| Message ID | 20220623014917.199563-33-chenzhongjin@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | objtool: add base support for arm64 | expand |
On 2022-06-23 02:49, Chen Zhongjin wrote: > Using unreachable() at default of switch generates an extra branch at > end of the function, and compiler won't generate a ret to close this > branch because it knows it's unreachable. > > If there's no instruction in this branch, compiler will generate a NOP, > And it will confuse objtool to warn this NOP as a fall through branch. > > In fact these branches are actually unreachable, so we can replace > unreachable() with returning a -EINVAL value. > > Signed-off-by: Chen Zhongjin <chenzhongjin@huawei.com> > --- > arch/arm64/kvm/hyp/vgic-v3-sr.c | 7 +++---- > drivers/irqchip/irq-gic-v3.c | 2 +- > 2 files changed, 4 insertions(+), 5 deletions(-) Basic courtesy would have been to Cc the maintainers of this code. > > diff --git a/arch/arm64/kvm/hyp/vgic-v3-sr.c > b/arch/arm64/kvm/hyp/vgic-v3-sr.c > index 4fb419f7b8b6..f3cee92c3038 100644 > --- a/arch/arm64/kvm/hyp/vgic-v3-sr.c > +++ b/arch/arm64/kvm/hyp/vgic-v3-sr.c > @@ -6,7 +6,6 @@ > > #include <hyp/adjust_pc.h> > > -#include <linux/compiler.h> > #include <linux/irqchip/arm-gic-v3.h> > #include <linux/kvm_host.h> > > @@ -55,7 +54,7 @@ static u64 __gic_v3_get_lr(unsigned int lr) > return read_gicreg(ICH_LR15_EL2); > } > > - unreachable(); > + return -EINVAL; NAK. That's absolutely *wrong*, and will hide future bugs. Nothing checks for -EINVAL, and we *never* expect to reach this, hence the perfectly valid annotation. If something needs fixing, it probably is your tooling. M.
Hi, Thanks for your review and patient. On 2022/6/23 16:13, Marc Zyngier wrote: > On 2022-06-23 02:49, Chen Zhongjin wrote: >> Using unreachable() at default of switch generates an extra branch at >> end of the function, and compiler won't generate a ret to close this >> branch because it knows it's unreachable. >> >> If there's no instruction in this branch, compiler will generate a NOP, >> And it will confuse objtool to warn this NOP as a fall through branch. >> >> In fact these branches are actually unreachable, so we can replace >> unreachable() with returning a -EINVAL value. >> >> Signed-off-by: Chen Zhongjin <chenzhongjin@huawei.com> >> --- >> arch/arm64/kvm/hyp/vgic-v3-sr.c | 7 +++---- >> drivers/irqchip/irq-gic-v3.c | 2 +- >> 2 files changed, 4 insertions(+), 5 deletions(-) > > Basic courtesy would have been to Cc the maintainers of this code. > Sorry for that. I'll cc everyone next time. >> >> diff --git a/arch/arm64/kvm/hyp/vgic-v3-sr.c b/arch/arm64/kvm/hyp/vgic-v3-sr.c >> index 4fb419f7b8b6..f3cee92c3038 100644 >> --- a/arch/arm64/kvm/hyp/vgic-v3-sr.c >> +++ b/arch/arm64/kvm/hyp/vgic-v3-sr.c >> @@ -6,7 +6,6 @@ >> >> #include <hyp/adjust_pc.h> >> >> -#include <linux/compiler.h> >> #include <linux/irqchip/arm-gic-v3.h> >> #include <linux/kvm_host.h> >> >> @@ -55,7 +54,7 @@ static u64 __gic_v3_get_lr(unsigned int lr) >> return read_gicreg(ICH_LR15_EL2); >> } >> >> - unreachable(); >> + return -EINVAL; > > NAK. That's absolutely *wrong*, and will hide future bugs. > Nothing checks for -EINVAL, and we *never* expect to > reach this, hence the perfectly valid annotation. > > If something needs fixing, it probably is your tooling. > > M. You are right. Essentially, this is because objtool does not anticipate that the compiler will generate additional instructions when marking unreachable instructions. I'll fix this problem or add a specific check for this state. Best, Chen
diff --git a/arch/arm64/kvm/hyp/vgic-v3-sr.c b/arch/arm64/kvm/hyp/vgic-v3-sr.c index 4fb419f7b8b6..f3cee92c3038 100644 --- a/arch/arm64/kvm/hyp/vgic-v3-sr.c +++ b/arch/arm64/kvm/hyp/vgic-v3-sr.c @@ -6,7 +6,6 @@ #include <hyp/adjust_pc.h> -#include <linux/compiler.h> #include <linux/irqchip/arm-gic-v3.h> #include <linux/kvm_host.h> @@ -55,7 +54,7 @@ static u64 __gic_v3_get_lr(unsigned int lr) return read_gicreg(ICH_LR15_EL2); } - unreachable(); + return -EINVAL; } static void __gic_v3_set_lr(u64 val, int lr) @@ -166,7 +165,7 @@ static u32 __vgic_v3_read_ap0rn(int n) val = read_gicreg(ICH_AP0R3_EL2); break; default: - unreachable(); + val = -EINVAL; } return val; @@ -190,7 +189,7 @@ static u32 __vgic_v3_read_ap1rn(int n) val = read_gicreg(ICH_AP1R3_EL2); break; default: - unreachable(); + val = -EINVAL; } return val; diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c index b252d5534547..2ef98e32d257 100644 --- a/drivers/irqchip/irq-gic-v3.c +++ b/drivers/irqchip/irq-gic-v3.c @@ -475,7 +475,7 @@ static u32 __gic_get_ppi_index(irq_hw_number_t hwirq) case EPPI_RANGE: return hwirq - EPPI_BASE_INTID + 16; default: - unreachable(); + return -EINVAL; } }
Using unreachable() at default of switch generates an extra branch at end of the function, and compiler won't generate a ret to close this branch because it knows it's unreachable. If there's no instruction in this branch, compiler will generate a NOP, And it will confuse objtool to warn this NOP as a fall through branch. In fact these branches are actually unreachable, so we can replace unreachable() with returning a -EINVAL value. Signed-off-by: Chen Zhongjin <chenzhongjin@huawei.com> --- arch/arm64/kvm/hyp/vgic-v3-sr.c | 7 +++---- drivers/irqchip/irq-gic-v3.c | 2 +- 2 files changed, 4 insertions(+), 5 deletions(-)