[BUG]Sof-logger ABI version is mismatch with .ldc file built by xcc

Describe the bug
Error: abi version in /etc/sof/sof-apl.ldc file does not coincide with abi version used by logger.
logger ABI Version is 4:2:0
ldc_file ABI Version is 3:14:0

To Reproduce

Reboot via "sudo reboot"
Check with command "sudo sof-logger -l /etc/sof/sof-apl.ldc -t"

Reproduction Rate
100%

Expected behavior
Can see sof trace print normal

Impact
No sof trace because sof-logger ABI mismatch with .ldc file

Environment

Branch name and commit hash of the 2 repositories: sof (firmware/topology) and linux (kernel driver).
- Kernel: sof-dev ac3abc5
- SOF: master 6cd8acc
Name of the topology file
- Topology: sof-apl-pcm512x.tplg
Name of the platform(s) on which the bug is observed.
- Platform: APL-UP2 with codec PCM512X in I2S mode

check-sof-logger.sh.txt
dmesg.txt

on below platforms also has this paoblem
GLK Chrome with onboard codec DA7219 in I2S mode
CML laptop with onboard codec ALC3204 in HDA mode
ICL-U RVP with onboard codec ALC700 in HDA mode
TGL-U RVP with onboard codec ALC711 in HDA mode
CFL-S RVP with onboard codec ALC700 in HDA mode

@mmaka1

Bi-sect result: caused by master 332e150 , @mmaka1 please help to check this

Rebuild and upgrade the logger when you see a version mismatch.

@xiulipan

@xiulipan Are you sure you did a clean build? I am using xcc all the time. This is from my system:

$ hexdump -C build_fw_apl/sof-apl.ldc |head
00000000  4c 6f 67 73 00 00 00 20  54 27 01 00 4c 00 00 00  |Logs... T'..L...|
00000010  3c 00 00 00 01 00 04 00  01 00 00 00 00 00 00 00  |<...............|
00000020  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
00000030  00 00 30 34 36 38 61 00  00 20 00 04 00 00 00 00  |..0468a.. ......|

You should also see the current dbg abi version when running rimage as Liam mentioned:

Firmware manifest and signing completed !
fw abi main version: 3:14:0
fw abi dbg version: 4:2:0

@mmaka1

@mmaka1 @lgirdwood I did some dig in and find this is related to maybe compiler. Here is my analysis:

First to easily dump info from ELF file I add some debug patch to change the ENUM

+       SOF_IPC_EXT_WINDOW              = 0x1f,
+       SOF_IPC_EXT_CC_INFO             = 0x2e,
+       SOF_IPC_EXT_PROBE_INFO          = 0x3d,
+       SOF_IPC_EXT_USER_ABI_INFO       = 0x4c,
 };

Then I add some logs into rimage:

diff --git a/rimage/file_simple.c b/rimage/file_simple.c
index b68483f2ebab..cb0c4024dcd5 100644
--- a/rimage/file_simple.c
+++ b/rimage/file_simple.c
@@ -119,15 +119,21 @@ static int fw_version_copy(struct snd_sof_logs_header *header,
         */
        ext_hdr = buffer + ((struct sof_ipc_fw_ready *)buffer)->hdr.size;
        while ((uintptr_t)ext_hdr < (uintptr_t)buffer + section->size) {
+               fprintf(stdout, "PXL: fw abi dbg type: %x, %x\n", ext_hdr->type, ext_hdr->hdr.size);
                if (ext_hdr->type == SOF_IPC_EXT_USER_ABI_INFO) {
+                       fprintf(stdout, "PXL: fw abi dbg version: %x\n",
+                               ((struct sof_ipc_user_abi_version *)
+                                                       ext_hdr)->abi_dbg_version);
                        header->version.abi_version =

And here is what I got

Firmware manifest and signing completed !
fw abi main version: 3:14:0
PXL: fw abi dbg type: 1f, b8
PXL: fw abi dbg type: 2e, 50
PXL: fw abi dbg type: 0, 70000000
fw abi dbg version: 3:14:0
logs dictionary: size 75680

Dump from elf file
Note the size for the SOF_IPC_EXT_CC_INFO cc_versionis is 0x50, but actually the size is 0x4c. Not sure if this issue is that my CC_DESC here is RG-2017.8-linux and this is 16-char length (a space at the beginning)

00043c60  00 **50** 00 00 00 00 00 00  70 2e 00 00 00 0c 00 00  |.P......p.......|
00043c70  00 00 00 00 00 08 00 00  00 00 00 00 00 00 00 00  |................|
00043c80  00 00 00 00 00 00 00 00  00 58 43 43 00 00 00 00  |.........XCC....|
00043c90  00 00 00 00 00 00 00 00  00 4f 32 00 00 20 52 47  |.........O2.. RG|
00043ca0  2d 32 30 31 37 2e 38 2d  6c 69 6e 75 78 1c 00 00  |-2017.8-linux...| // here 1c is the start of the SOF_IPC_EXT_PROBE_INFO          
00043cb0  00 00 00 00 70 3d 00 00  00 00 00 00 00 00 00 00  |....p=..........|
00043cc0  00 00 00 00 00 00 00 00  00 10 00 00 00 00 00 00  |................|
00043cd0  70 4c 00 00 00 00 20 00  04 00 00 00 00 00 00 00  |pL.... .........|

@macchian

@macchian @lgirdwood @ktrzcinx I have a fix here to change flex array into static one to avoid issues between different compiler. #2522
I can also think solution like, move the flex array to the end of the fw_ready_metadata section.

diff --git a/src/ipc/CMakeLists.txt b/src/ipc/CMakeLists.txt
index 7423a2eac72b..cf579da81854 100644
--- a/src/ipc/CMakeLists.txt
+++ b/src/ipc/CMakeLists.txt
@@ -36,9 +36,9 @@ set_property(TARGET data_structs APPEND
             CC_OPTIMIZE_FLAGS="${optimization_flag}")

 add_local_sources(data_structs
-       cc_version.c
        probe_support.c
-       user_abi_version.c)
+       user_abi_version.c
+       cc_version.c)

 target_link_libraries(data_structs sof_options)
 target_link_libraries(sof_static_libraries INTERFACE data_structs)

rimage logs will becomes to

fw abi main version: 3:14:0
PXL: fw abi dbg type: 1f, b8
PXL: fw abi dbg type: 3d, 1c
PXL: fw abi dbg type: 4c, 10
PXL: fw abi dbg version: 4002000
fw abi dbg version: 4:2:0

As I see, there is a some limitation in XCC because RG-2017.8-linux string should be terminated with '\0' but I can't see it. Moreover this terminating zero is included in sizeof(CC_DESC), so there is also a solution like this:

const struct sof_ipc_cc_version cc_version
	__section(".fw_ready_metadata") = {
	.ext_hdr = {
		.hdr.cmd = SOF_IPC_FW_READY,
		.hdr.size = ALIGN_UP(sizeof(struct sof_ipc_cc_version)
				     + sizeof(CC_DESC) - 1, 4),

After this change:

fw abi main version: 3:14:0
PXL: fw abi dbg type: 1, b8
PXL: fw abi dbg type: 2, 4c
PXL: fw abi dbg type: 3, 1c
PXL: fw abi dbg type: 4, 10
PXL: fw abi dbg version: 4002000
fw abi dbg version: 4:2:0

Another possible issue in future is related with lack of memory alignment, SOF_IPC_EXT_PROBE_INFO starts just after SOF_IPC_EXT_CC_INFO, but there should be padding to '4

@kakulesza

@kakulesza Yep, I think this is because that the 0 at the end maybe merged by the all four sections in __section(".fw_ready_metadata") is merged in ELF.
Your solution will result a overflow as the CC_DESC STRING has no termination. It will result in issues in the later use.

@xiulipan

@xiulipan

Not sure if this issue is that my CC_DESC here is RG-2017.8-linux

It seems to work on my system since the beginning because I am compiling with the Windows version of the tools and this section looks like (note 4c and two extra zeroes because of shorter description):

 be0561b4 00000000 4c000000 00000070 02000000  ....L......p....
 be0561c4 0c000000 00000000 08000000 00000000  ................
 be0561d4 00000000 00000000 00000000 58434300  ............XCC.
 be0561e4 00000000 00000000 00000000 4f320000  ............O2..
 be0561f4 2052472d 32303137 2e382d77 696e0000   RG-2017.8-win..

@mmaka1

@mmaka1 Oh, that explain why we have different result. The issue seems to be more related to the cc_version.c .
@jajanusz @tlauda @lgirdwood Do you know any issue like this here?
It seems the flex length structure in ELF may not have the right layout as we expected. So I think we should avoid flex length structure in ELF file #2522

@jajanusz

@jajanusz @tlauda @lgirdwood Do you know any issue like this here?

Do you mean on linux version of compiler? For me it's weird that same version of compiler gives different output just cos of os, but anyway, it's better to w/a issue with having fixed sizes on problematic fields.

@jajanusz

@jajanusz After debug, it seems to be related to the flex length structure when assemble to section in ELF. The appending length will disappear.

Update with compare test between GCC and XCC, it seems to be some issue related to compiler linker.

diff --git a/src/arch/xtensa/include/arch/compiler_info.h b/src/arch/xtensa/include/arch/compiler_info.h
index d59da46a7a1f..fc4a3a8b6ce0 100644
--- a/src/arch/xtensa/include/arch/compiler_info.h
+++ b/src/arch/xtensa/include/arch/compiler_info.h
@@ -23,6 +23,6 @@
 #define CC_MINOR ((XTHAL_RELEASE_MAJOR % 1000) / 10)
 #define CC_MICRO XTHAL_RELEASE_MINOR
 #define CC_NAME "XCC"
-#define CC_DESC " " XCC_TOOLS_VERSION
+#define CC_DESC "0123456789ABCDEF"

 #endif /* __ARCH_COMPILER_INFO_H__ */

XCC got header size 0x50, actual binary size 0x4c

00043c60  00 50 00 00 00 00 00 00  70 02 00 00 00 0c 00 00  |.P......p.......|
00043c70  00 00 00 00 00 08 00 00  00 00 00 00 00 00 00 00  |................|
00043c80  00 00 00 00 00 00 00 00  00 58 43 43 00 00 00 00  |.........XCC....|
00043c90  00 00 00 00 00 00 00 00  00 4f 32 00 00 30 31 32  |.........O2..012|
00043ca0  33 34 35 36 37 38 39 41  42 43 44 45 46 1c 00 00  |3456789ABCDEF...|

GCC got both size is 0x50

0002a5e0  00 00 00 00 50 00 00 00  00 00 00 70 02 00 00 00  |....P......p....|
0002a5f0  0c 00 00 00 00 00 00 00  08 00 00 00 00 00 00 00  |................|
0002a600  00 00 00 00 00 00 00 00  00 00 00 00 58 43 43 00  |............XCC.|
0002a610  00 00 00 00 00 00 00 00  00 00 00 00 4f 32 00 00  |............O2..|
0002a620  30 31 32 33 34 35 36 37  38 39 41 42 43 44 45 46  |0123456789ABCDEF|
0002a630  00 00 00 00 1c 00 00 00  00 00 00 70 03 00 00 00  |...........p....|

worked around by #2647

Error not seen because PR #2647 merged，closed it

Navigation Menu

Repository navigation

[BUG]Sof-logger ABI version is mismatch with .ldc file built by xcc #2507

6 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[BUG]Sof-logger ABI version is mismatch with .ldc file built by xcc #2507

Description

Activity

Liviali155 commented on Mar 9, 2020

dengyangchao commented on Mar 9, 2020

mmaka1 commented on Mar 9, 2020

6 remaining items

mmaka1 commented on Mar 10, 2020

xiulipan commented on Mar 11, 2020

xiulipan commented on Mar 11, 2020

ktrzcinx commented on Mar 11, 2020

xiulipan commented on Mar 11, 2020

mmaka1 commented on Mar 11, 2020

xiulipan commented on Mar 12, 2020

jajanusz commented on Mar 13, 2020

xiulipan commented on Mar 16, 2020

xiulipan commented on Mar 16, 2020

mengdonglin commented on Mar 31, 2020

Liviali155 commented on Mar 31, 2020

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions