On 2025-08-13 09:41, Dmytro Prokopchuk1 wrote:
On 8/9/25 00:40, Nicola Vetrini wrote:
The rule states: "Initializer lists shall not contain persistent side
effects".
The specific way in which the 'mrs' instruction is used does not lead
to
visible side effects for the surrounding code.
Signed-off-by: Nicola Vetrini <nicola.vetr...@bugseng.com>
---
Not yet tested on the Xen ECLAIR runner, as the syntax used in the
deviation
is only supported after updating the runner.
What the tool is reporting is that due to the '=r' constraint and the
semantics of the instruction, there is the side effect of writing to
'_r',
but this is not observable outside the stmt expr. The deviation ends
up being
a bit too general for my taste, but the restriction on the actual
istruction
should be enough to limit applicability to cases that are arguably
safe in
practice.
An alternative approach would be represented by stating that side
effects in
'READ_SYSREG64' are safe, but this is not true in general.
---
automation/eclair_analysis/ECLAIR/deviations.ecl | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/automation/eclair_analysis/ECLAIR/deviations.ecl
b/automation/eclair_analysis/ECLAIR/deviations.ecl
index ec0cac797e5f..6b492e38505d 100644
--- a/automation/eclair_analysis/ECLAIR/deviations.ecl
+++ b/automation/eclair_analysis/ECLAIR/deviations.ecl
@@ -437,6 +437,10 @@ write or not"
# Series 13
#
+-doc_begin="Consider the asm instruction to read an Arm system
register to have no side effects."
+-asm_properties+={"asm(any())&&child(text,
ast_field(value,^mrs\\s+%0.*$))", {no_side_effect}}
+-doc_end
+
-doc_begin="All developers and reviewers can be safely assumed to be
well aware
of the short-circuit evaluation strategy of such logical operators."
-config=MC3A2.R13.5,reports+={disapplied,"any()"}
I think it's worth to add example of macro expansion in the commit
description or asm_properties doc:
uint64_t _r; asm volatile("mrs %0, ""TPIDR_EL2" : "=r" (_r));
This uses the 'mrs' instruction to read from the TPIDR_EL2 register.
While this read operation accesses a system register, reading itself
doesn't cause any persistent side effects, as no program state is
modified.
Definitely not in the -doc_begin, perhaps in deviations.rst, though in
reality it is a single case this currently applies to. Reading the
register is not the reason why this deviation was requested, but the
write with the "=r" constraint on "_r", as that is the side effect the
tool is pointing at.
--
Nicola Vetrini, B.Sc.
Software Engineer
BUGSENG (https://bugseng.com)
LinkedIn: https://www.linkedin.com/in/nicola-vetrini-a42471253
