
mtmsrd r11 ,1
2009-07-24 03:15:59 +04:00
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
2005-10-10 16:36:14 +04:00
ld r9 ,T I _ F L A G S ( r12 )
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
li r11 ,- _ L A S T _ E R R N O
2006-03-08 05:24:22 +03:00
andi. r0 ,r9 ,( _ T I F _ S Y S C A L L _ T _ O R _ A | _ T I F _ S I N G L E S T E P | _ T I F _ U S E R _ W O R K _ M A S K | _ T I F _ P E R S Y S C A L L _ M A S K )
2005-10-10 16:36:14 +04:00
bne- s y s c a l l _ e x i t _ w o r k
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
cmpld r3 ,r11
ld r5 ,_ C C R ( r1 )
bge- s y s c a l l _ e r r o r
2012-04-04 22:23:27 +04:00
.Lsyscall_error_cont :
2005-10-10 16:36:14 +04:00
ld r7 ,_ N I P ( r1 )
2010-08-11 05:40:27 +04:00
BEGIN_ F T R _ S E C T I O N
2005-10-10 16:36:14 +04:00
stdcx. r0 ,0 ,r1 / * t o c l e a r t h e r e s e r v a t i o n * /
2010-08-11 05:40:27 +04:00
END_ F T R _ S E C T I O N _ I F C L R ( C P U _ F T R _ S T C X _ C H E C K S _ A D D R E S S )
2005-10-10 16:36:14 +04:00
andi. r6 ,r8 ,M S R _ P R
ld r4 ,_ L I N K ( r1 )
2009-07-24 03:15:59 +04:00
powerpc: Implement accurate task and CPU time accounting
This implements accurate task and cpu time accounting for 64-bit
powerpc kernels. Instead of accounting a whole jiffy of time to a
task on a timer interrupt because that task happened to be running at
the time, we now account time in units of timebase ticks according to
the actual time spent by the task in user mode and kernel mode. We
also count the time spent processing hardware and software interrupts
accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If
that is not set, we do tick-based approximate accounting as before.
To get this accurate information, we read either the PURR (processor
utilization of resources register) on POWER5 machines, or the timebase
on other machines on
* each entry to the kernel from usermode
* each exit to usermode
* transitions between process context, hard irq context and soft irq
context in kernel mode
* context switches.
On POWER5 systems with shared-processor logical partitioning we also
read both the PURR and the timebase at each timer interrupt and
context switch in order to determine how much time has been taken by
the hypervisor to run other partitions ("steal" time). Unfortunately,
since we need values of the PURR on both threads at the same time to
accurately calculate the steal time, and since we can only calculate
steal time on a per-core basis, the apportioning of the steal time
between idle time (time which we ceded to the hypervisor in the idle
loop) and actual stolen time is somewhat approximate at the moment.
This is all based quite heavily on what s390 does, and it uses the
generic interfaces that were added by the s390 developers,
i.e. account_system_time(), account_user_time(), etc.
This patch doesn't add any new interfaces between the kernel and
userspace, and doesn't change the units in which time is reported to
userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(),
times(), etc. Internally the various task and cpu times are stored in
timebase units, but they are converted to USER_HZ units (1/100th of a
second) when reported to userspace. Some precision is therefore lost
but there should not be any accumulating error, since the internal
accumulation is at full precision.
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-24 02:06:59 +03:00
beq- 1 f
ACCOUNT_ C P U _ U S E R _ E X I T ( r11 , r12 )
2012-12-07 01:51:04 +04:00
HMT_ M E D I U M _ L O W _ H A S _ P P R
powerpc: Implement accurate task and CPU time accounting
This implements accurate task and cpu time accounting for 64-bit
powerpc kernels. Instead of accounting a whole jiffy of time to a
task on a timer interrupt because that task happened to be running at
the time, we now account time in units of timebase ticks according to
the actual time spent by the task in user mode and kernel mode. We
also count the time spent processing hardware and software interrupts
accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If
that is not set, we do tick-based approximate accounting as before.
To get this accurate information, we read either the PURR (processor
utilization of resources register) on POWER5 machines, or the timebase
on other machines on
* each entry to the kernel from usermode
* each exit to usermode
* transitions between process context, hard irq context and soft irq
context in kernel mode
* context switches.
On POWER5 systems with shared-processor logical partitioning we also
read both the PURR and the timebase at each timer interrupt and
context switch in order to determine how much time has been taken by
the hypervisor to run other partitions ("steal" time). Unfortunately,
since we need values of the PURR on both threads at the same time to
accurately calculate the steal time, and since we can only calculate
steal time on a per-core basis, the apportioning of the steal time
between idle time (time which we ceded to the hypervisor in the idle
loop) and actual stolen time is somewhat approximate at the moment.
This is all based quite heavily on what s390 does, and it uses the
generic interfaces that were added by the s390 developers,
i.e. account_system_time(), account_user_time(), etc.
This patch doesn't add any new interfaces between the kernel and
userspace, and doesn't change the units in which time is reported to
userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(),
times(), etc. Internally the various task and cpu times are stored in
timebase units, but they are converted to USER_HZ units (1/100th of a
second) when reported to userspace. Some precision is therefore lost
but there should not be any accumulating error, since the internal
accumulation is at full precision.
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-24 02:06:59 +03:00
ld r13 ,G P R 1 3 ( r1 ) / * o n l y r e s t o r e r13 i f r e t u r n i n g t o u s e r m o d e * /
2005-10-10 16:36:14 +04:00
1 : ld r2 ,G P R 2 ( r1 )
ld r1 ,G P R 1 ( r1 )
mtlr r4
mtcr r5
mtspr S P R N _ S R R 0 ,r7
mtspr S P R N _ S R R 1 ,r8
2009-07-24 03:15:59 +04:00
RFI
2005-10-10 16:36:14 +04:00
b . / * p r e v e n t s p e c u l a t i v e e x e c u t i o n * /
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
syscall_error :
2005-10-10 16:36:14 +04:00
oris r5 ,r5 ,0 x10 0 0 / * S e t S O b i t i n C R * /
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
neg r3 ,r3
2005-10-10 16:36:14 +04:00
std r5 ,_ C C R ( r1 )
2012-04-04 22:23:27 +04:00
b . L s y s c a l l _ e r r o r _ c o n t
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
2005-10-10 16:36:14 +04:00
/* Traced system call support */
syscall_dotrace :
bl . s a v e _ n v g p r s
addi r3 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D
bl . d o _ s y s c a l l _ t r a c e _ e n t e r
2008-07-27 10:51:03 +04:00
/ *
* Restore a r g u m e n t r e g i s t e r s p o s s i b l y j u s t c h a n g e d .
* We u s e t h e r e t u r n v a l u e o f d o _ s y s c a l l _ t r a c e _ e n t e r
* for t h e c a l l n u m b e r t o l o o k u p i n t h e t a b l e ( r0 ) .
* /
mr r0 ,r3
2005-10-10 16:36:14 +04:00
ld r3 ,G P R 3 ( r1 )
ld r4 ,G P R 4 ( r1 )
ld r5 ,G P R 5 ( r1 )
ld r6 ,G P R 6 ( r1 )
ld r7 ,G P R 7 ( r1 )
ld r8 ,G P R 8 ( r1 )
addi r9 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D
2012-07-05 08:41:35 +04:00
CURRENT_ T H R E A D _ I N F O ( r10 , r1 )
2005-10-10 16:36:14 +04:00
ld r10 ,T I _ F L A G S ( r10 )
2012-04-04 22:23:27 +04:00
b . L s y s c a l l _ d o t r a c e _ c o n t
2005-10-10 16:36:14 +04:00
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
syscall_enosys :
li r3 ,- E N O S Y S
b s y s c a l l _ e x i t
syscall_exit_work :
2012-05-29 16:22:00 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 S
mtmsrd r10 ,1 / * R e s t o r e R I * /
# endif
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
/ * If T I F _ R E S T O R E A L L i s s e t , d o n ' t s c r i b b l e o n e i t h e r r3 o r c c r .
If T I F _ N O E R R O R i s s e t , j u s t s a v e r3 a s i t i s . * /
andi. r0 ,r9 ,_ T I F _ R E S T O R E A L L
2006-03-08 05:24:22 +03:00
beq+ 0 f
REST_ N V G P R S ( r1 )
b 2 f
0 : cmpld r3 ,r11 / * r10 i s - L A S T _ E R R N O * /
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
blt+ 1 f
andi. r0 ,r9 ,_ T I F _ N O E R R O R
bne- 1 f
ld r5 ,_ C C R ( r1 )
neg r3 ,r3
oris r5 ,r5 ,0 x10 0 0 / * S e t S O b i t i n C R * /
std r5 ,_ C C R ( r1 )
1 : std r3 ,G P R 3 ( r1 )
2 : andi. r0 ,r9 ,( _ T I F _ P E R S Y S C A L L _ M A S K )
beq 4 f
2006-03-08 05:24:22 +03:00
/* Clear per-syscall TIF flags if any are set. */
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
li r11 ,_ T I F _ P E R S Y S C A L L _ M A S K
addi r12 ,r12 ,T I _ F L A G S
3 : ldarx r10 ,0 ,r12
andc r10 ,r10 ,r11
stdcx. r10 ,0 ,r12
bne- 3 b
subi r12 ,r12 ,T I _ F L A G S
2006-03-08 05:24:22 +03:00
4 : /* Anything else left to do? */
2013-04-15 05:44:14 +04:00
SET_ D E F A U L T _ T H R E A D _ P P R ( r3 , r10 ) / * S e t t h r e a d . p p r = 3 * /
2006-03-08 05:24:22 +03:00
andi. r0 ,r9 ,( _ T I F _ S Y S C A L L _ T _ O R _ A | _ T I F _ S I N G L E S T E P )
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
beq . r e t _ f r o m _ e x c e p t _ l i t e
/* Re-enable interrupts */
2009-07-24 03:15:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
wrteei 1
# else
2012-03-01 08:40:23 +04:00
ld r10 ,P A C A K M S R ( r13 )
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
ori r10 ,r10 ,M S R _ E E
mtmsrd r10 ,1
2009-07-24 03:15:59 +04:00
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
2006-03-08 05:24:22 +03:00
bl . s a v e _ n v g p r s
2005-10-10 16:36:14 +04:00
addi r3 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D
bl . d o _ s y s c a l l _ t r a c e _ l e a v e
2006-03-08 05:24:22 +03:00
b . r e t _ f r o m _ e x c e p t
2005-10-10 16:36:14 +04:00
/* Save non-volatile GPRs, if not already saved. */
_ GLOBAL( s a v e _ n v g p r s )
ld r11 ,_ T R A P ( r1 )
andi. r0 ,r11 ,1
beqlr-
SAVE_ N V G P R S ( r1 )
clrrdi r0 ,r11 ,1
std r0 ,_ T R A P ( r1 )
blr
[PATCH] syscall entry/exit revamp
This cleanup patch speeds up the null syscall path on ppc64 by about 3%,
and brings the ppc32 and ppc64 code slightly closer together.
The ppc64 code was checking current_thread_info()->flags twice in the
syscall exit path; once for TIF_SYSCALL_T_OR_A before disabling
interrupts, and then again for TIF_SIGPENDING|TIF_NEED_RESCHED etc after
disabling interrupts. Now we do the same as ppc32 -- check the flags
only once in the fast path, and re-enable interrupts if necessary in the
ptrace case.
The patch abolishes the 'syscall_noerror' member of struct thread_info
and replaces it with a TIF_NOERROR bit in the flags, which is handled in
the slow path. This shortens the syscall entry code, which no longer
needs to clear syscall_noerror.
The patch adds a TIF_SAVE_NVGPRS flag which causes the syscall exit slow
path to save the non-volatile GPRs into a signal frame. This removes the
need for the assembly wrappers around sys_sigsuspend(),
sys_rt_sigsuspend(), et al which existed solely to save those registers
in advance. It also means I don't have to add new wrappers for ppoll()
and pselect(), which is what I was supposed to be doing when I got
distracted into this...
Finally, it unifies the ppc64 and ppc32 methods of handling syscall exit
directly into a signal handler (as required by sigsuspend et al) by
introducing a TIF_RESTOREALL flag which causes _all_ the registers to be
reloaded from the pt_regs by taking the ret_from_exception path, instead
of the normal syscall exit path which stomps on the callee-saved GPRs.
It appears to pass an LTP test run on ppc64, and passes basic testing on
ppc32 too. Brief tests of ptrace functionality with strace and gdb also
appear OK. I wouldn't send it to Linus for 2.6.15 just yet though :)
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2005-11-15 21:52:18 +03:00
2005-10-10 16:36:14 +04:00
/ *
* The s i g s u s p e n d a n d r t _ s i g s u s p e n d s y s t e m c a l l s c a n c a l l d o _ s i g n a l
* and t h u s p u t t h e p r o c e s s i n t o t h e s t o p p e d s t a t e w h e r e w e m i g h t
* want t o e x a m i n e i t s u s e r s t a t e w i t h p t r a c e . T h e r e f o r e w e n e e d
* to s a v e a l l t h e n o n v o l a t i l e r e g i s t e r s ( r14 - r31 ) b e f o r e c a l l i n g
* the C c o d e . S i m i l a r l y , f o r k , v f o r k a n d c l o n e n e e d t h e f u l l
* register s t a t e o n t h e s t a c k s o t h a t i t c a n b e c o p i e d t o t h e c h i l d .
* /
_ GLOBAL( p p c _ f o r k )
bl . s a v e _ n v g p r s
bl . s y s _ f o r k
b s y s c a l l _ e x i t
_ GLOBAL( p p c _ v f o r k )
bl . s a v e _ n v g p r s
bl . s y s _ v f o r k
b s y s c a l l _ e x i t
_ GLOBAL( p p c _ c l o n e )
bl . s a v e _ n v g p r s
bl . s y s _ c l o n e
b s y s c a l l _ e x i t
2006-03-08 05:24:22 +03:00
_ GLOBAL( p p c32 _ s w a p c o n t e x t )
bl . s a v e _ n v g p r s
bl . c o m p a t _ s y s _ s w a p c o n t e x t
b s y s c a l l _ e x i t
_ GLOBAL( p p c64 _ s w a p c o n t e x t )
bl . s a v e _ n v g p r s
bl . s y s _ s w a p c o n t e x t
b s y s c a l l _ e x i t
2005-10-10 16:36:14 +04:00
_ GLOBAL( r e t _ f r o m _ f o r k )
bl . s c h e d u l e _ t a i l
REST_ N V G P R S ( r1 )
li r3 ,0
b s y s c a l l _ e x i t
2012-09-13 02:32:42 +04:00
_ GLOBAL( r e t _ f r o m _ k e r n e l _ t h r e a d )
bl . s c h e d u l e _ t a i l
REST_ N V G P R S ( r1 )
2012-10-22 00:50:34 +04:00
ld r14 , 0 ( r14 )
2012-09-13 02:32:42 +04:00
mtlr r14
mr r3 ,r15
blrl
li r3 ,0
2012-08-31 23:48:05 +04:00
b s y s c a l l _ e x i t
2012-09-03 20:51:10 +04:00
.section " .toc " , " aw"
DSCR_DEFAULT :
.tc _ d e f a u l t [ T C ] ,d s c r _ d e f a u l t
.section " .text "
2005-10-10 16:36:14 +04:00
/ *
* This r o u t i n e s w i t c h e s b e t w e e n t w o d i f f e r e n t t a s k s . T h e p r o c e s s
* state o f o n e i s s a v e d o n i t s k e r n e l s t a c k . T h e n t h e s t a t e
* of t h e o t h e r i s r e s t o r e d f r o m i t s k e r n e l s t a c k . T h e m e m o r y
* management h a r d w a r e i s u p d a t e d t o t h e s e c o n d p r o c e s s ' s s t a t e .
* Finally, w e c a n r e t u r n t o t h e s e c o n d p r o c e s s , v i a r e t _ f r o m _ e x c e p t .
* On e n t r y , r3 p o i n t s t o t h e T H R E A D f o r t h e c u r r e n t t a s k , r4
* points t o t h e T H R E A D f o r t h e n e w t a s k .
*
* Note : there a r e t w o w a y s t o g e t t o t h e " g o i n g o u t " p o r t i o n
* of t h i s c o d e ; either by coming in via the entry (_switch)
* or v i a " f o r k " w h i c h m u s t s e t u p a n e n v i r o n m e n t e q u i v a l e n t
* to t h e " _ s w i t c h " p a t h . I f y o u c h a n g e t h i s y o u ' l l h a v e t o c h a n g e
* the f o r k c o d e a l s o .
*
* The c o d e w h i c h c r e a t e s t h e n e w t a s k c o n t e x t i s i n ' c o p y _ t h r e a d '
2006-01-23 19:58:20 +03:00
* in a r c h / p o w e r p c / k e r n e l / p r o c e s s . c
2005-10-10 16:36:14 +04:00
* /
.align 7
_ GLOBAL( _ s w i t c h )
mflr r0
std r0 ,1 6 ( r1 )
stdu r1 ,- S W I T C H _ F R A M E _ S I Z E ( r1 )
/* r3-r13 are caller saved -- Cort */
SAVE_ 8 G P R S ( 1 4 , r1 )
SAVE_ 1 0 G P R S ( 2 2 , r1 )
mflr r20 / * R e t u r n t o s w i t c h c a l l e r * /
mfmsr r22
li r0 , M S R _ F P
2008-06-25 08:07:18 +04:00
# ifdef C O N F I G _ V S X
BEGIN_ F T R _ S E C T I O N
oris r0 ,r0 ,M S R _ V S X @h /* Disable VSX */
END_ F T R _ S E C T I O N _ I F S E T ( C P U _ F T R _ V S X )
# endif / * C O N F I G _ V S X * /
2005-10-10 16:36:14 +04:00
# ifdef C O N F I G _ A L T I V E C
BEGIN_ F T R _ S E C T I O N
oris r0 ,r0 ,M S R _ V E C @h /* Disable altivec */
mfspr r24 ,S P R N _ V R S A V E / * s a v e v r s a v e r e g i s t e r v a l u e * /
std r24 ,T H R E A D _ V R S A V E ( r3 )
END_ F T R _ S E C T I O N _ I F S E T ( C P U _ F T R _ A L T I V E C )
# endif / * C O N F I G _ A L T I V E C * /
2011-03-02 18:18:48 +03:00
# ifdef C O N F I G _ P P C 6 4
BEGIN_ F T R _ S E C T I O N
mfspr r25 ,S P R N _ D S C R
std r25 ,T H R E A D _ D S C R ( r3 )
END_ F T R _ S E C T I O N _ I F S E T ( C P U _ F T R _ D S C R )
# endif
2005-10-10 16:36:14 +04:00
and. r0 ,r0 ,r22
beq+ 1 f
andc r22 ,r22 ,r0
2009-07-24 03:15:59 +04:00
MTMSRD( r22 )
2005-10-10 16:36:14 +04:00
isync
1 : std r20 ,_ N I P ( r1 )
mfcr r23
std r23 ,_ C C R ( r1 )
std r1 ,K S P ( r3 ) / * S e t o l d s t a c k p o i n t e r * /
2013-02-07 19:46:58 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 S _ 6 4
BEGIN_ F T R _ S E C T I O N
2013-05-01 00:17:04 +04:00
/* Event based branch registers */
mfspr r0 , S P R N _ B E S C R
std r0 , T H R E A D _ B E S C R ( r3 )
mfspr r0 , S P R N _ E B B H R
std r0 , T H R E A D _ E B B H R ( r3 )
mfspr r0 , S P R N _ E B B R R
std r0 , T H R E A D _ E B B R R ( r3 )
2013-05-01 00:17:02 +04:00
END_ F T R _ S E C T I O N _ I F S E T ( C P U _ F T R _ A R C H _ 2 0 7 S )
2013-02-07 19:46:58 +04:00
# endif
2005-10-10 16:36:14 +04:00
# ifdef C O N F I G _ S M P
/ * We n e e d a s y n c s o m e w h e r e h e r e t o m a k e s u r e t h a t i f t h e
* previous t a s k g e t s r e s c h e d u l e d o n a n o t h e r C P U , i t s e e s a l l
* stores i t h a s p e r f o r m e d o n t h i s o n e .
* /
sync
# endif / * C O N F I G _ S M P * /
2010-08-11 05:40:27 +04:00
/ *
* If w e o p t i m i s e a w a y t h e c l e a r o f t h e r e s e r v a t i o n i n s y s t e m
* calls b e c a u s e w e k n o w t h e C P U t r a c k s t h e a d d r e s s o f t h e
* reservation, t h e n w e n e e d t o c l e a r i t h e r e t o c o v e r t h e
* case t h a t t h e k e r n e l c o n t e x t s w i t c h p a t h h a s n o l a r x
* instructions.
* /
BEGIN_ F T R _ S E C T I O N
ldarx r6 ,0 ,r1
END_ F T R _ S E C T I O N _ I F S E T ( C P U _ F T R _ S T C X _ C H E C K S _ A D D R E S S )
2013-05-29 23:34:27 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 S
/ * Cancel a l l e x p l i c t u s e r s t r e a m s a s t h e y w i l l h a v e n o u s e a f t e r c o n t e x t
* switch a n d w i l l s t o p t h e H W f r o m c r e a t i n g s t r e a m s i t s e l f
* /
DCBT_ S T O P _ A L L _ S T R E A M _ I D S ( r6 )
# endif
2005-10-10 16:36:14 +04:00
addi r6 ,r4 ,- T H R E A D / * C o n v e r t T H R E A D t o ' c u r r e n t ' * /
std r6 ,P A C A C U R R E N T ( r13 ) / * S e t n e w ' c u r r e n t ' * /
ld r8 ,K S P ( r4 ) / * n e w s t a c k p o i n t e r * /
2009-07-24 03:15:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 S
2007-10-11 14:37:10 +04:00
BEGIN_ F T R _ S E C T I O N
2008-06-24 05:33:05 +04:00
BEGIN_ F T R _ S E C T I O N _ N E S T E D ( 9 5 )
2005-10-10 16:36:14 +04:00
clrrdi r6 ,r8 ,2 8 / * g e t i t s E S I D * /
clrrdi r9 ,r1 ,2 8 / * g e t c u r r e n t s p E S I D * /
2008-06-24 05:33:05 +04:00
FTR_ S E C T I O N _ E L S E _ N E S T E D ( 9 5 )
2007-10-11 14:37:10 +04:00
clrrdi r6 ,r8 ,4 0 / * g e t i t s 1 T E S I D * /
clrrdi r9 ,r1 ,4 0 / * g e t c u r r e n t s p 1 T E S I D * /
2011-04-06 23:48:50 +04:00
ALT_ M M U _ F T R _ S E C T I O N _ E N D _ N E S T E D _ I F C L R ( M M U _ F T R _ 1 T _ S E G M E N T , 9 5 )
2008-06-24 05:33:05 +04:00
FTR_ S E C T I O N _ E L S E
b 2 f
2011-04-06 23:48:50 +04:00
ALT_ M M U _ F T R _ S E C T I O N _ E N D _ I F S E T ( M M U _ F T R _ S L B )
2005-10-10 16:36:14 +04:00
clrldi. r0 ,r6 ,2 / * i s n e w E S I D c00 0 0 0 0 0 0 ? * /
cmpd c r1 ,r6 ,r9 / * o r i s n e w E S I D t h e s a m e a s c u r r e n t E S I D ? * /
cror e q ,4 * c r1 + e q ,e q
beq 2 f / * i f y e s , d o n ' t s l b i e i t * /
/* Bolt in the new stack SLB entry */
ld r7 ,K S P _ V S I D ( r4 ) / * G e t n e w s t a c k ' s V S I D * /
oris r0 ,r6 ,( S L B _ E S I D _ V ) @h
ori r0 ,r0 ,( S L B _ N U M _ B O L T E D - 1 ) @l
2007-10-11 14:37:10 +04:00
BEGIN_ F T R _ S E C T I O N
li r9 ,M M U _ S E G S I Z E _ 1 T / * i n s e r t B f i e l d * /
oris r6 ,r6 ,( M M U _ S E G S I Z E _ 1 T < < S L B I E _ S S I Z E _ S H I F T ) @h
rldimi r7 ,r9 ,S L B _ V S I D _ S S I Z E _ S H I F T ,0
2011-04-06 23:48:50 +04:00
END_ M M U _ F T R _ S E C T I O N _ I F S E T ( M M U _ F T R _ 1 T _ S E G M E N T )
2006-08-07 10:19:19 +04:00
2007-08-24 10:58:37 +04:00
/ * Update t h e l a s t b o l t e d S L B . N o w r i t e b a r r i e r s a r e n e e d e d
* here, p r o v i d e d w e o n l y u p d a t e t h e c u r r e n t C P U ' s S L B s h a d o w
* buffer.
* /
2006-08-07 10:19:19 +04:00
ld r9 ,P A C A _ S L B S H A D O W P T R ( r13 )
2006-08-09 11:00:30 +04:00
li r12 ,0
2013-08-06 20:01:46 +04:00
std r12 ,S L B S H A D O W _ S T A C K E S I D ( r9 ) / * C l e a r E S I D * /
li r12 ,S L B S H A D O W _ S T A C K V S I D
STDX_ B E r7 ,r12 ,r9 / * S a v e V S I D * /
li r12 ,S L B S H A D O W _ S T A C K E S I D
STDX_ B E r0 ,r12 ,r9 / * S a v e E S I D * /
2006-08-07 10:19:19 +04:00
2011-04-06 23:48:50 +04:00
/ * No n e e d t o c h e c k f o r M M U _ F T R _ N O _ S L B I E _ B h e r e , s i n c e w h e n
2007-10-15 18:58:59 +04:00
* we h a v e 1 T B s e g m e n t s , t h e o n l y C P U s k n o w n t o h a v e t h e e r r a t a
* only s u p p o r t l e s s t h a n 1 T B o f s y s t e m m e m o r y a n d w e ' l l n e v e r
* actually h i t t h i s c o d e p a t h .
* /
2005-10-10 16:36:14 +04:00
slbie r6
slbie r6 / * W o r k a r o u n d P O W E R 5 < D D 2 . 1 i s s u e * /
slbmte r7 ,r0
isync
2 :
2009-07-24 03:15:59 +04:00
# endif / * ! C O N F I G _ P P C _ B O O K 3 S * /
2012-07-05 08:41:35 +04:00
CURRENT_ T H R E A D _ I N F O ( r7 , r8 ) / * b a s e o f n e w s t a c k * /
2005-10-10 16:36:14 +04:00
/ * Note : this u s e s S W I T C H _ F R A M E _ S I Z E r a t h e r t h a n I N T _ F R A M E _ S I Z E
because w e d o n ' t n e e d t o l e a v e t h e 2 8 8 - b y t e A B I g a p a t t h e
top o f t h e k e r n e l s t a c k . * /
addi r7 ,r7 ,T H R E A D _ S I Z E - S W I T C H _ F R A M E _ S I Z E
mr r1 ,r8 / * s t a r t u s i n g n e w s t a c k p o i n t e r * /
std r7 ,P A C A K S A V E ( r13 )
2013-02-07 19:46:58 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 S _ 6 4
BEGIN_ F T R _ S E C T I O N
2013-05-01 00:17:04 +04:00
/* Event based branch registers */
ld r0 , T H R E A D _ B E S C R ( r4 )
mtspr S P R N _ B E S C R , r0
ld r0 , T H R E A D _ E B B H R ( r4 )
mtspr S P R N _ E B B H R , r0
ld r0 , T H R E A D _ E B B R R ( r4 )
mtspr S P R N _ E B B R R , r0
2013-02-07 19:46:58 +04:00
ld r0 ,T H R E A D _ T A R ( r4 )
mtspr S P R N _ T A R ,r0
2013-05-01 00:17:02 +04:00
END_ F T R _ S E C T I O N _ I F S E T ( C P U _ F T R _ A R C H _ 2 0 7 S )
2013-02-07 19:46:58 +04:00
# endif
2005-10-10 16:36:14 +04:00
# ifdef C O N F I G _ A L T I V E C
BEGIN_ F T R _ S E C T I O N
ld r0 ,T H R E A D _ V R S A V E ( r4 )
mtspr S P R N _ V R S A V E ,r0 / * i f G 4 , r e s t o r e V R S A V E r e g * /
END_ F T R _ S E C T I O N _ I F S E T ( C P U _ F T R _ A L T I V E C )
# endif / * C O N F I G _ A L T I V E C * /
2011-03-02 18:18:48 +03:00
# ifdef C O N F I G _ P P C 6 4
BEGIN_ F T R _ S E C T I O N
2012-09-03 20:51:10 +04:00
lwz r6 ,T H R E A D _ D S C R _ I N H E R I T ( r4 )
ld r7 ,D S C R _ D E F A U L T @toc(2)
2011-03-02 18:18:48 +03:00
ld r0 ,T H R E A D _ D S C R ( r4 )
2012-09-03 20:51:10 +04:00
cmpwi r6 ,0
bne 1 f
ld r0 ,0 ( r7 )
2013-08-09 11:29:29 +04:00
1 :
2013-08-26 07:55:57 +04:00
BEGIN_ F T R _ S E C T I O N _ N E S T E D ( 7 0 )
mfspr r8 , S P R N _ F S C R
rldimi r8 , r6 , F S C R _ D S C R _ L G , ( 6 3 - F S C R _ D S C R _ L G )
mtspr S P R N _ F S C R , r8
END_ F T R _ S E C T I O N _ N E S T E D ( C P U _ F T R _ A R C H _ 2 0 7 S , C P U _ F T R _ A R C H _ 2 0 7 S , 7 0 )
cmpd r0 ,r25
2012-09-03 20:51:10 +04:00
beq 2 f
2011-03-02 18:18:48 +03:00
mtspr S P R N _ D S C R ,r0
2012-09-03 20:51:10 +04:00
2 :
2011-03-02 18:18:48 +03:00
END_ F T R _ S E C T I O N _ I F S E T ( C P U _ F T R _ D S C R )
# endif
2005-10-10 16:36:14 +04:00
2012-09-03 20:51:10 +04:00
ld r6 ,_ C C R ( r1 )
mtcrf 0 x F F ,r6
2005-10-10 16:36:14 +04:00
/* r3-r13 are destroyed -- Cort */
REST_ 8 G P R S ( 1 4 , r1 )
REST_ 1 0 G P R S ( 2 2 , r1 )
/* convert old thread to its task_struct for return value */
addi r3 ,r3 ,- T H R E A D
ld r7 ,_ N I P ( r1 ) / * R e t u r n t o _ s w i t c h c a l l e r i n n e w t a s k * /
mtlr r7
addi r1 ,r1 ,S W I T C H _ F R A M E _ S I Z E
blr
.align 7
_ GLOBAL( r e t _ f r o m _ e x c e p t )
ld r11 ,_ T R A P ( r1 )
andi. r0 ,r11 ,1
bne . r e t _ f r o m _ e x c e p t _ l i t e
REST_ N V G P R S ( r1 )
_ GLOBAL( r e t _ f r o m _ e x c e p t _ l i t e )
/ *
* Disable i n t e r r u p t s s o t h a t c u r r e n t _ t h r e a d _ i n f o ( ) - > f l a g s
* can' t c h a n g e b e t w e e n w h e n w e t e s t i t a n d w h e n w e r e t u r n
* from t h e i n t e r r u p t .
* /
2009-07-24 03:15:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
wrteei 0
# else
2012-03-02 04:33:52 +04:00
ld r10 ,P A C A K M S R ( r13 ) / * G e t k e r n e l M S R w i t h o u t E E * /
mtmsrd r10 ,1 / * U p d a t e m a c h i n e s t a t e * /
2009-07-24 03:15:59 +04:00
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
2005-10-10 16:36:14 +04:00
2012-07-05 08:41:35 +04:00
CURRENT_ T H R E A D _ I N F O ( r9 , r1 )
2005-10-10 16:36:14 +04:00
ld r3 ,_ M S R ( r1 )
2013-05-22 08:20:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
ld r10 ,P A C A C U R R E N T ( r13 )
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
2005-10-10 16:36:14 +04:00
ld r4 ,T I _ F L A G S ( r9 )
andi. r3 ,r3 ,M S R _ P R
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
beq r e s u m e _ k e r n e l
2013-05-22 08:20:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
lwz r3 ,( T H R E A D + T H R E A D _ D B C R 0 ) ( r10 )
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
2005-10-10 16:36:14 +04:00
/* Check current_thread_info()->flags */
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
andi. r0 ,r4 ,_ T I F _ U S E R _ W O R K _ M A S K
2013-05-22 08:20:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
bne 1 f
/ *
* Check t o s e e i f t h e d b c r0 r e g i s t e r i s s e t u p t o d e b u g .
* Use t h e i n t e r n a l d e b u g m o d e b i t t o d o t h i s .
* /
andis. r0 ,r3 ,D B C R 0 _ I D M @h
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
beq r e s t o r e
2013-05-22 08:20:59 +04:00
mfmsr r0
rlwinm r0 ,r0 ,0 ,~ M S R _ D E / * C l e a r M S R . D E * /
mtmsr r0
mtspr S P R N _ D B C R 0 ,r3
li r10 , - 1
mtspr S P R N _ D B S R ,r10
b r e s t o r e
# else
beq r e s t o r e
# endif
1 : andi. r0 ,r4 ,_ T I F _ N E E D _ R E S C H E D
beq 2 f
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
bl . r e s t o r e _ i n t e r r u p t s
2013-05-13 20:16:43 +04:00
SCHEDULE_ U S E R
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
b . r e t _ f r o m _ e x c e p t _ l i t e
2013-05-22 08:20:59 +04:00
2 : bl . s a v e _ n v g p r s
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
bl . r e s t o r e _ i n t e r r u p t s
addi r3 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D
bl . d o _ n o t i f y _ r e s u m e
b . r e t _ f r o m _ e x c e p t
resume_kernel :
2012-09-17 03:54:30 +04:00
/* check current_thread_info, _TIF_EMULATE_STACK_STORE */
CURRENT_ T H R E A D _ I N F O ( r9 , r1 )
ld r8 ,T I _ F L A G S ( r9 )
andis. r8 ,r8 ,_ T I F _ E M U L A T E _ S T A C K _ S T O R E @h
beq+ 1 f
addi r8 ,r1 ,I N T _ F R A M E _ S I Z E / * G e t t h e k p r o b e d f u n c t i o n e n t r y * /
lwz r3 ,G P R 1 ( r1 )
subi r3 ,r3 ,I N T _ F R A M E _ S I Z E / * d s t : A l l o c a t e a t r a m p o l i n e e x c e p t i o n f r a m e * /
mr r4 ,r1 / * s r c : c u r r e n t e x c e p t i o n f r a m e * /
mr r1 ,r3 / * R e r o u t e t h e t r a m p o l i n e f r a m e t o r1 * /
/* Copy from the original to the trampoline. */
li r5 ,I N T _ F R A M E _ S I Z E / 8 / * s i z e : I N T _ F R A M E _ S I Z E * /
li r6 ,0 / * s t a r t o f f s e t : 0 * /
mtctr r5
2 : ldx r0 ,r6 ,r4
stdx r0 ,r6 ,r3
addi r6 ,r6 ,8
bdnz 2 b
/* Do real store operation to complete stwu */
lwz r5 ,G P R 1 ( r1 )
std r8 ,0 ( r5 )
/* Clear _TIF_EMULATE_STACK_STORE flag */
lis r11 ,_ T I F _ E M U L A T E _ S T A C K _ S T O R E @h
addi r5 ,r9 ,T I _ F L A G S
2013-04-10 02:31:24 +04:00
0 : ldarx r4 ,0 ,r5
2012-09-17 03:54:30 +04:00
andc r4 ,r4 ,r11
stdcx. r4 ,0 ,r5
bne- 0 b
1 :
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
# ifdef C O N F I G _ P R E E M P T
/* Check if we need to preempt */
andi. r0 ,r4 ,_ T I F _ N E E D _ R E S C H E D
beq+ r e s t o r e
/* Check that preempt_count() == 0 and interrupts are enabled */
lwz r8 ,T I _ P R E E M P T ( r9 )
cmpwi c r1 ,r8 ,0
ld r0 ,S O F T E ( r1 )
cmpdi r0 ,0
crandc e q ,c r1 * 4 + e q ,e q
bne r e s t o r e
/ *
* Here w e a r e p r e e m p t i n g t h e c u r r e n t t a s k . W e w a n t t o m a k e
2013-07-16 07:09:30 +04:00
* sure w e a r e s o f t - d i s a b l e d f i r s t a n d r e c o n c i l e i r q s t a t e .
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
* /
2013-07-16 07:09:30 +04:00
RECONCILE_ I R Q _ S T A T E ( r3 ,r4 )
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
1 : bl . p r e e m p t _ s c h e d u l e _ i r q
/* Re-test flags and eventually loop */
2012-07-05 08:41:35 +04:00
CURRENT_ T H R E A D _ I N F O ( r9 , r1 )
2005-10-10 16:36:14 +04:00
ld r4 ,T I _ F L A G S ( r9 )
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
andi. r0 ,r4 ,_ T I F _ N E E D _ R E S C H E D
bne 1 b
2013-01-06 04:49:34 +04:00
/ *
* arch_ l o c a l _ i r q _ r e s t o r e ( ) f r o m p r e e m p t _ s c h e d u l e _ i r q a b o v e m a y
* enable h a r d i n t e r r u p t b u t w e r e a l l y s h o u l d d i s a b l e i n t e r r u p t s
* when w e r e t u r n f r o m t h e i n t e r r u p t , a n d s o t h a t w e d o n ' t g e t
* interrupted a f t e r l o a d i n g S R R 0 / 1 .
* /
# ifdef C O N F I G _ P P C _ B O O K 3 E
wrteei 0
# else
ld r10 ,P A C A K M S R ( r13 ) / * G e t k e r n e l M S R w i t h o u t E E * /
mtmsrd r10 ,1 / * U p d a t e m a c h i n e s t a t e * /
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
ppc64: fix missing to check all bits of _TIF_USER_WORK_MASK in preempt
In entry_64.S version of ret_from_except_lite, you'll notice that
in the !preempt case, after we've checked MSR_PR we test for any
TIF flag in _TIF_USER_WORK_MASK to decide whether to go to do_work
or not. However, in the preempt case, we do a convoluted trick to
test SIGPENDING only if PR was set and always test NEED_RESCHED ...
but we forget to test any other bit of _TIF_USER_WORK_MASK !!! So
that means that with preempt, we completely fail to test for things
like single step, syscall tracing, etc...
This should be fixed as the following path:
- Test PR. If not set, go to resume_kernel, else continue.
- If go resume_kernel, to do that original do_work.
- If else, then always test for _TIF_USER_WORK_MASK to decide to do
that original user_work, else restore directly.
Signed-off-by: Tiejun Chen <tiejun.chen@windriver.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-06-07 00:56:43 +04:00
# endif / * C O N F I G _ P R E E M P T * /
2005-10-10 16:36:14 +04:00
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
.globl fast_exc_return_irq
fast_exc_return_irq :
2005-10-10 16:36:14 +04:00
restore :
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
/ *
2012-05-10 20:12:38 +04:00
* This i s t h e m a i n k e r n e l e x i t p a t h . F i r s t w e c h e c k i f w e
* are a b o u t t o r e - e n a b l e i n t e r r u p t s
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
* /
2008-07-16 08:21:34 +04:00
ld r5 ,S O F T E ( r1 )
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
lbz r6 ,P A C A S O F T I R Q E N ( r13 )
2012-05-10 20:12:38 +04:00
cmpwi c r0 ,r5 ,0
beq r e s t o r e _ i r q _ o f f
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
2012-05-10 20:12:38 +04:00
/* We are enabling, were we already enabled ? Yes, just return */
cmpwi c r0 ,r6 ,1
beq c r0 ,d o _ r e s t o r e
2005-10-10 16:36:14 +04:00
2012-05-10 20:12:38 +04:00
/ *
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
* We a r e a b o u t t o s o f t - e n a b l e i n t e r r u p t s ( w e a r e h a r d d i s a b l e d
* at t h i s p o i n t ) . W e c h e c k i f t h e r e ' s a n y t h i n g t h a t n e e d s t o
* be r e p l a y e d f i r s t .
* /
lbz r0 ,P A C A I R Q H A P P E N E D ( r13 )
cmpwi c r0 ,r0 ,0
bne- r e s t o r e _ c h e c k _ i r q _ r e p l a y
2007-02-07 05:13:26 +03:00
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
/ *
* Get h e r e w h e n n o t h i n g h a p p e n e d w h i l e s o f t - d i s a b l e d , j u s t
* soft- e n a b l e a n d m o v e - o n . W e w i l l h a r d - e n a b l e a s a s i d e
* effect o f r f i
* /
restore_no_replay :
TRACE_ E N A B L E _ I N T S
li r0 ,1
stb r0 ,P A C A S O F T I R Q E N ( r13 ) ;
/ *
* Final r e t u r n p a t h . B o o k E i s h a n d l e d i n a d i f f e r e n t f i l e
* /
2012-05-10 20:12:38 +04:00
do_restore :
2009-07-24 03:15:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
b . e x c e p t i o n _ r e t u r n _ b o o k 3 e
# else
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
/ *
* Clear t h e r e s e r v a t i o n . I f w e k n o w t h e C P U t r a c k s t h e a d d r e s s o f
* the r e s e r v a t i o n t h e n w e c a n p o t e n t i a l l y s a v e s o m e c y c l e s a n d u s e
* a l a r x . O n P O W E R 6 a n d P O W E R 7 t h i s i s s i g n i f i c a n t l y f a s t e r .
* /
BEGIN_ F T R _ S E C T I O N
stdcx. r0 ,0 ,r1 / * t o c l e a r t h e r e s e r v a t i o n * /
FTR_ S E C T I O N _ E L S E
ldarx r4 ,0 ,r1
ALT_ F T R _ S E C T I O N _ E N D _ I F C L R ( C P U _ F T R _ S T C X _ C H E C K S _ A D D R E S S )
/ *
* Some c o d e p a t h s u c h a s l o a d _ u p _ f p u o r a l t i v e c r e t u r n d i r e c t l y
* here. T h e y r u n e n t i r e l y h a r d d i s a b l e d a n d d o n o t a l t e r t h e
* interrupt s t a t e . T h e y a l s o d o n ' t u s e l w a r x / s t w c x . a n d t h u s
* are k n o w n n o t t o l e a v e d a n g l i n g r e s e r v a t i o n s .
* /
.globl fast_exception_return
fast_exception_return :
ld r3 ,_ M S R ( r1 )
2007-02-07 05:13:26 +03:00
ld r4 ,_ C T R ( r1 )
ld r0 ,_ L I N K ( r1 )
mtctr r4
mtlr r0
ld r4 ,_ X E R ( r1 )
mtspr S P R N _ X E R ,r4
REST_ 8 G P R S ( 5 , r1 )
2005-10-10 16:36:14 +04:00
andi. r0 ,r3 ,M S R _ R I
beq- u n r e c o v _ r e s t o r e
2007-02-07 05:13:26 +03:00
/ *
* Clear R I b e f o r e r e s t o r i n g r13 . I f w e a r e r e t u r n i n g t o
* userspace a n d w e t a k e a n e x c e p t i o n a f t e r r e s t o r i n g r13 ,
* we e n d u p c o r r u p t i n g t h e u s e r s p a c e r13 v a l u e .
* /
2012-03-02 04:33:52 +04:00
ld r4 ,P A C A K M S R ( r13 ) / * G e t k e r n e l M S R w i t h o u t E E * /
andc r4 ,r4 ,r0 / * r0 c o n t a i n s M S R _ R I h e r e * /
2007-02-07 05:13:26 +03:00
mtmsrd r4 ,1
2005-10-10 16:36:14 +04:00
2013-02-13 20:21:34 +04:00
# ifdef C O N F I G _ P P C _ T R A N S A C T I O N A L _ M E M
/* TM debug */
std r3 , P A C A T M S C R A T C H ( r13 ) / * S t a s h r e t u r n e d - t o M S R * /
# endif
2005-10-10 16:36:14 +04:00
/ *
* r1 3 i s o u r p e r c p u a r e a , o n l y r e s t o r e i t i f w e a r e r e t u r n i n g t o
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
* userspace t h e v a l u e s t o r e d i n t h e s t a c k f r a m e m a y b e l o n g t o
* another C P U .
2005-10-10 16:36:14 +04:00
* /
2007-02-07 05:13:26 +03:00
andi. r0 ,r3 ,M S R _ P R
2005-10-10 16:36:14 +04:00
beq 1 f
2007-02-07 05:13:26 +03:00
ACCOUNT_ C P U _ U S E R _ E X I T ( r2 , r4 )
2012-12-07 01:51:04 +04:00
RESTORE_ P P R ( r2 , r4 )
2005-10-10 16:36:14 +04:00
REST_ G P R ( 1 3 , r1 )
1 :
2007-02-07 05:13:26 +03:00
mtspr S P R N _ S R R 1 ,r3
2005-10-10 16:36:14 +04:00
ld r2 ,_ C C R ( r1 )
mtcrf 0 x F F ,r2
ld r2 ,_ N I P ( r1 )
mtspr S P R N _ S R R 0 ,r2
ld r0 ,G P R 0 ( r1 )
ld r2 ,G P R 2 ( r1 )
ld r3 ,G P R 3 ( r1 )
ld r4 ,G P R 4 ( r1 )
ld r1 ,G P R 1 ( r1 )
rfid
b . / * p r e v e n t s p e c u l a t i v e e x e c u t i o n * /
2009-07-24 03:15:59 +04:00
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
2012-05-10 20:12:38 +04:00
/ *
* We a r e r e t u r n i n g t o a c o n t e x t w i t h i n t e r r u p t s s o f t d i s a b l e d .
*
* However, w e m a y a l s o a b o u t t o h a r d e n a b l e , s o w e n e e d t o
* make s u r e t h a t i n t h i s c a s e , w e a l s o c l e a r P A C A _ I R Q _ H A R D _ D I S
* or t h a t b i t c a n g e t o u t o f s y n c a n d b a d t h i n g s w i l l h a p p e n
* /
restore_irq_off :
ld r3 ,_ M S R ( r1 )
lbz r7 ,P A C A I R Q H A P P E N E D ( r13 )
andi. r0 ,r3 ,M S R _ E E
beq 1 f
rlwinm r7 ,r7 ,0 ,~ P A C A _ I R Q _ H A R D _ D I S
stb r7 ,P A C A I R Q H A P P E N E D ( r13 )
1 : li r0 ,0
stb r0 ,P A C A S O F T I R Q E N ( r13 ) ;
TRACE_ D I S A B L E _ I N T S
b d o _ r e s t o r e
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
/ *
* Something d i d h a p p e n , c h e c k i f a r e - e m i t i s n e e d e d
* ( this a l s o c l e a r s p a c a - > i r q _ h a p p e n e d )
* /
restore_check_irq_replay :
/ * XXX : We c o u l d i m p l e m e n t a f a s t p a t h h e r e w h e r e w e c h e c k
* for i r q _ h a p p e n e d b e i n g j u s t 0 x01 , i n w h i c h c a s e w e c a n
* clear i t a n d r e t u r n . T h a t m e a n s t h a t w e w o u l d p o t e n t i a l l y
* miss a d e c r e m e n t e r h a v i n g w r a p p e d a l l t h e w a y a r o u n d .
*
* Still, t h i s m i g h t b e u s e f u l f o r t h i n g s l i k e h a s h _ p a g e
* /
bl . _ _ c h e c k _ i r q _ r e p l a y
cmpwi c r0 ,r3 ,0
beq r e s t o r e _ n o _ r e p l a y
/ *
* We n e e d t o r e - e m i t a n i n t e r r u p t . W e d o s o b y r e - u s i n g o u r
* existing e x c e p t i o n f r a m e . W e f i r s t c h a n g e t h e t r a p v a l u e ,
* but w e n e e d t o e n s u r e w e p r e s e r v e t h e l o w n i b b l e o f i t
* /
ld r4 ,_ T R A P ( r1 )
clrldi r4 ,r4 ,6 0
or r4 ,r4 ,r3
std r4 ,_ T R A P ( r1 )
/ *
* Then f i n d t h e r i g h t h a n d l e r a n d c a l l i t . I n t e r r u p t s a r e
* still s o f t - d i s a b l e d a n d w e k e e p t h e m t h a t w a y .
* /
cmpwi c r0 ,r3 ,0 x50 0
bne 1 f
addi r3 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D ;
bl . d o _ I R Q
b . r e t _ f r o m _ e x c e p t
1 : cmpwi c r0 ,r3 ,0 x90 0
bne 1 f
addi r3 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D ;
bl . t i m e r _ i n t e r r u p t
b . r e t _ f r o m _ e x c e p t
2012-11-14 22:49:48 +04:00
# ifdef C O N F I G _ P P C _ D O O R B E L L
1 :
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
2012-11-14 22:49:48 +04:00
cmpwi c r0 ,r3 ,0 x28 0
# else
BEGIN_ F T R _ S E C T I O N
cmpwi c r0 ,r3 ,0 x e 8 0
FTR_ S E C T I O N _ E L S E
cmpwi c r0 ,r3 ,0 x a00
ALT_ F T R _ S E C T I O N _ E N D _ I F S E T ( C P U _ F T R _ H V M O D E )
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
bne 1 f
addi r3 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D ;
bl . d o o r b e l l _ e x c e p t i o n
b . r e t _ f r o m _ e x c e p t
2012-11-14 22:49:48 +04:00
# endif / * C O N F I G _ P P C _ D O O R B E L L * /
powerpc: Rework lazy-interrupt handling
The current implementation of lazy interrupts handling has some
issues that this tries to address.
We don't do the various workarounds we need to do when re-enabling
interrupts in some cases such as when returning from an interrupt
and thus we may still lose or get delayed decrementer or doorbell
interrupts.
The current scheme also makes it much harder to handle the external
"edge" interrupts provided by some BookE processors when using the
EPR facility (External Proxy) and the Freescale Hypervisor.
Additionally, we tend to keep interrupts hard disabled in a number
of cases, such as decrementer interrupts, external interrupts, or
when a masked decrementer interrupt is pending. This is sub-optimal.
This is an attempt at fixing it all in one go by reworking the way
we do the lazy interrupt disabling from the ground up.
The base idea is to replace the "hard_enabled" field with a
"irq_happened" field in which we store a bit mask of what interrupt
occurred while soft-disabled.
When re-enabling, either via arch_local_irq_restore() or when returning
from an interrupt, we can now decide what to do by testing bits in that
field.
We then implement replaying of the missed interrupts either by
re-using the existing exception frame (in exception exit case) or via
the creation of a new one from an assembly trampoline (in the
arch_local_irq_enable case).
This removes the need to play with the decrementer to try to create
fake interrupts, among others.
In addition, this adds a few refinements:
- We no longer hard disable decrementer interrupts that occur
while soft-disabled. We now simply bump the decrementer back to max
(on BookS) or leave it stopped (on BookE) and continue with hard interrupts
enabled, which means that we'll potentially get better sample quality from
performance monitor interrupts.
- Timer, decrementer and doorbell interrupts now hard-enable
shortly after removing the source of the interrupt, which means
they no longer run entirely hard disabled. Again, this will improve
perf sample quality.
- On Book3E 64-bit, we now make the performance monitor interrupt
act as an NMI like Book3S (the necessary C code for that to work
appear to already be present in the FSL perf code, notably calling
nmi_enter instead of irq_enter). (This also fixes a bug where BookE
perfmon interrupts could clobber r14 ... oops)
- We could make "masked" decrementer interrupts act as NMIs when doing
timer-based perf sampling to improve the sample quality.
Signed-off-by-yet: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
v2:
- Add hard-enable to decrementer, timer and doorbells
- Fix CR clobber in masked irq handling on BookE
- Make embedded perf interrupt act as an NMI
- Add a PACA_HAPPENED_EE_EDGE for use by FSL if they want
to retrigger an interrupt without preventing hard-enable
v3:
- Fix or vs. ori bug on Book3E
- Fix enabling of interrupts for some exceptions on Book3E
v4:
- Fix resend of doorbells on return from interrupt on Book3E
v5:
- Rebased on top of my latest series, which involves some significant
rework of some aspects of the patch.
v6:
- 32-bit compile fix
- more compile fixes with various .config combos
- factor out the asm code to soft-disable interrupts
- remove the C wrapper around preempt_schedule_irq
v7:
- Fix a bug with hard irq state tracking on native power7
2012-03-06 11:27:59 +04:00
1 : b . r e t _ f r o m _ e x c e p t / * W h a t e l s e t o d o h e r e ? * /
2005-10-10 16:36:14 +04:00
unrecov_restore :
addi r3 ,r1 ,S T A C K _ F R A M E _ O V E R H E A D
bl . u n r e c o v e r a b l e _ e x c e p t i o n
b u n r e c o v _ r e s t o r e
# ifdef C O N F I G _ P P C _ R T A S
/ *
* On C H R P , t h e R u n - T i m e A b s t r a c t i o n S e r v i c e s ( R T A S ) h a v e t o b e
* called w i t h t h e M M U o f f .
*
* In a d d i t i o n , w e n e e d t o b e i n 3 2 b m o d e , a t l e a s t f o r n o w .
*
* Note : r3 i s a n i n p u t p a r a m e t e r t o r t a s , s o d o n ' t t r a s h i t . . .
* /
_ GLOBAL( e n t e r _ r t a s )
mflr r0
std r0 ,1 6 ( r1 )
stdu r1 ,- R T A S _ F R A M E _ S I Z E ( r1 ) / * S a v e S P a n d c r e a t e s t a c k s p a c e . * /
/ * Because R T A S i s r u n n i n g i n 3 2 b m o d e , i t c l o b b e r s t h e h i g h o r d e r h a l f
* of a l l r e g i s t e r s t h a t i t s a v e s . W e t h e r e f o r e s a v e t h o s e r e g i s t e r s
* RTAS m i g h t t o u c h t o t h e s t a c k . ( r0 , r3 - r13 a r e c a l l e r s a v e d )
* /
SAVE_ G P R ( 2 , r1 ) / * S a v e t h e T O C * /
SAVE_ G P R ( 1 3 , r1 ) / * S a v e p a c a * /
SAVE_ 8 G P R S ( 1 4 , r1 ) / * S a v e t h e n o n - v o l a t i l e s * /
SAVE_ 1 0 G P R S ( 2 2 , r1 ) / * d i t t o * /
mfcr r4
std r4 ,_ C C R ( r1 )
mfctr r5
std r5 ,_ C T R ( r1 )
mfspr r6 ,S P R N _ X E R
std r6 ,_ X E R ( r1 )
mfdar r7
std r7 ,_ D A R ( r1 )
mfdsisr r8
std r8 ,_ D S I S R ( r1 )
2006-03-28 03:20:00 +04:00
/ * Temporary w o r k a r o u n d t o c l e a r C R u n t i l R T A S c a n b e m o d i f i e d t o
* ignore a l l b i t s .
* /
li r0 ,0
mtcr r0
2007-01-01 21:45:34 +03:00
# ifdef C O N F I G _ B U G
2005-10-10 16:36:14 +04:00
/ * There i s n o w a y i t i s a c c e p t a b l e t o g e t h e r e w i t h i n t e r r u p t s e n a b l e d ,
* check i t w i t h t h e a s m e q u i v a l e n t o f W A R N _ O N
* /
[POWERPC] Lazy interrupt disabling for 64-bit machines
This implements a lazy strategy for disabling interrupts. This means
that local_irq_disable() et al. just clear the 'interrupts are
enabled' flag in the paca. If an interrupt comes along, the interrupt
entry code notices that interrupts are supposed to be disabled, and
clears the EE bit in SRR1, clears the 'interrupts are hard-enabled'
flag in the paca, and returns. This means that interrupts only
actually get disabled in the processor when an interrupt comes along.
When interrupts are enabled by local_irq_enable() et al., the code
sets the interrupts-enabled flag in the paca, and then checks whether
interrupts got hard-disabled. If so, it also sets the EE bit in the
MSR to hard-enable the interrupts.
This has the potential to improve performance, and also makes it
easier to make a kernel that can boot on iSeries and on other 64-bit
machines, since this lazy-disable strategy is very similar to the
soft-disable strategy that iSeries already uses.
This version renames paca->proc_enabled to paca->soft_enabled, and
changes a couple of soft-disables in the kexec code to hard-disables,
which should fix the crash that Michael Ellerman saw. This doesn't
yet use a reserved CR field for the soft_enabled and hard_enabled
flags. This applies on top of Stephen Rothwell's patches to make it
possible to build a combined iSeries/other kernel.
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-10-04 10:47:49 +04:00
lbz r0 ,P A C A S O F T I R Q E N ( r13 )
2005-10-10 16:36:14 +04:00
1 : tdnei r0 ,0
2007-01-01 21:45:34 +03:00
EMIT_ B U G _ E N T R Y 1 b ,_ _ F I L E _ _ ,_ _ L I N E _ _ ,B U G F L A G _ W A R N I N G
# endif
[POWERPC] Lazy interrupt disabling for 64-bit machines
This implements a lazy strategy for disabling interrupts. This means
that local_irq_disable() et al. just clear the 'interrupts are
enabled' flag in the paca. If an interrupt comes along, the interrupt
entry code notices that interrupts are supposed to be disabled, and
clears the EE bit in SRR1, clears the 'interrupts are hard-enabled'
flag in the paca, and returns. This means that interrupts only
actually get disabled in the processor when an interrupt comes along.
When interrupts are enabled by local_irq_enable() et al., the code
sets the interrupts-enabled flag in the paca, and then checks whether
interrupts got hard-disabled. If so, it also sets the EE bit in the
MSR to hard-enable the interrupts.
This has the potential to improve performance, and also makes it
easier to make a kernel that can boot on iSeries and on other 64-bit
machines, since this lazy-disable strategy is very similar to the
soft-disable strategy that iSeries already uses.
This version renames paca->proc_enabled to paca->soft_enabled, and
changes a couple of soft-disables in the kexec code to hard-disables,
which should fix the crash that Michael Ellerman saw. This doesn't
yet use a reserved CR field for the soft_enabled and hard_enabled
flags. This applies on top of Stephen Rothwell's patches to make it
possible to build a combined iSeries/other kernel.
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-10-04 10:47:49 +04:00
/* Hard-disable interrupts */
mfmsr r6
rldicl r7 ,r6 ,4 8 ,1
rotldi r7 ,r7 ,1 6
mtmsrd r7 ,1
2005-10-10 16:36:14 +04:00
/ * Unfortunately, t h e s t a c k p o i n t e r a n d t h e M S R a r e a l s o c l o b b e r e d ,
* so t h e y a r e s a v e d i n t h e P A C A w h i c h a l l o w s u s t o r e s t o r e
* our o r i g i n a l s t a t e a f t e r R T A S r e t u r n s .
* /
std r1 ,P A C A R 1 ( r13 )
std r6 ,P A C A S A V E D M S R ( r13 )
/* Setup our real return addr */
2006-01-13 06:56:25 +03:00
LOAD_ R E G _ A D D R ( r4 ,. r t a s _ r e t u r n _ l o c )
clrldi r4 ,r4 ,2 / * c o n v e r t t o r e a l m o d e a d d r e s s * /
2005-10-10 16:36:14 +04:00
mtlr r4
li r0 ,0
ori r0 ,r0 ,M S R _ E E | M S R _ S E | M S R _ B E | M S R _ R I
andc r0 ,r6 ,r0
li r9 ,1
rldicr r9 ,r9 ,M S R _ S F _ L G ,( 6 3 - M S R _ S F _ L G )
2013-09-23 06:04:45 +04:00
ori r9 ,r9 ,M S R _ I R | M S R _ D R | M S R _ F E 0 | M S R _ F E 1 | M S R _ F P | M S R _ R I | M S R _ L E
2005-10-10 16:36:14 +04:00
andc r6 ,r0 ,r9
sync / * d i s a b l e i n t e r r u p t s s o S R R 0 / 1 * /
mtmsrd r0 / * d o n ' t g e t t r a s h e d * /
2006-01-13 06:56:25 +03:00
LOAD_ R E G _ A D D R ( r4 , r t a s )
2005-10-10 16:36:14 +04:00
ld r5 ,R T A S E N T R Y ( r4 ) / * g e t t h e r t a s - > e n t r y v a l u e * /
ld r4 ,R T A S B A S E ( r4 ) / * g e t t h e r t a s - > b a s e v a l u e * /
mtspr S P R N _ S R R 0 ,r5
mtspr S P R N _ S R R 1 ,r6
rfid
b . / * p r e v e n t s p e c u l a t i v e e x e c u t i o n * /
_ STATIC( r t a s _ r e t u r n _ l o c )
2013-09-23 06:04:45 +04:00
FIXUP_ E N D I A N
2005-10-10 16:36:14 +04:00
/* relocation is off at this point */
2011-01-20 09:50:21 +03:00
GET_ P A C A ( r4 )
2006-01-13 06:56:25 +03:00
clrldi r4 ,r4 ,2 / * c o n v e r t t o r e a l m o d e a d d r e s s * /
2005-10-10 16:36:14 +04:00
2008-08-30 05:41:12 +04:00
bcl 2 0 ,3 1 ,$ + 4
0 : mflr r3
ld r3 ,( 1 f - 0 b ) ( r3 ) / * g e t & . r t a s _ r e s t o r e _ r e g s * /
2005-10-10 16:36:14 +04:00
mfmsr r6
li r0 ,M S R _ R I
andc r6 ,r6 ,r0
sync
mtmsrd r6
ld r1 ,P A C A R 1 ( r4 ) / * R e s t o r e o u r S P * /
ld r4 ,P A C A S A V E D M S R ( r4 ) / * R e s t o r e o u r M S R * /
mtspr S P R N _ S R R 0 ,r3
mtspr S P R N _ S R R 1 ,r4
rfid
b . / * p r e v e n t s p e c u l a t i v e e x e c u t i o n * /
2008-08-30 05:41:12 +04:00
.align 3
1 : .llong . r t a s _ r e s t o r e _ r e g s
2005-10-10 16:36:14 +04:00
_ STATIC( r t a s _ r e s t o r e _ r e g s )
/* relocation is on at this point */
REST_ G P R ( 2 , r1 ) / * R e s t o r e t h e T O C * /
REST_ G P R ( 1 3 , r1 ) / * R e s t o r e p a c a * /
REST_ 8 G P R S ( 1 4 , r1 ) / * R e s t o r e t h e n o n - v o l a t i l e s * /
REST_ 1 0 G P R S ( 2 2 , r1 ) / * d i t t o * /
2011-01-20 09:50:21 +03:00
GET_ P A C A ( r13 )
2005-10-10 16:36:14 +04:00
ld r4 ,_ C C R ( r1 )
mtcr r4
ld r5 ,_ C T R ( r1 )
mtctr r5
ld r6 ,_ X E R ( r1 )
mtspr S P R N _ X E R ,r6
ld r7 ,_ D A R ( r1 )
mtdar r7
ld r8 ,_ D S I S R ( r1 )
mtdsisr r8
addi r1 ,r1 ,R T A S _ F R A M E _ S I Z E / * U n s t a c k o u r f r a m e * /
ld r0 ,1 6 ( r1 ) / * g e t r e t u r n a d d r e s s * /
mtlr r0
blr / * r e t u r n t o c a l l e r * /
# endif / * C O N F I G _ P P C _ R T A S * /
_ GLOBAL( e n t e r _ p r o m )
mflr r0
std r0 ,1 6 ( r1 )
stdu r1 ,- P R O M _ F R A M E _ S I Z E ( r1 ) / * S a v e S P a n d c r e a t e s t a c k s p a c e * /
/ * Because P R O M i s r u n n i n g i n 3 2 b m o d e , i t c l o b b e r s t h e h i g h o r d e r h a l f
* of a l l r e g i s t e r s t h a t i t s a v e s . W e t h e r e f o r e s a v e t h o s e r e g i s t e r s
* PROM m i g h t t o u c h t o t h e s t a c k . ( r0 , r3 - r13 a r e c a l l e r s a v e d )
* /
2009-07-24 03:15:07 +04:00
SAVE_ G P R ( 2 , r1 )
2005-10-10 16:36:14 +04:00
SAVE_ G P R ( 1 3 , r1 )
SAVE_ 8 G P R S ( 1 4 , r1 )
SAVE_ 1 0 G P R S ( 2 2 , r1 )
2009-07-24 03:15:07 +04:00
mfcr r10
2005-10-10 16:36:14 +04:00
mfmsr r11
2009-07-24 03:15:07 +04:00
std r10 ,_ C C R ( r1 )
2005-10-10 16:36:14 +04:00
std r11 ,_ M S R ( r1 )
2013-09-23 06:04:45 +04:00
/* Put PROM address in SRR0 */
mtsrr0 r4
/* Setup our trampoline return addr in LR */
bcl 2 0 ,3 1 ,$ + 4
0 : mflr r4
addi r4 ,r4 ,( 1 f - 0 b )
mtlr r4
2005-10-10 16:36:14 +04:00
2013-09-23 06:04:45 +04:00
/ * Prepare a 3 2 - b i t m o d e b i g e n d i a n M S R
2005-10-10 16:36:14 +04:00
* /
2009-07-24 03:15:59 +04:00
# ifdef C O N F I G _ P P C _ B O O K 3 E
rlwinm r11 ,r11 ,0 ,1 ,3 1
2013-09-23 06:04:45 +04:00
mtsrr1 r11
rfi
2009-07-24 03:15:59 +04:00
# else / * C O N F I G _ P P C _ B O O K 3 E * /
2013-09-23 06:04:45 +04:00
LOAD_ R E G _ I M M E D I A T E ( r12 , M S R _ S F | M S R _ I S F | M S R _ L E )
andc r11 ,r11 ,r12
mtsrr1 r11
rfid
2009-07-24 03:15:59 +04:00
# endif / * C O N F I G _ P P C _ B O O K 3 E * /
2005-10-10 16:36:14 +04:00
2013-09-23 06:04:45 +04:00
1 : /* Return from OF */
FIXUP_ E N D I A N
2005-10-10 16:36:14 +04:00
/ * Just m a k e s u r e t h a t r1 t o p 3 2 b i t s d i d n ' t g e t
* corrupt b y O F
* /
rldicl r1 ,r1 ,0 ,3 2
/* Restore the MSR (back to 64 bits) */
ld r0 ,_ M S R ( r1 )
2009-07-24 03:15:07 +04:00
MTMSRD( r0 )
2005-10-10 16:36:14 +04:00
isync
/* Restore other registers */
REST_ G P R ( 2 , r1 )
REST_ G P R ( 1 3 , r1 )
REST_ 8 G P R S ( 1 4 , r1 )
REST_ 1 0 G P R S ( 2 2 , r1 )
ld r4 ,_ C C R ( r1 )
mtcr r4
addi r1 ,r1 ,P R O M _ F R A M E _ S I Z E
ld r0 ,1 6 ( r1 )
mtlr r0
blr
2008-05-15 07:49:44 +04:00
2008-10-07 03:06:12 +04:00
# ifdef C O N F I G _ F U N C T I O N _ T R A C E R
2008-05-15 07:49:44 +04:00
# ifdef C O N F I G _ D Y N A M I C _ F T R A C E
_ GLOBAL( m c o u n t )
_ GLOBAL( _ m c o u n t )
blr
_ GLOBAL( f t r a c e _ c a l l e r )
/* Taken from output of objdump from lib64/glibc */
mflr r3
ld r11 , 0 ( r1 )
stdu r1 , - 1 1 2 ( r1 )
std r3 , 1 2 8 ( r1 )
ld r4 , 1 6 ( r11 )
2008-06-21 22:17:27 +04:00
subi r3 , r3 , M C O U N T _ I N S N _ S I Z E
2008-05-15 07:49:44 +04:00
.globl ftrace_call
ftrace_call :
bl f t r a c e _ s t u b
nop
2009-02-11 09:19:54 +03:00
# ifdef C O N F I G _ F U N C T I O N _ G R A P H _ T R A C E R
.globl ftrace_graph_call
ftrace_graph_call :
b f t r a c e _ g r a p h _ s t u b
_ GLOBAL( f t r a c e _ g r a p h _ s t u b )
# endif
2008-05-15 07:49:44 +04:00
ld r0 , 1 2 8 ( r1 )
mtlr r0
addi r1 , r1 , 1 1 2
_ GLOBAL( f t r a c e _ s t u b )
blr
# else
_ GLOBAL( m c o u n t )
blr
_ GLOBAL( _ m c o u n t )
/* Taken from output of objdump from lib64/glibc */
mflr r3
ld r11 , 0 ( r1 )
stdu r1 , - 1 1 2 ( r1 )
std r3 , 1 2 8 ( r1 )
ld r4 , 1 6 ( r11 )
2008-06-21 22:17:27 +04:00
subi r3 , r3 , M C O U N T _ I N S N _ S I Z E
2008-05-15 07:49:44 +04:00
LOAD_ R E G _ A D D R ( r5 ,f t r a c e _ t r a c e _ f u n c t i o n )
ld r5 ,0 ( r5 )
ld r5 ,0 ( r5 )
mtctr r5
bctrl
nop
2009-02-10 08:10:27 +03:00
# ifdef C O N F I G _ F U N C T I O N _ G R A P H _ T R A C E R
b f t r a c e _ g r a p h _ c a l l e r
# endif
2008-05-15 07:49:44 +04:00
ld r0 , 1 2 8 ( r1 )
mtlr r0
addi r1 , r1 , 1 1 2
_ GLOBAL( f t r a c e _ s t u b )
blr
2009-02-10 08:10:27 +03:00
# endif / * C O N F I G _ D Y N A M I C _ F T R A C E * /
# ifdef C O N F I G _ F U N C T I O N _ G R A P H _ T R A C E R
2009-02-11 09:19:54 +03:00
_ GLOBAL( f t r a c e _ g r a p h _ c a l l e r )
2009-02-10 08:10:27 +03:00
/* load r4 with local address */
ld r4 , 1 2 8 ( r1 )
subi r4 , r4 , M C O U N T _ I N S N _ S I Z E
/* get the parent address */
ld r11 , 1 1 2 ( r1 )
addi r3 , r11 , 1 6
bl . p r e p a r e _ f t r a c e _ r e t u r n
nop
ld r0 , 1 2 8 ( r1 )
mtlr r0
addi r1 , r1 , 1 1 2
blr
_ GLOBAL( r e t u r n _ t o _ h a n d l e r )
2009-02-11 23:45:49 +03:00
/* need to save return values */
std r4 , - 2 4 ( r1 )
std r3 , - 1 6 ( r1 )
std r31 , - 8 ( r1 )
mr r31 , r1
stdu r1 , - 1 1 2 ( r1 )
bl . f t r a c e _ r e t u r n _ t o _ h a n d l e r
nop
/* return value has real return address */
mtlr r3
ld r1 , 0 ( r1 )
ld r4 , - 2 4 ( r1 )
ld r3 , - 1 6 ( r1 )
ld r31 , - 8 ( r1 )
/* Jump back to real return address */
blr
_ GLOBAL( m o d _ r e t u r n _ t o _ h a n d l e r )
2009-02-10 08:10:27 +03:00
/* need to save return values */
std r4 , - 3 2 ( r1 )
std r3 , - 2 4 ( r1 )
/* save TOC */
std r2 , - 1 6 ( r1 )
std r31 , - 8 ( r1 )
mr r31 , r1
stdu r1 , - 1 1 2 ( r1 )
2009-02-11 23:45:49 +03:00
/ *
* We a r e i n a m o d u l e u s i n g t h e m o d u l e ' s T O C .
* Switch t o o u r T O C t o r u n i n s i d e t h e c o r e k e r n e l .
* /
2009-09-15 19:30:14 +04:00
ld r2 , P A C A T O C ( r13 )
2009-02-10 08:10:27 +03:00
bl . f t r a c e _ r e t u r n _ t o _ h a n d l e r
nop
/* return value has real return address */
mtlr r3
ld r1 , 0 ( r1 )
ld r4 , - 3 2 ( r1 )
ld r3 , - 2 4 ( r1 )
ld r2 , - 1 6 ( r1 )
ld r31 , - 8 ( r1 )
/* Jump back to real return address */
blr
# endif / * C O N F I G _ F U N C T I O N _ G R A P H _ T R A C E R * /
# endif / * C O N F I G _ F U N C T I O N _ T R A C E R * /
