On Tue, Dec 19, 2023 at 12:18:45PM -0500, Hugo Villeneuve wrote: > From: Hugo Villeneuve <hvilleneuve@xxxxxxxxxxxx> > > If a problem with a device occurs during probing, and we subsequently > try to remove the driver, we get the following error: > > $ rmmod sc16is7xx > [ 62.783166] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000040 > [ 62.994226] Call trace: > [ 62.996672] serial_core_unregister_port+0x58/0x2b0 > [ 63.001558] serial_ctrl_unregister_port+0x18/0x30 > [ 63.006354] uart_remove_one_port+0x18/0x30 > [ 63.010542] sc16is7xx_spi_remove+0xc0/0x190 [sc16is7xx] > Segmentation fault > > Tested on a custom board with two SC16IS742 ICs, and by simulating a fault > when probing channel (port) B of the second device. > > The reason is that uart_remove_one_port() has already been called during > probe in the out_ports: error path, and is called again in > sc16is7xx_remove(). > > Fix the problem by clearing the device drvdata in probe error path to > indicate that all resources have been deallocated. And only deallocate > resources in sc16is7xx_remove() if device drvdata is non-null. ... > + dev_set_drvdata(dev, NULL); I believe this is wrong approach to fix the issue as this one is prone to be cleaned up in the future as we don't do this call explicitly for the past ~15 years. -- With Best Regards, Andy Shevchenko
