kernel_optimize_test/arch/tile/mm/migrate_64.S
Chris Metcalf 51007004f4 arch/tile: use interrupt critical sections less
In general we want to avoid ever touching memory while within an
interrupt critical section, since the page fault path goes through
a different path from the hypervisor when in an interrupt critical
section, and we carefully decided with tilegx that we didn't need
to support this path in the kernel.  (On tilepro we did implement
that path as part of supporting atomic instructions in software.)

In practice we always need to touch the kernel stack, since that's
where we store the interrupt state before releasing the critical
section, but this change cleans up a few things.  The IRQ_ENABLE
macro is split up so that when we want to enable interrupts in a
deferred way (e.g. for cpu_idle or for interrupt return) we can
read the per-cpu enable mask before entering the critical section.
The cache-migration code is changed to use interrupt masking instead
of interrupt critical sections.  And, the interrupt-entry code is
changed so that we defer loading "tp" from per-cpu data until after
we have released the interrupt critical section.

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
2012-05-25 12:48:20 -04:00

168 lines
3.5 KiB
ArmAsm

/*
* Copyright 2011 Tilera Corporation. All Rights Reserved.
*
* 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, version 2.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
* NON INFRINGEMENT. See the GNU General Public License for
* more details.
*
* This routine is a helper for migrating the home of a set of pages to
* a new cpu. See the documentation in homecache.c for more information.
*/
#include <linux/linkage.h>
#include <linux/threads.h>
#include <asm/page.h>
#include <asm/thread_info.h>
#include <asm/types.h>
#include <asm/asm-offsets.h>
#include <hv/hypervisor.h>
.text
/*
* First, some definitions that apply to all the code in the file.
*/
/* Locals (caller-save) */
#define r_tmp r10
#define r_save_sp r11
/* What we save where in the stack frame; must include all callee-saves. */
#define FRAME_SP 8
#define FRAME_R30 16
#define FRAME_R31 24
#define FRAME_R32 32
#define FRAME_SIZE 40
/*
* On entry:
*
* r0 the new context PA to install (moved to r_context)
* r1 PTE to use for context access (moved to r_access)
* r2 ASID to use for new context (moved to r_asid)
* r3 pointer to cpumask with just this cpu set in it (r_my_cpumask)
*/
/* Arguments (caller-save) */
#define r_context_in r0
#define r_access_in r1
#define r_asid_in r2
#define r_my_cpumask r3
/* Locals (callee-save); must not be more than FRAME_xxx above. */
#define r_context r30
#define r_access r31
#define r_asid r32
/*
* Caller-save locals and frame constants are the same as
* for homecache_migrate_stack_and_flush.
*/
STD_ENTRY(flush_and_install_context)
/*
* Create a stack frame; we can't touch it once we flush the
* cache until we install the new page table and flush the TLB.
*/
{
move r_save_sp, sp
st sp, lr
addi sp, sp, -FRAME_SIZE
}
addi r_tmp, sp, FRAME_SP
{
st r_tmp, r_save_sp
addi r_tmp, sp, FRAME_R30
}
{
st r_tmp, r30
addi r_tmp, sp, FRAME_R31
}
{
st r_tmp, r31
addi r_tmp, sp, FRAME_R32
}
st r_tmp, r32
/* Move some arguments to callee-save registers. */
{
move r_context, r_context_in
move r_access, r_access_in
}
move r_asid, r_asid_in
/* First, flush our L2 cache. */
{
move r0, zero /* cache_pa */
moveli r1, hw2_last(HV_FLUSH_EVICT_L2) /* cache_control */
}
{
shl16insli r1, r1, hw1(HV_FLUSH_EVICT_L2)
move r2, r_my_cpumask /* cache_cpumask */
}
{
shl16insli r1, r1, hw0(HV_FLUSH_EVICT_L2)
move r3, zero /* tlb_va */
}
{
move r4, zero /* tlb_length */
move r5, zero /* tlb_pgsize */
}
{
move r6, zero /* tlb_cpumask */
move r7, zero /* asids */
}
{
move r8, zero /* asidcount */
jal hv_flush_remote
}
bnez r0, 1f
/* Now install the new page table. */
{
move r0, r_context
move r1, r_access
}
{
move r2, r_asid
moveli r3, HV_CTX_DIRECTIO | CTX_PAGE_FLAG
}
jal hv_install_context
bnez r0, 1f
/* Finally, flush the TLB. */
{
movei r0, 0 /* preserve_global */
jal hv_flush_all
}
1: /* Restore the callee-saved registers and return. */
addli lr, sp, FRAME_SIZE
{
ld lr, lr
addli r_tmp, sp, FRAME_R30
}
{
ld r30, r_tmp
addli r_tmp, sp, FRAME_R31
}
{
ld r31, r_tmp
addli r_tmp, sp, FRAME_R32
}
{
ld r32, r_tmp
addi sp, sp, FRAME_SIZE
}
jrp lr
STD_ENDPROC(flush_and_install_context)