target/ppc: fix unreachable code in do_ldst_quad()
v2 cover target/ppc: fix unreachable code in do_ldst_quad()
changes from v1:
- do not use ifdefs
- returned to the code that was in placed before the macro
- v1 link: https://lists.gnu.org/archive/html/qemu-devel/2022-07/msg03697.html
Daniel Henrique Barboza (1):
target/ppc: fix unreachable code in do_ldst_quad()
target/ppc/translate/fixedpoint-impl.c.inc | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
Signature · 1 line
2.36.1
PATCH v2 1/1 target/ppc: fix unreachable code in do_ldst_quad()
Coverity reports that commit fc34e81acd51 ("target/ppc: add macros to check privilege level") turned the following code unreachable:
if (!prefixed && !(ctx->insns_flags2 & PPC2_LSQ_ISA207)) {
/* lq and stq were privileged prior to V. 2.07 */
REQUIRE_SV(ctx);Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
CID 1490757: Control flow issues (UNREACHABLE) This code cannot be reached: "if (
ctx->le_mode) {
if (ctx->le_mode) {
gen_align_no_le(ctx);
return true;
}
}This happens because the macro REQUIRE_SV(), in CONFIG_USER_MODE, will always result in a 'return true' statement. In fact, all REQUIRE_* macros for target/ppc/translate.c behave the same way: if a condition isn't met, an exception is generated and a 'return' statement is issued.
The difference is that all other callers are using it in insns that are not implemented in user mode. do_ldst_quad(), on the other hand, is user mode compatible.
Fixes include wrapping these lines in "if !defined(CONFIG_USER_MODE)", making it explicit that these lines are not user mode anymore. Another fix would be, for example, to change REQUIRE_SV() to not issue a 'return' and check if we're running in privileged mode or not by hand, but this would change all other callers of the macro that are using it in an adequate manner.
The code that was in place before fc34e81acd51 was good enough, so let's go back to that: open code the ctx->pr condition and fire the exception if we're not privileged. The difference from the code back then to what we're doing now is an 'unlikely' compiler hint to ctx->pr and the use of gen_priv_opc() instead of gen_priv_exception().
--- target/ppc/translate/fixedpoint-impl.c.inc | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
target/ppc/translate/fixedpoint-impl.c.inc +5 -2
@@ -79,8 +79,11 @@ static bool do_ldst_quad(DisasContext *ctx, arg_D *a, bool store, bool prefixed) REQUIRE_INSNS_FLAGS(ctx, 64BX); if (!prefixed && !(ctx->insns_flags2 & PPC2_LSQ_ISA207)) { - /* lq and stq were privileged prior to V. 2.07 */ - REQUIRE_SV(ctx); + if (unlikely(ctx->pr)) { + /* lq and stq were privileged prior to V. 2.07 */ + gen_priv_opc(ctx); + return true; + } if (ctx->le_mode) { gen_align_no_le(ctx);
Signature · 1 line
2.36.1
Quoted 63 lines · Daniel Henrique Barboza
Coverity reports that commit
fc34e81acd51("target/ppc: add macros to check privilege level") turned the following code unreachable:if (!prefixed && !(ctx->insns_flags2 & PPC2_LSQ_ISA207)) { /* lq and stq were privileged prior to V. 2.07 */ REQUIRE_SV(ctx);Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
CID 1490757: Control flow issues (UNREACHABLE) This code cannot be reached: "if (
ctx->le_mode) {if (ctx->le_mode) { gen_align_no_le(ctx); return true; } }This happens because the macro
REQUIRE_SV(), inCONFIG_USER_MODE, will always result in a 'return true' statement. In fact, all REQUIRE_* macros fortarget/ppc/translate.cbehave the same way: if a condition isn't met, an exception is generated and a 'return' statement is issued.The difference is that all other callers are using it in insns that are not implemented in user mode.
do_ldst_quad(), on the other hand, is user mode compatible.Fixes include wrapping these lines in "if !defined(CONFIG_USER_MODE)", making it explicit that these lines are not user mode anymore. Another fix would be, for example, to change
REQUIRE_SV()to not issue a 'return' and check if we're running in privileged mode or not by hand, but this would change all other callers of the macro that are using it in an adequate manner.The code that was in place before
fc34e81acd51was good enough, so let's go back to that: open code thectx->prcondition and fire the exception if we're not privileged. The difference from the code back then to what we're doing now is an 'unlikely' compiler hint toctx->prand the use ofgen_priv_opc()instead ofgen_priv_exception().Fixes Coverity CID 1490757Cc Matheus Ferst <[email protected]>Signed-off-by Daniel Henrique Barboza <[email protected]>--- target/ppc/translate/fixedpoint-impl.c.inc | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)target/ppc/translate/fixedpoint-impl.c.inc +5
-3@@ -79,8 +79,11 @@ static bool do_ldst_quad(DisasContext *ctx, arg_D *a, bool store, bool prefixed) REQUIRE_INSNS_FLAGS(ctx, 64BX); if (!prefixed && !(ctx->insns_flags2 & PPC2_LSQ_ISA207)) { - /* lq and stq were privileged prior to V. 2.07 */ - REQUIRE_SV(ctx); + if (unlikely(ctx->pr)) { + /* lq and stq were privileged prior to V. 2.07 */ + gen_priv_opc(ctx); + return true; + } if (ctx->le_mode) { gen_align_no_le(ctx); --
Quoted 2 lines
2.36.1
Since the remaining code in this branch is dead code in user-mode, I'd personally prefer the v1 approach, but the difference is unlikely to have any meaningful impact, so either way is good.
Thanks,
Matheus K. Ferst
Instituto de Pesquisas ELDORADO <http://www.eldorado.org.br/>
Analista de Software
Aviso Legal - Disclaimer <https://www.eldorado.org.br/disclaimer.html>
Quoted 16 lines · Daniel Henrique Barboza
Coverity reports that commit
fc34e81acd51("target/ppc: add macros to check privilege level") turned the following code unreachable:if (!prefixed && !(ctx->insns_flags2 & PPC2_LSQ_ISA207)) { /* lq and stq were privileged prior to V. 2.07 */ REQUIRE_SV(ctx);Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
CID 1490757: Control flow issues (UNREACHABLE) This code cannot be reached: "if (
ctx->le_mode) {if (ctx->le_mode) { gen_align_no_le(ctx); return true; } }
Quoted 8 lines
This happens because the macro
REQUIRE_SV(), inCONFIG_USER_MODE, will always result in a 'return true' statement. In fact, all REQUIRE_* macros fortarget/ppc/translate.cbehave the same way: if a condition isn't met, an exception is generated and a 'return' statement is issued.The difference is that all other callers are using it in insns that are not implemented in user mode.
do_ldst_quad(), on the other hand, is user mode compatible.
This is a Coverity false positive, and I'd already marked it that way in the Coverity UI back on the 20th. Coverity gets confused sometimes by ifdeffery.
So you don't need this patch, unless you think the code is genuinely better (more readable to humans, etc) this way.
thanks
-- PMM
Quoted 27 lines · Peter Maydell
Quoted 16 lines · Daniel Henrique Barboza
Coverity reports that commit
fc34e81acd51("target/ppc: add macros to check privilege level") turned the following code unreachable:if (!prefixed && !(ctx->insns_flags2 & PPC2_LSQ_ISA207)) { /* lq and stq were privileged prior to V. 2.07 */ REQUIRE_SV(ctx);Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
CID 1490757: Control flow issues (UNREACHABLE) This code cannot be reached: "if (
ctx->le_mode) {if (ctx->le_mode) { gen_align_no_le(ctx); return true; } }Quoted 8 lines
This happens because the macro
REQUIRE_SV(), inCONFIG_USER_MODE, will always result in a 'return true' statement. In fact, all REQUIRE_* macros fortarget/ppc/translate.cbehave the same way: if a condition isn't met, an exception is generated and a 'return' statement is issued.The difference is that all other callers are using it in insns that are not implemented in user mode.
do_ldst_quad(), on the other hand, is user mode compatible.
Quoted 6 lines
This is a Coverity false positive, and I'd already marked it that way in the Coverity UI back on the 20th. Coverity gets confused sometimes by ifdeffery.
So you don't need this patch, unless you think the code is genuinely better (more readable to humans, etc) this way.
The idea was to make Coverity happy. If Coverity is already happy then let's drop this patch - there's no particular improvement made here that justifies it.
Thanks,
Daniel
Quoted 3 lines
thanks
-- PMM