On Mon, Apr 22, 2019 at 12:25:00AM +0900, Akira Yokosawa wrote:
> Hi Paul,
>
> The sentence shown below as a diff looks incomplete to me.
> What is your intention here?
Not that! Thank you for catching this!!! Please see below for an
attempted fix. Thoughts?
I have queued your four patches. My problems are (1) force of habit
and (2) failing to remember that comment-only lines are discarded.
I am not all that happy with the #ifdef, but it beats the alternative.
But what I did in this case was to just remove the fprintf() in favor
of the BUG_ON(). ;-)
Thanx, Paul
------------------------------------------------------------------------
commit 72e774fda87eaead9637d887426b701ec229664d
Author: Paul E. McKenney <paulmck@xxxxxxxxxxxxx>
Date: Sun Apr 21 11:33:49 2019 -0700
defer/hazptr: Fix broken statement of restart location constraints
Reported-by: Akira Yokosawa <akiyks@xxxxxxxxx>
Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxx>
diff --git a/defer/hazptr.tex b/defer/hazptr.tex
index 4b61d34d334c..b491c590ae2d 100644
--- a/defer/hazptr.tex
+++ b/defer/hazptr.tex
@@ -239,9 +239,12 @@ Which is a very good thing, because B's successor is the now-freed
element~C, which means that Thread~0's subsequent accesses might have
resulted in arbitrarily horrible memory corruption, especially if the
memory for element~C had since been re-allocated for some other purpose.
-Therefore, statically allocated global variables), hazard-pointer
-readers must typically restart the full traversal in the face of a
-concurrent deletion.
+Therefore, hazard-pointer readers must typically restart the full
+traversal in the face of a concurrent deletion.
+Often the restart must go back to some global (and thus immortal) pointer,
+but it is sometimes possible to restart at some intermediate location
+if that location is guaranteed to still be live, for example, due to
+the current thread holding a lock, a reference count, etc.
\QuickQuiz{}
Readers must ``typically'' restart?
