某SPI机器设备驱动器造成的自动开关机稳定性测试卡死难题一例

硬件系统：某ARM SoC 软件系统：Linux 难题状况：商品做自动开关机稳定性测试，产生卡死。

自然环境

硬件系统：某ARM SoC

软件系统：Linux

难题状况：商品做自动开关机稳定性测试，产生卡死。

剖析

用crash专用工具分析2次卡死dump信息内容，获得卡死前的log以下。2次卡死的backtrace略有不同，但死机原因相近：最终全是在启用 complete 的全过程中浏览空指针造成 kernel panic。

log 1:
[    1.092790] c2 Unable to handle kernel NULL pointer dereference at virtual address 00000004
[    1.092796] c2 pgd = c0004000
[    1.092802] c0 [00000004] *pgd=00000000
[    1.092812] c2 Internal error: Oops: 5 [#1] PREEMPT SMP ARM
...
[    1.094412] c0 Backtrace:
[    1.094430] c2 [<c07e971c>] (_raw_spin_lock_irqsave) from [<c006e2a8>] (complete c01c/c04c)
[    1.094446] c2 [<c006e28c>] (complete) from [<c04318d4>] (spi_complete c010/c014)
[    1.094468] c2 [<c04318c4>] (spi_complete) from [<c0432328>] (spi_finalize_current_message c0228/c026c)
[    1.094479] c2 [<c0432100>] (spi_finalize_current_message) from [<c04327c8>] (spi_transfer_one_message c045c/c0484)
[    1.094503] c2 [<c043236c>] (spi_transfer_one_message) from [<c0432e98>] (__spi_pump_messages c06a8/c06e8)
[    1.094531] c2 [<c04327f0>] (__spi_pump_messages) from [<c04330fc>] (__spi_sync c0208/c0270)
[    1.094559] c2 [<c0432ef4>] (__spi_sync) from [<c0433178>] (spi_sync c014/c018)
[    1.094586] c2 [<c0433164>] (spi_sync) from [<c0440744>] (dm9051_r_reg c04c/c07c)
[    1.094595] c2 [<c04406f8>] (dm9051_r_reg) from [<c0441e30>] (dm9051_probe c06e4/0x82c)
[    1.094605] c2 [<c044174c>] (dm9051_probe) from [<c0431764>] (spi_drv_probe c08c/c0a8)

log 2:
[    1.271300] c3 Unable to handle kernel NULL pointer dereference at virtual address 00000004
[    1.271305] c3 pgd = c0004000
[    1.271312] c0 [00000004] *pgd=00000000
[    1.271323] c3 Internal error: Oops: 5 [#1] PREEMPT SMP ARM
...
[    1.414585] c0 Backtrace:
[    1.414603] c3 [<c07e97f4>] (_raw_spin_lock_irqsave) from [<c006e2a8>] (complete c01c/c04c)
[    1.414618] c3 [<c006e28c>] (complete) from [<c0431fb8>] (spi_complete c028/c034)
[    1.414643] c3 [<c0431f90>] (spi_complete) from [<c0432350>] (spi_finalize_current_message c0230/c0278)
[    1.414666] c3 [<c0432120>] (spi_finalize_current_message) from [<c04327f4>] (spi_transfer_one_message c045c/c0484)
[    1.414693] c3 [<c0432398>] (spi_transfer_one_message) from [<c0432ec4>] (__spi_pump_messages c06a8/c06e8)
[    1.414725] c3 [<c043281c>] (__spi_pump_messages) from [<c0432f1c>] (spi_pump_messages c018/c01c)
[    1.414759] c3 [<c0432f04>] (spi_pump_messages) from [<c0042abc>] (kthread_worker_fn c0c0/c014c)
[    1.414771] c3 [<c00429fc>] (kthread_worker_fn) from [<c00429e8>] (kthread 0x110/0x124)
[    1.414798] c3 [<c00428d8>] (kthread) from [<c000f208>] (ret_from_fork c014/c02c)

整理kernel spi驱动器编码：drivers/spi/spi.c

 //启用 complete

static void spi_complete(void *arg)
{
	complete(arg);
}

//给 spi_message 的 complete 与 context 组员取值

static int __spi_sync(struct spi_device *spi, struct spi_message *message,
		      int bus_locked)
{
	DECLARE_COMPLETION_ONSTACK(done);
...
	message->complete = spi_complete;
	message->context = &done;
	message->spi = spi;

	if (status == 0) {
...
			__spi_pump_messages(master, false);
...

		wait_for_completion(&done);
		status = message->status;
	}
	message->context = NULL;
	return status;
}

//回调函数 spi_message 的 complete 涵数，即 spi_complete，传到的主要参数是 spi_message 的 context，即 __spi_sync 涵数里边界定的 completion 自变量 “done”。

void spi_finalize_current_message(struct spi_master *master)
{
...
	spin_lock_irqsave(&master->queue_lock, flags);
	mesg = master->cur_msg;
	spin_unlock_irqrestore(&master->queue_lock, flags);

...
	if (mesg->complete)
		mesg->complete(mesg->context);
}

 因而基本上能够明确是实行 mesg->complete(mesg->context); 时，传到的主要参数 mesg->context 为 NULL 造成难题。继续看卡死前的一部分log：

//cpu2 上进行一次 spi 传送并进行

[    1.271151] c2 70a00000.spi: __spi_sync 
[    1.271172] c2 spi_master spi0: spi_finalize_current_message 
[    1.271180] c2 [spi_complete] 

//cpu2 上再度进行一次 spi 传送并进行

[    1.271192] c2 70a00000.spi: __spi_sync 
[    1.271212] c2 spi_master spi0: spi_finalize_current_message 
[    1.271219] c2 [spi_complete]

//cpu2 上再度进行一次 spi 传送，但在进行前，cpu1 进行了新的 spi 传送

[    1.271227] c2 70a00000.spi: __spi_sync 
[    1.271247] c2 spi_master spi0: spi_finalize_current_message
[    1.271249] c1 70a00000.spi: __spi_sync  //cpu1 进行新的传送
[    1.271261] c2 [spi_complete]
[    1.271281] c1 dm9051_r_reg: spi_sync() failed

//cpu3 上实行传送进行的 completion 时，传到的 mesg->context 为 NULL 造成卡死

[    1.271288] c3 spi_master spi0: spi_finalize_current_message 
[    1.271293] c3 [spi_complete]
[    1.271300] c3 Unable to handle kernel NULL pointer dereference at virtual address 00000004

__spi_sync 涵数中，有以下句子：

__spi_pump_messages(master, false); // 解决spi message，进行传送
wait_for_completion(&done); // 等候传送进行
message->context = NULL; //将该 spi message 的 context 置为 NULL

换句话说，__spi_sync 直到 completion 之后将此次已传送进行的 spi message 的 context 置为NULL。

假如在cpu1上进行新的传送时，传到的 spi message 自变量详细地址与cpu2上的是同一个，那麼cpu1就会有很有可能浏览到这一 NULL context。

spi message 自变量是dm9051驱动器传送出来的，查询dm9051驱动器，其启用spi的编码以下，能够见到它应用了局部变量 dm->spi_msg1 来结构 spi message。因而导致了难题。

static u8 dm9051_r_reg(struct dm9051_net *dm, u16 reg_addr)
{
    struct spi_transfer *xfer = &dm->spi_xfer1;
    struct spi_message *msg = &dm->spi_msg1;
    u8 r_cmd[2] = {reg_addr, 0x00};
    u8 r_data[2] = {0x00, 0x00};
    int ret;

    xfer->tx_buf = r_cmd;
    xfer->rx_buf = r_data;
    xfer->len = 2;

    ret = spi_sync(dm->spidev, msg);
    if (ret < 0)
        DM_MSG0("dm9051_r_reg: spi_sync() failed\n");

    return r_data[1];
}

static int dm9051_probe(struct spi_device *spi)
{
...
    spi_message_init(&dm->spi_msg1);
    spi_message_add_tail(&dm->spi_xfer1, &dm->spi_msg1);
...
}

处理

模仿 include/linux/spi/spi.h 中 spi_read 和 spi_write 的完成，每一次读写能力前都结构新的 spi message。

static inline int
spi_read(struct spi_device *spi, void *buf, size_t len)
{
	struct spi_transfer	t = {
			.rx_buf		= buf,
			.len		= len,
		};
	struct spi_message	m;

	spi_message_init(&m);
	spi_message_add_tail(&t, &m);
	return spi_sync(spi, &m);
}

————————————————-

创作者：bigfish99

blog：https://www.cnblogs.com/bigfish0506/

微信公众号：大魚内嵌式