Device drivers may encounter errors when communicating with OS kernel and hardware. However, error handling code often gets insufficient attention in driver development and testing, because these errors rarely occur in real execution. For this reason, many bugs are hidden in error handling code. Previous approaches for testing error handling code often neglect the characteristics of device drivers, which limits their efficiency and accuracy. In this paper, we first study the source code of Linux drivers to find useful characteristics of error handling code. Then we use these characteristics in fault injection testing, and propose a practical approach named EH-Test, which can automatically and efficiently test error handling code in drivers. To improve the representativeness of injected faults, we design a pattern-based extraction strategy to automatically and accurately extract target functions which can actually fail and trigger error handling code. During execution, we use a monitor to record runtime information and pair checkers to check resource usages. We have evaluated EH-Test on 15 real Linux device drivers and found 50 new bugs in Linux 3.17.2. The code coverage is also effectively increased. Comparison experiments to some previous approaches also show the availability of EH-Test.