On Tue, Sep 10, 2019 at 07:51:58PM -0700, Matthew Wilcox wrote: > > If we encounter an error on a page, we can lose the error if we've > dropped the mmap_sem while we wait for the I/O. That can result in > taking the fault multiple times, and retrying the read multiple times. > Spotted by inspection. But does the patch make any difference? Looking into x86 code, VM_FAULT_RETRY makes it retry the fault or return to userspace before it checks for VM_FAULT_SIGBUS bit. -- Kirill A. Shutemov
