Skip to content

L
linux
Commit ed3a88d7 authored by Stafford Horne's avatar Stafford Horne
Browse files
Options

openrisc: Remove unused IMMU tlb workardound 

This looks to be some historical code that was used to convert TLB
misses on branches from l.bf, l.jal, l.j etc all to a trampoline
using l.jr (jump register).  I don't see this being used and I don't
know the history of it so remove it.
Signed-off-by: default avatarStafford Horne <shorne@gmail.com>
parent d8fee3f6
Hide whitespace changes
Inline Side-by-side
Showing with 0 additions and 209 deletions
arch/openrisc/kernel/head.S
View file @ ed3a88d7
......@@ -1325,215 +1325,6 @@ i_pte_not_present:
/* =================================================[ debugging aids ]=== */
.align 64
_immu_trampoline:
.space 64
_immu_trampoline_top:
#define TRAMP_SLOT_0 (0x0)
#define TRAMP_SLOT_1 (0x4)
#define TRAMP_SLOT_2 (0x8)
#define TRAMP_SLOT_3 (0xc)
#define TRAMP_SLOT_4 (0x10)
#define TRAMP_SLOT_5 (0x14)
#define TRAMP_FRAME_SIZE (0x18)
ENTRY(_immu_trampoline_workaround)
// r2 EEA
// r6 is physical EEA
tophys(r6,r2)
LOAD_SYMBOL_2_GPR(r5,_immu_trampoline)
tophys (r3,r5) // r3 is trampoline (physical)
LOAD_SYMBOL_2_GPR(r4,0x15000000)
l.sw TRAMP_SLOT_0(r3),r4
l.sw TRAMP_SLOT_1(r3),r4
l.sw TRAMP_SLOT_4(r3),r4
l.sw TRAMP_SLOT_5(r3),r4
// EPC = EEA - 0x4
l.lwz r4,0x0(r6) // load op @ EEA + 0x0 (fc address)
l.sw TRAMP_SLOT_3(r3),r4 // store it to _immu_trampoline_data
l.lwz r4,-0x4(r6) // load op @ EEA - 0x4 (f8 address)
l.sw TRAMP_SLOT_2(r3),r4 // store it to _immu_trampoline_data
l.srli r5,r4,26 // check opcode for write access
l.sfeqi r5,0 // l.j
l.bf 0f
l.sfeqi r5,0x11 // l.jr
l.bf 1f
l.sfeqi r5,1 // l.jal
l.bf 2f
l.sfeqi r5,0x12 // l.jalr
l.bf 3f
l.sfeqi r5,3 // l.bnf
l.bf 4f
l.sfeqi r5,4 // l.bf
l.bf 5f
99:
l.nop
l.j 99b // should never happen
l.nop 1
// r2 is EEA
// r3 is trampoline address (physical)
// r4 is instruction
// r6 is physical(EEA)
//
// r5
2: // l.jal
/* 19 20 aa aa l.movhi r9,0xaaaa
* a9 29 bb bb l.ori r9,0xbbbb
*
* where 0xaaaabbbb is EEA + 0x4 shifted right 2
*/
l.addi r6,r2,0x4 // this is 0xaaaabbbb
// l.movhi r9,0xaaaa
l.ori r5,r0,0x1920 // 0x1920 == l.movhi r9
l.sh (TRAMP_SLOT_0+0x0)(r3),r5
l.srli r5,r6,16
l.sh (TRAMP_SLOT_0+0x2)(r3),r5
// l.ori r9,0xbbbb
l.ori r5,r0,0xa929 // 0xa929 == l.ori r9
l.sh (TRAMP_SLOT_1+0x0)(r3),r5
l.andi r5,r6,0xffff
l.sh (TRAMP_SLOT_1+0x2)(r3),r5
/* falthrough, need to set up new jump offset */
0: // l.j
l.slli r6,r4,6 // original offset shifted left 6 - 2
// l.srli r6,r6,6 // original offset shifted right 2
l.slli r4,r2,4 // old jump position: EEA shifted left 4
// l.srli r4,r4,6 // old jump position: shifted right 2
l.addi r5,r3,0xc // new jump position (physical)
l.slli r5,r5,4 // new jump position: shifted left 4
// calculate new jump offset
// new_off = old_off + (old_jump - new_jump)
l.sub r5,r4,r5 // old_jump - new_jump
l.add r5,r6,r5 // orig_off + (old_jump - new_jump)
l.srli r5,r5,6 // new offset shifted right 2
// r5 is new jump offset
// l.j has opcode 0x0...
l.sw TRAMP_SLOT_2(r3),r5 // write it back
l.j trampoline_out
l.nop
/* ----------------------------- */
3: // l.jalr
/* 19 20 aa aa l.movhi r9,0xaaaa
* a9 29 bb bb l.ori r9,0xbbbb
*
* where 0xaaaabbbb is EEA + 0x4 shifted right 2
*/
l.addi r6,r2,0x4 // this is 0xaaaabbbb
// l.movhi r9,0xaaaa
l.ori r5,r0,0x1920 // 0x1920 == l.movhi r9
l.sh (TRAMP_SLOT_0+0x0)(r3),r5
l.srli r5,r6,16
l.sh (TRAMP_SLOT_0+0x2)(r3),r5
// l.ori r9,0xbbbb
l.ori r5,r0,0xa929 // 0xa929 == l.ori r9
l.sh (TRAMP_SLOT_1+0x0)(r3),r5
l.andi r5,r6,0xffff
l.sh (TRAMP_SLOT_1+0x2)(r3),r5
l.lhz r5,(TRAMP_SLOT_2+0x0)(r3) // load hi part of jump instruction
l.andi r5,r5,0x3ff // clear out opcode part
l.ori r5,r5,0x4400 // opcode changed from l.jalr -> l.jr
l.sh (TRAMP_SLOT_2+0x0)(r3),r5 // write it back
/* falthrough */
1: // l.jr
l.j trampoline_out
l.nop
/* ----------------------------- */
4: // l.bnf
5: // l.bf
l.slli r6,r4,6 // original offset shifted left 6 - 2
// l.srli r6,r6,6 // original offset shifted right 2
l.slli r4,r2,4 // old jump position: EEA shifted left 4
// l.srli r4,r4,6 // old jump position: shifted right 2
l.addi r5,r3,0xc // new jump position (physical)
l.slli r5,r5,4 // new jump position: shifted left 4
// calculate new jump offset
// new_off = old_off + (old_jump - new_jump)
l.add r6,r6,r4 // (orig_off + old_jump)
l.sub r6,r6,r5 // (orig_off + old_jump) - new_jump
l.srli r6,r6,6 // new offset shifted right 2
// r6 is new jump offset
l.lwz r4,(TRAMP_SLOT_2+0x0)(r3) // load jump instruction
l.srli r4,r4,16
l.andi r4,r4,0xfc00 // get opcode part
l.slli r4,r4,16
l.or r6,r4,r6 // l.b(n)f new offset
l.sw TRAMP_SLOT_2(r3),r6 // write it back
/* we need to add l.j to EEA + 0x8 */
tophys (r4,r2) // may not be needed (due to shifts down_
l.addi r4,r4,(0x8 - 0x8) // jump target = r2 + 0x8 (compensate for 0x8)
// jump position = r5 + 0x8 (0x8 compensated)
l.sub r4,r4,r5 // jump offset = target - new_position + 0x8
l.slli r4,r4,4 // the amount of info in imediate of jump
l.srli r4,r4,6 // jump instruction with offset
l.sw TRAMP_SLOT_4(r3),r4 // write it to 4th slot
/* fallthrough */
trampoline_out:
// set up new EPC to point to our trampoline code
LOAD_SYMBOL_2_GPR(r5,_immu_trampoline)
l.mtspr r0,r5,SPR_EPCR_BASE
// immu_trampoline is (4x) CACHE_LINE aligned
// and only 6 instructions long,
// so we need to invalidate only 2 lines
/* Establish cache block size
If BS=0, 16;
If BS=1, 32;
r14 contain block size
*/
l.mfspr r21,r0,SPR_ICCFGR
l.andi r21,r21,SPR_ICCFGR_CBS
l.srli r21,r21,7
l.ori r23,r0,16
l.sll r14,r23,r21
l.mtspr r0,r5,SPR_ICBIR
l.add r5,r5,r14
l.mtspr r0,r5,SPR_ICBIR
l.jr r9
l.nop
/*
* DESC: Prints ASCII character stored in r7
*
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7