diff options
author | Stefan Roese <sr@denx.de> | 2007-12-06 05:58:43 +0100 |
---|---|---|
committer | Stefan Roese <sr@denx.de> | 2007-12-06 05:58:43 +0100 |
commit | a27044b14a9e93678a82d7b35f202b93e7687abc (patch) | |
tree | c4a0a54d0fcc9cbde8820ed77c61d3cff2e714f5 /board/amcc/yucca/yucca.c | |
parent | a90921f71d225bf9e0f0fc7b8beadeb8001bf78a (diff) | |
download | u-boot-imx-a27044b14a9e93678a82d7b35f202b93e7687abc.zip u-boot-imx-a27044b14a9e93678a82d7b35f202b93e7687abc.tar.gz u-boot-imx-a27044b14a9e93678a82d7b35f202b93e7687abc.tar.bz2 |
ppc4xx: Enable hardware-fix for PCI/DMA errata on AMCC 440SP/SPe boards
This patch enables the hardware-fix for the PCI/DMA errata's 19+22 by
setting the FIXD bit in the SDR0_MFR register. Here a description of the
symptoms:
Problem Description
------------------------------
If a DMA is performed between memory and PCI with the DMA 1 Controller
using prefetch, and as a result uses a special purpose buffer selected by
the PCIXn Bridge Options 1 Register (PCIXn_BRDGOPT1[RBP7] - bits 31-29),
the first part of the transfer sequence is performed twice. The
PPC440SPe PCI Controller requests more data than was needed such that in
the case of enforce memory protection, a host CPU exception can occur.
No data is corrupted, because data transfer is stopped in the PCI
Controller. Prefetch enable is specified by setting DMA Configuration
Register (I2O0_DMAx_CFG[DXEPD] - bit 31) to 0.
Behavior that may be observed in a running system
---------------------------------------------------------------------------
1. DMA performance is decreased because of the double access on the PCI bus
interface.
2. If an illegal access to some address on the PCI bus is detected at the
system level, a machine check or similar system error may occur.
Workarounds Available
----------------------------------
1. Do not program prefetch. Note that a prefetch command cannot be programmed
without selecting a special purpose buffer.
2. To avoid crossing a physical boundary of the PCI slave device, add 512
bytes of address to the PCI address range.
This patch was originally provided by Pravin M. Bathija <pbathija@amcc.com>
from AMCC and slighly changed.
Signed-off-by: Pravin M. Bathija <pbathija@amcc.com>
Signed-off-by: Stefan Roese <sr@denx.de>
Diffstat (limited to 'board/amcc/yucca/yucca.c')
-rw-r--r-- | board/amcc/yucca/yucca.c | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/board/amcc/yucca/yucca.c b/board/amcc/yucca/yucca.c index 17aaa26..52486cc 100644 --- a/board/amcc/yucca/yucca.c +++ b/board/amcc/yucca/yucca.c @@ -529,10 +529,10 @@ int board_early_init_f (void) mtdcr (uic0sr, 0x00000000); /* clear all interrupts */ mtdcr (uic0sr, 0xffffffff); /* clear all interrupts */ - /* SDR0_MFR should be part of Ethernet init */ - mfsdr (sdr_mfr, mfr); - mfr &= ~SDR0_MFR_ECS_MASK; - /*mtsdr(sdr_mfr, mfr);*/ + mfsdr(sdr_mfr, mfr); + mfr |= SDR0_MFR_FIXD; /* Workaround for PCI/DMA */ + mtsdr(sdr_mfr, mfr); + fpga_init(); return 0; |