xref: /openbmc/linux/arch/arc/mm/tlbex.S (revision 9cfc5c90) 
1/*
2 * TLB Exception Handling for ARC
3 *
4 * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
9 *
10 * Vineetg: April 2011 :
11 *  -MMU v1: moved out legacy code into a seperate file
12 *  -MMU v3: PD{0,1} bits layout changed: They don't overlap anymore,
13 *      helps avoid a shift when preparing PD0 from PTE
14 *
15 * Vineetg: July 2009
16 *  -For MMU V2, we need not do heuristics at the time of commiting a D-TLB
17 *   entry, so that it doesn't knock out it's I-TLB entry
18 *  -Some more fine tuning:
19 *   bmsk instead of add, asl.cc instead of branch, delay slot utilise etc
20 *
21 * Vineetg: July 2009
22 *  -Practically rewrote the I/D TLB Miss handlers
23 *   Now 40 and 135 instructions a peice as compared to 131 and 449 resp.
24 *   Hence Leaner by 1.5 K
25 *   Used Conditional arithmetic to replace excessive branching
26 *   Also used short instructions wherever possible
27 *
28 * Vineetg: Aug 13th 2008
29 *  -Passing ECR (Exception Cause REG) to do_page_fault( ) for printing
30 *   more information in case of a Fatality
31 *
32 * Vineetg: March 25th Bug #92690
33 *  -Added Debug Code to check if sw-ASID == hw-ASID
34
35 * Rahul Trivedi, Amit Bhor: Codito Technologies 2004
36 */
37
38#include <linux/linkage.h>
39#include <asm/entry.h>
40#include <asm/mmu.h>
41#include <asm/pgtable.h>
42#include <asm/arcregs.h>
43#include <asm/cache.h>
44#include <asm/processor.h>
45#include <asm/tlb-mmu1.h>
46
47#ifdef CONFIG_ISA_ARCOMPACT
48;-----------------------------------------------------------------
49; ARC700 Exception Handling doesn't auto-switch stack and it only provides
50; ONE scratch AUX reg "ARC_REG_SCRATCH_DATA0"
51;
52; For Non-SMP, the scratch AUX reg is repurposed to cache task PGD, so a
53; "global" is used to free-up FIRST core reg to be able to code the rest of
54; exception prologue (IRQ auto-disabled on Exceptions, so it's IRQ-safe).
55; Since the Fast Path TLB Miss handler is coded with 4 regs, the remaining 3
56; need to be saved as well by extending the "global" to be 4 words. Hence
57;	".size   ex_saved_reg1, 16"
58; [All of this dance is to avoid stack switching for each TLB Miss, since we
59; only need to save only a handful of regs, as opposed to complete reg file]
60;
61; For ARC700 SMP, the "global" obviously can't be used for free up the FIRST
62; core reg as it will not be SMP safe.
63; Thus scratch AUX reg is used (and no longer used to cache task PGD).
64; To save the rest of 3 regs - per cpu, the global is made "per-cpu".
65; Epilogue thus has to locate the "per-cpu" storage for regs.
66; To avoid cache line bouncing the per-cpu global is aligned/sized per
67; L1_CACHE_SHIFT, despite fundamentally needing to be 12 bytes only. Hence
68;	".size   ex_saved_reg1, (CONFIG_NR_CPUS << L1_CACHE_SHIFT)"
69
70; As simple as that....
71;--------------------------------------------------------------------------
72
73; scratch memory to save [r0-r3] used to code TLB refill Handler
74ARCFP_DATA ex_saved_reg1
75	.align 1 << L1_CACHE_SHIFT
76	.type   ex_saved_reg1, @object
77#ifdef CONFIG_SMP
78	.size   ex_saved_reg1, (CONFIG_NR_CPUS << L1_CACHE_SHIFT)
79ex_saved_reg1:
80	.zero (CONFIG_NR_CPUS << L1_CACHE_SHIFT)
81#else
82	.size   ex_saved_reg1, 16
83ex_saved_reg1:
84	.zero 16
85#endif
86
87.macro TLBMISS_FREEUP_REGS
88#ifdef CONFIG_SMP
89	sr  r0, [ARC_REG_SCRATCH_DATA0]	; freeup r0 to code with
90	GET_CPU_ID  r0			; get to per cpu scratch mem,
91	asl r0, r0, L1_CACHE_SHIFT	; cache line wide per cpu
92	add r0, @ex_saved_reg1, r0
93#else
94	st    r0, [@ex_saved_reg1]
95	mov_s r0, @ex_saved_reg1
96#endif
97	st_s  r1, [r0, 4]
98	st_s  r2, [r0, 8]
99	st_s  r3, [r0, 12]
100
101	; VERIFY if the ASID in MMU-PID Reg is same as
102	; one in Linux data structures
103
104	tlb_paranoid_check_asm
105.endm
106
107.macro TLBMISS_RESTORE_REGS
108#ifdef CONFIG_SMP
109	GET_CPU_ID  r0			; get to per cpu scratch mem
110	asl r0, r0, L1_CACHE_SHIFT	; each is cache line wide
111	add r0, @ex_saved_reg1, r0
112	ld_s  r3, [r0,12]
113	ld_s  r2, [r0, 8]
114	ld_s  r1, [r0, 4]
115	lr    r0, [ARC_REG_SCRATCH_DATA0]
116#else
117	mov_s r0, @ex_saved_reg1
118	ld_s  r3, [r0,12]
119	ld_s  r2, [r0, 8]
120	ld_s  r1, [r0, 4]
121	ld_s  r0, [r0]
122#endif
123.endm
124
125#else	/* ARCv2 */
126
127.macro TLBMISS_FREEUP_REGS
128	PUSH  r0
129	PUSH  r1
130	PUSH  r2
131	PUSH  r3
132.endm
133
134.macro TLBMISS_RESTORE_REGS
135	POP   r3
136	POP   r2
137	POP   r1
138	POP   r0
139.endm
140
141#endif
142
143;============================================================================
144;  Troubleshooting Stuff
145;============================================================================
146
147; Linux keeps ASID (Address Space ID) in task->active_mm->context.asid
148; When Creating TLB Entries, instead of doing 3 dependent loads from memory,
149; we use the MMU PID Reg to get current ASID.
150; In bizzare scenrios SW and HW ASID can get out-of-sync which is trouble.
151; So we try to detect this in TLB Mis shandler
152
153.macro tlb_paranoid_check_asm
154
155#ifdef CONFIG_ARC_DBG_TLB_PARANOIA
156
157	GET_CURR_TASK_ON_CPU  r3
158	ld r0, [r3, TASK_ACT_MM]
159	ld r0, [r0, MM_CTXT+MM_CTXT_ASID]
160	breq r0, 0, 55f	; Error if no ASID allocated
161
162	lr r1, [ARC_REG_PID]
163	and r1, r1, 0xFF
164
165	and r2, r0, 0xFF	; MMU PID bits only for comparison
166	breq r1, r2, 5f
167
16855:
169	; Error if H/w and S/w ASID don't match, but NOT if in kernel mode
170	lr  r2, [erstatus]
171	bbit0 r2, STATUS_U_BIT, 5f
172
173	; We sure are in troubled waters, Flag the error, but to do so
174	; need to switch to kernel mode stack to call error routine
175	GET_TSK_STACK_BASE   r3, sp
176
177	; Call printk to shoutout aloud
178	mov r2, 1
179	j print_asid_mismatch
180
1815:	; ASIDs match so proceed normally
182	nop
183
184#endif
185
186.endm
187
188;============================================================================
189;TLB Miss handling Code
190;============================================================================
191
192;-----------------------------------------------------------------------------
193; This macro does the page-table lookup for the faulting address.
194; OUT: r0 = PTE faulted on, r1 = ptr to PTE, r2 = Faulting V-address
195.macro LOAD_FAULT_PTE
196
197	lr  r2, [efa]
198
199#ifndef CONFIG_SMP
200	lr  r1, [ARC_REG_SCRATCH_DATA0] ; current pgd
201#else
202	GET_CURR_TASK_ON_CPU  r1
203	ld  r1, [r1, TASK_ACT_MM]
204	ld  r1, [r1, MM_PGD]
205#endif
206
207	lsr     r0, r2, PGDIR_SHIFT     ; Bits for indexing into PGD
208	ld.as   r3, [r1, r0]            ; PGD entry corresp to faulting addr
209	tst	r3, r3
210	bz	do_slow_path_pf         ; if no Page Table, do page fault
211
212#ifdef CONFIG_TRANSPARENT_HUGEPAGE
213	and.f	0, r3, _PAGE_HW_SZ	; Is this Huge PMD (thp)
214	add2.nz	r1, r1, r0
215	bnz.d	2f		; YES: PGD == PMD has THP PTE: stop pgd walk
216	mov.nz	r0, r3
217
218#endif
219	and	r1, r3, PAGE_MASK
220
221	; Get the PTE entry: The idea is
222	; (1) x = addr >> PAGE_SHIFT 	-> masks page-off bits from @fault-addr
223	; (2) y = x & (PTRS_PER_PTE - 1) -> to get index
224	; (3) z = (pgtbl + y * 4)
225
226#ifdef CONFIG_ARC_HAS_PAE40
227#define PTE_SIZE_LOG	3	/* 8 == 2 ^ 3 */
228#else
229#define PTE_SIZE_LOG	2	/* 4 == 2 ^ 2 */
230#endif
231
232	; multiply in step (3) above avoided by shifting lesser in step (1)
233	lsr     r0, r2, ( PAGE_SHIFT - PTE_SIZE_LOG )
234	and     r0, r0, ( (PTRS_PER_PTE - 1) << PTE_SIZE_LOG )
235	ld.aw   r0, [r1, r0]            ; r0: PTE (lower word only for PAE40)
236					; r1: PTE ptr
237
2382:
239
240#ifdef CONFIG_ARC_DBG_TLB_MISS_COUNT
241	and.f 0, r0, _PAGE_PRESENT
242	bz   1f
243	ld   r3, [num_pte_not_present]
244	add  r3, r3, 1
245	st   r3, [num_pte_not_present]
2461:
247#endif
248
249.endm
250
251;-----------------------------------------------------------------
252; Convert Linux PTE entry into TLB entry
253; A one-word PTE entry is programmed as two-word TLB Entry [PD0:PD1] in mmu
254;    (for PAE40, two-words PTE, while three-word TLB Entry [PD0:PD1:PD1HI])
255; IN: r0 = PTE, r1 = ptr to PTE
256
257.macro CONV_PTE_TO_TLB
258	and    r3, r0, PTE_BITS_RWX	;          r  w  x
259	asl    r2, r3, 3		; Kr Kw Kx 0  0  0 (GLOBAL, kernel only)
260	and.f  0,  r0, _PAGE_GLOBAL
261	or.z   r2, r2, r3		; Kr Kw Kx Ur Uw Ux (!GLOBAL, user page)
262
263	and r3, r0, PTE_BITS_NON_RWX_IN_PD1 ; Extract PFN+cache bits from PTE
264	or  r3, r3, r2
265
266	sr  r3, [ARC_REG_TLBPD1]    	; paddr[31..13] | Kr Kw Kx Ur Uw Ux | C
267#ifdef	CONFIG_ARC_HAS_PAE40
268	ld	r3, [r1, 4]		; paddr[39..32]
269	sr	r3, [ARC_REG_TLBPD1HI]
270#endif
271
272	and r2, r0, PTE_BITS_IN_PD0 ; Extract other PTE flags: (V)alid, (G)lb
273
274	lr  r3,[ARC_REG_TLBPD0]     ; MMU prepares PD0 with vaddr and asid
275
276	or  r3, r3, r2              ; S | vaddr | {sasid|asid}
277	sr  r3,[ARC_REG_TLBPD0]     ; rewrite PD0
278.endm
279
280;-----------------------------------------------------------------
281; Commit the TLB entry into MMU
282
283.macro COMMIT_ENTRY_TO_MMU
284#if (CONFIG_ARC_MMU_VER < 4)
285
286	/* Get free TLB slot: Set = computed from vaddr, way = random */
287	sr  TLBGetIndex, [ARC_REG_TLBCOMMAND]
288
289	/* Commit the Write */
290#if (CONFIG_ARC_MMU_VER >= 2)   /* introduced in v2 */
291	sr TLBWriteNI, [ARC_REG_TLBCOMMAND]
292#else
293	sr TLBWrite, [ARC_REG_TLBCOMMAND]
294#endif
295
296#else
297	sr TLBInsertEntry, [ARC_REG_TLBCOMMAND]
298#endif
299.endm
300
301
302ARCFP_CODE	;Fast Path Code, candidate for ICCM
303
304;-----------------------------------------------------------------------------
305; I-TLB Miss Exception Handler
306;-----------------------------------------------------------------------------
307
308ENTRY(EV_TLBMissI)
309
310	TLBMISS_FREEUP_REGS
311
312#ifdef CONFIG_ARC_DBG_TLB_MISS_COUNT
313	ld  r0, [@numitlb]
314	add r0, r0, 1
315	st  r0, [@numitlb]
316#endif
317
318	;----------------------------------------------------------------
319	; Get the PTE corresponding to V-addr accessed, r2 is setup with EFA
320	LOAD_FAULT_PTE
321
322	;----------------------------------------------------------------
323	; VERIFY_PTE: Check if PTE permissions approp for executing code
324	cmp_s   r2, VMALLOC_START
325	mov_s   r2, (_PAGE_PRESENT | _PAGE_EXECUTE)
326	or.hs   r2, r2, _PAGE_GLOBAL
327
328	and     r3, r0, r2  ; Mask out NON Flag bits from PTE
329	xor.f   r3, r3, r2  ; check ( ( pte & flags_test ) == flags_test )
330	bnz     do_slow_path_pf
331
332	; Let Linux VM know that the page was accessed
333	or      r0, r0, _PAGE_ACCESSED  ; set Accessed Bit
334	st_s    r0, [r1]                ; Write back PTE
335
336	CONV_PTE_TO_TLB
337	COMMIT_ENTRY_TO_MMU
338	TLBMISS_RESTORE_REGS
339EV_TLBMissI_fast_ret:	; additional label for VDK OS-kit instrumentation
340	rtie
341
342END(EV_TLBMissI)
343
344;-----------------------------------------------------------------------------
345; D-TLB Miss Exception Handler
346;-----------------------------------------------------------------------------
347
348ENTRY(EV_TLBMissD)
349
350	TLBMISS_FREEUP_REGS
351
352#ifdef CONFIG_ARC_DBG_TLB_MISS_COUNT
353	ld  r0, [@numdtlb]
354	add r0, r0, 1
355	st  r0, [@numdtlb]
356#endif
357
358	;----------------------------------------------------------------
359	; Get the PTE corresponding to V-addr accessed
360	; If PTE exists, it will setup, r0 = PTE, r1 = Ptr to PTE, r2 = EFA
361	LOAD_FAULT_PTE
362
363	;----------------------------------------------------------------
364	; VERIFY_PTE: Chk if PTE permissions approp for data access (R/W/R+W)
365
366	cmp_s	r2, VMALLOC_START
367	mov_s   r2, _PAGE_PRESENT	; common bit for K/U PTE
368	or.hs	r2, r2, _PAGE_GLOBAL	; kernel PTE only
369
370	; Linux PTE [RWX] bits are semantically overloaded:
371	; -If PAGE_GLOBAL set, they refer to kernel-only flags (vmalloc)
372	; -Otherwise they are user-mode permissions, and those are exactly
373	;  same for kernel mode as well (e.g. copy_(to|from)_user)
374
375	lr      r3, [ecr]
376	btst_s  r3, ECR_C_BIT_DTLB_LD_MISS	; Read Access
377	or.nz   r2, r2, _PAGE_READ      	; chk for Read flag in PTE
378	btst_s  r3, ECR_C_BIT_DTLB_ST_MISS	; Write Access
379	or.nz   r2, r2, _PAGE_WRITE     	; chk for Write flag in PTE
380	; Above laddering takes care of XCHG access (both R and W)
381
382	; By now, r2 setup with all the Flags we need to check in PTE
383	and     r3, r0, r2              ; Mask out NON Flag bits from PTE
384	brne.d  r3, r2, do_slow_path_pf ; is ((pte & flags_test) == flags_test)
385
386	;----------------------------------------------------------------
387	; UPDATE_PTE: Let Linux VM know that page was accessed/dirty
388	lr      r3, [ecr]
389	or      r0, r0, _PAGE_ACCESSED        ; Accessed bit always
390	btst_s  r3,  ECR_C_BIT_DTLB_ST_MISS   ; See if it was a Write Access ?
391	or.nz   r0, r0, _PAGE_DIRTY           ; if Write, set Dirty bit as well
392	st_s    r0, [r1]                      ; Write back PTE
393
394	CONV_PTE_TO_TLB
395
396#if (CONFIG_ARC_MMU_VER == 1)
397	; MMU with 2 way set assoc J-TLB, needs some help in pathetic case of
398	; memcpy where 3 parties contend for 2 ways, ensuing a livelock.
399	; But only for old MMU or one with Metal Fix
400	TLB_WRITE_HEURISTICS
401#endif
402
403	COMMIT_ENTRY_TO_MMU
404	TLBMISS_RESTORE_REGS
405EV_TLBMissD_fast_ret:	; additional label for VDK OS-kit instrumentation
406	rtie
407
408;-------- Common routine to call Linux Page Fault Handler -----------
409do_slow_path_pf:
410
411	; Restore the 4-scratch regs saved by fast path miss handler
412	TLBMISS_RESTORE_REGS
413
414	; Slow path TLB Miss handled as a regular ARC Exception
415	; (stack switching / save the complete reg-file).
416	b  call_do_page_fault
417END(EV_TLBMissD)
418