Message ID | 20210706162621.940924-1-linux@roeck-us.net (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | riscv: Disable STACKPROTECTOR_PER_TASK if GCC_PLUGIN_RANDSTRUCT is enabled | expand |
On Tue, 06 Jul 2021 09:26:21 PDT (-0700), linux@roeck-us.net wrote: > riscv uses the value of TSK_STACK_CANARY to set > stack-protector-guard-offset. With GCC_PLUGIN_RANDSTRUCT enabled, that > value is non-deterministic, and with riscv:allmodconfig often results > in build errors such as > > cc1: error: '8120' is not a valid offset in '-mstack-protector-guard-offset=' > > Enable STACKPROTECTOR_PER_TASK only if GCC_PLUGIN_RANDSTRUCT is disabled > to fix the problem. > > Fixes: fea2fed201ee5 ("riscv: Enable per-task stack canaries") > Cc: Guo Ren <guoren@linux.alibaba.com> > Signed-off-by: Guenter Roeck <linux@roeck-us.net> > --- > arch/riscv/Kconfig | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig > index 469a70bd8da6..3afb84fa2190 100644 > --- a/arch/riscv/Kconfig > +++ b/arch/riscv/Kconfig > @@ -486,6 +486,7 @@ config CC_HAVE_STACKPROTECTOR_TLS > > config STACKPROTECTOR_PER_TASK > def_bool y > + depends on !GCC_PLUGIN_RANDSTRUCT > depends on STACKPROTECTOR && CC_HAVE_STACKPROTECTOR_TLS > > config PHYS_RAM_BASE_FIXED Thanks, this is on fixes. I do an allmodconfig build and haven't seen these failures, is there something special I need to do to get this working? I know nothing about GCC plugins...
On 7/23/21 5:58 PM, Palmer Dabbelt wrote: > On Tue, 06 Jul 2021 09:26:21 PDT (-0700), linux@roeck-us.net wrote: >> riscv uses the value of TSK_STACK_CANARY to set >> stack-protector-guard-offset. With GCC_PLUGIN_RANDSTRUCT enabled, that >> value is non-deterministic, and with riscv:allmodconfig often results >> in build errors such as >> >> cc1: error: '8120' is not a valid offset in '-mstack-protector-guard-offset=' >> >> Enable STACKPROTECTOR_PER_TASK only if GCC_PLUGIN_RANDSTRUCT is disabled >> to fix the problem. >> >> Fixes: fea2fed201ee5 ("riscv: Enable per-task stack canaries") >> Cc: Guo Ren <guoren@linux.alibaba.com> >> Signed-off-by: Guenter Roeck <linux@roeck-us.net> >> --- >> arch/riscv/Kconfig | 1 + >> 1 file changed, 1 insertion(+) >> >> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig >> index 469a70bd8da6..3afb84fa2190 100644 >> --- a/arch/riscv/Kconfig >> +++ b/arch/riscv/Kconfig >> @@ -486,6 +486,7 @@ config CC_HAVE_STACKPROTECTOR_TLS >> >> config STACKPROTECTOR_PER_TASK >> def_bool y >> + depends on !GCC_PLUGIN_RANDSTRUCT >> depends on STACKPROTECTOR && CC_HAVE_STACKPROTECTOR_TLS >> >> config PHYS_RAM_BASE_FIXED > > Thanks, this is on fixes. I do an allmodconfig build and haven't seen these failures, is there something special I need to do to get this working? I know nothing about GCC plugins... The problem is seen only randomly. It depends on the value of TSK_STACK_CANARY in include/generated/asm-offsets.h. That in turn depends on the location of stack_canary in struct task_struct. You may have to try several times from a clean tree to see the problem. Guenter
On Fri, 23 Jul 2021 17:58:13 PDT (-0700), Palmer Dabbelt wrote: > On Tue, 06 Jul 2021 09:26:21 PDT (-0700), linux@roeck-us.net wrote: >> riscv uses the value of TSK_STACK_CANARY to set >> stack-protector-guard-offset. With GCC_PLUGIN_RANDSTRUCT enabled, that >> value is non-deterministic, and with riscv:allmodconfig often results >> in build errors such as >> >> cc1: error: '8120' is not a valid offset in '-mstack-protector-guard-offset=' >> >> Enable STACKPROTECTOR_PER_TASK only if GCC_PLUGIN_RANDSTRUCT is disabled >> to fix the problem. >> >> Fixes: fea2fed201ee5 ("riscv: Enable per-task stack canaries") >> Cc: Guo Ren <guoren@linux.alibaba.com> >> Signed-off-by: Guenter Roeck <linux@roeck-us.net> >> --- >> arch/riscv/Kconfig | 1 + >> 1 file changed, 1 insertion(+) >> >> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig >> index 469a70bd8da6..3afb84fa2190 100644 >> --- a/arch/riscv/Kconfig >> +++ b/arch/riscv/Kconfig >> @@ -486,6 +486,7 @@ config CC_HAVE_STACKPROTECTOR_TLS >> >> config STACKPROTECTOR_PER_TASK >> def_bool y >> + depends on !GCC_PLUGIN_RANDSTRUCT >> depends on STACKPROTECTOR && CC_HAVE_STACKPROTECTOR_TLS >> >> config PHYS_RAM_BASE_FIXED > > Thanks, this is on fixes. I do an allmodconfig build and haven't seen > these failures, is there something special I need to do to get this > working? I know nothing about GCC plugins... Sorry, looks like I screwed something up and I didn't actually merge this onto fixes. Not sure what happened, but this time it should actually be on fixes.
On 8/3/21 10:14 PM, Palmer Dabbelt wrote: > On Fri, 23 Jul 2021 17:58:13 PDT (-0700), Palmer Dabbelt wrote: >> On Tue, 06 Jul 2021 09:26:21 PDT (-0700), linux@roeck-us.net wrote: >>> riscv uses the value of TSK_STACK_CANARY to set >>> stack-protector-guard-offset. With GCC_PLUGIN_RANDSTRUCT enabled, that >>> value is non-deterministic, and with riscv:allmodconfig often results >>> in build errors such as >>> >>> cc1: error: '8120' is not a valid offset in '-mstack-protector-guard-offset=' >>> >>> Enable STACKPROTECTOR_PER_TASK only if GCC_PLUGIN_RANDSTRUCT is disabled >>> to fix the problem. >>> >>> Fixes: fea2fed201ee5 ("riscv: Enable per-task stack canaries") >>> Cc: Guo Ren <guoren@linux.alibaba.com> >>> Signed-off-by: Guenter Roeck <linux@roeck-us.net> >>> --- >>> arch/riscv/Kconfig | 1 + >>> 1 file changed, 1 insertion(+) >>> >>> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig >>> index 469a70bd8da6..3afb84fa2190 100644 >>> --- a/arch/riscv/Kconfig >>> +++ b/arch/riscv/Kconfig >>> @@ -486,6 +486,7 @@ config CC_HAVE_STACKPROTECTOR_TLS >>> >>> config STACKPROTECTOR_PER_TASK >>> def_bool y >>> + depends on !GCC_PLUGIN_RANDSTRUCT >>> depends on STACKPROTECTOR && CC_HAVE_STACKPROTECTOR_TLS >>> >>> config PHYS_RAM_BASE_FIXED >> >> Thanks, this is on fixes. I do an allmodconfig build and haven't seen >> these failures, is there something special I need to do to get this >> working? I know nothing about GCC plugins... > > Sorry, looks like I screwed something up and I didn't actually merge this onto fixes. Not sure what happened, but this time it should actually be on fixes. Ah, I thought you decided not to apply it for some reason. I changed my test scripts to always disable GCC_PLUGIN_RANDSTRUCT for riscv build tests and stopped reporting the problem. I mentioned this a couple of weeks ago in my test reports. Just in case you wondered. Guenter
diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig index 469a70bd8da6..3afb84fa2190 100644 --- a/arch/riscv/Kconfig +++ b/arch/riscv/Kconfig @@ -486,6 +486,7 @@ config CC_HAVE_STACKPROTECTOR_TLS config STACKPROTECTOR_PER_TASK def_bool y + depends on !GCC_PLUGIN_RANDSTRUCT depends on STACKPROTECTOR && CC_HAVE_STACKPROTECTOR_TLS config PHYS_RAM_BASE_FIXED
riscv uses the value of TSK_STACK_CANARY to set stack-protector-guard-offset. With GCC_PLUGIN_RANDSTRUCT enabled, that value is non-deterministic, and with riscv:allmodconfig often results in build errors such as cc1: error: '8120' is not a valid offset in '-mstack-protector-guard-offset=' Enable STACKPROTECTOR_PER_TASK only if GCC_PLUGIN_RANDSTRUCT is disabled to fix the problem. Fixes: fea2fed201ee5 ("riscv: Enable per-task stack canaries") Cc: Guo Ren <guoren@linux.alibaba.com> Signed-off-by: Guenter Roeck <linux@roeck-us.net> --- arch/riscv/Kconfig | 1 + 1 file changed, 1 insertion(+)