perf: reproducibility fixes for pmu-events.c

perf generates pmu-events.c as part of the build process. The
code that generates the events is doing tree walks and potentially
other non-determinstic things.

We'd rather not mess with that implementation, so we add a script
that knows how to read the pmu-events.c, sort the entries and then
copy it over the generated one.

With this, we should always have events in the same order, improving
reproducibility.

Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>

2 files changed, 109 insertions, 0 deletions

diff --git a/meta/recipes-kernel/perf/perf.bb b/meta/recipes-kernel/perf/perf.bb
index 2beb404c03..c7653e523c 100644
--- a/meta/recipes-kernel/perf/perf.bb
+++ b/meta/recipes-kernel/perf/perf.bb
@@ -250,6 +250,14 @@ do_configure_prepend () {
         # all the calls have YFLAGS, which contains prefix mapping information.
         sed -i -e 's,$(YACC),$(YACC) $(YFLAGS),g' ${S}/scripts/Makefile.host
     fi
+    if [ -e "${S}/tools/perf/pmu-events/Build" ]; then
+        target='$(OUTPUT)pmu-events/pmu-events.c $(V)'
+        replacement1='$(OUTPUT)pmu-events/pmu-events.c $(V)\n'
+        replacement2='\t$(srctree)/sort-pmuevents.py $(OUTPUT)pmu-events/pmu-events.c $(OUTPUT)pmu-events/pmu-events.c.new\n'
+        replacement3='\tcp $(OUTPUT)pmu-events/pmu-events.c.new $(OUTPUT)pmu-events/pmu-events.c'
+        sed -i -e "s,$target,$replacement1$replacement2$replacement3,g" \
+                       "${S}/tools/perf/pmu-events/Build"
+    fi
     # end reproducibility substitutions
 
     # We need to ensure the --sysroot option in CC is preserved
@@ -292,6 +300,14 @@ do_configure_prepend () {
     # so we copy it from the sysroot unistd.h to the perf unistd.h
     install -D -m0644 ${STAGING_INCDIR}/asm-generic/unistd.h ${S}/tools/include/uapi/asm-generic/unistd.h
     install -D -m0644 ${STAGING_INCDIR}/asm-generic/unistd.h ${S}/include/uapi/asm-generic/unistd.h
+
+    # the fetcher is inhibited by the 'inherit kernelsrc', so we do a quick check and
+    # copy for a helper script we need
+    for p in $(echo ${FILESPATH} | tr ':' '\n'); do
+	if [ -e $p/sort-pmuevents.py ]; then
+	    cp $p/sort-pmuevents.py ${S}
+	fi
+    done
 }
 
 python do_package_prepend() {
diff --git a/meta/recipes-kernel/perf/perf/sort-pmuevents.py b/meta/recipes-kernel/perf/perf/sort-pmuevents.py
new file mode 100755
index 0000000000..5ddf0f144f
--- /dev/null
+++ b/meta/recipes-kernel/perf/perf/sort-pmuevents.py
@@ -0,0 +1,93 @@
+#!/usr/bin/env python3
+
+# perf pmu-events sorting tool
+#
+# Copyright (C) 2021 Bruce Ashfield
+#
+# SPDX-License-Identifier: MIT
+#
+
+import sys
+import os
+import re
+from collections import OrderedDict
+
+if len(sys.argv) < 2:
+    print( "[ERROR]: input and output pmu files missing" )
+    sys.exit(1)
+
+if len(sys.argv) < 3:
+    print( "[ERROR]: output pmu file missing" )
+    sys.exit(1)
+
+infile = sys.argv[1]
+outfile = sys.argv[2]
+
+if not os.path.exists(infile):
+    print( "ERROR. input file does not exist: %s" % infile )
+    sys.exit(1)
+
+if os.path.exists(outfile):
+    print( "WARNING. output file will be overwritten: %s" % infile )
+
+with open(infile, 'r') as file:
+    data = file.read()
+
+preamble_regex = re.compile( '^(.*?)^struct', re.MULTILINE | re.DOTALL )
+
+preamble = re.search( preamble_regex, data )
+struct_block_regex = re.compile( '^struct.*?(\w+) (.*?)\[\] = {(.*?)^};', re.MULTILINE | re.DOTALL )
+field_regex =  re.compile( '{.*?},', re.MULTILINE | re.DOTALL )
+cpuid_regex = re.compile( '\.cpuid = (.*?),', re.MULTILINE | re.DOTALL )
+name_regex = re.compile( '\.name = (.*?),', re.MULTILINE | re.DOTALL )
+
+# create a dictionary structure to store all the structs, their
+# types and then their fields.
+entry_dict = {}
+for struct in re.findall( struct_block_regex, data ):
+    # print( "struct: %s %s" % (struct[0],struct[1]) )
+    entry_dict[struct[1]] = {}
+    entry_dict[struct[1]]['type'] = struct[0]
+    entry_dict[struct[1]]['fields'] = {}
+    for entry in re.findall( field_regex, struct[2] ):
+        #print( "    entry: %s" % entry )
+        cpuid = re.search( cpuid_regex, entry )
+        if cpuid:
+            #print( "    cpuid found: %s" % cpuid.group(1) )
+            entry_dict[struct[1]]['fields'][cpuid.group(1)] = entry
+            
+        name = re.search( name_regex, entry )
+        if name:
+            #print( "    name found: %s" % name.group(1) )
+            entry_dict[struct[1]]['fields'][name.group(1)] = entry
+        
+
+# created ordered dictionaries from the captured values. These are ordered by
+# a sorted() iteration of the keys. We don't care about the order we read
+# things, just the sorted order. Hency why we couldn't create these during
+# reading.
+#
+# yes, there's a more concise way to do this, but our nested dictionaries of
+# fields make it complex enough that it becomes unreadable.
+entry_dict_sorted = OrderedDict()
+for i in sorted(entry_dict.keys()):
+    entry_dict_sorted[i] = {}
+    entry_dict_sorted[i]['type'] = entry_dict[i]['type']
+    entry_dict_sorted[i]['fields'] = {}
+    for f in sorted(entry_dict[i]['fields'].keys()):
+        entry_dict_sorted[i]['fields'][f] = entry_dict[i]['fields'][f] 
+
+# dump the sorted elements to the outfile
+outf = open( outfile, 'w' )
+
+print( preamble.group(1) )
+outf.write( preamble.group(1) )
+for d in entry_dict_sorted:
+    outf.write( "struct %s %s[] = {\n" % (entry_dict_sorted[d]['type'],d) )
+    for f in entry_dict_sorted[d]['fields']:
+        outf.write( entry_dict_sorted[d]['fields'][f] + '\n' )
+
+    outf.write( "};\n" )
+
+outf.close()
+