aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorNicholas Piggin <npiggin@gmail.com>2016-11-23 11:41:37 -0500
committerMichal Marek <mmarek@suse.com>2016-11-29 09:53:19 -0500
commit7e2b37c971a2a20ec8a311a195a626c16c774031 (patch)
tree7a6758f281848f72ee464d0a7f92f3aebb849225
parent1001354ca34179f3db924eb66672442a173147dc (diff)
kbuild: kallsyms allow 3-pass generation if symbols size has changed
kallsyms generation is not foolproof, due to some linkers adding symbols (e.g., branch trampolines) when a binary size changes. Have it attempt a 3rd pass automatically if the kallsyms size changes in the 2nd pass. This allows powerpc64 allyesconfig to build without adding another pass when it's not required. This can be solved other ways by directing the linker not to add labels on branch stubs, or to move kallsyms near the end of the image. The former is undesirable for debugging/tracing, and the latter is a more significant change that requires more testing and review. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Marek <mmarek@suse.com>
-rwxr-xr-xscripts/link-vmlinux.sh19
1 files changed, 13 insertions, 6 deletions
diff --git a/scripts/link-vmlinux.sh b/scripts/link-vmlinux.sh
index f742c65108b9..6b94fe4e932d 100755
--- a/scripts/link-vmlinux.sh
+++ b/scripts/link-vmlinux.sh
@@ -246,10 +246,14 @@ if [ -n "${CONFIG_KALLSYMS}" ]; then
246 # the right size, but due to the added section, some 246 # the right size, but due to the added section, some
247 # addresses have shifted. 247 # addresses have shifted.
248 # From here, we generate a correct .tmp_kallsyms2.o 248 # From here, we generate a correct .tmp_kallsyms2.o
249 # 2a) We may use an extra pass as this has been necessary to 249 # 3) That link may have expanded the kernel image enough that
250 # woraround some alignment related bugs. 250 # more linker branch stubs / trampolines had to be added, which
251 # KALLSYMS_EXTRA_PASS=1 is used to trigger this. 251 # introduces new names, which further expands kallsyms. Do another
252 # 3) The correct ${kallsymso} is linked into the final vmlinux. 252 # pass if that is the case. In theory it's possible this results
253 # in even more stubs, but unlikely.
254 # KALLSYMS_EXTRA_PASS=1 may also used to debug or work around
255 # other bugs.
256 # 4) The correct ${kallsymso} is linked into the final vmlinux.
253 # 257 #
254 # a) Verify that the System.map from vmlinux matches the map from 258 # a) Verify that the System.map from vmlinux matches the map from
255 # ${kallsymso}. 259 # ${kallsymso}.
@@ -265,8 +269,11 @@ if [ -n "${CONFIG_KALLSYMS}" ]; then
265 vmlinux_link .tmp_kallsyms1.o .tmp_vmlinux2 269 vmlinux_link .tmp_kallsyms1.o .tmp_vmlinux2
266 kallsyms .tmp_vmlinux2 .tmp_kallsyms2.o 270 kallsyms .tmp_vmlinux2 .tmp_kallsyms2.o
267 271
268 # step 2a 272 # step 3
269 if [ -n "${KALLSYMS_EXTRA_PASS}" ]; then 273 size1=$(stat -c "%s" .tmp_kallsyms1.o)
274 size2=$(stat -c "%s" .tmp_kallsyms2.o)
275
276 if [ $size1 -ne $size2 ] || [ -n "${KALLSYMS_EXTRA_PASS}" ]; then
270 kallsymso=.tmp_kallsyms3.o 277 kallsymso=.tmp_kallsyms3.o
271 kallsyms_vmlinux=.tmp_vmlinux3 278 kallsyms_vmlinux=.tmp_vmlinux3
272 279