On 08/04/20(Wed) 12:11, Vitaliy Makkoveev wrote: > On Wed, Apr 08, 2020 at 09:51:45AM +0200, Martin Pieuchot wrote: > [...] > As I see (pseudo code): > [...] > So, I fixed this issue :)
We fix what we see. Bugs are always in the unseen :) This mess is clearly complicated. Mixing 3 different locks in different orders with different pieces of code is complex. So let's ask this question again: what is the NET_LOCK() protecting in this code? Is it needed? What is the `pppx_ifs_lk' protecting? >From whom? Are they needed or do they introduce more problem than they are solving? We see the problem of context switch breaking atomicity of code, the simplest way to get rid of this problem is to get rid of the locks, or at least reduce their number. So does one see how useless is `pppx_ifs_lk' since all the code it "protects" is executed under KERNEL_LOCK() and doesn't contain any context switch? Historically the NET_LOCK() has been "pushed" under all pseudo-device ioctl(2). This was a mechanical change until somebody figures out if it is needed or not. That's the real question. So let's ask it :o)
