xref: /openbmc/linux/arch/arc/mm/tlbex.S (revision de528723) 
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * TLB Exception Handling for ARC
4 *
5 * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
6 *
7 * Vineetg: April 2011 :
8 *  -MMU v1: moved out legacy code into a seperate file
9 *  -MMU v3: PD{0,1} bits layout changed: They don't overlap anymore,
10 *      helps avoid a shift when preparing PD0 from PTE
11 *
12 * Vineetg: July 2009
13 *  -For MMU V2, we need not do heuristics at the time of commiting a D-TLB
14 *   entry, so that it doesn't knock out it's I-TLB entry
15 *  -Some more fine tuning:
16 *   bmsk instead of add, asl.cc instead of branch, delay slot utilise etc
17 *
18 * Vineetg: July 2009
19 *  -Practically rewrote the I/D TLB Miss handlers
20 *   Now 40 and 135 instructions a peice as compared to 131 and 449 resp.
21 *   Hence Leaner by 1.5 K
22 *   Used Conditional arithmetic to replace excessive branching
23 *   Also used short instructions wherever possible
24 *
25 * Vineetg: Aug 13th 2008
26 *  -Passing ECR (Exception Cause REG) to do_page_fault( ) for printing
27 *   more information in case of a Fatality
28 *
29 * Vineetg: March 25th Bug #92690
30 *  -Added Debug Code to check if sw-ASID == hw-ASID
31
32 * Rahul Trivedi, Amit Bhor: Codito Technologies 2004
33 */
34
35#include <linux/linkage.h>
36#include <asm/entry.h>
37#include <asm/mmu.h>
38#include <asm/pgtable.h>
39#include <asm/arcregs.h>
40#include <asm/cache.h>
41#include <asm/processor.h>
42#include <asm/tlb-mmu1.h>
43
44#ifdef CONFIG_ISA_ARCOMPACT
45;-----------------------------------------------------------------
46; ARC700 Exception Handling doesn't auto-switch stack and it only provides
47; ONE scratch AUX reg "ARC_REG_SCRATCH_DATA0"
48;
49; For Non-SMP, the scratch AUX reg is repurposed to cache task PGD, so a
50; "global" is used to free-up FIRST core reg to be able to code the rest of
51; exception prologue (IRQ auto-disabled on Exceptions, so it's IRQ-safe).
52; Since the Fast Path TLB Miss handler is coded with 4 regs, the remaining 3
53; need to be saved as well by extending the "global" to be 4 words. Hence
54;	".size   ex_saved_reg1, 16"
55; [All of this dance is to avoid stack switching for each TLB Miss, since we
56; only need to save only a handful of regs, as opposed to complete reg file]
57;
58; For ARC700 SMP, the "global" obviously can't be used for free up the FIRST
59; core reg as it will not be SMP safe.
60; Thus scratch AUX reg is used (and no longer used to cache task PGD).
61; To save the rest of 3 regs - per cpu, the global is made "per-cpu".
62; Epilogue thus has to locate the "per-cpu" storage for regs.
63; To avoid cache line bouncing the per-cpu global is aligned/sized per
64; L1_CACHE_SHIFT, despite fundamentally needing to be 12 bytes only. Hence
65;	".size   ex_saved_reg1, (CONFIG_NR_CPUS << L1_CACHE_SHIFT)"
66
67; As simple as that....
68;--------------------------------------------------------------------------
69
70; scratch memory to save [r0-r3] used to code TLB refill Handler
71ARCFP_DATA ex_saved_reg1
72	.align 1 << L1_CACHE_SHIFT
73	.type   ex_saved_reg1, @object
74#ifdef CONFIG_SMP
75	.size   ex_saved_reg1, (CONFIG_NR_CPUS << L1_CACHE_SHIFT)
76ex_saved_reg1:
77	.zero (CONFIG_NR_CPUS << L1_CACHE_SHIFT)
78#else
79	.size   ex_saved_reg1, 16
80ex_saved_reg1:
81	.zero 16
82#endif
83
84.macro TLBMISS_FREEUP_REGS
85#ifdef CONFIG_SMP
86	sr  r0, [ARC_REG_SCRATCH_DATA0]	; freeup r0 to code with
87	GET_CPU_ID  r0			; get to per cpu scratch mem,
88	asl r0, r0, L1_CACHE_SHIFT	; cache line wide per cpu
89	add r0, @ex_saved_reg1, r0
90#else
91	st    r0, [@ex_saved_reg1]
92	mov_s r0, @ex_saved_reg1
93#endif
94	st_s  r1, [r0, 4]
95	st_s  r2, [r0, 8]
96	st_s  r3, [r0, 12]
97
98	; VERIFY if the ASID in MMU-PID Reg is same as
99	; one in Linux data structures
100
101	tlb_paranoid_check_asm
102.endm
103
104.macro TLBMISS_RESTORE_REGS
105#ifdef CONFIG_SMP
106	GET_CPU_ID  r0			; get to per cpu scratch mem
107	asl r0, r0, L1_CACHE_SHIFT	; each is cache line wide
108	add r0, @ex_saved_reg1, r0
109	ld_s  r3, [r0,12]
110	ld_s  r2, [r0, 8]
111	ld_s  r1, [r0, 4]
112	lr    r0, [ARC_REG_SCRATCH_DATA0]
113#else
114	mov_s r0, @ex_saved_reg1
115	ld_s  r3, [r0,12]
116	ld_s  r2, [r0, 8]
117	ld_s  r1, [r0, 4]
118	ld_s  r0, [r0]
119#endif
120.endm
121
122#else	/* ARCv2 */
123
124.macro TLBMISS_FREEUP_REGS
125	PUSH  r0
126	PUSH  r1
127	PUSH  r2
128	PUSH  r3
129.endm
130
131.macro TLBMISS_RESTORE_REGS
132	POP   r3
133	POP   r2
134	POP   r1
135	POP   r0
136.endm
137
138#endif
139
140;============================================================================
141;  Troubleshooting Stuff
142;============================================================================
143
144; Linux keeps ASID (Address Space ID) in task->active_mm->context.asid
145; When Creating TLB Entries, instead of doing 3 dependent loads from memory,
146; we use the MMU PID Reg to get current ASID.
147; In bizzare scenrios SW and HW ASID can get out-of-sync which is trouble.
148; So we try to detect this in TLB Mis shandler
149
150.macro tlb_paranoid_check_asm
151
152#ifdef CONFIG_ARC_DBG_TLB_PARANOIA
153
154	GET_CURR_TASK_ON_CPU  r3
155	ld r0, [r3, TASK_ACT_MM]
156	ld r0, [r0, MM_CTXT+MM_CTXT_ASID]
157	breq r0, 0, 55f	; Error if no ASID allocated
158
159	lr r1, [ARC_REG_PID]
160	and r1, r1, 0xFF
161
162	and r2, r0, 0xFF	; MMU PID bits only for comparison
163	breq r1, r2, 5f
164
16555:
166	; Error if H/w and S/w ASID don't match, but NOT if in kernel mode
167	lr  r2, [erstatus]
168	bbit0 r2, STATUS_U_BIT, 5f
169
170	; We sure are in troubled waters, Flag the error, but to do so
171	; need to switch to kernel mode stack to call error routine
172	GET_TSK_STACK_BASE   r3, sp
173
174	; Call printk to shoutout aloud
175	mov r2, 1
176	j print_asid_mismatch
177
1785:	; ASIDs match so proceed normally
179	nop
180
181#endif
182
183.endm
184
185;============================================================================
186;TLB Miss handling Code
187;============================================================================
188
189;-----------------------------------------------------------------------------
190; This macro does the page-table lookup for the faulting address.
191; OUT: r0 = PTE faulted on, r1 = ptr to PTE, r2 = Faulting V-address
192.macro LOAD_FAULT_PTE
193
194	lr  r2, [efa]
195
196#ifndef CONFIG_SMP
197	lr  r1, [ARC_REG_SCRATCH_DATA0] ; current pgd
198#else
199	GET_CURR_TASK_ON_CPU  r1
200	ld  r1, [r1, TASK_ACT_MM]
201	ld  r1, [r1, MM_PGD]
202#endif
203
204	lsr     r0, r2, PGDIR_SHIFT     ; Bits for indexing into PGD
205	ld.as   r3, [r1, r0]            ; PGD entry corresp to faulting addr
206	tst	r3, r3
207	bz	do_slow_path_pf         ; if no Page Table, do page fault
208
209#ifdef CONFIG_TRANSPARENT_HUGEPAGE
210	and.f	0, r3, _PAGE_HW_SZ	; Is this Huge PMD (thp)
211	add2.nz	r1, r1, r0
212	bnz.d	2f		; YES: PGD == PMD has THP PTE: stop pgd walk
213	mov.nz	r0, r3
214
215#endif
216	and	r1, r3, PAGE_MASK
217
218	; Get the PTE entry: The idea is
219	; (1) x = addr >> PAGE_SHIFT 	-> masks page-off bits from @fault-addr
220	; (2) y = x & (PTRS_PER_PTE - 1) -> to get index
221	; (3) z = (pgtbl + y * 4)
222
223#ifdef CONFIG_ARC_HAS_PAE40
224#define PTE_SIZE_LOG	3	/* 8 == 2 ^ 3 */
225#else
226#define PTE_SIZE_LOG	2	/* 4 == 2 ^ 2 */
227#endif
228
229	; multiply in step (3) above avoided by shifting lesser in step (1)
230	lsr     r0, r2, ( PAGE_SHIFT - PTE_SIZE_LOG )
231	and     r0, r0, ( (PTRS_PER_PTE - 1) << PTE_SIZE_LOG )
232	ld.aw   r0, [r1, r0]            ; r0: PTE (lower word only for PAE40)
233					; r1: PTE ptr
234
2352:
236
237.endm
238
239;-----------------------------------------------------------------
240; Convert Linux PTE entry into TLB entry
241; A one-word PTE entry is programmed as two-word TLB Entry [PD0:PD1] in mmu
242;    (for PAE40, two-words PTE, while three-word TLB Entry [PD0:PD1:PD1HI])
243; IN: r0 = PTE, r1 = ptr to PTE
244
245.macro CONV_PTE_TO_TLB
246	and    r3, r0, PTE_BITS_RWX	;          r  w  x
247	asl    r2, r3, 3		; Kr Kw Kx 0  0  0 (GLOBAL, kernel only)
248	and.f  0,  r0, _PAGE_GLOBAL
249	or.z   r2, r2, r3		; Kr Kw Kx Ur Uw Ux (!GLOBAL, user page)
250
251	and r3, r0, PTE_BITS_NON_RWX_IN_PD1 ; Extract PFN+cache bits from PTE
252	or  r3, r3, r2
253
254	sr  r3, [ARC_REG_TLBPD1]    	; paddr[31..13] | Kr Kw Kx Ur Uw Ux | C
255#ifdef	CONFIG_ARC_HAS_PAE40
256	ld	r3, [r1, 4]		; paddr[39..32]
257	sr	r3, [ARC_REG_TLBPD1HI]
258#endif
259
260	and r2, r0, PTE_BITS_IN_PD0 ; Extract other PTE flags: (V)alid, (G)lb
261
262	lr  r3,[ARC_REG_TLBPD0]     ; MMU prepares PD0 with vaddr and asid
263
264	or  r3, r3, r2              ; S | vaddr | {sasid|asid}
265	sr  r3,[ARC_REG_TLBPD0]     ; rewrite PD0
266.endm
267
268;-----------------------------------------------------------------
269; Commit the TLB entry into MMU
270
271.macro COMMIT_ENTRY_TO_MMU
272#if (CONFIG_ARC_MMU_VER < 4)
273
274#ifdef CONFIG_EZNPS_MTM_EXT
275	/* verify if entry for this vaddr+ASID already exists */
276	sr    TLBProbe, [ARC_REG_TLBCOMMAND]
277	lr    r0, [ARC_REG_TLBINDEX]
278	bbit0 r0, 31, 88f
279#endif
280
281	/* Get free TLB slot: Set = computed from vaddr, way = random */
282	sr  TLBGetIndex, [ARC_REG_TLBCOMMAND]
283
284	/* Commit the Write */
285#if (CONFIG_ARC_MMU_VER >= 2)   /* introduced in v2 */
286	sr TLBWriteNI, [ARC_REG_TLBCOMMAND]
287#else
288	sr TLBWrite, [ARC_REG_TLBCOMMAND]
289#endif
290
291#else
292	sr TLBInsertEntry, [ARC_REG_TLBCOMMAND]
293#endif
294
29588:
296.endm
297
298
299ARCFP_CODE	;Fast Path Code, candidate for ICCM
300
301;-----------------------------------------------------------------------------
302; I-TLB Miss Exception Handler
303;-----------------------------------------------------------------------------
304
305ENTRY(EV_TLBMissI)
306
307	TLBMISS_FREEUP_REGS
308
309	;----------------------------------------------------------------
310	; Get the PTE corresponding to V-addr accessed, r2 is setup with EFA
311	LOAD_FAULT_PTE
312
313	;----------------------------------------------------------------
314	; VERIFY_PTE: Check if PTE permissions approp for executing code
315	cmp_s   r2, VMALLOC_START
316	mov_s   r2, (_PAGE_PRESENT | _PAGE_EXECUTE)
317	or.hs   r2, r2, _PAGE_GLOBAL
318
319	and     r3, r0, r2  ; Mask out NON Flag bits from PTE
320	xor.f   r3, r3, r2  ; check ( ( pte & flags_test ) == flags_test )
321	bnz     do_slow_path_pf
322
323	; Let Linux VM know that the page was accessed
324	or      r0, r0, _PAGE_ACCESSED  ; set Accessed Bit
325	st_s    r0, [r1]                ; Write back PTE
326
327	CONV_PTE_TO_TLB
328	COMMIT_ENTRY_TO_MMU
329	TLBMISS_RESTORE_REGS
330EV_TLBMissI_fast_ret:	; additional label for VDK OS-kit instrumentation
331	rtie
332
333END(EV_TLBMissI)
334
335;-----------------------------------------------------------------------------
336; D-TLB Miss Exception Handler
337;-----------------------------------------------------------------------------
338
339ENTRY(EV_TLBMissD)
340
341	TLBMISS_FREEUP_REGS
342
343	;----------------------------------------------------------------
344	; Get the PTE corresponding to V-addr accessed
345	; If PTE exists, it will setup, r0 = PTE, r1 = Ptr to PTE, r2 = EFA
346	LOAD_FAULT_PTE
347
348	;----------------------------------------------------------------
349	; VERIFY_PTE: Chk if PTE permissions approp for data access (R/W/R+W)
350
351	cmp_s	r2, VMALLOC_START
352	mov_s   r2, _PAGE_PRESENT	; common bit for K/U PTE
353	or.hs	r2, r2, _PAGE_GLOBAL	; kernel PTE only
354
355	; Linux PTE [RWX] bits are semantically overloaded:
356	; -If PAGE_GLOBAL set, they refer to kernel-only flags (vmalloc)
357	; -Otherwise they are user-mode permissions, and those are exactly
358	;  same for kernel mode as well (e.g. copy_(to|from)_user)
359
360	lr      r3, [ecr]
361	btst_s  r3, ECR_C_BIT_DTLB_LD_MISS	; Read Access
362	or.nz   r2, r2, _PAGE_READ      	; chk for Read flag in PTE
363	btst_s  r3, ECR_C_BIT_DTLB_ST_MISS	; Write Access
364	or.nz   r2, r2, _PAGE_WRITE     	; chk for Write flag in PTE
365	; Above laddering takes care of XCHG access (both R and W)
366
367	; By now, r2 setup with all the Flags we need to check in PTE
368	and     r3, r0, r2              ; Mask out NON Flag bits from PTE
369	brne.d  r3, r2, do_slow_path_pf ; is ((pte & flags_test) == flags_test)
370
371	;----------------------------------------------------------------
372	; UPDATE_PTE: Let Linux VM know that page was accessed/dirty
373	lr      r3, [ecr]
374	or      r0, r0, _PAGE_ACCESSED        ; Accessed bit always
375	btst_s  r3,  ECR_C_BIT_DTLB_ST_MISS   ; See if it was a Write Access ?
376	or.nz   r0, r0, _PAGE_DIRTY           ; if Write, set Dirty bit as well
377	st_s    r0, [r1]                      ; Write back PTE
378
379	CONV_PTE_TO_TLB
380
381#if (CONFIG_ARC_MMU_VER == 1)
382	; MMU with 2 way set assoc J-TLB, needs some help in pathetic case of
383	; memcpy where 3 parties contend for 2 ways, ensuing a livelock.
384	; But only for old MMU or one with Metal Fix
385	TLB_WRITE_HEURISTICS
386#endif
387
388	COMMIT_ENTRY_TO_MMU
389	TLBMISS_RESTORE_REGS
390EV_TLBMissD_fast_ret:	; additional label for VDK OS-kit instrumentation
391	rtie
392
393;-------- Common routine to call Linux Page Fault Handler -----------
394do_slow_path_pf:
395
396#ifdef CONFIG_ISA_ARCV2
397	; Set Z flag if exception in U mode. Hardware micro-ops do this on any
398	; taken interrupt/exception, and thus is already the case at the entry
399	; above, but ensuing code would have already clobbered.
400	; EXCEPTION_PROLOGUE called in slow path, relies on correct Z flag set
401
402	lr	r2, [erstatus]
403	and	r2, r2, STATUS_U_MASK
404	bxor.f	0, r2, STATUS_U_BIT
405#endif
406
407	; Restore the 4-scratch regs saved by fast path miss handler
408	TLBMISS_RESTORE_REGS
409
410	; Slow path TLB Miss handled as a regular ARC Exception
411	; (stack switching / save the complete reg-file).
412	b  call_do_page_fault
413END(EV_TLBMissD)
414