[PATCH] tools/perf: Fix dso kernel load and symbol process to correctly map dso to its long_name, type and adjust_symbols

Athira Rajeev atrajeev at linux.vnet.ibm.com
Fri Aug 11 15:15:46 AEST 2023

Test "object cocde reading" fails sometimes for kernel address
as below:

    Reading object code for memory address: 0xc000000000004c3c
    File is: [kernel.kallsyms]
    On file address is: 0x14c3c
    dso__data_read_offset failed
    test child finished with -1
    ---- end ----
    Object code reading: FAILED!

Here the dso__data_read_offset fails for symbol address
0xc000000000004c3c. This is because, the dso long_name here
is [kernel.kallsyms] and hence open_dso fails to open this
file. There is an incorrect dso to map handling here. The
key points here is:
- dso long_name is set to [kernel.kallsyms]. This file is
  not present and hence returns error
- DSO binary type is set to DSO_BINARY_TYPE__NOT_FOUND
- dso adjust_symbols is set to zero

In the end dso__data_read_offset() returns -1 and the address
0x14c3c can not be resolved. Hence the test fails. But the
address actually maps to kernel dso

    # objdump -z -d --start-address=0xc000000000004c3c --stop-address=0xc000000000004cbc /home/athira/linux/vmlinux

    /home/athira/linux/vmlinux:     file format elf64-powerpcle

    Disassembly of section .head.text:

    c000000000004c3c <exc_virt_0x4c00_system_call+0x3c>:
    c000000000004c3c:	a6 02 9b 7d 	mfsrr1  r12
    c000000000004c40:	78 13 42 7c 	mr      r2,r2
    c000000000004c44:	18 00 4d e9 	ld      r10,24(r13)
    c000000000004c48:	60 c6 4a 61 	ori     r10,r10,50784
    c000000000004c4c:	a6 03 49 7d 	mtctr   r10

Fix the dso__process_kernel_symbol function to set the
binary_type and adjust_symbols. adjust_symbols is used by
function "map__rip_2objdump" which converts symbol start
address to objdump address. Also set the dso long_name during
dso__load_vmlinux function.

Suggested-by: Adrian Hunter <adrian.hunter at intel.com>
Signed-off-by: Athira Rajeev <atrajeev at linux.vnet.ibm.com>
Note: Found similar discussion here in thread:
      where Adrian proposed the fix, but looks like this was
      not added to the perf. Hence addeed Suggested-by from Adrian.

      Additional to the fix proposed by Adrian, the patch also
      adds setting of adjust_symbols which is needed for
      map__rip_2objdump to convert symbol start to objdump address.

 tools/perf/util/symbol-elf.c |  2 ++
 tools/perf/util/symbol.c     | 15 ++++++++++-----
 2 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/tools/perf/util/symbol-elf.c b/tools/perf/util/symbol-elf.c
index 252d26a59e64..9e7eeaf616b8 100644
--- a/tools/perf/util/symbol-elf.c
+++ b/tools/perf/util/symbol-elf.c
@@ -1440,6 +1440,8 @@ static int dso__process_kernel_symbol(struct dso *dso, struct map *map,
 		curr_dso->kernel = dso->kernel;
 		curr_dso->long_name = dso->long_name;
 		curr_dso->long_name_len = dso->long_name_len;
+		curr_dso->binary_type = dso->binary_type;
+		curr_dso->adjust_symbols = dso->adjust_symbols;
 		curr_map = map__new2(start, curr_dso);
 		if (curr_map == NULL)
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index f849f9ef68e6..3f36675b7c8f 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -2204,15 +2204,20 @@ int dso__load_vmlinux(struct dso *dso, struct map *map,
 	if (symsrc__init(&ss, dso, symfs_vmlinux, symtab_type))
 		return -1;
+	/*
+	 * dso__load_sym() may copy 'dso' which will result in the copies having
+	 * an incorrect long name unless we set it here first.
+	 */
+	dso__set_long_name(dso, vmlinux, vmlinux_allocated);
+	if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
+		dso->binary_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
+	else
+		dso->binary_type = DSO_BINARY_TYPE__VMLINUX;
 	err = dso__load_sym(dso, map, &ss, &ss, 0);
 	if (err > 0) {
-		if (dso->kernel == DSO_SPACE__KERNEL_GUEST)
-			dso->binary_type = DSO_BINARY_TYPE__GUEST_VMLINUX;
-		else
-			dso->binary_type = DSO_BINARY_TYPE__VMLINUX;
-		dso__set_long_name(dso, vmlinux, vmlinux_allocated);
 		pr_debug("Using %s for symbols\n", symfs_vmlinux);

More information about the Linuxppc-dev mailing list