| Message ID | 20220505095640.312-1-thunder.leizhen@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v3] arm64: add the printing of tpidr_elx in __show_regs() | expand |
On Thu, May 05, 2022 at 05:56:40PM +0800, Zhen Lei wrote: > Commit 7158627686f0 ("arm64: percpu: implement optimised pcpu access > using tpidr_el1") and commit 6d99b68933fb ("arm64: alternatives: use > tpidr_el2 on VHE hosts") use tpidr_elx to cache my_cpu_offset to optimize > pcpu access. However, when performing reverse execution based on the > registers and the memory contents in kdump, this information is sometimes > required if there is a pcpu access. > > Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> > --- > arch/arm64/kernel/process.c | 5 +++++ > 1 file changed, 5 insertions(+) > > v2 --> v3: > 1) Relace "switch (read_sysreg(CurrentEL))" statement with > "if (is_kernel_in_hyp_mode())" statement. > 2) Change the register name to lowercase. > > v1 --> v2: > Directly print the tpidr_elx register of the current exception level. > Avoid coupling with the implementation of 'my_cpu_offset'. > > diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c > index 7fa97df55e3ad3f..7b6bccce9721c36 100644 > --- a/arch/arm64/kernel/process.c > +++ b/arch/arm64/kernel/process.c > @@ -216,6 +216,11 @@ void __show_regs(struct pt_regs *regs) > show_regs_print_info(KERN_DEFAULT); > print_pstate(regs); > > + if (is_kernel_in_hyp_mode()) > + printk("tpidr_el2 : %016llx\n", read_sysreg(tpidr_el2)); > + else > + printk("tpidr_el1 : %016llx\n", read_sysreg(tpidr_el1)); > + I'd still much prefer if we only printed this information for exceptions taken in kernel mode. What use is this information for user faults? We may only store an offset in the register today, but that could easily change and nobody will come back and fix this code up. Thanks, Will
On Thu, May 05, 2022 at 05:56:40PM +0800, Zhen Lei wrote: > Commit 7158627686f0 ("arm64: percpu: implement optimised pcpu access > using tpidr_el1") and commit 6d99b68933fb ("arm64: alternatives: use > tpidr_el2 on VHE hosts") use tpidr_elx to cache my_cpu_offset to optimize > pcpu access. However, when performing reverse execution based on the > registers and the memory contents in kdump, this information is sometimes > required if there is a pcpu access. > > Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> > --- > arch/arm64/kernel/process.c | 5 +++++ > 1 file changed, 5 insertions(+) > > v2 --> v3: > 1) Relace "switch (read_sysreg(CurrentEL))" statement with > "if (is_kernel_in_hyp_mode())" statement. > 2) Change the register name to lowercase. > > v1 --> v2: > Directly print the tpidr_elx register of the current exception level. > Avoid coupling with the implementation of 'my_cpu_offset'. > > diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c > index 7fa97df55e3ad3f..7b6bccce9721c36 100644 > --- a/arch/arm64/kernel/process.c > +++ b/arch/arm64/kernel/process.c > @@ -216,6 +216,11 @@ void __show_regs(struct pt_regs *regs) > show_regs_print_info(KERN_DEFAULT); > print_pstate(regs); > > + if (is_kernel_in_hyp_mode()) > + printk("tpidr_el2 : %016llx\n", read_sysreg(tpidr_el2)); > + else > + printk("tpidr_el1 : %016llx\n", read_sysreg(tpidr_el1)); If we care about the offset specifically, this would be simpler as: printk("cpu offset : 0x%016lx\n", __my_cpu_offset()); ... which should do the right thing even if we repurpose the TPIDRs and move the offset elsewhere. As Will says, we should only log this for !user_mode(regs), so it could be placed in the block below, immediately before we print the kernel PC, i.e. if (!user_mode_regs) { printk("cpu offset : %016lx\n", __my_cpu_offset()); printk("pc : %pS\n", (void *)regs->pc); printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr)); ... } ... or in a separate block which checks the same condition. Thanks, Mark. > + > if (!user_mode(regs)) { > printk("pc : %pS\n", (void *)regs->pc); > printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr)); > -- > 2.25.1 >
On 2022/5/5 21:04, Mark Rutland wrote: > On Thu, May 05, 2022 at 05:56:40PM +0800, Zhen Lei wrote: >> Commit 7158627686f0 ("arm64: percpu: implement optimised pcpu access >> using tpidr_el1") and commit 6d99b68933fb ("arm64: alternatives: use >> tpidr_el2 on VHE hosts") use tpidr_elx to cache my_cpu_offset to optimize >> pcpu access. However, when performing reverse execution based on the >> registers and the memory contents in kdump, this information is sometimes >> required if there is a pcpu access. >> >> Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> >> --- >> arch/arm64/kernel/process.c | 5 +++++ >> 1 file changed, 5 insertions(+) >> >> v2 --> v3: >> 1) Relace "switch (read_sysreg(CurrentEL))" statement with >> "if (is_kernel_in_hyp_mode())" statement. >> 2) Change the register name to lowercase. >> >> v1 --> v2: >> Directly print the tpidr_elx register of the current exception level. >> Avoid coupling with the implementation of 'my_cpu_offset'. >> >> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c >> index 7fa97df55e3ad3f..7b6bccce9721c36 100644 >> --- a/arch/arm64/kernel/process.c >> +++ b/arch/arm64/kernel/process.c >> @@ -216,6 +216,11 @@ void __show_regs(struct pt_regs *regs) >> show_regs_print_info(KERN_DEFAULT); >> print_pstate(regs); >> >> + if (is_kernel_in_hyp_mode()) >> + printk("tpidr_el2 : %016llx\n", read_sysreg(tpidr_el2)); >> + else >> + printk("tpidr_el1 : %016llx\n", read_sysreg(tpidr_el1)); > > If we care about the offset specifically, this would be simpler as: > > printk("cpu offset : 0x%016lx\n", __my_cpu_offset()); The function name is __show_regs(), so not using register name may not be good. In fact, some other architectures may also have this problem. If we use my_cpu_offset, we may need to put it in a public. > > ... which should do the right thing even if we repurpose the TPIDRs and move the offset elsewhere. > > As Will says, we should only log this for !user_mode(regs), so it could > be placed in the block below, immediately before we print the kernel PC, i.e. > > if (!user_mode_regs) { > printk("cpu offset : %016lx\n", __my_cpu_offset()); > printk("pc : %pS\n", (void *)regs->pc); > printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr)); > ... > } > > ... or in a separate block which checks the same condition. > > Thanks, > Mark. > >> + >> if (!user_mode(regs)) { >> printk("pc : %pS\n", (void *)regs->pc); >> printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr)); >> -- >> 2.25.1 >> > . >
On 2022/5/5 21:26, Leizhen (ThunderTown) wrote: > > > On 2022/5/5 21:04, Mark Rutland wrote: >> On Thu, May 05, 2022 at 05:56:40PM +0800, Zhen Lei wrote: >>> Commit 7158627686f0 ("arm64: percpu: implement optimised pcpu access >>> using tpidr_el1") and commit 6d99b68933fb ("arm64: alternatives: use >>> tpidr_el2 on VHE hosts") use tpidr_elx to cache my_cpu_offset to optimize >>> pcpu access. However, when performing reverse execution based on the >>> registers and the memory contents in kdump, this information is sometimes >>> required if there is a pcpu access. >>> >>> Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> >>> --- >>> arch/arm64/kernel/process.c | 5 +++++ >>> 1 file changed, 5 insertions(+) >>> >>> v2 --> v3: >>> 1) Relace "switch (read_sysreg(CurrentEL))" statement with >>> "if (is_kernel_in_hyp_mode())" statement. >>> 2) Change the register name to lowercase. >>> >>> v1 --> v2: >>> Directly print the tpidr_elx register of the current exception level. >>> Avoid coupling with the implementation of 'my_cpu_offset'. >>> >>> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c >>> index 7fa97df55e3ad3f..7b6bccce9721c36 100644 >>> --- a/arch/arm64/kernel/process.c >>> +++ b/arch/arm64/kernel/process.c >>> @@ -216,6 +216,11 @@ void __show_regs(struct pt_regs *regs) >>> show_regs_print_info(KERN_DEFAULT); >>> print_pstate(regs); >>> >>> + if (is_kernel_in_hyp_mode()) >>> + printk("tpidr_el2 : %016llx\n", read_sysreg(tpidr_el2)); >>> + else >>> + printk("tpidr_el1 : %016llx\n", read_sysreg(tpidr_el1)); >> >> If we care about the offset specifically, this would be simpler as: >> >> printk("cpu offset : 0x%016lx\n", __my_cpu_offset()); > > The function name is __show_regs(), so not using register name may not be good. > In fact, some other architectures may also have this problem. If we use my_cpu_offset, > we may need to put it in a public. The other idea is to back up each my_cpu_offset in an array. In this way, the offset can be queried through vmcore even if it is not printed. > >> >> ... which should do the right thing even if we repurpose the TPIDRs and move the offset elsewhere. >> >> As Will says, we should only log this for !user_mode(regs), so it could >> be placed in the block below, immediately before we print the kernel PC, i.e. >> >> if (!user_mode_regs) { >> printk("cpu offset : %016lx\n", __my_cpu_offset()); >> printk("pc : %pS\n", (void *)regs->pc); >> printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr)); >> ... >> } >> >> ... or in a separate block which checks the same condition. >> >> Thanks, >> Mark. >> >>> + >>> if (!user_mode(regs)) { >>> printk("pc : %pS\n", (void *)regs->pc); >>> printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr)); >>> -- >>> 2.25.1 >>> >> . >> >
On 2022/5/5 21:34, Leizhen (ThunderTown) wrote: > > > On 2022/5/5 21:26, Leizhen (ThunderTown) wrote: >> >> >> On 2022/5/5 21:04, Mark Rutland wrote: >>> On Thu, May 05, 2022 at 05:56:40PM +0800, Zhen Lei wrote: >>>> Commit 7158627686f0 ("arm64: percpu: implement optimised pcpu access >>>> using tpidr_el1") and commit 6d99b68933fb ("arm64: alternatives: use >>>> tpidr_el2 on VHE hosts") use tpidr_elx to cache my_cpu_offset to optimize >>>> pcpu access. However, when performing reverse execution based on the >>>> registers and the memory contents in kdump, this information is sometimes >>>> required if there is a pcpu access. >>>> >>>> Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> >>>> --- >>>> arch/arm64/kernel/process.c | 5 +++++ >>>> 1 file changed, 5 insertions(+) >>>> >>>> v2 --> v3: >>>> 1) Relace "switch (read_sysreg(CurrentEL))" statement with >>>> "if (is_kernel_in_hyp_mode())" statement. >>>> 2) Change the register name to lowercase. >>>> >>>> v1 --> v2: >>>> Directly print the tpidr_elx register of the current exception level. >>>> Avoid coupling with the implementation of 'my_cpu_offset'. >>>> >>>> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c >>>> index 7fa97df55e3ad3f..7b6bccce9721c36 100644 >>>> --- a/arch/arm64/kernel/process.c >>>> +++ b/arch/arm64/kernel/process.c >>>> @@ -216,6 +216,11 @@ void __show_regs(struct pt_regs *regs) >>>> show_regs_print_info(KERN_DEFAULT); >>>> print_pstate(regs); >>>> >>>> + if (is_kernel_in_hyp_mode()) >>>> + printk("tpidr_el2 : %016llx\n", read_sysreg(tpidr_el2)); >>>> + else >>>> + printk("tpidr_el1 : %016llx\n", read_sysreg(tpidr_el1)); >>> >>> If we care about the offset specifically, this would be simpler as: >>> >>> printk("cpu offset : 0x%016lx\n", __my_cpu_offset()); >> >> The function name is __show_regs(), so not using register name may not be good. >> In fact, some other architectures may also have this problem. If we use my_cpu_offset, >> we may need to put it in a public. > > The other idea is to back up each my_cpu_offset in an array. In this way, the offset can > be queried through vmcore even if it is not printed. Sorry, __per_cpu_offset[NR_CPUS] is always defined. > >> >>> >>> ... which should do the right thing even if we repurpose the TPIDRs and move the offset elsewhere. >>> >>> As Will says, we should only log this for !user_mode(regs), so it could >>> be placed in the block below, immediately before we print the kernel PC, i.e. >>> >>> if (!user_mode_regs) { >>> printk("cpu offset : %016lx\n", __my_cpu_offset()); >>> printk("pc : %pS\n", (void *)regs->pc); >>> printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr)); >>> ... >>> } >>> >>> ... or in a separate block which checks the same condition. >>> >>> Thanks, >>> Mark. >>> >>>> + >>>> if (!user_mode(regs)) { >>>> printk("pc : %pS\n", (void *)regs->pc); >>>> printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr)); >>>> -- >>>> 2.25.1 >>>> >>> . >>> >> >
On Fri, May 06, 2022 at 04:16:55PM +0800, Leizhen (ThunderTown) wrote: > > > On 2022/5/5 21:34, Leizhen (ThunderTown) wrote: > > On 2022/5/5 21:26, Leizhen (ThunderTown) wrote: > >> On 2022/5/5 21:04, Mark Rutland wrote: > >>> On Thu, May 05, 2022 at 05:56:40PM +0800, Zhen Lei wrote: > >>>> Commit 7158627686f0 ("arm64: percpu: implement optimised pcpu access > >>>> using tpidr_el1") and commit 6d99b68933fb ("arm64: alternatives: use > >>>> tpidr_el2 on VHE hosts") use tpidr_elx to cache my_cpu_offset to optimize > >>>> pcpu access. However, when performing reverse execution based on the > >>>> registers and the memory contents in kdump, this information is sometimes > >>>> required if there is a pcpu access. > >>>> > >>>> Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> > >>>> --- > >>>> arch/arm64/kernel/process.c | 5 +++++ > >>>> 1 file changed, 5 insertions(+) > >>>> > >>>> v2 --> v3: > >>>> 1) Relace "switch (read_sysreg(CurrentEL))" statement with > >>>> "if (is_kernel_in_hyp_mode())" statement. > >>>> 2) Change the register name to lowercase. > >>>> > >>>> v1 --> v2: > >>>> Directly print the tpidr_elx register of the current exception level. > >>>> Avoid coupling with the implementation of 'my_cpu_offset'. > >>>> > >>>> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c > >>>> index 7fa97df55e3ad3f..7b6bccce9721c36 100644 > >>>> --- a/arch/arm64/kernel/process.c > >>>> +++ b/arch/arm64/kernel/process.c > >>>> @@ -216,6 +216,11 @@ void __show_regs(struct pt_regs *regs) > >>>> show_regs_print_info(KERN_DEFAULT); > >>>> print_pstate(regs); > >>>> > >>>> + if (is_kernel_in_hyp_mode()) > >>>> + printk("tpidr_el2 : %016llx\n", read_sysreg(tpidr_el2)); > >>>> + else > >>>> + printk("tpidr_el1 : %016llx\n", read_sysreg(tpidr_el1)); > >>> > >>> If we care about the offset specifically, this would be simpler as: > >>> > >>> printk("cpu offset : 0x%016lx\n", __my_cpu_offset()); > >> > >> The function name is __show_regs(), so not using register name may not be good. > >> In fact, some other architectures may also have this problem. If we use my_cpu_offset, > >> we may need to put it in a public. > > > > The other idea is to back up each my_cpu_offset in an array. In this way, the offset can > > be queried through vmcore even if it is not printed. > > Sorry, __per_cpu_offset[NR_CPUS] is always defined. Surely that's in the vmcore already? It's just data in memory. Thanks, Mark.
On 2022/5/6 18:09, Mark Rutland wrote: > On Fri, May 06, 2022 at 04:16:55PM +0800, Leizhen (ThunderTown) wrote: >> >> >> On 2022/5/5 21:34, Leizhen (ThunderTown) wrote: >>> On 2022/5/5 21:26, Leizhen (ThunderTown) wrote: >>>> On 2022/5/5 21:04, Mark Rutland wrote: >>>>> On Thu, May 05, 2022 at 05:56:40PM +0800, Zhen Lei wrote: >>>>>> Commit 7158627686f0 ("arm64: percpu: implement optimised pcpu access >>>>>> using tpidr_el1") and commit 6d99b68933fb ("arm64: alternatives: use >>>>>> tpidr_el2 on VHE hosts") use tpidr_elx to cache my_cpu_offset to optimize >>>>>> pcpu access. However, when performing reverse execution based on the >>>>>> registers and the memory contents in kdump, this information is sometimes >>>>>> required if there is a pcpu access. >>>>>> >>>>>> Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> >>>>>> --- >>>>>> arch/arm64/kernel/process.c | 5 +++++ >>>>>> 1 file changed, 5 insertions(+) >>>>>> >>>>>> v2 --> v3: >>>>>> 1) Relace "switch (read_sysreg(CurrentEL))" statement with >>>>>> "if (is_kernel_in_hyp_mode())" statement. >>>>>> 2) Change the register name to lowercase. >>>>>> >>>>>> v1 --> v2: >>>>>> Directly print the tpidr_elx register of the current exception level. >>>>>> Avoid coupling with the implementation of 'my_cpu_offset'. >>>>>> >>>>>> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c >>>>>> index 7fa97df55e3ad3f..7b6bccce9721c36 100644 >>>>>> --- a/arch/arm64/kernel/process.c >>>>>> +++ b/arch/arm64/kernel/process.c >>>>>> @@ -216,6 +216,11 @@ void __show_regs(struct pt_regs *regs) >>>>>> show_regs_print_info(KERN_DEFAULT); >>>>>> print_pstate(regs); >>>>>> >>>>>> + if (is_kernel_in_hyp_mode()) >>>>>> + printk("tpidr_el2 : %016llx\n", read_sysreg(tpidr_el2)); >>>>>> + else >>>>>> + printk("tpidr_el1 : %016llx\n", read_sysreg(tpidr_el1)); >>>>> >>>>> If we care about the offset specifically, this would be simpler as: >>>>> >>>>> printk("cpu offset : 0x%016lx\n", __my_cpu_offset()); >>>> >>>> The function name is __show_regs(), so not using register name may not be good. >>>> In fact, some other architectures may also have this problem. If we use my_cpu_offset, >>>> we may need to put it in a public. >>> >>> The other idea is to back up each my_cpu_offset in an array. In this way, the offset can >>> be queried through vmcore even if it is not printed. >> >> Sorry, __per_cpu_offset[NR_CPUS] is always defined. > > Surely that's in the vmcore already? It's just data in memory. Yes. There is also a definition in drivers/base/arch_numa.c. I didn't search the drivers directory at first. I thought there would be no definition when CONFIG_HAVE_SETUP_PER_CPU_AREA=y. crash> p -x __per_cpu_offset __per_cpu_offset = $1 = {0xffff8003f460d000, 0xffff8003f4621000, 0xffff8003f4635000, 0xffff8003f4649000 > > Thanks, > Mark. > . >
diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c index 7fa97df55e3ad3f..7b6bccce9721c36 100644 --- a/arch/arm64/kernel/process.c +++ b/arch/arm64/kernel/process.c @@ -216,6 +216,11 @@ void __show_regs(struct pt_regs *regs) show_regs_print_info(KERN_DEFAULT); print_pstate(regs); + if (is_kernel_in_hyp_mode()) + printk("tpidr_el2 : %016llx\n", read_sysreg(tpidr_el2)); + else + printk("tpidr_el1 : %016llx\n", read_sysreg(tpidr_el1)); + if (!user_mode(regs)) { printk("pc : %pS\n", (void *)regs->pc); printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr));
Commit 7158627686f0 ("arm64: percpu: implement optimised pcpu access using tpidr_el1") and commit 6d99b68933fb ("arm64: alternatives: use tpidr_el2 on VHE hosts") use tpidr_elx to cache my_cpu_offset to optimize pcpu access. However, when performing reverse execution based on the registers and the memory contents in kdump, this information is sometimes required if there is a pcpu access. Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com> --- arch/arm64/kernel/process.c | 5 +++++ 1 file changed, 5 insertions(+) v2 --> v3: 1) Relace "switch (read_sysreg(CurrentEL))" statement with "if (is_kernel_in_hyp_mode())" statement. 2) Change the register name to lowercase. v1 --> v2: Directly print the tpidr_elx register of the current exception level. Avoid coupling with the implementation of 'my_cpu_offset'.