On Fri, Feb 08, 2019 at 03:19:07PM +0000, Emmanuel Dreyfus wrote: > As I understand, that means SIGSEGV is not caused by userland > code, but by kernel code. I assume that if I do a SCSI command > that access unmapped memory, I would get something like this? > But no thread seems to be undergoing a tape I/O.
You need to check the details of the system call, e.g. if I do a write(2), and pass a buffer of, say, 16k, but use nbytes = 64k it will most likely cause a SIGSEGV when the kernel tries to access a page beyound the 16k allocated userland buffer. This access is on behalf of the user, so the kernel will signal the process, and this looks like what you see here. Martin
