Skip to contentSkip to conversation

Makefile: perform sanity checks on payload during build

PATCH New Tim Hutt 2 participants 4 messages Oct 4, 2024 – Oct 9, 2024 1 message not in the archive

lore
Share & tools
lore

Tim Hutt

author
#0 lore raw

PATCH 1/1 Makefile: perform sanity checks on payload during build

To make mistakes more obvious and debugging easier we can check that the payload is not an ELF file, and that the first byte is possibly the start of a valid instruction.

I would have preferred to not do this check in Bash but I didn't want to introduce any additional dependencies, and there isn't a proper language already in use.

I also made it an #error to not define FW_PAYLOAD_PATH since the build system will always do that. If the user doesn't specify one a default file is used.

---
 firmware/external_deps.mk |  3 ++-
 firmware/fw_payload.S     |  6 ++----
 scripts/check-payload.sh  | 43 +++++++++++++++++++++++++++++++++++++++
 3 files changed, 47 insertions(+), 5 deletions(-)
 create mode 100755 scripts/check-payload.sh
firmware/external_deps.mk +2 -1
index 6264005..d4c480c 100644
@@ -10,5 +10,6 @@
 $(platform_build_dir)/firmware/fw_dynamic.o: $(FW_FDT_PATH)
 $(platform_build_dir)/firmware/fw_jump.o: $(FW_FDT_PATH)
 $(platform_build_dir)/firmware/fw_payload.o: $(FW_FDT_PATH)
-
+	scripts/check-payload.sh $(FW_PAYLOAD_PATH_FINAL)
+	$(call compile_as,$@,$<)
 $(platform_build_dir)/firmware/fw_payload.o: $(FW_PAYLOAD_PATH_FINAL)
firmware/fw_payload.S +2 -4
index 3c8433e..0b0f342 100644
@@ -94,8 +94,6 @@ fw_options:
 	.globl payload_bin
 payload_bin:
 #ifndef FW_PAYLOAD_PATH
-	wfi
-	j	payload_bin
-#else
-	.incbin	FW_PAYLOAD_PATH
+#error FW_PAYLOAD_PATH should always be set by the build system
 #endif
+	.incbin	FW_PAYLOAD_PATH
scripts/check-payload.sh new file +43 -0

43 lines not on this page · raw message

Signature · 1 line
2.39.2
#1 lore raw

PATCH 1/1 Makefile: perform sanity checks on payload during build

Hi Tim,

Quoted 5 lines · Tim Hutt

To make mistakes more obvious and debugging easier we can check that the payload is not an ELF file, and that the first byte is possibly the start of a valid instruction.

I would have preferred to not do this check in Bash but I didn't want to introduce any additional dependencies, and there isn't a proper language already in use.

I also made it an #error to not define FW_PAYLOAD_PATH since the build system will always do that. If the user doesn't specify one a default file is used.

You need a Signed-off-by. Please also wrap the commit message to 72 characters.

Quoted 12 lines
---
 firmware/external_deps.mk |  3 ++-
 firmware/fw_payload.S     |  6 ++----
 scripts/check-payload.sh  | 43 +++++++++++++++++++++++++++++++++++++++
 3 files changed, 47 insertions(+), 5 deletions(-)
 create mode 100755 scripts/check-payload.sh
firmware/external_deps.mk +0 -0
index 6264005..d4c480c 100644

firmware/external_deps.mk line 14

Quoted 6 lines
firmware/external_deps.mk +2 -1
@@ -10,5 +10,6 @@
 $(platform_build_dir)/firmware/fw_dynamic.o: $(FW_FDT_PATH)
 $(platform_build_dir)/firmware/fw_jump.o: $(FW_FDT_PATH)
 $(platform_build_dir)/firmware/fw_payload.o: $(FW_FDT_PATH)
-
+	scripts/check-payload.sh $(FW_PAYLOAD_PATH_FINAL)
+	$(call compile_as,$@,$<)

I would suggest leaving the recipe for fw_payload.o alone, and using a separate, possibly phony, target for this check. It's surprising to have the recipe for assembling this one file defined in a separate file.

Quoted 22 lines
firmware/external_deps.mk +0 -0
 $(platform_build_dir)/firmware/fw_payload.o: $(FW_PAYLOAD_PATH_FINAL)
firmware/fw_payload.S +2 -4
index 3c8433e..0b0f342 100644
@@ -94,8 +94,6 @@ fw_options:
 	.globl payload_bin
 payload_bin:
 #ifndef FW_PAYLOAD_PATH
-	wfi
-	j	payload_bin
-#else
-	.incbin	FW_PAYLOAD_PATH
+#error FW_PAYLOAD_PATH should always be set by the build system
 #endif
+	.incbin	FW_PAYLOAD_PATH
scripts/check-payload.sh new file +0 -0
new file mode 100755
index 0000000..5752497

scripts/check-payload.sh line 1

Quoted 1 line
scripts/check-payload.sh new file +1 -0
@@ -0,0 +1,43 @@
+#!/usr/bin/env bash

With the below changes this script can be POSIX-compatible and use /bin/sh.

Quoted 4 lines
scripts/check-payload.sh new file +4 -0
+
+# Simple script to try and detect invalid payloads. The payload should be a flat
+# image and the first bytes should be valid instructions - OpenSBI will jump
+# to the start of the file. It shouldn't be an ELF file.

scripts/check-payload.sh line 15

Quoted 10 lines
scripts/check-payload.sh new file +10 -0
+#
+# Bash is a terrible language for this purpose but we can make it work.
+
+if [ -z "$1" ]; then
+    echo "Usage: $0 <file>"
+    exit 1
+fi
+
+# Check the payload is not an ELF file.
+readelf -h "$1" >/dev/null 2>&1 && {

This should get $READELF from Makefile; see how that file sets OBJCOPY.

scripts/check-payload.sh line 24

Quoted 9 lines
scripts/check-payload.sh new file +9 -0
+    echo "Error: The payload is an ELF file; it should be a flat executable instead."
+    exit 1
+}
+
+# Check the first bytes are a valid RISC-V instruction. Since we don't know
+# what is valid we can only do limited checks but this will check that it is at
+# least a 16 or 32-bit instruction. Larger instructions are currently unused.
+first_byte="$(head -c 1 "$1")"
+LC_ALL=C printf -v first_byte_dec %d \'$first_byte

This can be done portably with od(1):

first_byte=$(od -An -N1 -td1 "$1")

scripts/check-payload.sh line 37

Quoted 13 lines
scripts/check-payload.sh new file +13 -0
+
+# for byte in range(256):
+#     if not ((byte & 0b11) != 0b11 or (byte & 0b11100) != 0b11100):
+#         print(byte)
+[ \
+    "$first_byte_dec" == 31 -o \
+    "$first_byte_dec" == 63 -o \
+    "$first_byte_dec" == 95 -o \
+    "$first_byte_dec" == 127 -o \
+    "$first_byte_dec" == 159 -o \
+    "$first_byte_dec" == 191 -o \
+    "$first_byte_dec" == 223 -o \
+    "$first_byte_dec" == 255 \

You can use arithmetic expansion here (which is also POSIX):

[ $((first_byte & 31)) -eq 31 ]

scripts/check-payload.sh line 43

Quoted 6 lines
scripts/check-payload.sh new file +6 -0
+] && {
+    echo "Error: The payload's first byte is not a 16- or 32-bit RISC-V instruction. The payload should be a flat executable."
+    exit 1
+}
+
+exit 0

No need for an explicit "exit 0" at the end.

Regards,
Samuel

A message that is not in the archive (sent off-list or not yet received): CAKuVd4DNsOsLVp-R9kkTrtAeEjPkA-UwNo40CZOy6JPUrjHspg@mail.gmail.com

author
#3 lore raw

PATCH 1/1 Makefile: perform sanity checks on payload during build

Quoted 4 lines

I would suggest leaving the recipe for fw_payload.o alone, and using a separate, possibly phony, target for this check. It's surprising to have the recipe for assembling this one file defined in a separate file.

I think the check then needs to write to a file... something like this?

$(platform_build_dir)/firmware/fw_payload.o: $(FW_FDT_PATH) $(platform_build_dir)/firmware/fw_payload.o: $(FW_PAYLOAD_PATH_FINAL) $(platform_build_dir)/firmware/fw_payload.o:
$(platform_build_dir)/firmware/payload_check.stamp $(platform_build_dir)/firmware/payload_check.stamp: $(FW_PAYLOAD_PATH_FINAL)

    @scripts/check-payload.sh $<
    @touch $@
Quoted 1 line

This can be done portably with od(1):

Ah cool, I didn't know about od (and googling everyone uses xxi and hacky sed pipelines), thanks!

Quoted 1 line

No need for an explicit "exit 0" at the end.

Actually there is because the exit code will be the value from the last command, which is false with the way I've written the script. Try this script:

#!/bin/bash
false && {
echo "Failed"
exit 1
}

It will not print "Failed" and it will also exit with a failure exit code. I have change the script so it's like this:

true || { ... fail ... }

But maybe it's worth keeping the explicit exit 0 just because shell is crazy like this?

#4 lore raw

PATCH 1/1 Makefile: perform sanity checks on payload during build

Hi Tim,

Quoted 6 lines · Tim Hutt
Quoted 3 lines

I would suggest leaving the recipe for fw_payload.o alone, and using a separate, possibly phony, target for this check. It's surprising to have the recipe for assembling this one file defined in a separate file.

I think the check then needs to write to a file... something like this?

Quoted 7 lines

$(platform_build_dir)/firmware/fw_payload.o: $(FW_FDT_PATH) $(platform_build_dir)/firmware/fw_payload.o: $(FW_PAYLOAD_PATH_FINAL) $(platform_build_dir)/firmware/fw_payload.o:
$(platform_build_dir)/firmware/payload_check.stamp $(platform_build_dir)/firmware/payload_check.stamp: $(FW_PAYLOAD_PATH_FINAL)

    @scripts/check-payload.sh $<
    @touch $@

Yes, that would work.

Quoted 17 lines
Quoted 1 line

This can be done portably with od(1):

Ah cool, I didn't know about od (and googling everyone uses xxi and hacky sed pipelines), thanks!

Quoted 1 line

No need for an explicit "exit 0" at the end.

Actually there is because the exit code will be the value from the last command, which is false with the way I've written the script. Try this script:

#!/bin/bash
false && {
echo "Failed"
exit 1
}

Quoted 7 lines

It will not print "Failed" and it will also exit with a failure exit code. I have change the script so it's like this:

true || { ... fail ... }

But maybe it's worth keeping the explicit exit 0 just because shell is crazy like this?

You're right. Sorry for the misinformation. Keeping the exit 0 is fine, or you could put the touch command inside the script, so the last command is something that should succeed.

Regards,
Samuel