| Message ID | 20190806193434.965-1-cai@lca.pw (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] arm64/cache: fix -Woverride-init compiler warnings | expand |
On Tue, Aug 06, 2019 at 03:34:34PM -0400, Qian Cai wrote: > The commit 155433cb365e ("arm64: cache: Remove support for ASID-tagged > VIVT I-caches") introduced some compiation warnings from GCC (and > Clang), > > arch/arm64/kernel/cpuinfo.c:38:26: warning: initialized field > overwritten [-Woverride-init] > [ICACHE_POLICY_VIPT] = "VIPT", > ^~~~~~ > arch/arm64/kernel/cpuinfo.c:38:26: note: (near initialization for > 'icache_policy_str[2]') > arch/arm64/kernel/cpuinfo.c:39:26: warning: initialized field > overwritten [-Woverride-init] > [ICACHE_POLICY_PIPT] = "PIPT", > ^~~~~~ > arch/arm64/kernel/cpuinfo.c:39:26: note: (near initialization for > 'icache_policy_str[3]') > arch/arm64/kernel/cpuinfo.c:40:27: warning: initialized field > overwritten [-Woverride-init] > [ICACHE_POLICY_VPIPT] = "VPIPT", > ^~~~~~~ > arch/arm64/kernel/cpuinfo.c:40:27: note: (near initialization for > 'icache_policy_str[0]') > > because it initializes icache_policy_str[0 ... 3] twice. Since the array > is only used in cpuinfo_detect_icache_policy(), fix it by initializing > a specific field there just before using. > > Fixes: 155433cb365e ("arm64: cache: Remove support for ASID-tagged VIVT I-caches") > Signed-off-by: Qian Cai <cai@lca.pw> Rather than trying to "fix" correct code like this (and making it harder to read), could you instead look into where/whether the warning is actually useful? I had a look at an arm64 defconfig, where I see: [mark@lakrids:~/src/linux]% grep override-init err.log | grep -o '^[^[:space:]:]\+' | sort | uniq -c | sort -rn 434 arch/arm64/kernel/sys32.c // all benign 291 arch/arm64/kernel/sys.c // all benign 48 ./arch/arm64/include/asm/perf_event.h // all benign 37 arch/arm64/kernel/traps.c // all benign 21 arch/arm64/kvm/handle_exit.c // all benign 12 drivers/ata/ahci.h // all benign 6 arch/arm64/kernel/perf_event.c // all benign 4 kernel/time/hrtimer.c // all benign 3 arch/arm64/kernel/cpuinfo.c // all benign 2 drivers/pinctrl/tegra/pinctrl-tegra194.c // unclear to me 1 ./include/linux/blkdev.h // all benign 1 drivers/gpu/drm/sun4i/sun4i_drv.c // all benign 1 drivers/ata/sata_sil24.c // all benign 1 ./arch/arm64/include/asm/mmu.h // all benign ... so that's 862 warnings where at least 860 are unhelpful (and I suspect those last two are also fine, but I haven't untangled the set of macros). Given that, what's the point in enabling this warning? It forces us to write worse code that's harder to maintain, and it doesn't spot anything useful. > --- > > v2: Initialize a specific field in cpuinfo_detect_icache_policy(). > > arch/arm64/kernel/cpuinfo.c | 8 ++++---- > 1 file changed, 4 insertions(+), 4 deletions(-) > > diff --git a/arch/arm64/kernel/cpuinfo.c b/arch/arm64/kernel/cpuinfo.c > index 876055e37352..a0c495a3f4fd 100644 > --- a/arch/arm64/kernel/cpuinfo.c > +++ b/arch/arm64/kernel/cpuinfo.c > @@ -34,10 +34,7 @@ DEFINE_PER_CPU(struct cpuinfo_arm64, cpu_data); > static struct cpuinfo_arm64 boot_cpu_data; > > static char *icache_policy_str[] = { > - [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN", > - [ICACHE_POLICY_VIPT] = "VIPT", > - [ICACHE_POLICY_PIPT] = "PIPT", > - [ICACHE_POLICY_VPIPT] = "VPIPT", > + [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN" > }; > > unsigned long __icache_flags; > @@ -310,13 +307,16 @@ static void cpuinfo_detect_icache_policy(struct cpuinfo_arm64 *info) > > switch (l1ip) { > case ICACHE_POLICY_PIPT: > + icache_policy_str[ICACHE_POLICY_PIPT] = "PIPT"; > break; > case ICACHE_POLICY_VPIPT: > + icache_policy_str[ICACHE_POLICY_VPIPT] = "VPIPT"; > set_bit(ICACHEF_VPIPT, &__icache_flags); > break; > default: > /* Fallthrough */ > case ICACHE_POLICY_VIPT: > + icache_policy_str[ICACHE_POLICY_VIPT] = "VIPT"; > /* Assume aliasing */ > set_bit(ICACHEF_ALIASING, &__icache_flags); NAK to this. Please leave this code as-is. Thanks, Mark. > } > -- > 2.20.1 (Apple Git-117) >
On Tue, Aug 06, 2019 at 03:34:34PM -0400, Qian Cai wrote: > diff --git a/arch/arm64/kernel/cpuinfo.c b/arch/arm64/kernel/cpuinfo.c > index 876055e37352..a0c495a3f4fd 100644 > --- a/arch/arm64/kernel/cpuinfo.c > +++ b/arch/arm64/kernel/cpuinfo.c > @@ -34,10 +34,7 @@ DEFINE_PER_CPU(struct cpuinfo_arm64, cpu_data); > static struct cpuinfo_arm64 boot_cpu_data; > > static char *icache_policy_str[] = { > - [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN", > - [ICACHE_POLICY_VIPT] = "VIPT", > - [ICACHE_POLICY_PIPT] = "PIPT", > - [ICACHE_POLICY_VPIPT] = "VPIPT", > + [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN" > }; > > unsigned long __icache_flags; > @@ -310,13 +307,16 @@ static void cpuinfo_detect_icache_policy(struct cpuinfo_arm64 *info) > > switch (l1ip) { > case ICACHE_POLICY_PIPT: > + icache_policy_str[ICACHE_POLICY_PIPT] = "PIPT"; > break; > case ICACHE_POLICY_VPIPT: > + icache_policy_str[ICACHE_POLICY_VPIPT] = "VPIPT"; > set_bit(ICACHEF_VPIPT, &__icache_flags); > break; > default: > /* Fallthrough */ > case ICACHE_POLICY_VIPT: > + icache_policy_str[ICACHE_POLICY_VIPT] = "VIPT"; > /* Assume aliasing */ > set_bit(ICACHEF_ALIASING, &__icache_flags); I still think this is worse than the code in mainline. I don't think -Woverride-init should warn when overriding a field from a GCC range designated initialiser, since it makes them considerably less useful imo. Will
> On Aug 7, 2019, at 6:56 AM, Will Deacon <will@kernel.org> wrote: > > On Tue, Aug 06, 2019 at 03:34:34PM -0400, Qian Cai wrote: >> diff --git a/arch/arm64/kernel/cpuinfo.c b/arch/arm64/kernel/cpuinfo.c >> index 876055e37352..a0c495a3f4fd 100644 >> --- a/arch/arm64/kernel/cpuinfo.c >> +++ b/arch/arm64/kernel/cpuinfo.c >> @@ -34,10 +34,7 @@ DEFINE_PER_CPU(struct cpuinfo_arm64, cpu_data); >> static struct cpuinfo_arm64 boot_cpu_data; >> >> static char *icache_policy_str[] = { >> - [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN", >> - [ICACHE_POLICY_VIPT] = "VIPT", >> - [ICACHE_POLICY_PIPT] = "PIPT", >> - [ICACHE_POLICY_VPIPT] = "VPIPT", >> + [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN" >> }; >> >> unsigned long __icache_flags; >> @@ -310,13 +307,16 @@ static void cpuinfo_detect_icache_policy(struct cpuinfo_arm64 *info) >> >> switch (l1ip) { >> case ICACHE_POLICY_PIPT: >> + icache_policy_str[ICACHE_POLICY_PIPT] = "PIPT"; >> break; >> case ICACHE_POLICY_VPIPT: >> + icache_policy_str[ICACHE_POLICY_VPIPT] = "VPIPT"; >> set_bit(ICACHEF_VPIPT, &__icache_flags); >> break; >> default: >> /* Fallthrough */ >> case ICACHE_POLICY_VIPT: >> + icache_policy_str[ICACHE_POLICY_VIPT] = "VIPT"; >> /* Assume aliasing */ >> set_bit(ICACHEF_ALIASING, &__icache_flags); > > I still think this is worse than the code in mainline. I don't think > -Woverride-init should warn when overriding a field from a GCC range > designated initialiser, since it makes them considerably less useful > imo. Unfortunately, compiler people are moving into a different direction as Clang would warn those kind of usage too. It actually prove that those warnings are useful to find real issues. See, Fae5e033d65a (“mfd: rk808: Fix RK818_IRQ_DISCHG_ILIM initializer”) 32df34d875bb (“[media] rc: img-ir: jvc: Remove unused no-leader timings”) Especially, to find redundant initializations in large structures. e.g., e6ea0b917875 (“[media] dvb_frontend: Don't declare values twice at a table”)
On Wed, Aug 07, 2019 at 07:50:43AM -0400, Qian Cai wrote: > > > > On Aug 7, 2019, at 6:56 AM, Will Deacon <will@kernel.org> wrote: > > > > On Tue, Aug 06, 2019 at 03:34:34PM -0400, Qian Cai wrote: > >> diff --git a/arch/arm64/kernel/cpuinfo.c b/arch/arm64/kernel/cpuinfo.c > >> index 876055e37352..a0c495a3f4fd 100644 > >> --- a/arch/arm64/kernel/cpuinfo.c > >> +++ b/arch/arm64/kernel/cpuinfo.c > >> @@ -34,10 +34,7 @@ DEFINE_PER_CPU(struct cpuinfo_arm64, cpu_data); > >> static struct cpuinfo_arm64 boot_cpu_data; > >> > >> static char *icache_policy_str[] = { > >> - [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN", > >> - [ICACHE_POLICY_VIPT] = "VIPT", > >> - [ICACHE_POLICY_PIPT] = "PIPT", > >> - [ICACHE_POLICY_VPIPT] = "VPIPT", > >> + [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN" > >> }; > >> > >> unsigned long __icache_flags; > >> @@ -310,13 +307,16 @@ static void cpuinfo_detect_icache_policy(struct cpuinfo_arm64 *info) > >> > >> switch (l1ip) { > >> case ICACHE_POLICY_PIPT: > >> + icache_policy_str[ICACHE_POLICY_PIPT] = "PIPT"; > >> break; > >> case ICACHE_POLICY_VPIPT: > >> + icache_policy_str[ICACHE_POLICY_VPIPT] = "VPIPT"; > >> set_bit(ICACHEF_VPIPT, &__icache_flags); > >> break; > >> default: > >> /* Fallthrough */ > >> case ICACHE_POLICY_VIPT: > >> + icache_policy_str[ICACHE_POLICY_VIPT] = "VIPT"; > >> /* Assume aliasing */ > >> set_bit(ICACHEF_ALIASING, &__icache_flags); > > > > I still think this is worse than the code in mainline. I don't think > > -Woverride-init should warn when overriding a field from a GCC range > > designated initialiser, since it makes them considerably less useful > > imo. > > Unfortunately, compiler people are moving into a different direction as > Clang would warn those kind of usage too. > > It actually prove that those warnings are useful to find real issues. See, > > Fae5e033d65a (“mfd: rk808: Fix RK818_IRQ_DISCHG_ILIM initializer”) > 32df34d875bb (“[media] rc: img-ir: jvc: Remove unused no-leader timings”) > > Especially, to find redundant initializations in large structures. e.g., > > e6ea0b917875 (“[media] dvb_frontend: Don't declare values twice at a table”) None of these appear to use the range initialisers I was referring to, so I don't see why this is relevant to the discussion at hand. Will
diff --git a/arch/arm64/kernel/cpuinfo.c b/arch/arm64/kernel/cpuinfo.c index 876055e37352..a0c495a3f4fd 100644 --- a/arch/arm64/kernel/cpuinfo.c +++ b/arch/arm64/kernel/cpuinfo.c @@ -34,10 +34,7 @@ DEFINE_PER_CPU(struct cpuinfo_arm64, cpu_data); static struct cpuinfo_arm64 boot_cpu_data; static char *icache_policy_str[] = { - [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN", - [ICACHE_POLICY_VIPT] = "VIPT", - [ICACHE_POLICY_PIPT] = "PIPT", - [ICACHE_POLICY_VPIPT] = "VPIPT", + [0 ... ICACHE_POLICY_PIPT] = "RESERVED/UNKNOWN" }; unsigned long __icache_flags; @@ -310,13 +307,16 @@ static void cpuinfo_detect_icache_policy(struct cpuinfo_arm64 *info) switch (l1ip) { case ICACHE_POLICY_PIPT: + icache_policy_str[ICACHE_POLICY_PIPT] = "PIPT"; break; case ICACHE_POLICY_VPIPT: + icache_policy_str[ICACHE_POLICY_VPIPT] = "VPIPT"; set_bit(ICACHEF_VPIPT, &__icache_flags); break; default: /* Fallthrough */ case ICACHE_POLICY_VIPT: + icache_policy_str[ICACHE_POLICY_VIPT] = "VIPT"; /* Assume aliasing */ set_bit(ICACHEF_ALIASING, &__icache_flags); }
The commit 155433cb365e ("arm64: cache: Remove support for ASID-tagged VIVT I-caches") introduced some compiation warnings from GCC (and Clang), arch/arm64/kernel/cpuinfo.c:38:26: warning: initialized field overwritten [-Woverride-init] [ICACHE_POLICY_VIPT] = "VIPT", ^~~~~~ arch/arm64/kernel/cpuinfo.c:38:26: note: (near initialization for 'icache_policy_str[2]') arch/arm64/kernel/cpuinfo.c:39:26: warning: initialized field overwritten [-Woverride-init] [ICACHE_POLICY_PIPT] = "PIPT", ^~~~~~ arch/arm64/kernel/cpuinfo.c:39:26: note: (near initialization for 'icache_policy_str[3]') arch/arm64/kernel/cpuinfo.c:40:27: warning: initialized field overwritten [-Woverride-init] [ICACHE_POLICY_VPIPT] = "VPIPT", ^~~~~~~ arch/arm64/kernel/cpuinfo.c:40:27: note: (near initialization for 'icache_policy_str[0]') because it initializes icache_policy_str[0 ... 3] twice. Since the array is only used in cpuinfo_detect_icache_policy(), fix it by initializing a specific field there just before using. Fixes: 155433cb365e ("arm64: cache: Remove support for ASID-tagged VIVT I-caches") Signed-off-by: Qian Cai <cai@lca.pw> --- v2: Initialize a specific field in cpuinfo_detect_icache_policy(). arch/arm64/kernel/cpuinfo.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)