AArch64 registers are 64bit whereas AArch32 registers
are 32bit or 64bit. MSR/MRS are expecting 64bit values thus
we should get rid of helpers READ/WRITE_SYSREG32
in favour of using READ/WRITE_SYSREG.
We should also use register_t type when reading sysregs
which can correspond to uint64_t or uint32_t.
Even though many AArch64 registers have upper 32bit reserved
it does not mean that they can't be widen in the future.
Modify SCTLR_EL2 accesses to use READ/WRITE_SYSREG.
SCTLR_EL2 already has bits defined in the range [32:63].
The ARM ARM defines them as unknown if implemented.
By writing in head.S SCTLR_EL2_SET we are zeroing the upper
32bit half which is correct but referring to this sysreg
as 32bit is a latent bug because the top 32bit was not used
by Xen.
Signed-off-by: Michal Orzel <michal.orzel@arm.com>
---
Changes since v1:
-Update commit message with SCTLR_EL2 analysis
---
xen/arch/arm/mm.c | 2 +-
xen/arch/arm/traps.c | 2 +-
2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/xen/arch/arm/mm.c b/xen/arch/arm/mm.c
index 59f8a3f15f..0e07335291 100644
--- a/xen/arch/arm/mm.c
+++ b/xen/arch/arm/mm.c
@@ -613,7 +613,7 @@ void __init remove_early_mappings(void)
*/
static void xen_pt_enforce_wnx(void)
{
- WRITE_SYSREG32(READ_SYSREG32(SCTLR_EL2) | SCTLR_Axx_ELx_WXN, SCTLR_EL2);
+ WRITE_SYSREG(READ_SYSREG(SCTLR_EL2) | SCTLR_Axx_ELx_WXN, SCTLR_EL2);
/*
* The TLBs may cache SCTLR_EL2.WXN. So ensure it is synchronized
* before flushing the TLBs.
diff --git a/xen/arch/arm/traps.c b/xen/arch/arm/traps.c
index c7acdb2087..e7384381cc 100644
--- a/xen/arch/arm/traps.c
+++ b/xen/arch/arm/traps.c
@@ -915,7 +915,7 @@ static void _show_registers(const struct cpu_user_regs *regs,
printk(" VTTBR_EL2: %016"PRIx64"\n", ctxt->vttbr_el2);
printk("\n");
- printk(" SCTLR_EL2: %08"PRIx32"\n", READ_SYSREG32(SCTLR_EL2));
+ printk(" SCTLR_EL2: %"PRIregister"\n", READ_SYSREG(SCTLR_EL2));
printk(" HCR_EL2: %"PRIregister"\n", READ_SYSREG(HCR_EL2));
printk(" TTBR0_EL2: %016"PRIx64"\n", READ_SYSREG64(TTBR0_EL2));
printk("\n");
--
2.29.0
On 27/04/2021 10:35, Michal Orzel wrote:
> AArch64 registers are 64bit whereas AArch32 registers
> are 32bit or 64bit. MSR/MRS are expecting 64bit values thus
> we should get rid of helpers READ/WRITE_SYSREG32
> in favour of using READ/WRITE_SYSREG.
> We should also use register_t type when reading sysregs
> which can correspond to uint64_t or uint32_t.
> Even though many AArch64 registers have upper 32bit reserved
> it does not mean that they can't be widen in the future.
>
> Modify SCTLR_EL2 accesses to use READ/WRITE_SYSREG.
>
> SCTLR_EL2 already has bits defined in the range [32:63].
> The ARM ARM defines them as unknown if implemented.
This is a bit ambiguous.
> By writing in head.S SCTLR_EL2_SET we are zeroing the upper
> 32bit half which is correct but referring to this sysreg
> as 32bit is a latent bug because the top 32bit was not used
> by Xen.
This seems to suggest the patch below will call SCTLR_EL2_SET whereas
this is already existing code.
>
> Signed-off-by: Michal Orzel <michal.orzel@arm.com>
> ---
> Changes since v1:
> -Update commit message with SCTLR_EL2 analysis
> ---
> xen/arch/arm/mm.c | 2 +-
> xen/arch/arm/traps.c | 2 +-
> 2 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/xen/arch/arm/mm.c b/xen/arch/arm/mm.c
> index 59f8a3f15f..0e07335291 100644
> --- a/xen/arch/arm/mm.c
> +++ b/xen/arch/arm/mm.c
> @@ -613,7 +613,7 @@ void __init remove_early_mappings(void)
> */
> static void xen_pt_enforce_wnx(void)
> {
> - WRITE_SYSREG32(READ_SYSREG32(SCTLR_EL2) | SCTLR_Axx_ELx_WXN, SCTLR_EL2);
> + WRITE_SYSREG(READ_SYSREG(SCTLR_EL2) | SCTLR_Axx_ELx_WXN, SCTLR_EL2);
> /*
> * The TLBs may cache SCTLR_EL2.WXN. So ensure it is synchronized
> * before flushing the TLBs.
> diff --git a/xen/arch/arm/traps.c b/xen/arch/arm/traps.c
> index c7acdb2087..e7384381cc 100644
> --- a/xen/arch/arm/traps.c
> +++ b/xen/arch/arm/traps.c
> @@ -915,7 +915,7 @@ static void _show_registers(const struct cpu_user_regs *regs,
> printk(" VTTBR_EL2: %016"PRIx64"\n", ctxt->vttbr_el2);
> printk("\n");
>
> - printk(" SCTLR_EL2: %08"PRIx32"\n", READ_SYSREG32(SCTLR_EL2));
> + printk(" SCTLR_EL2: %"PRIregister"\n", READ_SYSREG(SCTLR_EL2));
Your commit title suggests that you will modify mm.c but you are also
modifying traps.c. So how about the following commit message:
"
xen/arm: Always access SCTLR_EL2 using {READ, WRITE}_SYSREG()
The Armv8 specification describes the system register as a 64-bit value
on AArch64 and 32-bit value on AArch32 (same as Armv7).
Unfortunately, Xen is accessing the system registers using {READ,
WRITE}_SYSREG32() which means the top 32-bit are clobbered.
This is only a latent bug so far because Xen will not yet use the top
32-bit.
There is also no change in behavior because arch/arm/arm64/head.S will
initialize SCTLR_EL2 to a sane value with the top 32-bit zeroed.
"
Cheers,
--
Julien Grall
Hi Julien,
On 27.04.2021 11:59, Julien Grall wrote:
>
>
> On 27/04/2021 10:35, Michal Orzel wrote:
>> AArch64 registers are 64bit whereas AArch32 registers
>> are 32bit or 64bit. MSR/MRS are expecting 64bit values thus
>> we should get rid of helpers READ/WRITE_SYSREG32
>> in favour of using READ/WRITE_SYSREG.
>> We should also use register_t type when reading sysregs
>> which can correspond to uint64_t or uint32_t.
>> Even though many AArch64 registers have upper 32bit reserved
>> it does not mean that they can't be widen in the future.
>>
>> Modify SCTLR_EL2 accesses to use READ/WRITE_SYSREG.
>>
>> SCTLR_EL2 already has bits defined in the range [32:63].
>> The ARM ARM defines them as unknown if implemented.
>
> This is a bit ambiguous.
>
>> By writing in head.S SCTLR_EL2_SET we are zeroing the upper
>> 32bit half which is correct but referring to this sysreg
>> as 32bit is a latent bug because the top 32bit was not used
>> by Xen.
>
> This seems to suggest the patch below will call SCTLR_EL2_SET whereas this is already existing code.
>
>>
>> Signed-off-by: Michal Orzel <michal.orzel@arm.com>
>> ---
>> Changes since v1:
>> -Update commit message with SCTLR_EL2 analysis
>> ---
>> xen/arch/arm/mm.c | 2 +-
>> xen/arch/arm/traps.c | 2 +-
>> 2 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/xen/arch/arm/mm.c b/xen/arch/arm/mm.c
>> index 59f8a3f15f..0e07335291 100644
>> --- a/xen/arch/arm/mm.c
>> +++ b/xen/arch/arm/mm.c
>> @@ -613,7 +613,7 @@ void __init remove_early_mappings(void)
>> */
>> static void xen_pt_enforce_wnx(void)
>> {
>> - WRITE_SYSREG32(READ_SYSREG32(SCTLR_EL2) | SCTLR_Axx_ELx_WXN, SCTLR_EL2);
>> + WRITE_SYSREG(READ_SYSREG(SCTLR_EL2) | SCTLR_Axx_ELx_WXN, SCTLR_EL2);
>> /*
>> * The TLBs may cache SCTLR_EL2.WXN. So ensure it is synchronized
>> * before flushing the TLBs.
>> diff --git a/xen/arch/arm/traps.c b/xen/arch/arm/traps.c
>> index c7acdb2087..e7384381cc 100644
>> --- a/xen/arch/arm/traps.c
>> +++ b/xen/arch/arm/traps.c
>> @@ -915,7 +915,7 @@ static void _show_registers(const struct cpu_user_regs *regs,
>> printk(" VTTBR_EL2: %016"PRIx64"\n", ctxt->vttbr_el2);
>> printk("\n");
>> - printk(" SCTLR_EL2: %08"PRIx32"\n", READ_SYSREG32(SCTLR_EL2));
>> + printk(" SCTLR_EL2: %"PRIregister"\n", READ_SYSREG(SCTLR_EL2));
>
> Your commit title suggests that you will modify mm.c but you are also modifying traps.c. So how about the following commit message:
>
> "
> xen/arm: Always access SCTLR_EL2 using {READ, WRITE}_SYSREG()
>
> The Armv8 specification describes the system register as a 64-bit value on AArch64 and 32-bit value on AArch32 (same as Armv7).
>
> Unfortunately, Xen is accessing the system registers using {READ, WRITE}_SYSREG32() which means the top 32-bit are clobbered.
>
> This is only a latent bug so far because Xen will not yet use the top 32-bit.
>
> There is also no change in behavior because arch/arm/arm64/head.S will initialize SCTLR_EL2 to a sane value with the top 32-bit zeroed.
> "
>
Thank you. I will modify the commit msg according to what you suggested.
> Cheers,
>
© 2016 - 2026 Red Hat, Inc.