From 460c2ce362e56890c2a029e2c3b1ff2796c7fc54 Mon Sep 17 00:00:00 2001 From: Wolfgang Denk Date: Mon, 21 Jun 2010 22:29:59 +0200 Subject: MPC5200: workaround data corruption for unaligned local bus accesses The MPC5200 has a nasty problem that will cause silent data corruption when performing unaligned 16 or 32 byte accesses when reading from the local bus - typically this affects reading from flash. The problem can be easily shown: => md fc0c0000 10 fc0c0000: 323e4337 01626f6f 74636d64 3d72756e 2>C7.bootcmd=run fc0c0010: 206e6574 5f6e6673 00626f6f 7464656c net_nfs.bootdel fc0c0020: 61793d35 00626175 64726174 653d3131 ay=5.baudrate=11 fc0c0030: 35323030 00707265 626f6f74 3d656368 5200.preboot=ech => md fc0c0001 10 fc0c0001: 65636801 00000074 0000003d 00000020 ech....t...=... fc0c0011: 0000005f 00000000 00000074 00000061 ..._.......t...a fc0c0021: 00000000 00000064 00000065 00000035 .......d...e...5 fc0c0031: 00000000 00000062 0000003d 0000006f .......b...=...o => md.w fc0c0001 10 fc0c0001: 0000 3701 0000 6f74 0000 643d 0000 6e20 ..7...ot..d=..n fc0c0011: 0000 745f 0000 7300 0000 6f74 0000 6c61 ..t_..s...ot..la This commit implements a workaround at least for the most blatant problem: using memcpy() from NOR flash. We rename the assembler routine into __memcpy() and provide a wrapper, which will use a byte-wise copy loop for unaligned source or target addresses when reading from NOR flash, and branch to the optimized __memcpy() in all other cases, thus minimizing the performance impact. Tested on lite5200b and TQM5200S. Signed-off-by: Wolfgang Denk Cc: Detlev Zundel --- arch/powerpc/cpu/mpc5xxx/Makefile | 5 +++ arch/powerpc/cpu/mpc5xxx/memcpy_mpc5200.c | 71 +++++++++++++++++++++++++++++++ 2 files changed, 76 insertions(+) create mode 100644 arch/powerpc/cpu/mpc5xxx/memcpy_mpc5200.c (limited to 'arch/powerpc/cpu/mpc5xxx') diff --git a/arch/powerpc/cpu/mpc5xxx/Makefile b/arch/powerpc/cpu/mpc5xxx/Makefile index 0ee0611550..4ab2b7be76 100644 --- a/arch/powerpc/cpu/mpc5xxx/Makefile +++ b/arch/powerpc/cpu/mpc5xxx/Makefile @@ -30,6 +30,11 @@ SOBJS = io.o firmware_sc_task_bestcomm.impl.o COBJS = i2c.o traps.o cpu.o cpu_init.o ide.o interrupts.o \ loadtask.o pci_mpc5200.o serial.o speed.o usb_ohci.o usb.o +# Workaround for local bus unaligned access problem on MPC5200 +#ifdef CONFIG_MPC5200 +COBJS += memcpy_mpc5200.o +#endif + SRCS := $(START:.o=.S) $(SOBJS:.o=.S) $(COBJS:.o=.c) OBJS := $(addprefix $(obj),$(SOBJS) $(COBJS)) START := $(addprefix $(obj),$(START)) diff --git a/arch/powerpc/cpu/mpc5xxx/memcpy_mpc5200.c b/arch/powerpc/cpu/mpc5xxx/memcpy_mpc5200.c new file mode 100644 index 0000000000..09503548aa --- /dev/null +++ b/arch/powerpc/cpu/mpc5xxx/memcpy_mpc5200.c @@ -0,0 +1,71 @@ +/* + * (C) Copyright 2010 + * Wolfgang Denk, DENX Software Engineering, wd@denx.de. + * + * See file CREDITS for list of people who contributed to this + * project. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License as + * published by the Free Software Foundation; either version 2 of + * the License, or (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, + * MA 02111-1307 USA + */ + +/* + * This is a workaround for issues on the MPC5200, where unaligned + * 32-bit-accesses to the local bus will deliver corrupted data. This + * happens for example when trying to use memcpy() from an odd NOR + * flash address; the behaviour can be also seen when using "md" on an + * odd NOR flash address (but there it is not a bug in U-Boot, which + * only shows the behaviour of this processor). + * + * For memcpy(), we test if either the source or the target address + * are not 32 bit aligned, and - if so - if the source address is in + * NOR flash: in this case we perform a byte-wise (slow) then; for + * aligned operations of non-flash areas we use the optimized (fast) + * real __memcpy(). This way we minimize the performance impact of + * this workaround. + * + */ + +#include +#include +#include + +void *memcpy(void *trg, const void *src, size_t len) +{ + extern void* __memcpy(void *, const void *, size_t); + char *s = (char *)src; + char *t = (char *)trg; + void *dest = (void *)src; + + /* + * Check is source address is in flash: + * If not, we use the fast assembler code + */ + if (((((unsigned long)s & 3) == 0) /* source aligned */ + && /* AND */ + (((unsigned long)t & 3) == 0)) /* target aligned, */ + || /* or */ + (addr2info((ulong)s) == NULL)) { /* source not in flash */ + return __memcpy(trg, src, len); + } + + /* + * Copying from flash, perform byte by byte copy. + */ + while (len-- > 0) + *t++ = *s++; + + return dest; +} -- cgit v1.2.1