问题描述：TQE9开发板下载带有图形界面的qnx-ifs系统镜像后总是无法启动，卡在Load QNX image from SDMMC...
分析问题之前，先分析一下sabreARD-graphics.build的启动脚本，镜像文件说明如下：
[image=0x10800000]
# For u-boot, IFS image should be uncompressed
[virtual=armle-v7,raw] .bootstrap = {
# For IPL, IFS image should be compressed
#[virtual=armle-v7,raw +compress] .bootstrap = {# Startup parameters:# '-b' -> enable BT (conflicts with SPI NOR and PCIe)# '-m' -> enable d-cache/MMU (improves boot time)# '-W' -> enable watchdog (wdtkick should be uncommented when using this option)# Note:only ONE of below option may be selected for NOR flash# '-n0' -> no NOR Flash (I2C3 enabled)# '-n1' -> parallel NOR Flash (I2C3 disabled)# '-n2' -> SPI NOR Flash (I2C3 disabled)## '-c' -> CAN startup option (conflicts with Ethernet)# '-s' -> Use SDMA to load IFS# '-r' -> to reserve the top 256M of RAM (0x80000000 -- 0x8FFFFFFF) so that# screen will work properly. There is an issue with the graphics processor# on the i.MX6 Quad Plus where the code will not run if allocated in this# block of memory. This is an interm fix until the problem is resolved by# NXP.# startup-imx6x-sabreARD -n0 -m -r 0x80000000,256M# Otherwise use the following version of startup to access the full range of RAMstartup-imx6x-sabreARD -n1 -m -W -sPATH=/proc/boot:/bin:/usr/bin:/opt/bin/sbin:/usr/sbin LD_LIBRARY_PATH=/proc/boot:/lib:/usr/lib:/lib/dll:/lib/dll/pci:/opt/lib procnto-smp-instr
}armle-v7：CPU type
raw：Create a binary image with an instruction sequence at its beginning to jump
that when you download a raw image to memory using a bootloader, you can
then instruct it to run right at the beginning of the image, rather than having
to figure out what the actual startup_vaddr is each time you modify the
startup code.
+compress: compress flag.
compress参数会指明要不要对镜像进行压缩。对于u-boot来讲，镜像是不可以被压缩的，否则设备无法启动。默认加载的镜像image地址是0x10800000，如果代码中加载镜像的地址与build中image值不一致，则需要使用-s参数，使用SDMA加载IFS。
分析我单板的问题：无论是否进行压缩，单板始终都无法正常启动，但同样生成的不带graphic的镜像却可以正常启动，经过层层定位分析：问题出现在接口fat_copy_file上，该接口的逻辑实现:如果当前存储数据的块是连续的，则会一次性把连续块上的内容全部读取出来；如果不连续，则只读当前块的内容。问题就是一次性把连续块上的内容都读取出来的时候存在异常，这个也不知道是emmc的问题还是bsp版本实现的问题。当我们强制让emmc分多次将连续块上的内容分多次读取出来的时候，大包便可以启动了。代码如下：
/* copy a file from to a memory location */
static int
fat_copy_file(unsigned cluster, unsigned size, unsigned char *buf)
{#if 1int result, txf;unsigned prev_c, next_c, curr_c;int sz = (int)size;int cbytes = fs_info.cluster_size*SECTOR_SIZE;int actual_len = 0;if(sdmmc_debug > 1){ser_putstr((char *)"fat_copy_file: cs 0x");ser_puthex(fs_info.cluster_size);ser_putstr((char *)" size 0x");ser_puthex(sz);ser_putstr((char *)"type 0x");ser_puthex(fs_info.fat_type );ser_putstr((char *)"\n");}#if 1cache_valid = 1;
#endifg_fat_sector = -1;/**Note that this impl assume the following:* 1) The max DMA transfer size is bigger than the max consolidate transfer size* Otherwise, we need to break down into smaller transfer.* 2) we always do at least one whole cluster transfer. This might overwrite the client buffer, but* since this is purely used for IPL, we don't care about that now.*/curr_c = cluster;while(sz>0){txf = cbytes;prev_c = curr_c;while(sz>txf){//try consolidate contigus entry;next_c = fat_get_fat_entry(curr_c);if(next_c == (curr_c+1) && txf > 20*SECTOR_SIZE){txf +=cbytes;curr_c = next_c;}else{curr_c = next_c;break;}}if(sdmmc_debug > 4){ser_putstr((char *)"blkcnt 0x");ser_puthex(txf/SECTOR_SIZE);ser_putstr((char *)" p 0x");ser_puthex(prev_c);ser_putstr((char *)" n 0x");ser_puthex(curr_c);ser_putstr((char *)"\n");}//read the contig cluster outresult= read_fsector(cluster2fsector(prev_c), buf, txf/SECTOR_SIZE) ;if (result != SDMMC_OK)return result;sz -= txf;buf += txf;actual_len += txf;}
#elseint sz = (int)size;while(!end_of_file(cluster) && (sz > 0)) {int txf = MIN(sz, fs_info.cluster_size * SECTOR_SIZE);if (SDMMC_OK != read_cluster(cluster, buf, txf)) {ser_putstr(" Error - read_clust(): clust buf txf = ");ser_puthex((unsigned int)clust); ser_putstr(" ");ser_puthex((unsigned int)buf); ser_putstr(" ");ser_puthex((unsigned int)txf); ser_putstr("\n");return SDMMC_ERROR;}ser_putstr((char *)"cluster %x");ser_puthex(cluster);ser_putstr((char *)"\n");sz -= txf;buf += txf;cluster = fat_get_fat_entry(cluster);}
#endifif(sdmmc_debug > 0){ser_putstr("Actual read fs size = 0x");ser_puthex(actual_len);ser_putstr("\n");}return SDMMC_OK;
}上述代码的修改地方是将if(next_c == (curr_c+1))改为if(next_c == (curr_c+1) && txf >20*SECTOR_SIZE)，即将连续块上超过10M的数据的分多次读写。这样修改完成之后，镜像可以正常启动。
经过试验后得出以下两条结论：
1、压缩情况下,只要地址不同（build文件中的image值和ipl文件main函数中的QNX_LOAD_ADDR值不同），启动参数无论是否包含-s，镜像均可以正常启动。（由于需要解压缩，所以大包加载的地址必须不能和build中的image地址不同，否则无法解压缩）
2、不压缩情况下，地址若相同，无论是否加-s，都可以正常启动；但若地址不同，只有不加-s的时候才可以正常启动。