linux-stable/arch/powerpc/kernel/misc_64.S
Michael Neuling 1fc711f7ff powerpc/kexec: Fix race in kexec shutdown
In kexec_prepare_cpus, the primary CPU IPIs the secondary CPUs to
kexec_smp_down().  kexec_smp_down() calls kexec_smp_wait() which sets
the hw_cpu_id() to -1.  The primary does this while leaving IRQs on
which means the primary can take a timer interrupt which can lead to
the IPIing one of the secondary CPUs (say, for a scheduler re-balance)
but since the secondary CPU now has a hw_cpu_id = -1, we IPI CPU
-1... Kaboom!

We are hitting this case regularly on POWER7 machines.

There is also a second race, where the primary will tear down the MMU
mappings before knowing the secondaries have entered real mode.

Also, the secondaries are clearing out any pending IPIs before
guaranteeing that no more will be received.

This changes kexec_prepare_cpus() so that we turn off IRQs in the
primary CPU much earlier.  It adds a paca flag to say that the
secondaries have entered the kexec_smp_down() IPI and turned off IRQs,
rather than overloading hw_cpu_id with -1.  This new paca flag is
again used to in indicate when the secondaries has entered real mode.

It also ensures that all CPUs have their IRQs off before we clear out
any pending IPI requests (in kexec_cpu_down()) to ensure there are no
trailing IPIs left unacknowledged.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2010-05-21 17:31:11 +10:00

631 lines
13 KiB
ArmAsm

/*
* This file contains miscellaneous low-level functions.
* Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
*
* Largely rewritten by Cort Dougan (cort@cs.nmt.edu)
* and Paul Mackerras.
* Adapted for iSeries by Mike Corrigan (mikejc@us.ibm.com)
* PPC64 updates by Dave Engebretsen (engebret@us.ibm.com)
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*
*/
#include <linux/sys.h>
#include <asm/unistd.h>
#include <asm/errno.h>
#include <asm/processor.h>
#include <asm/page.h>
#include <asm/cache.h>
#include <asm/ppc_asm.h>
#include <asm/asm-offsets.h>
#include <asm/cputable.h>
#include <asm/thread_info.h>
#include <asm/kexec.h>
.text
#ifdef CONFIG_IRQSTACKS
_GLOBAL(call_do_softirq)
mflr r0
std r0,16(r1)
stdu r1,THREAD_SIZE-STACK_FRAME_OVERHEAD(r3)
mr r1,r3
bl .__do_softirq
ld r1,0(r1)
ld r0,16(r1)
mtlr r0
blr
_GLOBAL(call_handle_irq)
ld r8,0(r6)
mflr r0
std r0,16(r1)
mtctr r8
stdu r1,THREAD_SIZE-STACK_FRAME_OVERHEAD(r5)
mr r1,r5
bctrl
ld r1,0(r1)
ld r0,16(r1)
mtlr r0
blr
#endif /* CONFIG_IRQSTACKS */
.section ".toc","aw"
PPC64_CACHES:
.tc ppc64_caches[TC],ppc64_caches
.section ".text"
/*
* Write any modified data cache blocks out to memory
* and invalidate the corresponding instruction cache blocks.
*
* flush_icache_range(unsigned long start, unsigned long stop)
*
* flush all bytes from start through stop-1 inclusive
*/
_KPROBE(__flush_icache_range)
/*
* Flush the data cache to memory
*
* Different systems have different cache line sizes
* and in some cases i-cache and d-cache line sizes differ from
* each other.
*/
ld r10,PPC64_CACHES@toc(r2)
lwz r7,DCACHEL1LINESIZE(r10)/* Get cache line size */
addi r5,r7,-1
andc r6,r3,r5 /* round low to line bdy */
subf r8,r6,r4 /* compute length */
add r8,r8,r5 /* ensure we get enough */
lwz r9,DCACHEL1LOGLINESIZE(r10) /* Get log-2 of cache line size */
srw. r8,r8,r9 /* compute line count */
beqlr /* nothing to do? */
mtctr r8
1: dcbst 0,r6
add r6,r6,r7
bdnz 1b
sync
/* Now invalidate the instruction cache */
lwz r7,ICACHEL1LINESIZE(r10) /* Get Icache line size */
addi r5,r7,-1
andc r6,r3,r5 /* round low to line bdy */
subf r8,r6,r4 /* compute length */
add r8,r8,r5
lwz r9,ICACHEL1LOGLINESIZE(r10) /* Get log-2 of Icache line size */
srw. r8,r8,r9 /* compute line count */
beqlr /* nothing to do? */
mtctr r8
2: icbi 0,r6
add r6,r6,r7
bdnz 2b
isync
blr
.previous .text
/*
* Like above, but only do the D-cache.
*
* flush_dcache_range(unsigned long start, unsigned long stop)
*
* flush all bytes from start to stop-1 inclusive
*/
_GLOBAL(flush_dcache_range)
/*
* Flush the data cache to memory
*
* Different systems have different cache line sizes
*/
ld r10,PPC64_CACHES@toc(r2)
lwz r7,DCACHEL1LINESIZE(r10) /* Get dcache line size */
addi r5,r7,-1
andc r6,r3,r5 /* round low to line bdy */
subf r8,r6,r4 /* compute length */
add r8,r8,r5 /* ensure we get enough */
lwz r9,DCACHEL1LOGLINESIZE(r10) /* Get log-2 of dcache line size */
srw. r8,r8,r9 /* compute line count */
beqlr /* nothing to do? */
mtctr r8
0: dcbst 0,r6
add r6,r6,r7
bdnz 0b
sync
blr
/*
* Like above, but works on non-mapped physical addresses.
* Use only for non-LPAR setups ! It also assumes real mode
* is cacheable. Used for flushing out the DART before using
* it as uncacheable memory
*
* flush_dcache_phys_range(unsigned long start, unsigned long stop)
*
* flush all bytes from start to stop-1 inclusive
*/
_GLOBAL(flush_dcache_phys_range)
ld r10,PPC64_CACHES@toc(r2)
lwz r7,DCACHEL1LINESIZE(r10) /* Get dcache line size */
addi r5,r7,-1
andc r6,r3,r5 /* round low to line bdy */
subf r8,r6,r4 /* compute length */
add r8,r8,r5 /* ensure we get enough */
lwz r9,DCACHEL1LOGLINESIZE(r10) /* Get log-2 of dcache line size */
srw. r8,r8,r9 /* compute line count */
beqlr /* nothing to do? */
mfmsr r5 /* Disable MMU Data Relocation */
ori r0,r5,MSR_DR
xori r0,r0,MSR_DR
sync
mtmsr r0
sync
isync
mtctr r8
0: dcbst 0,r6
add r6,r6,r7
bdnz 0b
sync
isync
mtmsr r5 /* Re-enable MMU Data Relocation */
sync
isync
blr
_GLOBAL(flush_inval_dcache_range)
ld r10,PPC64_CACHES@toc(r2)
lwz r7,DCACHEL1LINESIZE(r10) /* Get dcache line size */
addi r5,r7,-1
andc r6,r3,r5 /* round low to line bdy */
subf r8,r6,r4 /* compute length */
add r8,r8,r5 /* ensure we get enough */
lwz r9,DCACHEL1LOGLINESIZE(r10)/* Get log-2 of dcache line size */
srw. r8,r8,r9 /* compute line count */
beqlr /* nothing to do? */
sync
isync
mtctr r8
0: dcbf 0,r6
add r6,r6,r7
bdnz 0b
sync
isync
blr
/*
* Flush a particular page from the data cache to RAM.
* Note: this is necessary because the instruction cache does *not*
* snoop from the data cache.
*
* void __flush_dcache_icache(void *page)
*/
_GLOBAL(__flush_dcache_icache)
/*
* Flush the data cache to memory
*
* Different systems have different cache line sizes
*/
/* Flush the dcache */
ld r7,PPC64_CACHES@toc(r2)
clrrdi r3,r3,PAGE_SHIFT /* Page align */
lwz r4,DCACHEL1LINESPERPAGE(r7) /* Get # dcache lines per page */
lwz r5,DCACHEL1LINESIZE(r7) /* Get dcache line size */
mr r6,r3
mtctr r4
0: dcbst 0,r6
add r6,r6,r5
bdnz 0b
sync
/* Now invalidate the icache */
lwz r4,ICACHEL1LINESPERPAGE(r7) /* Get # icache lines per page */
lwz r5,ICACHEL1LINESIZE(r7) /* Get icache line size */
mtctr r4
1: icbi 0,r3
add r3,r3,r5
bdnz 1b
isync
blr
#if defined(CONFIG_PPC_PMAC) || defined(CONFIG_PPC_MAPLE)
/*
* Do an IO access in real mode
*/
_GLOBAL(real_readb)
mfmsr r7
ori r0,r7,MSR_DR
xori r0,r0,MSR_DR
sync
mtmsrd r0
sync
isync
mfspr r6,SPRN_HID4
rldicl r5,r6,32,0
ori r5,r5,0x100
rldicl r5,r5,32,0
sync
mtspr SPRN_HID4,r5
isync
slbia
isync
lbz r3,0(r3)
sync
mtspr SPRN_HID4,r6
isync
slbia
isync
mtmsrd r7
sync
isync
blr
/*
* Do an IO access in real mode
*/
_GLOBAL(real_writeb)
mfmsr r7
ori r0,r7,MSR_DR
xori r0,r0,MSR_DR
sync
mtmsrd r0
sync
isync
mfspr r6,SPRN_HID4
rldicl r5,r6,32,0
ori r5,r5,0x100
rldicl r5,r5,32,0
sync
mtspr SPRN_HID4,r5
isync
slbia
isync
stb r3,0(r4)
sync
mtspr SPRN_HID4,r6
isync
slbia
isync
mtmsrd r7
sync
isync
blr
#endif /* defined(CONFIG_PPC_PMAC) || defined(CONFIG_PPC_MAPLE) */
#ifdef CONFIG_PPC_PASEMI
/* No support in all binutils for these yet, so use defines */
#define LBZCIX(RT,RA,RB) .long (0x7c0006aa|(RT<<21)|(RA<<16)|(RB << 11))
#define STBCIX(RS,RA,RB) .long (0x7c0007aa|(RS<<21)|(RA<<16)|(RB << 11))
_GLOBAL(real_205_readb)
mfmsr r7
ori r0,r7,MSR_DR
xori r0,r0,MSR_DR
sync
mtmsrd r0
sync
isync
LBZCIX(r3,0,r3)
isync
mtmsrd r7
sync
isync
blr
_GLOBAL(real_205_writeb)
mfmsr r7
ori r0,r7,MSR_DR
xori r0,r0,MSR_DR
sync
mtmsrd r0
sync
isync
STBCIX(r3,0,r4)
isync
mtmsrd r7
sync
isync
blr
#endif /* CONFIG_PPC_PASEMI */
#ifdef CONFIG_CPU_FREQ_PMAC64
/*
* SCOM access functions for 970 (FX only for now)
*
* unsigned long scom970_read(unsigned int address);
* void scom970_write(unsigned int address, unsigned long value);
*
* The address passed in is the 24 bits register address. This code
* is 970 specific and will not check the status bits, so you should
* know what you are doing.
*/
_GLOBAL(scom970_read)
/* interrupts off */
mfmsr r4
ori r0,r4,MSR_EE
xori r0,r0,MSR_EE
mtmsrd r0,1
/* rotate 24 bits SCOM address 8 bits left and mask out it's low 8 bits
* (including parity). On current CPUs they must be 0'd,
* and finally or in RW bit
*/
rlwinm r3,r3,8,0,15
ori r3,r3,0x8000
/* do the actual scom read */
sync
mtspr SPRN_SCOMC,r3
isync
mfspr r3,SPRN_SCOMD
isync
mfspr r0,SPRN_SCOMC
isync
/* XXX: fixup result on some buggy 970's (ouch ! we lost a bit, bah
* that's the best we can do). Not implemented yet as we don't use
* the scom on any of the bogus CPUs yet, but may have to be done
* ultimately
*/
/* restore interrupts */
mtmsrd r4,1
blr
_GLOBAL(scom970_write)
/* interrupts off */
mfmsr r5
ori r0,r5,MSR_EE
xori r0,r0,MSR_EE
mtmsrd r0,1
/* rotate 24 bits SCOM address 8 bits left and mask out it's low 8 bits
* (including parity). On current CPUs they must be 0'd.
*/
rlwinm r3,r3,8,0,15
sync
mtspr SPRN_SCOMD,r4 /* write data */
isync
mtspr SPRN_SCOMC,r3 /* write command */
isync
mfspr 3,SPRN_SCOMC
isync
/* restore interrupts */
mtmsrd r5,1
blr
#endif /* CONFIG_CPU_FREQ_PMAC64 */
/*
* Create a kernel thread
* kernel_thread(fn, arg, flags)
*/
_GLOBAL(kernel_thread)
std r29,-24(r1)
std r30,-16(r1)
stdu r1,-STACK_FRAME_OVERHEAD(r1)
mr r29,r3
mr r30,r4
ori r3,r5,CLONE_VM /* flags */
oris r3,r3,(CLONE_UNTRACED>>16)
li r4,0 /* new sp (unused) */
li r0,__NR_clone
sc
bns+ 1f /* did system call indicate error? */
neg r3,r3 /* if so, make return code negative */
1: cmpdi 0,r3,0 /* parent or child? */
bne 2f /* return if parent */
li r0,0
stdu r0,-STACK_FRAME_OVERHEAD(r1)
ld r2,8(r29)
ld r29,0(r29)
mtlr r29 /* fn addr in lr */
mr r3,r30 /* load arg and call fn */
blrl
li r0,__NR_exit /* exit after child exits */
li r3,0
sc
2: addi r1,r1,STACK_FRAME_OVERHEAD
ld r29,-24(r1)
ld r30,-16(r1)
blr
/*
* disable_kernel_fp()
* Disable the FPU.
*/
_GLOBAL(disable_kernel_fp)
mfmsr r3
rldicl r0,r3,(63-MSR_FP_LG),1
rldicl r3,r0,(MSR_FP_LG+1),0
mtmsrd r3 /* disable use of fpu now */
isync
blr
/* kexec_wait(phys_cpu)
*
* wait for the flag to change, indicating this kernel is going away but
* the slave code for the next one is at addresses 0 to 100.
*
* This is used by all slaves.
*
* Physical (hardware) cpu id should be in r3.
*/
_GLOBAL(kexec_wait)
bl 1f
1: mflr r5
addi r5,r5,kexec_flag-1b
li r4,KEXEC_STATE_REAL_MODE
stb r4,PACAKEXECSTATE(r13)
SYNC
99: HMT_LOW
#ifdef CONFIG_KEXEC /* use no memory without kexec */
lwz r4,0(r5)
cmpwi 0,r4,0
bnea 0x60
#endif
b 99b
/* this can be in text because we won't change it until we are
* running in real anyways
*/
kexec_flag:
.long 0
#ifdef CONFIG_KEXEC
/* kexec_smp_wait(void)
*
* call with interrupts off
* note: this is a terminal routine, it does not save lr
*
* get phys id from paca
* switch to real mode
* join other cpus in kexec_wait(phys_id)
*/
_GLOBAL(kexec_smp_wait)
lhz r3,PACAHWCPUID(r13)
bl real_mode
b .kexec_wait
/*
* switch to real mode (turn mmu off)
* we use the early kernel trick that the hardware ignores bits
* 0 and 1 (big endian) of the effective address in real mode
*
* don't overwrite r3 here, it is live for kexec_wait above.
*/
real_mode: /* assume normal blr return */
1: li r9,MSR_RI
li r10,MSR_DR|MSR_IR
mflr r11 /* return address to SRR0 */
mfmsr r12
andc r9,r12,r9
andc r10,r12,r10
mtmsrd r9,1
mtspr SPRN_SRR1,r10
mtspr SPRN_SRR0,r11
rfid
/*
* kexec_sequence(newstack, start, image, control, clear_all())
*
* does the grungy work with stack switching and real mode switches
* also does simple calls to other code
*/
_GLOBAL(kexec_sequence)
mflr r0
std r0,16(r1)
/* switch stacks to newstack -- &kexec_stack.stack */
stdu r1,THREAD_SIZE-STACK_FRAME_OVERHEAD(r3)
mr r1,r3
li r0,0
std r0,16(r1)
/* save regs for local vars on new stack.
* yes, we won't go back, but ...
*/
std r31,-8(r1)
std r30,-16(r1)
std r29,-24(r1)
std r28,-32(r1)
std r27,-40(r1)
std r26,-48(r1)
std r25,-56(r1)
stdu r1,-STACK_FRAME_OVERHEAD-64(r1)
/* save args into preserved regs */
mr r31,r3 /* newstack (both) */
mr r30,r4 /* start (real) */
mr r29,r5 /* image (virt) */
mr r28,r6 /* control, unused */
mr r27,r7 /* clear_all() fn desc */
mr r26,r8 /* spare */
lhz r25,PACAHWCPUID(r13) /* get our phys cpu from paca */
/* disable interrupts, we are overwriting kernel data next */
mfmsr r3
rlwinm r3,r3,0,17,15
mtmsrd r3,1
/* copy dest pages, flush whole dest image */
mr r3,r29
bl .kexec_copy_flush /* (image) */
/* turn off mmu */
bl real_mode
/* copy 0x100 bytes starting at start to 0 */
li r3,0
mr r4,r30 /* start, aka phys mem offset */
li r5,0x100
li r6,0
bl .copy_and_flush /* (dest, src, copy limit, start offset) */
1: /* assume normal blr return */
/* release other cpus to the new kernel secondary start at 0x60 */
mflr r5
li r6,1
stw r6,kexec_flag-1b(5)
/* clear out hardware hash page table and tlb */
ld r5,0(r27) /* deref function descriptor */
mtctr r5
bctrl /* ppc_md.hpte_clear_all(void); */
/*
* kexec image calling is:
* the first 0x100 bytes of the entry point are copied to 0
*
* all slaves branch to slave = 0x60 (absolute)
* slave(phys_cpu_id);
*
* master goes to start = entry point
* start(phys_cpu_id, start, 0);
*
*
* a wrapper is needed to call existing kernels, here is an approximate
* description of one method:
*
* v2: (2.6.10)
* start will be near the boot_block (maybe 0x100 bytes before it?)
* it will have a 0x60, which will b to boot_block, where it will wait
* and 0 will store phys into struct boot-block and load r3 from there,
* copy kernel 0-0x100 and tell slaves to back down to 0x60 again
*
* v1: (2.6.9)
* boot block will have all cpus scanning device tree to see if they
* are the boot cpu ?????
* other device tree differences (prop sizes, va vs pa, etc)...
*/
mr r3,r25 # my phys cpu
mr r4,r30 # start, aka phys mem offset
mtlr 4
li r5,0
blr /* image->start(physid, image->start, 0); */
#endif /* CONFIG_KEXEC */