| Message ID | 1484909410-11673-2-git-send-email-sudeep.holla@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Fri, Jan 20, 2017 at 10:50:10AM +0000, Sudeep Holla wrote: > csselr and ccsidr are treated as 64-bit values already elsewhere in the > kernel. It also aligns well with the architecture extensions that allow > 64-bit format for ccsidr. > > This patch upgrades the existing accesses to csselr and ccsidr from > 32-bit to 64-bit in preparation to add support to those extensions. > > Cc: Christoffer Dall <christoffer.dall@linaro.org> > Cc: Marc Zyngier <marc.zyngier@arm.com> > Signed-off-by: Sudeep Holla <sudeep.holla@arm.com> > --- > arch/arm64/kvm/sys_regs.c | 18 +++++++++--------- > 1 file changed, 9 insertions(+), 9 deletions(-) > > diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c > index 5dca1f10340f..a3559a8a2b0c 100644 > --- a/arch/arm64/kvm/sys_regs.c > +++ b/arch/arm64/kvm/sys_regs.c > @@ -58,15 +58,15 @@ > */ > > /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */ > -static u32 cache_levels; > +static u64 cache_levels; > > /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */ > #define CSSELR_MAX ((MAX_CACHE_LEVEL - 1) >> 1) > > /* Which cache CCSIDR represents depends on CSSELR value. */ > -static u32 get_ccsidr(u32 csselr) > +static u64 get_ccsidr(u64 csselr) > { > - u32 ccsidr; > + u64 ccsidr; > > /* Make sure noone else changes CSSELR during this! */ > local_irq_disable(); > @@ -1952,9 +1952,9 @@ static int set_invariant_sys_reg(u64 id, void __user *uaddr) > return 0; > } > > -static bool is_valid_cache(u32 val) > +static bool is_valid_cache(u64 val) > { > - u32 level, ctype; > + u64 level, ctype; > > if (val >= CSSELR_MAX) > return false; > @@ -1979,8 +1979,8 @@ static bool is_valid_cache(u32 val) > > static int demux_c15_get(u64 id, void __user *uaddr) > { > - u32 val; > - u32 __user *uval = uaddr; > + u64 val; > + u64 __user *uval = uaddr; > > /* Fail if we have unknown bits set. */ > if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK > @@ -2004,8 +2004,8 @@ static int demux_c15_get(u64 id, void __user *uaddr) > > static int demux_c15_set(u64 id, void __user *uaddr) > { > - u32 val, newval; > - u32 __user *uval = uaddr; > + u64 val, newval; > + u64 __user *uval = uaddr; Doesn't converting these uval pointers to u64 cause us to break the ABI as we'll now be reading/writing 64-bit values to userspace with the get_user and put_user following the declarations? > > /* Fail if we have unknown bits set. */ > if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK > -- > 2.7.4 > Thanks, -Christoffer
On 23/01/17 21:08, Christoffer Dall wrote: > On Fri, Jan 20, 2017 at 10:50:10AM +0000, Sudeep Holla wrote: >> csselr and ccsidr are treated as 64-bit values already elsewhere in the >> kernel. It also aligns well with the architecture extensions that allow >> 64-bit format for ccsidr. >> >> This patch upgrades the existing accesses to csselr and ccsidr from >> 32-bit to 64-bit in preparation to add support to those extensions. >> >> Cc: Christoffer Dall <christoffer.dall@linaro.org> >> Cc: Marc Zyngier <marc.zyngier@arm.com> >> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com> >> --- >> arch/arm64/kvm/sys_regs.c | 18 +++++++++--------- >> 1 file changed, 9 insertions(+), 9 deletions(-) >> >> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c >> index 5dca1f10340f..a3559a8a2b0c 100644 >> --- a/arch/arm64/kvm/sys_regs.c >> +++ b/arch/arm64/kvm/sys_regs.c [..] >> @@ -2004,8 +2004,8 @@ static int demux_c15_get(u64 id, void __user *uaddr) >> >> static int demux_c15_set(u64 id, void __user *uaddr) >> { >> - u32 val, newval; >> - u32 __user *uval = uaddr; >> + u64 val, newval; >> + u64 __user *uval = uaddr; > > Doesn't converting these uval pointers to u64 cause us to break the ABI > as we'll now be reading/writing 64-bit values to userspace with the > get_user and put_user following the declarations? > Yes, I too have similar concern. IIUC it is always read via kvm_one_reg structure. I could not find any specific user for this register to cross check.
On Tue, Jan 24, 2017 at 10:15:38AM +0000, Sudeep Holla wrote: > > > On 23/01/17 21:08, Christoffer Dall wrote: > > On Fri, Jan 20, 2017 at 10:50:10AM +0000, Sudeep Holla wrote: > >> csselr and ccsidr are treated as 64-bit values already elsewhere in the > >> kernel. It also aligns well with the architecture extensions that allow > >> 64-bit format for ccsidr. > >> > >> This patch upgrades the existing accesses to csselr and ccsidr from > >> 32-bit to 64-bit in preparation to add support to those extensions. > >> > >> Cc: Christoffer Dall <christoffer.dall@linaro.org> > >> Cc: Marc Zyngier <marc.zyngier@arm.com> > >> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com> > >> --- > >> arch/arm64/kvm/sys_regs.c | 18 +++++++++--------- > >> 1 file changed, 9 insertions(+), 9 deletions(-) > >> > >> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c > >> index 5dca1f10340f..a3559a8a2b0c 100644 > >> --- a/arch/arm64/kvm/sys_regs.c > >> +++ b/arch/arm64/kvm/sys_regs.c > > [..] > > >> @@ -2004,8 +2004,8 @@ static int demux_c15_get(u64 id, void __user *uaddr) > >> > >> static int demux_c15_set(u64 id, void __user *uaddr) > >> { > >> - u32 val, newval; > >> - u32 __user *uval = uaddr; > >> + u64 val, newval; > >> + u64 __user *uval = uaddr; > > > > Doesn't converting these uval pointers to u64 cause us to break the ABI > > as we'll now be reading/writing 64-bit values to userspace with the > > get_user and put_user following the declarations? > > > > Yes, I too have similar concern. IIUC it is always read via kvm_one_reg > structure. I could not find any specific user for this register to cross > check. > Not sure it matters which interface we get the userspace pointer from? This patch is definitely changing the write from a 32-bit write to a 64-bit write and there's a specific check prior to the put_user() call which checks that userspace intended a 32-bit value and presumably provided a 32-bit pointer. So I think the only way to return 64-bit AArch32 system register values to userspace (if that is the intention) is to define a new ID for 64-bit CCSIDR registers and handle them separately. Thanks, -Christoffer
On 24/01/17 10:30, Christoffer Dall wrote: > On Tue, Jan 24, 2017 at 10:15:38AM +0000, Sudeep Holla wrote: >> >> >> On 23/01/17 21:08, Christoffer Dall wrote: >>> On Fri, Jan 20, 2017 at 10:50:10AM +0000, Sudeep Holla wrote: >>>> csselr and ccsidr are treated as 64-bit values already elsewhere in the >>>> kernel. It also aligns well with the architecture extensions that allow >>>> 64-bit format for ccsidr. >>>> >>>> This patch upgrades the existing accesses to csselr and ccsidr from >>>> 32-bit to 64-bit in preparation to add support to those extensions. >>>> >>>> Cc: Christoffer Dall <christoffer.dall@linaro.org> >>>> Cc: Marc Zyngier <marc.zyngier@arm.com> >>>> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com> >>>> --- >>>> arch/arm64/kvm/sys_regs.c | 18 +++++++++--------- >>>> 1 file changed, 9 insertions(+), 9 deletions(-) >>>> >>>> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c >>>> index 5dca1f10340f..a3559a8a2b0c 100644 >>>> --- a/arch/arm64/kvm/sys_regs.c >>>> +++ b/arch/arm64/kvm/sys_regs.c >> >> [..] >> >>>> @@ -2004,8 +2004,8 @@ static int demux_c15_get(u64 id, void __user *uaddr) >>>> >>>> static int demux_c15_set(u64 id, void __user *uaddr) >>>> { >>>> - u32 val, newval; >>>> - u32 __user *uval = uaddr; >>>> + u64 val, newval; >>>> + u64 __user *uval = uaddr; >>> >>> Doesn't converting these uval pointers to u64 cause us to break the ABI >>> as we'll now be reading/writing 64-bit values to userspace with the >>> get_user and put_user following the declarations? >>> >> >> Yes, I too have similar concern. IIUC it is always read via kvm_one_reg >> structure. I could not find any specific user for this register to cross >> check. >> > > Not sure it matters which interface we get the userspace pointer from? > Agreed. > This patch is definitely changing the write from a 32-bit write to a > 64-bit write and there's a specific check prior to the put_user() call > which checks that userspace intended a 32-bit value and presumably > provided a 32-bit pointer. > I see you point, I missed to see that check(just to be sure KVM_REG_SIZE check right ?). > So I think the only way to return 64-bit AArch32 system register values > to userspace (if that is the intention) is to define a new ID for 64-bit > CCSIDR registers and handle them separately. > I will add KVM_REG_ARM_DEMUX_ID_CCSIDR_64B or something similar. Thanks for the review.
On Tue, Jan 24, 2017 at 10:55:24AM +0000, Sudeep Holla wrote: > > > On 24/01/17 10:30, Christoffer Dall wrote: > > On Tue, Jan 24, 2017 at 10:15:38AM +0000, Sudeep Holla wrote: > >> > >> > >> On 23/01/17 21:08, Christoffer Dall wrote: > >>> On Fri, Jan 20, 2017 at 10:50:10AM +0000, Sudeep Holla wrote: > >>>> csselr and ccsidr are treated as 64-bit values already elsewhere in the > >>>> kernel. It also aligns well with the architecture extensions that allow > >>>> 64-bit format for ccsidr. > >>>> > >>>> This patch upgrades the existing accesses to csselr and ccsidr from > >>>> 32-bit to 64-bit in preparation to add support to those extensions. > >>>> > >>>> Cc: Christoffer Dall <christoffer.dall@linaro.org> > >>>> Cc: Marc Zyngier <marc.zyngier@arm.com> > >>>> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com> > >>>> --- > >>>> arch/arm64/kvm/sys_regs.c | 18 +++++++++--------- > >>>> 1 file changed, 9 insertions(+), 9 deletions(-) > >>>> > >>>> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c > >>>> index 5dca1f10340f..a3559a8a2b0c 100644 > >>>> --- a/arch/arm64/kvm/sys_regs.c > >>>> +++ b/arch/arm64/kvm/sys_regs.c > >> > >> [..] > >> > >>>> @@ -2004,8 +2004,8 @@ static int demux_c15_get(u64 id, void __user *uaddr) > >>>> > >>>> static int demux_c15_set(u64 id, void __user *uaddr) > >>>> { > >>>> - u32 val, newval; > >>>> - u32 __user *uval = uaddr; > >>>> + u64 val, newval; > >>>> + u64 __user *uval = uaddr; > >>> > >>> Doesn't converting these uval pointers to u64 cause us to break the ABI > >>> as we'll now be reading/writing 64-bit values to userspace with the > >>> get_user and put_user following the declarations? > >>> > >> > >> Yes, I too have similar concern. IIUC it is always read via kvm_one_reg > >> structure. I could not find any specific user for this register to cross > >> check. > >> > > > > Not sure it matters which interface we get the userspace pointer from? > > > > Agreed. > > > This patch is definitely changing the write from a 32-bit write to a > > 64-bit write and there's a specific check prior to the put_user() call > > which checks that userspace intended a 32-bit value and presumably > > provided a 32-bit pointer. > > > > I see you point, I missed to see that check(just to be sure KVM_REG_SIZE > check right ?). yes. > > > So I think the only way to return 64-bit AArch32 system register values > > to userspace (if that is the intention) is to define a new ID for 64-bit > > CCSIDR registers and handle them separately. > > > > I will add KVM_REG_ARM_DEMUX_ID_CCSIDR_64B or something similar. > Thanks for the review. > Cool, thanks. -Christoffer
diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c index 5dca1f10340f..a3559a8a2b0c 100644 --- a/arch/arm64/kvm/sys_regs.c +++ b/arch/arm64/kvm/sys_regs.c @@ -58,15 +58,15 @@ */ /* 3 bits per cache level, as per CLIDR, but non-existent caches always 0 */ -static u32 cache_levels; +static u64 cache_levels; /* CSSELR values; used to index KVM_REG_ARM_DEMUX_ID_CCSIDR */ #define CSSELR_MAX ((MAX_CACHE_LEVEL - 1) >> 1) /* Which cache CCSIDR represents depends on CSSELR value. */ -static u32 get_ccsidr(u32 csselr) +static u64 get_ccsidr(u64 csselr) { - u32 ccsidr; + u64 ccsidr; /* Make sure noone else changes CSSELR during this! */ local_irq_disable(); @@ -1952,9 +1952,9 @@ static int set_invariant_sys_reg(u64 id, void __user *uaddr) return 0; } -static bool is_valid_cache(u32 val) +static bool is_valid_cache(u64 val) { - u32 level, ctype; + u64 level, ctype; if (val >= CSSELR_MAX) return false; @@ -1979,8 +1979,8 @@ static bool is_valid_cache(u32 val) static int demux_c15_get(u64 id, void __user *uaddr) { - u32 val; - u32 __user *uval = uaddr; + u64 val; + u64 __user *uval = uaddr; /* Fail if we have unknown bits set. */ if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK @@ -2004,8 +2004,8 @@ static int demux_c15_get(u64 id, void __user *uaddr) static int demux_c15_set(u64 id, void __user *uaddr) { - u32 val, newval; - u32 __user *uval = uaddr; + u64 val, newval; + u64 __user *uval = uaddr; /* Fail if we have unknown bits set. */ if (id & ~(KVM_REG_ARCH_MASK|KVM_REG_SIZE_MASK|KVM_REG_ARM_COPROC_MASK
csselr and ccsidr are treated as 64-bit values already elsewhere in the kernel. It also aligns well with the architecture extensions that allow 64-bit format for ccsidr. This patch upgrades the existing accesses to csselr and ccsidr from 32-bit to 64-bit in preparation to add support to those extensions. Cc: Christoffer Dall <christoffer.dall@linaro.org> Cc: Marc Zyngier <marc.zyngier@arm.com> Signed-off-by: Sudeep Holla <sudeep.holla@arm.com> --- arch/arm64/kvm/sys_regs.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-)