Messages in this thread | | | From | "Jin, Zhi" <> | Subject | RE: [PATCH] stm class: increase pos if no free channels found | Date | Mon, 15 Oct 2018 08:16:26 +0000 |
| |
> -----Original Message----- > From: Alexander Shishkin [mailto:alexander.shishkin@linux.intel.com] > Sent: Monday, October 15, 2018 3:43 PM > To: Jin, Zhi <zhi.jin@intel.com> > Cc: linux-kernel@vger.kernel.org; Jin, Zhi <zhi.jin@intel.com> > Subject: Re: [PATCH] stm class: increase pos if no free channels found > > Zhi Jin <zhi.jin@intel.com> writes: > > > Considering this case in find_free_channels(): > > > > bitmap: > > +------------------+-+-+-+-+-+-+-+-+-+-+ > > | ...... |0|0|0|0|0|0|0|1|0|0| > > +------------------+-+-+-+-+-+-+-+-+-+-+ > > > > 1. Channel #2 has been occupied, so bit #2 is 1, and the others > > are all 0. > > 2. Another thread tries to find 4 free channels from #0. > > 3. In the 1st loop, pos starts from 0, and then it checks if the > > following 4 bits are all 0, but fails, as bit#2 is 1. > > 4. In the 2st loop, pos is not updated, and still starts from 0, > > so nothing changes against loop #1. > > 5. Dead loop ... > > > > This patch is to update the pos in step #3 to avoid the issue. > > The description is slightly confusing, but the patch looks correct and > the original code is clearly wrong. Thank you for finding this! > > Basically, if you request 1 channel 3 times, release the first two and > then request 4 channels, you'll be stuck, right?
Yes, you are right. But the real case that I reproduced the issue is a little different: I have 2 stp-policy: "console": masters "256 259" channels "7 10" "user" : masters "256 1024" channels "0 127" I understand the policies should not be overlapped, which is caused by some other issues. So if someone uses "console" to request a channel (who will get Channel #7) and then another uses "user" to request more than 8 channels, it will be stuck. The commit message is what I trying to abstract the above case, sorry for the confusion.
> > Thanks, > -- > Alex
| |