kernel_optimize_test/arch/sparc64/kernel/tsb.S
David S. Miller 314ef68597 [SPARC64]: Refine register window trap handling.
When saving and restoing trap state, do the window spill/fill
handling inline so that we never trap deeper than 2 trap levels.
This is important for chips like Niagara.

The window fixup code is massively simplified, and many more
improvements are now possible.

Signed-off-by: David S. Miller <davem@davemloft.net>
2006-03-20 01:11:36 -08:00

214 lines
4.5 KiB
ArmAsm

/* tsb.S: Sparc64 TSB table handling.
*
* Copyright (C) 2006 David S. Miller <davem@davemloft.net>
*/
#include <asm/tsb.h>
.text
.align 32
/* Invoked from TLB miss handler, we are in the
* MMU global registers and they are setup like
* this:
*
* %g1: TSB entry pointer
* %g2: available temporary
* %g3: FAULT_CODE_{D,I}TLB
* %g4: available temporary
* %g5: available temporary
* %g6: TAG TARGET
* %g7: physical address base of the linux page
* tables for the current address space
*/
.globl tsb_miss_dtlb
tsb_miss_dtlb:
mov TLB_TAG_ACCESS, %g4
ldxa [%g4] ASI_DMMU, %g4
ba,pt %xcc, tsb_miss_page_table_walk
nop
.globl tsb_miss_itlb
tsb_miss_itlb:
mov TLB_TAG_ACCESS, %g4
ldxa [%g4] ASI_IMMU, %g4
ba,pt %xcc, tsb_miss_page_table_walk
nop
tsb_miss_page_table_walk:
TRAP_LOAD_PGD_PHYS(%g7, %g5)
USER_PGTABLE_WALK_TL1(%g4, %g7, %g5, %g2, tsb_do_fault)
tsb_reload:
TSB_LOCK_TAG(%g1, %g2, %g4)
/* Load and check PTE. */
ldxa [%g5] ASI_PHYS_USE_EC, %g5
brgez,a,pn %g5, tsb_do_fault
TSB_STORE(%g1, %g0)
/* If it is larger than the base page size, don't
* bother putting it into the TSB.
*/
srlx %g5, 32, %g2
sethi %hi(_PAGE_ALL_SZ_BITS >> 32), %g4
sethi %hi(_PAGE_SZBITS >> 32), %g7
and %g2, %g4, %g2
cmp %g2, %g7
bne,a,pn %xcc, tsb_tlb_reload
TSB_STORE(%g1, %g0)
TSB_WRITE(%g1, %g5, %g6)
/* Finally, load TLB and return from trap. */
tsb_tlb_reload:
cmp %g3, FAULT_CODE_DTLB
bne,pn %xcc, tsb_itlb_load
nop
tsb_dtlb_load:
stxa %g5, [%g0] ASI_DTLB_DATA_IN
retry
tsb_itlb_load:
stxa %g5, [%g0] ASI_ITLB_DATA_IN
retry
/* No valid entry in the page tables, do full fault
* processing.
*/
.globl tsb_do_fault
tsb_do_fault:
cmp %g3, FAULT_CODE_DTLB
rdpr %pstate, %g5
bne,pn %xcc, tsb_do_itlb_fault
wrpr %g5, PSTATE_AG | PSTATE_MG, %pstate
tsb_do_dtlb_fault:
rdpr %tl, %g4
cmp %g4, 1
mov TLB_TAG_ACCESS, %g4
ldxa [%g4] ASI_DMMU, %g5
be,pt %xcc, sparc64_realfault_common
mov FAULT_CODE_DTLB, %g4
ba,pt %xcc, winfix_trampoline
nop
tsb_do_itlb_fault:
rdpr %tpc, %g5
ba,pt %xcc, sparc64_realfault_common
mov FAULT_CODE_ITLB, %g4
.globl sparc64_realfault_common
sparc64_realfault_common:
/* fault code in %g4, fault address in %g5, etrap will
* preserve these two values in %l4 and %l5 respectively
*/
ba,pt %xcc, etrap ! Save trap state
1: rd %pc, %g7 ! ...
stb %l4, [%g6 + TI_FAULT_CODE] ! Save fault code
stx %l5, [%g6 + TI_FAULT_ADDR] ! Save fault address
call do_sparc64_fault ! Call fault handler
add %sp, PTREGS_OFF, %o0 ! Compute pt_regs arg
ba,pt %xcc, rtrap_clr_l6 ! Restore cpu state
nop ! Delay slot (fill me)
winfix_trampoline:
rdpr %tpc, %g3 ! Prepare winfixup TNPC
or %g3, 0x7c, %g3 ! Compute branch offset
wrpr %g3, %tnpc ! Write it into TNPC
done ! Trap return
/* Insert an entry into the TSB.
*
* %o0: TSB entry pointer (virt or phys address)
* %o1: tag
* %o2: pte
*/
.align 32
.globl __tsb_insert
__tsb_insert:
rdpr %pstate, %o5
wrpr %o5, PSTATE_IE, %pstate
TSB_LOCK_TAG(%o0, %g2, %g3)
TSB_WRITE(%o0, %o2, %o1)
wrpr %o5, %pstate
retl
nop
/* Flush the given TSB entry if it has the matching
* tag.
*
* %o0: TSB entry pointer (virt or phys address)
* %o1: tag
*/
.align 32
.globl tsb_flush
tsb_flush:
sethi %hi(TSB_TAG_LOCK_HIGH), %g2
1: TSB_LOAD_TAG(%o0, %g1)
srlx %g1, 32, %o3
andcc %o3, %g2, %g0
bne,pn %icc, 1b
membar #LoadLoad
cmp %g1, %o1
bne,pt %xcc, 2f
clr %o3
TSB_CAS_TAG(%o0, %g1, %o3)
cmp %g1, %o3
bne,pn %xcc, 1b
nop
2: retl
TSB_MEMBAR
/* Reload MMU related context switch state at
* schedule() time.
*
* %o0: page table physical address
* %o1: TSB register value
* %o2: TSB virtual address
* %o3: TSB mapping locked PTE
*
* We have to run this whole thing with interrupts
* disabled so that the current cpu doesn't change
* due to preemption.
*/
.align 32
.globl __tsb_context_switch
__tsb_context_switch:
rdpr %pstate, %o5
wrpr %o5, PSTATE_IE, %pstate
ldub [%g6 + TI_CPU], %g1
sethi %hi(trap_block), %g2
sllx %g1, TRAP_BLOCK_SZ_SHIFT, %g1
or %g2, %lo(trap_block), %g2
add %g2, %g1, %g2
stx %o0, [%g2 + TRAP_PER_CPU_PGD_PADDR]
mov TSB_REG, %g1
stxa %o1, [%g1] ASI_DMMU
membar #Sync
stxa %o1, [%g1] ASI_IMMU
membar #Sync
brz %o2, 9f
nop
sethi %hi(sparc64_highest_unlocked_tlb_ent), %o4
mov TLB_TAG_ACCESS, %g1
lduw [%o4 + %lo(sparc64_highest_unlocked_tlb_ent)], %g2
stxa %o2, [%g1] ASI_DMMU
membar #Sync
sllx %g2, 3, %g2
stxa %o3, [%g2] ASI_DTLB_DATA_ACCESS
membar #Sync
9:
wrpr %o5, %pstate
retl
nop