使用密码引擎来做一个简单的文本字符串的Sa1会伤害微芯片吗？

你能提供从闪存读取DMA的设置吗？在FRM中，我设置了一些设置安全关联和缓冲描述符的片段。执行任何算法之间唯一的真正区别是安全关联中的ALGO字段，告诉它要执行什么。然后，在缓冲区描述符中设置正确的指针，以便将输出放在哪里。最困难的部分是确保SAS和BDS不会被缓存，或者刷新缓存，这样引擎就可以接收到更改并执行它。我来看看我是否可以为初学者工具包设置一个简单的例子。



      以下为原文

    Can you provide your setup of the DMA for reading from Flash?
 
In the FRM, I put some snippets of setting up the Security Association and the Buffer Descriptors. The only real difference between executing any of the algorithms is the ALGO field in the Security Association, to tell it what to execute. That, and setting up the correct pointers in the Buffer Descriptor for where to put the output.
 
The hard part is making sure the SAs and BDs don't end up getting cached, or that you flush the cache, so that the engine can pick up the change and execute it. I'll see if I can set up a "simple" example for the starter kit.

您好EdCeCrouy，您提到过，您还想要实现没有协调的LAN连接（TCP/IP）吗？当做



      以下为原文

    Hi Edocecrous,
Did you mention that you also want to implement LAN connection (TCP/IP) without Harmony ?
Regards

听起来像是缓存一致性问题？你能分享一个小代码示例来演示这个问题吗？你确信你的数据是直接从ROM访问的吗？你提到“静态”，而不是“const”。如果它是静态的，它显然是“在RAM中”，由C启动代码初始化，但是从硬件的角度来看，它可能还没有从数据缓存刷新到RAM。DMA引擎只直接访问RAM，而不是通过DCACH。查找1600：http://WW1. Microchip .com /…AppNOTE/000 00 1600 C.PDF



      以下为原文

    Sounds like a cache coherency problem? Can you share a small code example demonstrating the problem?
 
Are you sure your data is being accessed directly from ROM? You mention "static", but not "const". If it's static only, it'll apparently be "in RAM", initialised by the C startup code, but from a hardware perspective it may well not have been flushed from data cache to RAM yet. The DMA engine only accesses RAM directly, not through dcache.
 
Look up AN1600: http://ww1.microchip.com/...AppNotes/00001600C.pdf

您好，首先感谢您的回答和即时帮助！霍华德：谢谢你的文件，我会研究并尝试不同的设置。我写的是静态的，但我的意思是静态const。所以我的代码尝试使用DMA从Flash复制到RAM。我的源操作数在9D3030E8，它通过ANP；0x1FFFFFF转换为1D3030E8，我的目的地是800万C34，转换为00000 C34。那么，缓存（如果启用）只与RAM一起使用，还是与Flash一起使用？RISC：不，我刚才提到我有一个板，上面有LAN子板。在这个时候，我不需要实现网络（但如果我这样做，它将不和谐）。@拉里：试图附加一个文件，但得到“不允许访问服务器…”错误，我的帖子也丢失了。我的DMA“复制”函数如下：[ValueDMACPEP（Valu*DST，Value*Src，未签名int size）{无符号int SDSCADR；未签名int dSTADDR；SRCADRD=（无符号INT）SRC＆AMP；0x1FFFFFFF；/ /物理地址SRCDSTADDR=（无符号INT）DST和0x1FFFFFFF；/ /物理地址DST//SRCADRDR= KVAYToEPA（SRC）；//DSTADDR＝KVAY-TYPA（DST）；IEC1CLR＝0x000 010000；/ /禁用DMA通道0中断SIFTIFSF1CLR＝0X000 010000；/ /清除现有DMA通道0中断FLALDMACON＝0x8000；//DMA模块ONDCH0CONC＝0x3；//DMA通道禁用，在块传输完成后禁用，优先级3（最高）DCH0ECON＝0；/DMA传输不是由IRQ触发的，必须强制，没有模式MatCHDCCRON＝0；//CRC模块OFDCHCH0IN＝0；/ /中断DeabDeDCH0SSA＝SRCADDR；//源启动ADSTRESDCH0DSA＝DSTADDR；/ /目的地起始地址ADSDRESCH0SSIZ＝大小；/ /源SsieDeCH0DSIZ＝大小；/ /目的地n SsieDeCH0csiz＝size；//单元大小字节，每个事件转移为0CdCrr= 0x000 FFFFF；//清除事件DC0CONSETSET＝0x80；//信道ONDCH00EcCEART＝0x80；//传输（CCHORE＝1），（DCH0INT＆AM.0x4）；int dMAFLAGS＝DCH0int；}[/COD] ]和我的全局变量是这样设置的：静态UTIN 8T屏幕[1024 ]；静态const Unt8t背景[4×1024 ]＝{0x44，0x1c，…} [代码]，我将查找FRM，一旦找出它是什么。并等待着你的“简单”的例子，关于初学者工具包的缓存。Csaba Bacskay（EdcCrrod）



      以下为原文

    Hello,
First of all, thank you for the answers and instant help!
 
@Howard: Thanks for the document, i'll study it and try different settings. I wrote static, but i meant static const.
So my code tries to use DMA to copy from FLASH to RAM i guess. My source operand is at 9D0030E8, that gets translated via & 0x1FFFFFFF to 1D0030E8, my destination is 80000C34 that gets translated to 00000C34. So then CACHE (if enabled) is used with RAM only, or also with FLASH?
 
@RISC: No, i just mentioned that i have a board which has the LAN daughterboard on it. At this moment i don't have to implement networking (but if i do, it will go without Harmony).
 
@Larry: Tried to attach a file but got a 'No permission to access to server..' error and my post also got lost. My DMA 'Copy' function is as follows:
[code]
void DMACopy(void * dst, void * src, unsigned int size){
unsigned int srcAddr;
unsigned int dstAddr;
srcAddr = (unsigned int) src & 0x1FFFFFFF;     //Physical address of src
dstAddr = (unsigned int) dst & 0x1FFFFFFF;     //Physical address of dst
//srcAddr=KVA_TO_PA(src);
//dstAddr=KVA_TO_PA(dst);
IEC1CLR=0x00010000;         // disable DMA channel 0 interrupts
IFS1CLR=0x00010000;         // clear existing DMA channel 0 interrupt flag
DMACON = 0x8000;            // dma module on
DCH0CON = 0x3;       //DMA Channel Disable, turns disabled after block transfer complete, priority 3 (highest)
DCH0ECON = 0;          // dma transfer is not triggered by irq, has to be forced, no pattern match
DCRCCON = 0;                // crc module off
DCH0INT = 0;                // interrupts disabled
DCH0SSA = srcAddr;       // source start address
DCH0DSA = dstAddr;        // destination start address
DCH0SSIZ = size;               // source size
DCH0DSIZ = size;               // destination size
DCH0CSIZ = size;               // cell size bytes transfered per event
DCH0INTCLR=0x00ff00ff;      //clear events and disable interrupts
DCH0CONSET = 0x80;      //channel on
DCH0ECONSET = 0x80;     //transfer on (CFORCE=1)
while (DCH0INT&0x4);
int DMAFlags=DCH0INT;
}
[/code]
And my global variables are set up like this:
[code]
static uint8_t screen[1024];
static const uint8_t background[4*1024] = {0x44, 0x1C, ...}
 [/code]
I'll look up FRM once figure out what is it. And waiting eagerly for your 'simple' example regarding caching for the starter kit.
 
To all, thank you for the help!
 
Csaba Bacskay
(Edocecrous)
 
 
 

简化了，在KSEG1内存范围之外的所有东西都使用缓存。更详细地说，CP0.CONFIG寄存器中的K0位控制KSGE0是否被缓存，页表属性控制映射内存范围的一致性。



      以下为原文

   
Simplified, everything outside the KSEG1 memory range uses the cache. In further detail, the K0 bits in the CP0.Config register controls whether KSEG0 is cached or not, and page table attributes controls coherency for the mapped memory ranges.

MZ设备中有没有一些关于闪存和密码的勘误表？



      以下为原文

    Don't some of the MZ devices have an errata regarding Flash and crypto?

在EC上，输出是大的Endiad格式，因此您必须在代码中交换输出。在EF上添加了一个比特，以便引擎自动地进行交换。在DA上，当密码引擎的DMA访问Flash时，总线有问题，并且我能围绕它工作的唯一方法是增加Flash等待状态。



      以下为原文

    On EC, the output was in big-endian format, so you had to swap the output in code. On EF a bit was added so that the engine would do the swap automatically.
 
On DA, there is an issue with the bus when the crypto engine's DMA accesses Flash, and the only way I could work around it was to increase the Flash wait states.

这是一个有趣的线索。很高兴拉里跟随，因为他对这个主题的输入可能是有用的。我想在PIC32 MZ（EFM变体）上设置加密的引导加载程序。这些都有硬件密码。每一个PIC32都具有一些独特性（用于启动的MAC地址），它可以用于一些关键信息。这将使HEX文件分散到更安全的更新设备上。不过，如果Bootloader空间中有足够的NVRAM，我会努力解决这个问题。



      以下为原文

    This is an interesting thread.  Glad Larry is following, because his input on this topic might be useful.

I want to set up an encrypted bootloader on a PIC32MZ ( EFM Variants ).  These have the hardware crypto in them. Each PIC32 has some uniqueness built in ( The mac address for a start ) that coudl be used for some key information.  This would make distriubtion of the hex file to update devices much more secure..

I'm trying to work out though if there is enough nvram in the bootloader space to pull this off.
 

您好，DMA问题是由缓存引起的。副本是从Flash到RAM，目的地是800万C34虚拟的，C34物理的，目的数据与缓存内容不匹配，因此在检查变量时它显示了先前的状态。一旦我将目的地分配到KSGE1中[使用了γ-AdviTyth*（（相干））]，副本就成功了。我还想尝试一个缓存无效的例程，但是首先我必须找出在我的设置中使用的缓存方案，并且需要对这个主题进行更多的阅读。关于密码引擎，我知道SA中的ALGO字段控制使用什么，但是设置所有BUF。针对不同的算法和处理拉里提出的缓存问题确实存在问题，所以我希望他能写出他提到的例子。现在我唯一需要从密码引擎来做的是SHA-1。THX，爱德华·艾尔利克



      以下为原文

    Hi,
 
The DMA problem was caused by the cache. The copy was from Flash to RAM, destination was 80000C34 virtual, C34 physical.The destination data didn't match the cache contents, so when checking the variable it showed the previous state.
 
Once i allocated the destination to be in KSEG1 [used __attribute__((coherent))], the copy was successful. I also would like to experiment with a cache invalidate routine, but first i have to find out which caching scheme is used in my setup right now and need more reading-up on the subject.
 
Regarding the Crypto engine, i know that the ALGO field in the SA controls what to use, but setting up all the buffers for the different algorithms and dealing with the caching issue Larry stated does pose a problem, so i hope he can write the examples he mentioned.
 
Right now the only thing i'd need from the Crypto engine is to do the SHA-1.
 
thx,
Edo
 

下面是我在String上运行SHA-1的方法。注意，每次运行一个新的散列时，引擎都必须重置。我尝试更新缓冲区描述符并重新运行它，但总是产生不同的结果。



      以下为原文

    Here's what I came up with for running a SHA-1 on a string.
 
Note that the engine seems to have to be reset every time a new hash is run. I tried just updating the buffer descriptor and rerunning it, but it always came up with a different result.
   Attachment(s)

crypto_sha1.c (9.08 KB) - downloaded 77 times

谢谢，太棒了！！！！它在我的32 MZ上运行良好：通过WOLFSSL进行PoPhink，看起来它们也在哈希之后重置引擎。



      以下为原文

    Thanks, that is great!!!
 
It works fine on my 32MZ :P
 
Poking through the wolfSSL it seems they also reset the engine after hashing.
 
Edo