View Bug Activity | Format For Printing
*** Bug 6433 has been marked as a duplicate of this bug. ***
*** Bug 6434 has been marked as a duplicate of this bug. ***
*** Bug 6435 has been marked as a duplicate of this bug. ***
Created an attachment (id=2757) Use probe_kernel functions in deref macros This changes the deref macros to use probe_kernel, but there's a performance hit, probably due to no longer inlining the dereferencing activity. Simple benchmark scripts to follow...
Created an attachment (id=2758) Test the overhead of dereferencing numbers Run as 'stap -gt deref-long.stp', and add '-DTEST_DIRECT' for direct memory access. My results on 2.6.26-rc3 x86_64: OLD: direct: 1057min/1268avg/10905max deref: 1073min/1305avg/9405max NEW: direct: 1050min/1235avg/9443max deref: 1238min/1473avg/8850max
Created an attachment (id=2759) Test the overhead of dereferencing strings Run as 'stap -gt deref-string.stp', and add '-DTEST_DIRECT' for direct memory access. My results on 2.6.26-rc3 x86_64: OLD: direct: 1725min/2072avg/10283max deref: 1778min/2085avg/10950max NEW: direct: 1650min/2012avg/10613max deref: 2152min/2476avg/12563max
So given the performance disadvantage, is this still a change you want to make? Or can anyone see a way to improve the efficiency?
Josh, thanks for trying this out. Please commit the patch, but #if-0'd out, so that some day we can easily try again, and resolve/wontfix this bug.
Committed in 2cda5f46cba978e19590efd7fee0b3075f8eb8df with the autoconf test disabled. One can use -DSTAPCONF_PROBE_KERNEL to enable it manually.