修复kernel编译栈帧大小异常问题error: the frame size of 1928 bytes is larger than 1024 bytes

上图为编译异常时：

在原本代码上增加了一个spin原子锁，然后就编译不过

复制代码

diff --git a/nvidia/drivers/optcamera_xdma/cdev_sgdma.c b/nvidia/drivers/optcamera_xdma/cdev_sgdma.c
index 94bdf43e9..5c8c912b5 100755
--- a/nvidia/drivers/optcamera_xdma/cdev_sgdma.c
+++ b/nvidia/drivers/optcamera_xdma/cdev_sgdma.c
@@ -575,6 +575,7 @@ static ssize_t char_sgdma_read_write(struct file *file, const char __user *buf,
 	struct xdma_dev *xdev;
 	struct xdma_engine *engine;
 	struct xdma_io_cb cb;
+//	unsigned long irq_flag;  //自旋锁 中断状态
 
 #ifdef  DBG_CHAR_SGDMA_READ_WRITE_TIMES
 	ktime_t curTime, curTime_wait_flag, curTime_align, curTime_memset, curTime_map, curTime_submit, curTime_unmap;
@@ -587,7 +588,8 @@ static ssize_t char_sgdma_read_write(struct file *file, const char __user *buf,
 		return rv;
 	xdev = xcdev->xdev;
 	engine = xcdev->engine;
-
+	
+//	spin_lock_irqsave(&xcdev->lock, irq_flag); //保存中断状态，禁止本地中断，并获取自旋锁
 	clean_fpga_irq();  //add by tab to call xdma
 	
 #ifdef  DBG_CHAR_SGDMA_READ_WRITE_TIMES
@@ -673,6 +675,7 @@ static ssize_t char_sgdma_read_write(struct file *file, const char __user *buf,
 	printk(" %s wait_ts = %lld us, align_ts = %lld us, memset_ts = %lld us, map_ts = %lld us, submits_ts = %lld us, umap_ts = %lld us \n", __func__, curTime_wait_flagst, curTime_alignst, curTime_memsetst, curTime_mapst, curTime_submitst, curTime_unmapst);
 #endif
 	//printk("enter %s end, line =%d \n", __func__, __LINE__);
+//	spin_unlock_irqrestore(&xcdev->lock, irq_flag);   //
 
 	return res;
 }

加入修改好，则编译异常:

1、报错信息：

error: the frame size of 1928 bytes is larger than 1024 bytes [-Werror=frame-larger-than=]

那么解决方案：

2、报错打印分析

(1)编译器编译内核时，发现栈帧大小是1024字节，但是栈使用是1928字节，超过了默认栈帧大小；

(2)默认编译器是报警告信息，上面报错误信息是因为在编译器选项中添加了-Wall，把警告当错误处理；

3、报错解决

3.1、报错原因分析

(1)栈内存申请过大，或者函数调用层次太深都会导致栈溢出，引起系统崩溃，在编译时会去检查栈使用大小是否超过配置的栈大小；

(2)内核编译时会通过"-Wframe-larger-than=xxx"选项，传递给编译器栈大小，当编译器检测到栈使用大于阈值时，会产生一条警告信息；

转自大佬解说：

编译时内核栈溢出：the frame size of 1928 bytes is larger than 1024 bytes_the frame size of 2048 bytes is larger than 1024 b-CSDN博客

具体解决方案：

复制代码

yahboom@yahboom-vm:~/r32.7.2_sources-sdk/Linux_for_Tegra/source/src_out/kernel_src_build/kernel$ git diff   kernel-4.9/arch/arm64/configs/tegra_defconfig
diff --git a/kernel-4.9/arch/arm64/configs/tegra_defconfig b/kernel-4.9/arch/arm64/configs/tegra_defconfig
index a3904fa79..5fa971b20 100644
--- a/kernel-4.9/arch/arm64/configs/tegra_defconfig
+++ b/kernel-4.9/arch/arm64/configs/tegra_defconfig
@@ -1209,3 +1209,4 @@ CONFIG_ARCH_TEGRA_23x_SOC=y
 CONFIG_TEGRA_SAFETY=y
 CONFIG_VIDEO_OPTCAMERA_XDMA=y
 CONFIG_OPT_IDE1_DIO=y
+CONFIG_FRAME_WARN=4096
yahboom@yahboom-vm:~/r32.7.2_sources-sdk/Linux_for_Tegra/source/src_out/kernel_src_build/kernel$

3.3、栈大小的影响

3.3.1、栈比较大

比如栈大小使用8K

(1)优点：栈比较大，不容易导致栈溢出；

(2)缺点：浪费内存，有些时候根本用不到这么大的栈内存；并且内存是4K分页，创建一个内核栈就需要申请连续2块的4K页，内存紧张的时候，申请8K的连续内存比申请4K困难的多；

3.3.2、栈比较小

比如栈大小1K

(1)优点：节省内存，都占用不到一个4K页，创建内核栈时比较容易；

(2)缺点：因为栈比较小，容易栈溢出；

3.4解决措施

(1)修改内核配置项，把"CONFIG_FRAME_WARN"配置项改大一点；

(2)修改程序，不要超过内核配置的栈大小；

补充：目前我所接触到的内核配置，32位的系统配置的栈大小一般是1K，64位的系统栈大小一般是4K；