On 21.11.24 17:28, Suren Baghdasaryan wrote:
Add raw_seqcount_try_begin() to opens a read critical section of the given
seqcount_t if the counter is even. This enables eliding the critical
section entirely if the counter is odd, instead of doing the speculation
knowing it will fail.
Suggested-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Signed-off-by: Suren Baghdasaryan <surenb@xxxxxxxxxx>
---
Applies over Linus' ToT
include/linux/seqlock.h | 22 ++++++++++++++++++++++
1 file changed, 22 insertions(+)
diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
index 5298765d6ca4..22c2c48b4265 100644
--- a/include/linux/seqlock.h
+++ b/include/linux/seqlock.h
@@ -318,6 +318,28 @@ SEQCOUNT_LOCKNAME(mutex, struct mutex, true, mutex)
__seq; \
})
+/**
+ * raw_seqcount_try_begin() - begin a seqcount_t read critical section
+ * w/o lockdep and w/o counter stabilization
+ * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
+ *
+ * Similar to raw_seqcount_begin(), except it enables eliding the critical
+ * section entirely if odd, instead of doing the speculation knowing it will
+ * fail.
+ *
+ * Useful when counter stabilization is more or less equivalent to taking
+ * the lock and there is a slowpath that does that.
+ *
+ * If true, start will be set to the (even) sequence count read.
+ *
+ * Return: true when a read critical section is started.
+ */
+#define raw_seqcount_try_begin(s, start) \
+({ \
+ start = raw_read_seqcount(s); \
+ !(start & 1); \
+})
In gup_fast(), we simply do
seq = raw_read_seqcount(¤t->mm->write_protect_seq);
if (seq & 1)
return 0;
Should we be using that there as well?
if (!raw_seqcount_try_begin(¤t->mm->write_protect_seqs, seq))
return 0;
--
Cheers,
David / dhildenb
