Shouldn't we have waited for buffer busy waits while waiting CBC latch？

With regular logical  IOs the buffer contents are not read while holding the CBC(Cache Buffer Chian) latch:

1.  1#  Take CBC latch into shared mode


2.  2# Walk the buffer hash chain until you find the relevant buffer header


3.  3# Upgrade the CBC latch to Exclusive mode


4.  4# Pin the buffer header


5.  5 # Release the CBC latch


6.  6 # Now access the buffer data( call transaction,  data layer etc)   -- if someone else wants to pin the buffer now , they'd wait for buffer busy waits


7.  7# Take the CBC latch again(in shared mode)


8.  8$ Unpin the buffer header


9.  10# Release the CBC latch

Sometimes "short" logical IOs can skip a few steps
with "short" LIOs like unique index lookup LIO(etc) Oracle can avoid the buffer pinning codepath:

1.  1# Take CBC altch in shared mode


2.  2# Walk the buffer hash chain until you find the relevant buffer header -- This show up as consistent reads - examination  conter in v$sesstat


3.  3# Now access the buffer data


4.  4# Release the CBC latch

 
from TanelPoder's