2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
6 * Copyright (C) 2002, 2004, 2007 by Ralf Baechle
14 * Another R4600 erratum. Due to the lack of errata information the exact
15 * technical details aren't known. I've experimentally found that disabling
16 * interrupts during indexed I-cache flushes seems to be sufficient to deal
19 #ifndef R4600_V1_INDEX_ICACHEOP_WAR
20 #error Check setting of R4600_V1_INDEX_ICACHEOP_WAR for your platform
24 * Pleasures of the R4600 V1.x. Cite from the IDT R4600 V1.7 errata:
26 * 18. The CACHE instructions Hit_Writeback_Invalidate_D, Hit_Writeback_D,
27 * Hit_Invalidate_D and Create_Dirty_Excl_D should only be
28 * executed if there is no other dcache activity. If the dcache is
29 * accessed for another instruction immeidately preceding when these
30 * cache instructions are executing, it is possible that the dcache
31 * tag match outputs used by these cache instructions will be
32 * incorrect. These cache instructions should be preceded by at least
33 * four instructions that are not any kind of load or store
36 * This is not allowed: lw
40 * cache Hit_Writeback_Invalidate_D
47 * cache Hit_Writeback_Invalidate_D
49 #ifndef R4600_V1_HIT_CACHEOP_WAR
50 #error Check setting of R4600_V1_HIT_CACHEOP_WAR for your platform
55 * Writeback and invalidate the primary cache dcache before DMA.
57 * R4600 v2.0 bug: "The CACHE instructions Hit_Writeback_Inv_D,
58 * Hit_Writeback_D, Hit_Invalidate_D and Create_Dirty_Exclusive_D will only
59 * operate correctly if the internal data cache refill buffer is empty. These
60 * CACHE instructions should be separated from any potential data cache miss
61 * by a load instruction to an uncached address to empty the response buffer."
62 * (Revision 2.0 device errata from IDT available on http://www.idt.com/
65 #ifndef R4600_V2_HIT_CACHEOP_WAR
66 #error Check setting of R4600_V2_HIT_CACHEOP_WAR for your platform
70 * When an interrupt happens on a CP0 register read instruction, CPU may
71 * lock up or read corrupted values of CP0 registers after it enters
72 * the exception handler.
74 * This workaround makes sure that we read a "safe" CP0 register as the
75 * first thing in the exception handler, which breaks one of the
76 * pre-conditions for this problem.
78 #ifndef R5432_CP0_INTERRUPT_WAR
79 #error Check setting of R5432_CP0_INTERRUPT_WAR for your platform
83 * Workaround for the Sibyte M3 errata the text of which can be found at
85 * http://sibyte.broadcom.com/hw/bcm1250/docs/pass2errata.txt
87 * This will enable the use of a special TLB refill handler which does a
88 * consistency check on the information in c0_badvaddr and c0_entryhi and
89 * will just return and take the exception again if the information was
90 * found to be inconsistent.
92 #ifndef BCM1250_M3_WAR
93 #error Check setting of BCM1250_M3_WAR for your platform
97 * This is a DUART workaround related to glitches around register accesses
99 #ifndef SIBYTE_1956_WAR
100 #error Check setting of SIBYTE_1956_WAR for your platform
104 * Fill buffers not flushed on CACHE instructions
106 * Hit_Invalidate_I cacheops invalidate an icache line but the refill
107 * for that line can get stale data from the fill buffer instead of
108 * accessing memory if the previous icache miss was also to that line.
110 * Workaround: generate an icache refill from a different line
113 * MIPS 4K RTL revision <3.0, PRID revision <4
115 #ifndef MIPS4K_ICACHE_REFILL_WAR
116 #error Check setting of MIPS4K_ICACHE_REFILL_WAR for your platform
120 * Missing implicit forced flush of evictions caused by CACHE
123 * Evictions caused by a CACHE instructions are not forced on to the
124 * bus. The BIU gives higher priority to fetches than to the data from
125 * the eviction buffer and no collision detection is performed between
126 * fetches and pending data from the eviction buffer.
128 * Workaround: Execute a SYNC instruction after the cache instruction
131 * MIPS 5Kc,5Kf RTL revision <2.3, PRID revision <8
132 * MIPS 20Kc RTL revision <4.0, PRID revision <?
134 #ifndef MIPS_CACHE_SYNC_WAR
135 #error Check setting of MIPS_CACHE_SYNC_WAR for your platform
139 * From TX49/H2 manual: "If the instruction (i.e. CACHE) is issued for
140 * the line which this instruction itself exists, the following
141 * operation is not guaranteed."
143 * Workaround: do two phase flushing for Index_Invalidate_I
145 #ifndef TX49XX_ICACHE_INDEX_INV_WAR
146 #error Check setting of TX49XX_ICACHE_INDEX_INV_WAR for your platform
150 * On the RM9000 there is a problem which makes the CreateDirtyExclusive
151 * eache operation unusable on SMP systems.
153 #ifndef RM9000_CDEX_SMP_WAR
154 #error Check setting of RM9000_CDEX_SMP_WAR for your platform
158 * The RM7000 processors and the E9000 cores have a bug (though PMC-Sierra
159 * opposes it being called that) where invalid instructions in the same
160 * I-cache line worth of instructions being fetched may case spurious
163 #ifndef ICACHE_REFILLS_WORKAROUND_WAR
164 #error Check setting of ICACHE_REFILLS_WORKAROUND_WAR for your platform
168 * On the R10000 upto version 2.6 (not sure about 2.7) there is a bug that
169 * may cause ll / sc and lld / scd sequences to execute non-atomically.
171 #ifndef R10000_LLSC_WAR
172 #error Check setting of R10000_LLSC_WAR for your platform
176 * 34K core erratum: "Problems Executing the TLBR Instruction"
178 #ifndef MIPS34K_MISSED_ITLB_WAR
179 #error Check setting of MIPS34K_MISSED_ITLB_WAR for your platform
182 #endif /* _ASM_WAR_H */