Using OpenOCD with BeagleBoard-xM

from http://www.jonmasters.org/blog/2010/12/13/using-openocd-with-beagleboard-xm/

Using OpenOCD with BeagleBoard-xM

NOTE: This is an extremely technical blog post intended only for very serious embedded Linux developers. It was written mostly for the benefit of Google. I wish that this posting had existed over the weekend when I was looking at this.

So I now have several Texas Instruments BeagleBoard-xM boards based upon the TI DM37x series Cortex-A8 based SoC. The “xM” is an improvement on the original BeagleBoard that features the improved DM3730 in place of the original OMAP3530. Like its predecessor, the DM37x series includes Texas Instruments’ on-chip “ICEpick” JTAG TAP controller that can be used to selectively expose various additional JTAG TAPs provided by other on-chip devices. This facilities selective exposure of these devices (which may not be present in the case of the “AM” version of the chip, or otherwise might be in a low-power or otherwise disabled state), because it is nice to be able to ignore devices we don’t want to play with today. The ICEpick itself can handle up to 16 devices, although the majority of those are “reserved” for the scarily more complex days of tomorrow. If you want some of the details, refer to chapter 27 of the DaVinci Digital Media Processor Technical Reference Manual (TRM) (under “User Guides”).

The full possible DM37x JTAG chain contains the following devices (in OpenOCD ordering, from TDO to TDI – the inverse of the physical ordering of the bus, which starts with the ICEpick at 0 and ends with the DAP at 3): DAP (Debug Access Port), Sequencer (ARM968), DSP, D2D, ICEpick. It’s the last device in the chain that we really care about because it is through the Debug Access Port that we will poke at the system memory address space and registers within the Cortex-A8 processor. At Power ON, the default JTAG chain configuration will depend upon the strapping of the EMU0 and EMU1 lines (which are exposed on the Flyswatter as jumpers adjacent to the JTAG ribbon cable). If both of these lines are high (pins 1-2 not 0-1 on the Rev-B Flyswatter, i.e. nearest to the ribbon cable) – which they had better be if you want an “out of the box experience” – then the only device to appear in the chain will be the ICEpick. OpenOCD is expecting this, because it contains special macro functions that will instruct the ICEpick through its control command registers to expose any additional TAPs after initialization (we actually only want the DAP in the default case, which is how OpenOCD will behave for you if you don’t change it). Each time a new TAP is exposed, the OpenOCD JTAG logic will do the necessary state transition for it to take effect (for the chain to lengthen or contract according to devices appearing and disappearing in the chain).

The default upstream version of OpenOCD does not work with BeagleBoard-xM at the moment (as of v0.4.0-651-gc6e0705 – last modification on December 8th) because some logic was recently added to master that attempts to handle busticated DAP ROM addresses on a Freescale (not Texas Instruments) IMX51 processor, which is very similar to the DM37x. Due to this “fixup”, the DM37x’s correctly provided ROM table information will be “fixed” to use the wrong DAP address, which won’t work. For the moment, find the following loop in src/target/arm_adi_v5.c:
/* Some CPUs are messed up, so fixup if needed. */
for (i = 0; i < sizeof(broken_cpus)/sizeof(struct broken_cpu); i++)
if (broken_cpus[i].dbgbase == dbgbase &&
broken_cpus[i].apid == apid) {
LOG_WARNING("Found broken CPU (%s), trying to fixup "
"ROM Table location from 0x%08x to 0x%08x",
broken_cpus[i].model, dbgbase,
broken_cpus[i].correct_dbgbase);
dbgbase = broken_cpus[i].correct_dbgbase;
break;
}

Wrap it with a #if 0 (and don't forget to also comment or define away the definition of "i") such that it won't take effect. Then, if you have a Rev-B xM you will also need to edit the file tcl/target/amdm37x.cfg to add a new TAPID for the undocumented revision recently made to the ICEpick on these TI parts (this is backward compatible with the Rev-A because it will now look for both):
...
switch $CHIPTYPE {
dm37x {
# Primary TAP: ICEpick-C (JTAG route controller) and boundary scan
set _JRC_TAPID 0x0b89102f
set _JRC_TAPID1 0x1b89102f
}
...
...
# Primary TAP: ICEpick - it is closest to TDI so last in the chain
jtag newtap $_CHIPNAME jrc -irlen 6 -ircapture 0x1 -irmask 0x3f /
-expected-id $_JRC_TAPID -expected-id $_JRC_TAPID1

Then make and install the OpenOCD binaries (ensure you have the free FTDI libraries installed, and configure with “configure –enable-maintainer-mode –enable-ft2232_libftdi”). You can now use OpenOCD with the xM revA or B.

Jon.

深度学习是机器学习的一个子领域,它基于人工神经网络的研究,特别是利用多层次的神经网络来进行学习和模式识别。深度学习模型能够学习数据的高层次特征,这些特征对于图像和语音识别、自然语言处理、医学图像分析等应用至关重要。以下是深度学习的一些关键概念和组成部分: 1. **神经网络(Neural Networks)**:深度学习的基础是人工神经网络,它是由多个层组成的网络结构,包括输入层、隐藏层和输出层。每个层由多个神经元组成,神经元之间通过权重连接。 2. **前馈神经网络(Feedforward Neural Networks)**:这是最常见的神经网络类型,信息从输入层流向隐藏层,最终到达输出层。 3. **卷积神经网络(Convolutional Neural Networks, CNNs)**:这种网络特别适合处理具有网格结构的数据,如图像。它们使用卷积层来提取图像的特征。 4. **循环神经网络(Recurrent Neural Networks, RNNs)**:这种网络能够处理序列数据,如时间序列或自然语言,因为它们具有记忆功能,能够捕捉数据中的时间依赖性。 5. **长短期记忆网络(Long Short-Term Memory, LSTM)**:LSTM 是一种特殊的 RNN,它能够学习长期依赖关系,非常适合复杂的序列预测任务。 6. **生成对抗网络(Generative Adversarial Networks, GANs)**:由两个网络组成,一个生成器和一个判别器,它们相互竞争,生成器生成数据,判别器评估数据的真实性。 7. **深度学习框架**:如 TensorFlow、Keras、PyTorch 等,这些框架提供了构建、训练和部署深度学习模型的工具和库。 8. **激活函数(Activation Functions)**:如 ReLU、Sigmoid、Tanh 等,它们在神经网络中用于添加非线性,使得网络能够学习复杂的函数。 9. **损失函数(Loss Functions)**:用于评估模型的预测与真实值之间的差异,常见的损失函数包括均方误差(MSE)、交叉熵(Cross-Entropy)等。 10. **优化算法(Optimization Algorithms)**:如梯度下降(Gradient Descent)、随机梯度下降(SGD)、Adam 等,用于更新网络权重,以最小化损失函数。 11. **正则化(Regularization)**:技术如 Dropout、L1/L2 正则化等,用于防止模型过拟合。 12. **迁移学习(Transfer Learning)**:利用在一个任务上训练好的模型来提高另一个相关任务的性能。 深度学习在许多领域都取得了显著的成就,但它也面临着一些挑战,如对大量数据的依赖、模型的解释性差、计算资源消耗大等。研究人员正在不断探索新的方法来解决这些问题。
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值