* [PATCH 1/2] x86/pkeys: Change caller of update_pkru_in_sigframe()
@ 2024-11-13 18:14 Aruna Ramakrishna
2024-11-13 18:14 ` [PATCH 2/2] x86/pkeys: Set XINUSE[PKRU] to 1 so that PKRU is XRSTOR'd correctly Aruna Ramakrishna
0 siblings, 1 reply; 4+ messages in thread
From: Aruna Ramakrishna @ 2024-11-13 18:14 UTC (permalink / raw)
To: linux-kernel; +Cc: x86, dave.hansen, tglx, mingo, rudi.horn, joe.jin
The commit 70044df250d0 ("x86/pkeys: Update PKRU to enable all pkeys
before XSAVE") enables all pkeys before XSAVE to allow writing data
to the sigaltstack (i.e. it does a wrpkru(0)).
However, when PKRU is set to 0, it has the side effect of setting
XINUSE[PKRU] to 0 on AMD systems, but not on Intel. If XINUSE[i]
is set to 0, then XRSTOR will not restore the component i from the
XSAVE area - effectively ignoring the value of PKRU updated on the
sigframe by update_pkru_in_sigframe(). This means that PKRU stays
at 0 when control is returned back to the user after signal
handling, which is unexpected.
To fix this, and to ensure that PKRU is always restored correctly
from the XSAVE data on both Intel and AMD systems, explicitly set
XINUSE[PKRU] to 1.
This change is split into 2 patches - the first one merely changes
the caller of update_pkru_in_sigframe(), but the behavior remains
the same. The next patch will update the XINUSE bit for PKRU.
No functional changes.
Signed-off-by: Aruna Ramakrishna <aruna.ramakrishna@oracle.com>
---
arch/x86/kernel/fpu/signal.c | 20 ++------------------
arch/x86/kernel/fpu/xstate.h | 15 ++++++++++++++-
2 files changed, 16 insertions(+), 19 deletions(-)
diff --git a/arch/x86/kernel/fpu/signal.c b/arch/x86/kernel/fpu/signal.c
index 1065ab995305..8f62e0666dea 100644
--- a/arch/x86/kernel/fpu/signal.c
+++ b/arch/x86/kernel/fpu/signal.c
@@ -63,16 +63,6 @@ static inline bool check_xstate_in_sigframe(struct fxregs_state __user *fxbuf,
return true;
}
-/*
- * Update the value of PKRU register that was already pushed onto the signal frame.
- */
-static inline int update_pkru_in_sigframe(struct xregs_state __user *buf, u32 pkru)
-{
- if (unlikely(!cpu_feature_enabled(X86_FEATURE_OSPKE)))
- return 0;
- return __put_user(pkru, (unsigned int __user *)get_xsave_addr_user(buf, XFEATURE_PKRU));
-}
-
/*
* Signal frame handlers.
*/
@@ -168,14 +158,8 @@ static inline bool save_xstate_epilog(void __user *buf, int ia32_frame,
static inline int copy_fpregs_to_sigframe(struct xregs_state __user *buf, u32 pkru)
{
- int err = 0;
-
- if (use_xsave()) {
- err = xsave_to_user_sigframe(buf);
- if (!err)
- err = update_pkru_in_sigframe(buf, pkru);
- return err;
- }
+ if (use_xsave())
+ return xsave_to_user_sigframe(buf, pkru);
if (use_fxsr())
return fxsave_to_user_sigframe((struct fxregs_state __user *) buf);
diff --git a/arch/x86/kernel/fpu/xstate.h b/arch/x86/kernel/fpu/xstate.h
index 0b86a5002c84..6b2924fbe5b8 100644
--- a/arch/x86/kernel/fpu/xstate.h
+++ b/arch/x86/kernel/fpu/xstate.h
@@ -69,6 +69,16 @@ static inline u64 xfeatures_mask_independent(void)
return fpu_kernel_cfg.independent_features;
}
+/*
+ * Update the value of PKRU register that was already pushed onto the signal frame.
+ */
+static inline int update_pkru_in_sigframe(struct xregs_state __user *buf, u32 pkru)
+{
+ if (unlikely(!cpu_feature_enabled(X86_FEATURE_OSPKE)))
+ return 0;
+ return __put_user(pkru, (unsigned int __user *)get_xsave_addr_user(buf, XFEATURE_PKRU));
+}
+
/* XSAVE/XRSTOR wrapper functions */
#ifdef CONFIG_X86_64
@@ -256,7 +266,7 @@ static inline u64 xfeatures_need_sigframe_write(void)
* The caller has to zero buf::header before calling this because XSAVE*
* does not touch the reserved fields in the header.
*/
-static inline int xsave_to_user_sigframe(struct xregs_state __user *buf)
+static inline int xsave_to_user_sigframe(struct xregs_state __user *buf, u32 pkru)
{
/*
* Include the features which are not xsaved/rstored by the kernel
@@ -281,6 +291,9 @@ static inline int xsave_to_user_sigframe(struct xregs_state __user *buf)
XSTATE_OP(XSAVE, buf, lmask, hmask, err);
clac();
+ if (!err)
+ err = update_pkru_in_sigframe(buf, pkru);
+
return err;
}
--
2.43.5
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH 2/2] x86/pkeys: Set XINUSE[PKRU] to 1 so that PKRU is XRSTOR'd correctly
2024-11-13 18:14 [PATCH 1/2] x86/pkeys: Change caller of update_pkru_in_sigframe() Aruna Ramakrishna
@ 2024-11-13 18:14 ` Aruna Ramakrishna
2024-11-13 19:10 ` Dave Hansen
0 siblings, 1 reply; 4+ messages in thread
From: Aruna Ramakrishna @ 2024-11-13 18:14 UTC (permalink / raw)
To: linux-kernel; +Cc: x86, dave.hansen, tglx, mingo, rudi.horn, joe.jin
PKRU value is not XRSTOR'd from the XSAVE area if the corresponding
XINUSE[i] bit is 0. When PKRU value is set to 0, it sets XINUSE[PKRU]
to 0 on AMD systems, which means the value updated on the sigframe
later (after a wrpkru(0)) is ignored.
To make this behavior consistent across Intel and AMD systems, and to
ensure that the PKRU value updated on the sigframe is always restored
correctly, explicitly set XINUSE[PKRU] to 1 in
update_pkru_in_sigframe().
Fixes: 70044df250d0 ("x86/pkeys: Update PKRU to enable all pkeys before XSAVE")
Signed-off-by: Aruna Ramakrishna <aruna.ramakrishna@oracle.com>
Suggested-by: Rudi Horn <rudi.horn@oracle.com>
---
arch/x86/kernel/fpu/xstate.h | 16 ++++++++++++++--
1 file changed, 14 insertions(+), 2 deletions(-)
diff --git a/arch/x86/kernel/fpu/xstate.h b/arch/x86/kernel/fpu/xstate.h
index 6b2924fbe5b8..b5056fcdd21e 100644
--- a/arch/x86/kernel/fpu/xstate.h
+++ b/arch/x86/kernel/fpu/xstate.h
@@ -72,10 +72,22 @@ static inline u64 xfeatures_mask_independent(void)
/*
* Update the value of PKRU register that was already pushed onto the signal frame.
*/
-static inline int update_pkru_in_sigframe(struct xregs_state __user *buf, u32 pkru)
+static inline int update_pkru_in_sigframe(struct xregs_state __user *buf, u64 mask, u32 pkru)
{
+ u64 xinuse;
+ int err;
+
if (unlikely(!cpu_feature_enabled(X86_FEATURE_OSPKE)))
return 0;
+
+ /* Mark PKRU as in-use so that it is restored correctly. */
+ xinuse = (mask & xfeatures_in_use()) | XFEATURE_MASK_PKRU;
+
+ err = __put_user(xinuse, &buf->header.xfeatures);
+ if (err)
+ return err;
+
+ /* Update PKRU value in the userspace xsave buffer. */
return __put_user(pkru, (unsigned int __user *)get_xsave_addr_user(buf, XFEATURE_PKRU));
}
@@ -292,7 +304,7 @@ static inline int xsave_to_user_sigframe(struct xregs_state __user *buf, u32 pkr
clac();
if (!err)
- err = update_pkru_in_sigframe(buf, pkru);
+ err = update_pkru_in_sigframe(buf, mask, pkru);
return err;
}
--
2.43.5
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH 2/2] x86/pkeys: Set XINUSE[PKRU] to 1 so that PKRU is XRSTOR'd correctly
2024-11-13 18:14 ` [PATCH 2/2] x86/pkeys: Set XINUSE[PKRU] to 1 so that PKRU is XRSTOR'd correctly Aruna Ramakrishna
@ 2024-11-13 19:10 ` Dave Hansen
2024-11-13 20:04 ` Aruna Ramakrishna
0 siblings, 1 reply; 4+ messages in thread
From: Dave Hansen @ 2024-11-13 19:10 UTC (permalink / raw)
To: Aruna Ramakrishna, linux-kernel
Cc: x86, dave.hansen, tglx, mingo, rudi.horn, joe.jin
On 11/13/24 10:14, Aruna Ramakrishna wrote:
> PKRU value is not XRSTOR'd from the XSAVE area if the corresponding
> XINUSE[i] bit is 0. When PKRU value is set to 0, it sets XINUSE[PKRU]
> to 0 on AMD systems, which means the value updated on the sigframe
> later (after a wrpkru(0)) is ignored.
I think this is confusing XINUSE and XSTATE_BV.
XINUSE is really internal to the CPU and is partially exposed in
xgetbv(1). But XINUSE is monolithic; it includes user and supervisor state.
XSTATE_BV is the actual memory location in the XSAVE buffer.
So I think it's incorrect to say that XRSTOR behavior depends on XINUSE.
XRSTOR behavior depends on XSTATE_BV.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH 2/2] x86/pkeys: Set XINUSE[PKRU] to 1 so that PKRU is XRSTOR'd correctly
2024-11-13 19:10 ` Dave Hansen
@ 2024-11-13 20:04 ` Aruna Ramakrishna
0 siblings, 0 replies; 4+ messages in thread
From: Aruna Ramakrishna @ 2024-11-13 20:04 UTC (permalink / raw)
To: Dave Hansen
Cc: linux-kernel, x86, dave.hansen, Thomas Gleixner, Ingo Molnar,
Rudi Horn, Joe Jin
> On Nov 13, 2024, at 11:10 AM, Dave Hansen <dave.hansen@intel.com> wrote:
>
> On 11/13/24 10:14, Aruna Ramakrishna wrote:
>> PKRU value is not XRSTOR'd from the XSAVE area if the corresponding
>> XINUSE[i] bit is 0. When PKRU value is set to 0, it sets XINUSE[PKRU]
>> to 0 on AMD systems, which means the value updated on the sigframe
>> later (after a wrpkru(0)) is ignored.
>
> I think this is confusing XINUSE and XSTATE_BV.
>
> XINUSE is really internal to the CPU and is partially exposed in
> xgetbv(1). But XINUSE is monolithic; it includes user and supervisor state.
>
> XSTATE_BV is the actual memory location in the XSAVE buffer.
>
> So I think it's incorrect to say that XRSTOR behavior depends on XINUSE.
> XRSTOR behavior depends on XSTATE_BV.
You’re right; XSTATE_BV[i] is set during XSAVE based on XINUSE[i], and
XRSTOR restores based on XSTATE_BV. I will reword that (and change the
variable name in update_pkru_in_sigframe()) for v2.
Thanks,
Aruna
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2024-11-13 20:05 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-11-13 18:14 [PATCH 1/2] x86/pkeys: Change caller of update_pkru_in_sigframe() Aruna Ramakrishna
2024-11-13 18:14 ` [PATCH 2/2] x86/pkeys: Set XINUSE[PKRU] to 1 so that PKRU is XRSTOR'd correctly Aruna Ramakrishna
2024-11-13 19:10 ` Dave Hansen
2024-11-13 20:04 ` Aruna Ramakrishna
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®