AMD CPUs currently execute WBINVD in the host when unregistering SEV
guest memory or when deactivating SEV guests. Such cache maintenance is
performed to prevent data corruption, wherein the encrypted (C=1)
version of a dirty cache line might otherwise only be written back
after the memory is written in a different context (ex: C=0), yielding
corruption. However, WBINVD is performance-costly, especially because
it invalidates processor caches.

Strictly-speaking, unless the SEV ASID is being recycled (meaning all
existing cache lines with the recycled ASID must be flushed), the
cache invalidation triggered by WBINVD is unnecessary; only the
writeback is needed to prevent data corruption in remaining scenarios.

To improve performance in these scenarios, use WBNOINVD when available
instead of WBINVD. WBNOINVD still writes back all dirty lines
(preventing host data corruption by SEV guests) but does *not*
invalidate processor caches.

First, provide helper functions to use WBNOINVD similar to how WBINVD
is invoked. Second, check for WBNOINVD support and execute WBNOINVD if
possible in lieu of WBINVD to avoid cache invalidations.

Note that I have *not* rebased this series atop proposed targeted
flushing optimizations [0], since the optimizations do not yet appear
to be finalized. However, I'm happy to do a rebase if that would be
helpful.

[0] https://lore.kernel.org/kvm/85frlcvjyo.fsf@amd.com/T/

Changelog
---
v4:
- add comments to wbnoinvd() for clarity on when to use and behavior
v3:
- rebase to tip @ e6609f8bea4a
- use WBINVD in wbnoinvd() if X86_FEATURE_WBNOINVD is not present
- provide sev_writeback_caches() wrapper function in anticipation of
  aforementioned [0] targeted flushing optimizations
- add Reviewed-by from Mingwei
- reword commits/comments
v2:
- rebase to tip @ dffeaed35cef
- drop unnecessary Xen changes
- reword commits/comments
---
Kevin Loughlin (2):
  x86, lib: Add WBNOINVD helper functions
  KVM: SEV: Prefer WBNOINVD over WBINVD for cache maintenance efficiency

 arch/x86/include/asm/smp.h           |  7 +++++
 arch/x86/include/asm/special_insns.h | 15 +++++++++-
 arch/x86/kvm/svm/sev.c               | 41 ++++++++++++++--------------
 arch/x86/lib/cache-smp.c             | 12 ++++++++
 4 files changed, 54 insertions(+), 21 deletions(-)

-- 
2.48.1.262.g85cc9f2d1e-goog

AMD CPUs currently execute WBINVD in the host when unregistering SEV
guest memory or when deactivating SEV guests. Such cache maintenance is
performed to prevent data corruption, wherein the encrypted (C=1)
version of a dirty cache line might otherwise only be written back
after the memory is written in a different context (ex: C=0), yielding
corruption. However, WBINVD is performance-costly, especially because
it invalidates processor caches.

Strictly-speaking, unless the SEV ASID is being recycled (meaning all
existing cache lines with the recycled ASID must be flushed), the
cache invalidation triggered by WBINVD is unnecessary; only the
writeback is needed to prevent data corruption in remaining scenarios.

To improve performance in these scenarios, use WBNOINVD when available
instead of WBINVD. WBNOINVD still writes back all dirty lines
(preventing host data corruption by SEV guests) but does *not*
invalidate processor caches.

First, provide helper functions to use WBNOINVD similar to how WBINVD
is invoked. Second, check for WBNOINVD support and execute WBNOINVD if
possible in lieu of WBINVD to avoid cache invalidations.

Note that I have *not* rebased this series atop proposed targeted
flushing optimizations [0], since the optimizations do not yet appear
to be finalized. However, I'm happy to do a rebase if that would be
helpful.

[0] https://lore.kernel.org/kvm/85frlcvjyo.fsf@amd.com/T/

Changelog
---
v5:
- explicitly encode wbnoinvd as 0xf3 0x0f 0x09 for binutils backwards compatibility
v4:
- add comments to wbnoinvd() for clarity on when to use and behavior
v3:
- rebase to tip @ e6609f8bea4a
- use WBINVD in wbnoinvd() if X86_FEATURE_WBNOINVD is not present
- provide sev_writeback_caches() wrapper function in anticipation of
  aforementioned [0] targeted flushing optimizations
- add Reviewed-by from Mingwei
- reword commits/comments
v2:
- rebase to tip @ dffeaed35cef
- drop unnecessary Xen changes
- reword commits/comments
---
Kevin Loughlin (2):
  x86, lib: Add WBNOINVD helper functions
  KVM: SEV: Prefer WBNOINVD over WBINVD for cache maintenance efficiency

 arch/x86/include/asm/smp.h           |  7 +++++
 arch/x86/include/asm/special_insns.h | 20 +++++++++++++-
 arch/x86/kvm/svm/sev.c               | 41 ++++++++++++++--------------
 arch/x86/lib/cache-smp.c             | 12 ++++++++
 4 files changed, 59 insertions(+), 21 deletions(-)

-- 
2.48.1.262.g85cc9f2d1e-goog

AMD CPUs currently execute WBINVD in the host when unregistering SEV
guest memory or when deactivating SEV guests. Such cache maintenance is
performed to prevent data corruption, wherein the encrypted (C=1)
version of a dirty cache line might otherwise only be written back
after the memory is written in a different context (ex: C=0), yielding
corruption. However, WBINVD is performance-costly, especially because
it invalidates processor caches.

Strictly-speaking, unless the SEV ASID is being recycled (meaning all
existing cache lines with the recycled ASID must be flushed), the
cache invalidation triggered by WBINVD is unnecessary; only the
writeback is needed to prevent data corruption in remaining scenarios.

To improve performance in these scenarios, use WBNOINVD when available
instead of WBINVD. WBNOINVD still writes back all dirty lines
(preventing host data corruption by SEV guests) but does *not*
invalidate processor caches.

First, provide helper functions to use WBNOINVD similar to how WBINVD
is invoked. Second, check for WBNOINVD support and execute WBNOINVD if
possible in lieu of WBINVD to avoid cache invalidations.

Note that I have *not* rebased this series atop proposed targeted
flushing optimizations [0], since the optimizations do not yet appear
to be finalized. However, I'm happy to do a rebase if that would be
helpful.

[0] https://lore.kernel.org/kvm/85frlcvjyo.fsf@amd.com/T/

Changelog
---
v6:
- hoist WBNONINVD encoding into macro for readability; shorten comment
v5:
- explicitly encode wbnoinvd as 0xf3 0x0f 0x09 for binutils backwards compatibility
v4:
- add comments to wbnoinvd() for clarity on when to use and behavior
v3:
- rebase to tip @ e6609f8bea4a
- use WBINVD in wbnoinvd() if X86_FEATURE_WBNOINVD is not present
- provide sev_writeback_caches() wrapper function in anticipation of
  aforementioned [0] targeted flushing optimizations
- add Reviewed-by from Mingwei
- reword commits/comments
v2:
- rebase to tip @ dffeaed35cef
- drop unnecessary Xen changes
- reword commits/comments
---
Kevin Loughlin (2):
  x86, lib: Add WBNOINVD helper functions
  KVM: SEV: Prefer WBNOINVD over WBINVD for cache maintenance efficiency

 arch/x86/include/asm/smp.h           |  7 +++++
 arch/x86/include/asm/special_insns.h | 19 ++++++++++++-
 arch/x86/kvm/svm/sev.c               | 41 ++++++++++++++--------------
 arch/x86/lib/cache-smp.c             | 12 ++++++++
 4 files changed, 58 insertions(+), 21 deletions(-)

-- 
2.48.1.362.g079036d154-goog

In line with WBINVD usage, add WBONINVD helper functions. For the
wbnoinvd() helper, fall back to WBINVD if via alternative() if
X86_FEATURE_WBNOINVD is not present. alternative() ensures
compatibility with early boot code if needed.

Signed-off-by: Kevin Loughlin <kevinloughlin@google.com>
---
 arch/x86/include/asm/smp.h           |  7 +++++++
 arch/x86/include/asm/special_insns.h | 19 ++++++++++++++++++-
 arch/x86/lib/cache-smp.c             | 12 ++++++++++++
 3 files changed, 37 insertions(+), 1 deletion(-)

diff --git a/arch/x86/include/asm/smp.h b/arch/x86/include/asm/smp.h
index ca073f40698f..ecf93a243b83 100644
--- a/arch/x86/include/asm/smp.h
+++ b/arch/x86/include/asm/smp.h
@@ -112,6 +112,7 @@ void native_play_dead(void);
 void play_dead_common(void);
 void wbinvd_on_cpu(int cpu);
 int wbinvd_on_all_cpus(void);
+int wbnoinvd_on_all_cpus(void);
 
 void smp_kick_mwait_play_dead(void);
 
@@ -160,6 +161,12 @@ static inline int wbinvd_on_all_cpus(void)
 	return 0;
 }
 
+static inline int wbnoinvd_on_all_cpus(void)
+{
+	wbnoinvd();
+	return 0;
+}
+
 static inline struct cpumask *cpu_llc_shared_mask(int cpu)
 {
 	return (struct cpumask *)cpumask_of(0);
diff --git a/arch/x86/include/asm/special_insns.h b/arch/x86/include/asm/special_insns.h
index 03e7c2d49559..86a903742139 100644
--- a/arch/x86/include/asm/special_insns.h
+++ b/arch/x86/include/asm/special_insns.h
@@ -117,7 +117,24 @@ static inline void wrpkru(u32 pkru)
 
 static __always_inline void wbinvd(void)
 {
-	asm volatile("wbinvd": : :"memory");
+	asm volatile("wbinvd" : : : "memory");
+}
+
+/* Instruction encoding provided for binutils backwards compatibility. */
+#define WBNOINVD ".byte 0xf3,0x0f,0x09"
+
+/*
+ * Cheaper version of wbinvd(). Call when caches
+ * need to be written back but not invalidated.
+ */
+static __always_inline void wbnoinvd(void)
+{
+	/*
+	 * If WBNOINVD is unavailable, fall back to the compatible but
+	 * more destructive WBINVD (which still writes the caches back
+	 * but also invalidates them).
+	 */
+	alternative("wbinvd", WBNOINVD, X86_FEATURE_WBNOINVD);
 }
 
 static inline unsigned long __read_cr4(void)
diff --git a/arch/x86/lib/cache-smp.c b/arch/x86/lib/cache-smp.c
index 7af743bd3b13..7ac5cca53031 100644
--- a/arch/x86/lib/cache-smp.c
+++ b/arch/x86/lib/cache-smp.c
@@ -20,3 +20,15 @@ int wbinvd_on_all_cpus(void)
 	return 0;
 }
 EXPORT_SYMBOL(wbinvd_on_all_cpus);
+
+static void __wbnoinvd(void *dummy)
+{
+	wbnoinvd();
+}
+
+int wbnoinvd_on_all_cpus(void)
+{
+	on_each_cpu(__wbnoinvd, NULL, 1);
+	return 0;
+}
+EXPORT_SYMBOL(wbnoinvd_on_all_cpus);
-- 
2.48.1.362.g079036d154-goog

On Sat, Feb 01, 2025, Kevin Loughlin wrote:
> +static inline int wbnoinvd_on_all_cpus(void)
> +{
> +	wbnoinvd();
> +	return 0;

Returning anything is silly.  I'll prepend a patch (I'm going to send a combined
version of this and the targeted flushing series) to remove the return value
from wbinvd_on_all_cpus(), which I'm guessing is the source of the silliness.

>  static inline struct cpumask *cpu_llc_shared_mask(int cpu)
>  {
>  	return (struct cpumask *)cpumask_of(0);
> diff --git a/arch/x86/include/asm/special_insns.h b/arch/x86/include/asm/special_insns.h
> index 03e7c2d49559..86a903742139 100644
> --- a/arch/x86/include/asm/special_insns.h
> +++ b/arch/x86/include/asm/special_insns.h
> @@ -117,7 +117,24 @@ static inline void wrpkru(u32 pkru)
>  
>  static __always_inline void wbinvd(void)
>  {
> -	asm volatile("wbinvd": : :"memory");
> +	asm volatile("wbinvd" : : : "memory");
> +}
> +
> +/* Instruction encoding provided for binutils backwards compatibility. */
> +#define WBNOINVD ".byte 0xf3,0x0f,0x09"

Argh.  This causes problems for KVM, because KVM's newfangled CPUID macros heavily
use token pasting with X86_FEATURE_xxx, and so KVM's usage of:

	F(WBNOINVD)

causes explosions.  Somewhat amusingly, this is the second time today I ran into
this problem, as WRMSRNS as the safe issue.

Dave (and others),

Any thoughts on the best way forward?  I hacked around a similar collision in
commit 8d862c270bf1 ("KVM: x86: #undef SPEC_CTRL_SSBD in cpuid.c to avoid macro
collisions"), but (a) that scares me less because KVM should never use the
SPEC_CTRL_SSBD macro, and (b) I really, really don't want that to be the long-term
solution.  The only reason I committed the hack was because it was the only
blocking issue for a massive rework, and I couldn't get traction on renaming
the MSR macro.

For WBNOINVD, WRMSRNS, and any other instructions that come along, what about this?
Quite ugly, but it's at least descriptive.  And more importantly, the chances of
unwanted behavior are quite low.


/* Instruction encoding provided for binutils backwards compatibility. */
#define ASM_WBNOINVD ".byte 0xf3,0x0f,0x09"

/*
 * Cheaper version of wbinvd(). Call when caches need to be written back but
 * not invalidated.
 */
static __always_inline void wbnoinvd(void)
{
	/*
	 * If WBNOINVD is unavailable, fall back to the compatible but
	 * more destructive WBINVD (which still writes the caches back
	 * but also invalidates them).
	 */
	alternative("wbinvd", ASM_WBNOINVD, X86_FEATURE_WBNOINVD);
}

On Tue, Feb 25, 2025 at 05:26:53PM -0800, Sean Christopherson wrote:
> /* Instruction encoding provided for binutils backwards compatibility. */
> #define ASM_WBNOINVD ".byte 0xf3,0x0f,0x09"

Yah, or simply INSN_WBNOINVD as that name basically tells you what it is. And
we have already

arch/x86/include/asm/bug.h:13:#define INSN_UD2  0x0b0f

oh and

arch/x86/include/asm/bug.h:12:#define ASM_UD2           ".byte 0x0f, 0x0b"

Pff.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

On 1/31/25 18:02, Kevin Loughlin wrote:
> In line with WBINVD usage, add WBONINVD helper functions. For the
> wbnoinvd() helper, fall back to WBINVD if via alternative() if
> X86_FEATURE_WBNOINVD is not present. alternative() ensures
> compatibility with early boot code if needed.
> 
> Signed-off-by: Kevin Loughlin <kevinloughlin@google.com>

Reviewed-by: Tom Lendacky <thomas.lendacky@amd.com>

> ---
>  arch/x86/include/asm/smp.h           |  7 +++++++
>  arch/x86/include/asm/special_insns.h | 19 ++++++++++++++++++-
>  arch/x86/lib/cache-smp.c             | 12 ++++++++++++
>  3 files changed, 37 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/include/asm/smp.h b/arch/x86/include/asm/smp.h
> index ca073f40698f..ecf93a243b83 100644
> --- a/arch/x86/include/asm/smp.h
> +++ b/arch/x86/include/asm/smp.h
> @@ -112,6 +112,7 @@ void native_play_dead(void);
>  void play_dead_common(void);
>  void wbinvd_on_cpu(int cpu);
>  int wbinvd_on_all_cpus(void);
> +int wbnoinvd_on_all_cpus(void);
>  
>  void smp_kick_mwait_play_dead(void);
>  
> @@ -160,6 +161,12 @@ static inline int wbinvd_on_all_cpus(void)
>  	return 0;
>  }
>  
> +static inline int wbnoinvd_on_all_cpus(void)
> +{
> +	wbnoinvd();
> +	return 0;
> +}
> +
>  static inline struct cpumask *cpu_llc_shared_mask(int cpu)
>  {
>  	return (struct cpumask *)cpumask_of(0);
> diff --git a/arch/x86/include/asm/special_insns.h b/arch/x86/include/asm/special_insns.h
> index 03e7c2d49559..86a903742139 100644
> --- a/arch/x86/include/asm/special_insns.h
> +++ b/arch/x86/include/asm/special_insns.h
> @@ -117,7 +117,24 @@ static inline void wrpkru(u32 pkru)
>  
>  static __always_inline void wbinvd(void)
>  {
> -	asm volatile("wbinvd": : :"memory");
> +	asm volatile("wbinvd" : : : "memory");
> +}
> +
> +/* Instruction encoding provided for binutils backwards compatibility. */
> +#define WBNOINVD ".byte 0xf3,0x0f,0x09"
> +
> +/*
> + * Cheaper version of wbinvd(). Call when caches
> + * need to be written back but not invalidated.
> + */
> +static __always_inline void wbnoinvd(void)
> +{
> +	/*
> +	 * If WBNOINVD is unavailable, fall back to the compatible but
> +	 * more destructive WBINVD (which still writes the caches back
> +	 * but also invalidates them).
> +	 */
> +	alternative("wbinvd", WBNOINVD, X86_FEATURE_WBNOINVD);
>  }
>  
>  static inline unsigned long __read_cr4(void)
> diff --git a/arch/x86/lib/cache-smp.c b/arch/x86/lib/cache-smp.c
> index 7af743bd3b13..7ac5cca53031 100644
> --- a/arch/x86/lib/cache-smp.c
> +++ b/arch/x86/lib/cache-smp.c
> @@ -20,3 +20,15 @@ int wbinvd_on_all_cpus(void)
>  	return 0;
>  }
>  EXPORT_SYMBOL(wbinvd_on_all_cpus);
> +
> +static void __wbnoinvd(void *dummy)
> +{
> +	wbnoinvd();
> +}
> +
> +int wbnoinvd_on_all_cpus(void)
> +{
> +	on_each_cpu(__wbnoinvd, NULL, 1);
> +	return 0;
> +}
> +EXPORT_SYMBOL(wbnoinvd_on_all_cpus);

AMD CPUs currently execute WBINVD in the host when unregistering SEV
guest memory or when deactivating SEV guests. Such cache maintenance is
performed to prevent data corruption, wherein the encrypted (C=1)
version of a dirty cache line might otherwise only be written back
after the memory is written in a different context (ex: C=0), yielding
corruption. However, WBINVD is performance-costly, especially because
it invalidates processor caches.

Strictly-speaking, unless the SEV ASID is being recycled (meaning the
SNP firmware requires the use of WBINVD prior to DF_FLUSH), the cache
invalidation triggered by WBINVD is unnecessary; only the writeback is
needed to prevent data corruption in remaining scenarios.

To improve performance in these scenarios, use WBNOINVD when available
instead of WBINVD. WBNOINVD still writes back all dirty lines
(preventing host data corruption by SEV guests) but does *not*
invalidate processor caches. Note that the implementation of wbnoinvd()
ensures fall back to WBINVD if WBNOINVD is unavailable.

In anticipation of forthcoming optimizations to limit the WBNOINVD only
to physical CPUs that have executed SEV guests, place the call to
wbnoinvd_on_all_cpus() in a wrapper function sev_writeback_caches().

Signed-off-by: Kevin Loughlin <kevinloughlin@google.com>
Reviewed-by: Mingwei Zhang <mizhang@google.com>
---
 arch/x86/kvm/svm/sev.c | 41 +++++++++++++++++++++--------------------
 1 file changed, 21 insertions(+), 20 deletions(-)

diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
index fe6cc763fd51..f10f1c53345e 100644
--- a/arch/x86/kvm/svm/sev.c
+++ b/arch/x86/kvm/svm/sev.c
@@ -116,6 +116,7 @@ static int sev_flush_asids(unsigned int min_asid, unsigned int max_asid)
 	 */
 	down_write(&sev_deactivate_lock);
 
+	/* SNP firmware requires use of WBINVD for ASID recycling. */
 	wbinvd_on_all_cpus();
 
 	if (sev_snp_enabled)
@@ -710,6 +711,16 @@ static void sev_clflush_pages(struct page *pages[], unsigned long npages)
 	}
 }
 
+static inline void sev_writeback_caches(void)
+{
+	/*
+	 * Ensure that all dirty guest tagged cache entries are written back
+	 * before releasing the pages back to the system for use. CLFLUSH will
+	 * not do this without SME_COHERENT, so issue a WBNOINVD.
+	 */
+	wbnoinvd_on_all_cpus();
+}
+
 static unsigned long get_num_contig_pages(unsigned long idx,
 				struct page **inpages, unsigned long npages)
 {
@@ -2773,12 +2784,7 @@ int sev_mem_enc_unregister_region(struct kvm *kvm,
 		goto failed;
 	}
 
-	/*
-	 * Ensure that all guest tagged cache entries are flushed before
-	 * releasing the pages back to the system for use. CLFLUSH will
-	 * not do this, so issue a WBINVD.
-	 */
-	wbinvd_on_all_cpus();
+	sev_writeback_caches();
 
 	__unregister_enc_region_locked(kvm, region);
 
@@ -2899,12 +2905,7 @@ void sev_vm_destroy(struct kvm *kvm)
 		return;
 	}
 
-	/*
-	 * Ensure that all guest tagged cache entries are flushed before
-	 * releasing the pages back to the system for use. CLFLUSH will
-	 * not do this, so issue a WBINVD.
-	 */
-	wbinvd_on_all_cpus();
+	sev_writeback_caches();
 
 	/*
 	 * if userspace was terminated before unregistering the memory regions
@@ -3126,16 +3127,16 @@ static void sev_flush_encrypted_page(struct kvm_vcpu *vcpu, void *va)
 
 	/*
 	 * VM Page Flush takes a host virtual address and a guest ASID.  Fall
-	 * back to WBINVD if this faults so as not to make any problems worse
+	 * back to WBNOINVD if this faults so as not to make any problems worse
 	 * by leaving stale encrypted data in the cache.
 	 */
 	if (WARN_ON_ONCE(wrmsrl_safe(MSR_AMD64_VM_PAGE_FLUSH, addr | asid)))
-		goto do_wbinvd;
+		goto do_sev_writeback_caches;
 
 	return;
 
-do_wbinvd:
-	wbinvd_on_all_cpus();
+do_sev_writeback_caches:
+	sev_writeback_caches();
 }
 
 void sev_guest_memory_reclaimed(struct kvm *kvm)
@@ -3144,12 +3145,12 @@ void sev_guest_memory_reclaimed(struct kvm *kvm)
 	 * With SNP+gmem, private/encrypted memory is unreachable via the
 	 * hva-based mmu notifiers, so these events are only actually
 	 * pertaining to shared pages where there is no need to perform
-	 * the WBINVD to flush associated caches.
+	 * the WBNOINVD to flush associated caches.
 	 */
 	if (!sev_guest(kvm) || sev_snp_guest(kvm))
 		return;
 
-	wbinvd_on_all_cpus();
+	sev_writeback_caches();
 }
 
 void sev_free_vcpu(struct kvm_vcpu *vcpu)
@@ -3858,7 +3859,7 @@ static int __sev_snp_update_protected_guest_state(struct kvm_vcpu *vcpu)
 		 * guest-mapped page rather than the initial one allocated
 		 * by KVM in svm->sev_es.vmsa. In theory, svm->sev_es.vmsa
 		 * could be free'd and cleaned up here, but that involves
-		 * cleanups like wbinvd_on_all_cpus() which would ideally
+		 * cleanups like sev_writeback_caches() which would ideally
 		 * be handled during teardown rather than guest boot.
 		 * Deferring that also allows the existing logic for SEV-ES
 		 * VMSAs to be re-used with minimal SNP-specific changes.
@@ -4910,7 +4911,7 @@ void sev_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end)
 
 		/*
 		 * SEV-ES avoids host/guest cache coherency issues through
-		 * WBINVD hooks issued via MMU notifiers during run-time, and
+		 * WBNOINVD hooks issued via MMU notifiers during run-time, and
 		 * KVM's VM destroy path at shutdown. Those MMU notifier events
 		 * don't cover gmem since there is no requirement to map pages
 		 * to a HVA in order to use them for a running guest. While the
-- 
2.48.1.362.g079036d154-goog

On 1/31/25 18:02, Kevin Loughlin wrote:
> AMD CPUs currently execute WBINVD in the host when unregistering SEV
> guest memory or when deactivating SEV guests. Such cache maintenance is
> performed to prevent data corruption, wherein the encrypted (C=1)
> version of a dirty cache line might otherwise only be written back
> after the memory is written in a different context (ex: C=0), yielding
> corruption. However, WBINVD is performance-costly, especially because
> it invalidates processor caches.
> 
> Strictly-speaking, unless the SEV ASID is being recycled (meaning the
> SNP firmware requires the use of WBINVD prior to DF_FLUSH), the cache
> invalidation triggered by WBINVD is unnecessary; only the writeback is
> needed to prevent data corruption in remaining scenarios.
> 
> To improve performance in these scenarios, use WBNOINVD when available
> instead of WBINVD. WBNOINVD still writes back all dirty lines
> (preventing host data corruption by SEV guests) but does *not*
> invalidate processor caches. Note that the implementation of wbnoinvd()
> ensures fall back to WBINVD if WBNOINVD is unavailable.
> 
> In anticipation of forthcoming optimizations to limit the WBNOINVD only
> to physical CPUs that have executed SEV guests, place the call to
> wbnoinvd_on_all_cpus() in a wrapper function sev_writeback_caches().
> 
> Signed-off-by: Kevin Loughlin <kevinloughlin@google.com>
> Reviewed-by: Mingwei Zhang <mizhang@google.com>

Reviewed-by: Tom Lendacky <thomas.lendacky@amd.com>

> ---
>  arch/x86/kvm/svm/sev.c | 41 +++++++++++++++++++++--------------------
>  1 file changed, 21 insertions(+), 20 deletions(-)
> 
> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> index fe6cc763fd51..f10f1c53345e 100644
> --- a/arch/x86/kvm/svm/sev.c
> +++ b/arch/x86/kvm/svm/sev.c
> @@ -116,6 +116,7 @@ static int sev_flush_asids(unsigned int min_asid, unsigned int max_asid)
>  	 */
>  	down_write(&sev_deactivate_lock);
>  
> +	/* SNP firmware requires use of WBINVD for ASID recycling. */
>  	wbinvd_on_all_cpus();
>  
>  	if (sev_snp_enabled)
> @@ -710,6 +711,16 @@ static void sev_clflush_pages(struct page *pages[], unsigned long npages)
>  	}
>  }
>  
> +static inline void sev_writeback_caches(void)
> +{
> +	/*
> +	 * Ensure that all dirty guest tagged cache entries are written back
> +	 * before releasing the pages back to the system for use. CLFLUSH will
> +	 * not do this without SME_COHERENT, so issue a WBNOINVD.
> +	 */
> +	wbnoinvd_on_all_cpus();
> +}
> +
>  static unsigned long get_num_contig_pages(unsigned long idx,
>  				struct page **inpages, unsigned long npages)
>  {
> @@ -2773,12 +2784,7 @@ int sev_mem_enc_unregister_region(struct kvm *kvm,
>  		goto failed;
>  	}
>  
> -	/*
> -	 * Ensure that all guest tagged cache entries are flushed before
> -	 * releasing the pages back to the system for use. CLFLUSH will
> -	 * not do this, so issue a WBINVD.
> -	 */
> -	wbinvd_on_all_cpus();
> +	sev_writeback_caches();
>  
>  	__unregister_enc_region_locked(kvm, region);
>  
> @@ -2899,12 +2905,7 @@ void sev_vm_destroy(struct kvm *kvm)
>  		return;
>  	}
>  
> -	/*
> -	 * Ensure that all guest tagged cache entries are flushed before
> -	 * releasing the pages back to the system for use. CLFLUSH will
> -	 * not do this, so issue a WBINVD.
> -	 */
> -	wbinvd_on_all_cpus();
> +	sev_writeback_caches();
>  
>  	/*
>  	 * if userspace was terminated before unregistering the memory regions
> @@ -3126,16 +3127,16 @@ static void sev_flush_encrypted_page(struct kvm_vcpu *vcpu, void *va)
>  
>  	/*
>  	 * VM Page Flush takes a host virtual address and a guest ASID.  Fall
> -	 * back to WBINVD if this faults so as not to make any problems worse
> +	 * back to WBNOINVD if this faults so as not to make any problems worse
>  	 * by leaving stale encrypted data in the cache.
>  	 */
>  	if (WARN_ON_ONCE(wrmsrl_safe(MSR_AMD64_VM_PAGE_FLUSH, addr | asid)))
> -		goto do_wbinvd;
> +		goto do_sev_writeback_caches;
>  
>  	return;
>  
> -do_wbinvd:
> -	wbinvd_on_all_cpus();
> +do_sev_writeback_caches:
> +	sev_writeback_caches();
>  }
>  
>  void sev_guest_memory_reclaimed(struct kvm *kvm)
> @@ -3144,12 +3145,12 @@ void sev_guest_memory_reclaimed(struct kvm *kvm)
>  	 * With SNP+gmem, private/encrypted memory is unreachable via the
>  	 * hva-based mmu notifiers, so these events are only actually
>  	 * pertaining to shared pages where there is no need to perform
> -	 * the WBINVD to flush associated caches.
> +	 * the WBNOINVD to flush associated caches.
>  	 */
>  	if (!sev_guest(kvm) || sev_snp_guest(kvm))
>  		return;
>  
> -	wbinvd_on_all_cpus();
> +	sev_writeback_caches();
>  }
>  
>  void sev_free_vcpu(struct kvm_vcpu *vcpu)
> @@ -3858,7 +3859,7 @@ static int __sev_snp_update_protected_guest_state(struct kvm_vcpu *vcpu)
>  		 * guest-mapped page rather than the initial one allocated
>  		 * by KVM in svm->sev_es.vmsa. In theory, svm->sev_es.vmsa
>  		 * could be free'd and cleaned up here, but that involves
> -		 * cleanups like wbinvd_on_all_cpus() which would ideally
> +		 * cleanups like sev_writeback_caches() which would ideally
>  		 * be handled during teardown rather than guest boot.
>  		 * Deferring that also allows the existing logic for SEV-ES
>  		 * VMSAs to be re-used with minimal SNP-specific changes.
> @@ -4910,7 +4911,7 @@ void sev_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end)
>  
>  		/*
>  		 * SEV-ES avoids host/guest cache coherency issues through
> -		 * WBINVD hooks issued via MMU notifiers during run-time, and
> +		 * WBNOINVD hooks issued via MMU notifiers during run-time, and
>  		 * KVM's VM destroy path at shutdown. Those MMU notifier events
>  		 * don't cover gmem since there is no requirement to map pages
>  		 * to a HVA in order to use them for a running guest. While the

On Thu, Jan 23, 2025, Kevin Loughlin wrote:
> v5:
> - explicitly encode wbnoinvd as 0xf3 0x0f 0x09 for binutils backwards compatibility

Please, please, please do not send new series with In-Reply-To.  Trying to sort
through the different versions in my workflow was painful.  From
Documentation/process/maintainer-kvm-x86.rst:

Links
~~~~~
Do not explicitly reference bug reports, prior versions of a patch/series, etc.
via ``In-Reply-To:`` headers.  Using ``In-Reply-To:`` becomes an unholy mess
for large series and/or when the version count gets high, and ``In-Reply-To:``
is useless for anyone that doesn't have the original message, e.g. if someone
wasn't Cc'd on the bug report or if the list of recipients changes between
versions.

To link to a bug report, previous version, or anything of interest, use lore
links.  For referencing previous version(s), generally speaking do not include
a Link: in the changelog as there is no need to record the history in git, i.e.
put the link in the cover letter or in the section git ignores.  Do provide a
formal Link: for bug reports and/or discussions that led to the patch.  The
context of why a change was made is highly valuable for future readers.

In line with WBINVD usage, add WBONINVD helper functions. For the
wbnoinvd() helper, fall back to WBINVD if X86_FEATURE_WBNOINVD is not
present.

Signed-off-by: Kevin Loughlin <kevinloughlin@google.com>
---
 arch/x86/include/asm/smp.h           |  7 +++++++
 arch/x86/include/asm/special_insns.h | 20 +++++++++++++++++++-
 arch/x86/lib/cache-smp.c             | 12 ++++++++++++
 3 files changed, 38 insertions(+), 1 deletion(-)

diff --git a/arch/x86/include/asm/smp.h b/arch/x86/include/asm/smp.h
index ca073f40698f..ecf93a243b83 100644
--- a/arch/x86/include/asm/smp.h
+++ b/arch/x86/include/asm/smp.h
@@ -112,6 +112,7 @@ void native_play_dead(void);
 void play_dead_common(void);
 void wbinvd_on_cpu(int cpu);
 int wbinvd_on_all_cpus(void);
+int wbnoinvd_on_all_cpus(void);
 
 void smp_kick_mwait_play_dead(void);
 
@@ -160,6 +161,12 @@ static inline int wbinvd_on_all_cpus(void)
 	return 0;
 }
 
+static inline int wbnoinvd_on_all_cpus(void)
+{
+	wbnoinvd();
+	return 0;
+}
+
 static inline struct cpumask *cpu_llc_shared_mask(int cpu)
 {
 	return (struct cpumask *)cpumask_of(0);
diff --git a/arch/x86/include/asm/special_insns.h b/arch/x86/include/asm/special_insns.h
index 03e7c2d49559..fb3ddb2ddea9 100644
--- a/arch/x86/include/asm/special_insns.h
+++ b/arch/x86/include/asm/special_insns.h
@@ -117,7 +117,25 @@ static inline void wrpkru(u32 pkru)
 
 static __always_inline void wbinvd(void)
 {
-	asm volatile("wbinvd": : :"memory");
+	asm volatile("wbinvd" : : : "memory");
+}
+
+/*
+ * Cheaper version of wbinvd(). Call when caches
+ * need to be written back but not invalidated.
+ */
+static __always_inline void wbnoinvd(void)
+{
+	/*
+	 * WBNOINVD is encoded as 0xf3 0x0f 0x09. Making this
+	 * encoding explicit ensures compatibility with older versions of
+	 * binutils, which may not know about WBNOINVD.
+	 *
+	 * If WBNOINVD is unavailable, fall back to the compatible but
+	 * more destructive WBINVD (which still writes the caches back
+	 * but also invalidates them).
+	 */
+	alternative("wbinvd", ".byte 0xf3,0x0f,0x09", X86_FEATURE_WBNOINVD);
 }
 
 static inline unsigned long __read_cr4(void)
diff --git a/arch/x86/lib/cache-smp.c b/arch/x86/lib/cache-smp.c
index 7af743bd3b13..7ac5cca53031 100644
--- a/arch/x86/lib/cache-smp.c
+++ b/arch/x86/lib/cache-smp.c
@@ -20,3 +20,15 @@ int wbinvd_on_all_cpus(void)
 	return 0;
 }
 EXPORT_SYMBOL(wbinvd_on_all_cpus);
+
+static void __wbnoinvd(void *dummy)
+{
+	wbnoinvd();
+}
+
+int wbnoinvd_on_all_cpus(void)
+{
+	on_each_cpu(__wbnoinvd, NULL, 1);
+	return 0;
+}
+EXPORT_SYMBOL(wbnoinvd_on_all_cpus);
-- 
2.48.1.262.g85cc9f2d1e-goog

On 1/22/25 16:24, Kevin Loughlin wrote:
> +static __always_inline void wbnoinvd(void)
> +{
> +	/*
> +	 * WBNOINVD is encoded as 0xf3 0x0f 0x09. Making this
> +	 * encoding explicit ensures compatibility with older versions of
> +	 * binutils, which may not know about WBNOINVD.

This kinda pokes at one of my pet peeves. It's writing a comment where
code would do. I'd *much* rather write a function that explains to you
in code that "WBNOINVD is encoded as 0xf3 0x0f 0x09":

static __always_inline void native_wbnoinvd(void)
{
        asm volatile(".byte 0xf3,0x0f,0x09\n\t": : :"memory");
}

instead of writing out a comment. It's kinda silly to have to write out
the encoding explicitly in a comment and then have to rewrite it in the
code.

On Wed, Jan 22, 2025 at 4:36 PM Dave Hansen <dave.hansen@intel.com> wrote:
>
> On 1/22/25 16:24, Kevin Loughlin wrote:
> > +static __always_inline void wbnoinvd(void)
> > +{
> > +     /*
> > +      * WBNOINVD is encoded as 0xf3 0x0f 0x09. Making this
> > +      * encoding explicit ensures compatibility with older versions of
> > +      * binutils, which may not know about WBNOINVD.
>
> This kinda pokes at one of my pet peeves. It's writing a comment where
> code would do. I'd *much* rather write a function that explains to you
> in code that "WBNOINVD is encoded as 0xf3 0x0f 0x09":
>
> static __always_inline void native_wbnoinvd(void)
> {
>         asm volatile(".byte 0xf3,0x0f,0x09\n\t": : :"memory");
> }
>
> instead of writing out a comment. It's kinda silly to have to write out
> the encoding explicitly in a comment and then have to rewrite it in the
> code.

Yeah, I see your point. I will add this native_wbnoinvd() wrapper in v6; thanks!

AMD CPUs currently execute WBINVD in the host when unregistering SEV
guest memory or when deactivating SEV guests. Such cache maintenance is
performed to prevent data corruption, wherein the encrypted (C=1)
version of a dirty cache line might otherwise only be written back
after the memory is written in a different context (ex: C=0), yielding
corruption. However, WBINVD is performance-costly, especially because
it invalidates processor caches.

Strictly-speaking, unless the SEV ASID is being recycled (meaning the
SNP firmware requires the use of WBINVD prior to DF_FLUSH), the cache
invalidation triggered by WBINVD is unnecessary; only the writeback is
needed to prevent data corruption in remaining scenarios.

To improve performance in these scenarios, use WBNOINVD when available
instead of WBINVD. WBNOINVD still writes back all dirty lines
(preventing host data corruption by SEV guests) but does *not*
invalidate processor caches. Note that the implementation of wbnoinvd()
ensures fall back to WBINVD if WBNOINVD is unavailable.

In anticipation of forthcoming optimizations to limit the WBNOINVD only
to physical CPUs that have executed SEV guests, place the call to
wbnoinvd_on_all_cpus() in a wrapper function sev_writeback_caches().

Signed-off-by: Kevin Loughlin <kevinloughlin@google.com>
Reviewed-by: Mingwei Zhang <mizhang@google.com>
---
 arch/x86/kvm/svm/sev.c | 41 +++++++++++++++++++++--------------------
 1 file changed, 21 insertions(+), 20 deletions(-)

diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
index fe6cc763fd51..f10f1c53345e 100644
--- a/arch/x86/kvm/svm/sev.c
+++ b/arch/x86/kvm/svm/sev.c
@@ -116,6 +116,7 @@ static int sev_flush_asids(unsigned int min_asid, unsigned int max_asid)
 	 */
 	down_write(&sev_deactivate_lock);
 
+	/* SNP firmware requires use of WBINVD for ASID recycling. */
 	wbinvd_on_all_cpus();
 
 	if (sev_snp_enabled)
@@ -710,6 +711,16 @@ static void sev_clflush_pages(struct page *pages[], unsigned long npages)
 	}
 }
 
+static inline void sev_writeback_caches(void)
+{
+	/*
+	 * Ensure that all dirty guest tagged cache entries are written back
+	 * before releasing the pages back to the system for use. CLFLUSH will
+	 * not do this without SME_COHERENT, so issue a WBNOINVD.
+	 */
+	wbnoinvd_on_all_cpus();
+}
+
 static unsigned long get_num_contig_pages(unsigned long idx,
 				struct page **inpages, unsigned long npages)
 {
@@ -2773,12 +2784,7 @@ int sev_mem_enc_unregister_region(struct kvm *kvm,
 		goto failed;
 	}
 
-	/*
-	 * Ensure that all guest tagged cache entries are flushed before
-	 * releasing the pages back to the system for use. CLFLUSH will
-	 * not do this, so issue a WBINVD.
-	 */
-	wbinvd_on_all_cpus();
+	sev_writeback_caches();
 
 	__unregister_enc_region_locked(kvm, region);
 
@@ -2899,12 +2905,7 @@ void sev_vm_destroy(struct kvm *kvm)
 		return;
 	}
 
-	/*
-	 * Ensure that all guest tagged cache entries are flushed before
-	 * releasing the pages back to the system for use. CLFLUSH will
-	 * not do this, so issue a WBINVD.
-	 */
-	wbinvd_on_all_cpus();
+	sev_writeback_caches();
 
 	/*
 	 * if userspace was terminated before unregistering the memory regions
@@ -3126,16 +3127,16 @@ static void sev_flush_encrypted_page(struct kvm_vcpu *vcpu, void *va)
 
 	/*
 	 * VM Page Flush takes a host virtual address and a guest ASID.  Fall
-	 * back to WBINVD if this faults so as not to make any problems worse
+	 * back to WBNOINVD if this faults so as not to make any problems worse
 	 * by leaving stale encrypted data in the cache.
 	 */
 	if (WARN_ON_ONCE(wrmsrl_safe(MSR_AMD64_VM_PAGE_FLUSH, addr | asid)))
-		goto do_wbinvd;
+		goto do_sev_writeback_caches;
 
 	return;
 
-do_wbinvd:
-	wbinvd_on_all_cpus();
+do_sev_writeback_caches:
+	sev_writeback_caches();
 }
 
 void sev_guest_memory_reclaimed(struct kvm *kvm)
@@ -3144,12 +3145,12 @@ void sev_guest_memory_reclaimed(struct kvm *kvm)
 	 * With SNP+gmem, private/encrypted memory is unreachable via the
 	 * hva-based mmu notifiers, so these events are only actually
 	 * pertaining to shared pages where there is no need to perform
-	 * the WBINVD to flush associated caches.
+	 * the WBNOINVD to flush associated caches.
 	 */
 	if (!sev_guest(kvm) || sev_snp_guest(kvm))
 		return;
 
-	wbinvd_on_all_cpus();
+	sev_writeback_caches();
 }
 
 void sev_free_vcpu(struct kvm_vcpu *vcpu)
@@ -3858,7 +3859,7 @@ static int __sev_snp_update_protected_guest_state(struct kvm_vcpu *vcpu)
 		 * guest-mapped page rather than the initial one allocated
 		 * by KVM in svm->sev_es.vmsa. In theory, svm->sev_es.vmsa
 		 * could be free'd and cleaned up here, but that involves
-		 * cleanups like wbinvd_on_all_cpus() which would ideally
+		 * cleanups like sev_writeback_caches() which would ideally
 		 * be handled during teardown rather than guest boot.
 		 * Deferring that also allows the existing logic for SEV-ES
 		 * VMSAs to be re-used with minimal SNP-specific changes.
@@ -4910,7 +4911,7 @@ void sev_gmem_invalidate(kvm_pfn_t start, kvm_pfn_t end)
 
 		/*
 		 * SEV-ES avoids host/guest cache coherency issues through
-		 * WBINVD hooks issued via MMU notifiers during run-time, and
+		 * WBNOINVD hooks issued via MMU notifiers during run-time, and
 		 * KVM's VM destroy path at shutdown. Those MMU notifier events
 		 * don't cover gmem since there is no requirement to map pages
 		 * to a HVA in order to use them for a running guest. While the
-- 
2.48.1.262.g85cc9f2d1e-goog

On Thu, Jan 23, 2025, Kevin Loughlin wrote:
> +static inline void sev_writeback_caches(void)
> +{
> +	/*
> +	 * Ensure that all dirty guest tagged cache entries are written back
> +	 * before releasing the pages back to the system for use. CLFLUSH will
> +	 * not do this without SME_COHERENT, so issue a WBNOINVD.

This is somewhat misleading.  A naive reading of this would interpret the above
as saying that KVM _should_ do CLFLUSH on SME_COHERENT CPUs, which begs the
question of why KVM _doesn't_ do that.  I also think this is the right place to
call out that WBNOINVD support isn't guaranteed.

	 * Ensure that all dirty guest tagged cache entries are written back
	 * before releasing the pages back to the system for use.  CLFLUSH will
	 * not do this without SME_COHERENT, and flushing many cache lines
	 * individually is slower than blasting WBINVD for large VMs, so issue
	 * WBNOINVD (or WBINVD if the "no invalidate" variant is unsupported)
	 * on CPUs that have done VMRUN, i.e. may have dirtied data using the
	 * VM's ASID.

> +	 */
> +	wbnoinvd_on_all_cpus();
> +}
> +
>  static unsigned long get_num_contig_pages(unsigned long idx,
>  				struct page **inpages, unsigned long npages)
>  {
> @@ -2773,12 +2784,7 @@ int sev_mem_enc_unregister_region(struct kvm *kvm,
>  		goto failed;
>  	}
>  
> -	/*
> -	 * Ensure that all guest tagged cache entries are flushed before
> -	 * releasing the pages back to the system for use. CLFLUSH will
> -	 * not do this, so issue a WBINVD.
> -	 */
> -	wbinvd_on_all_cpus();
> +	sev_writeback_caches();
>  
>  	__unregister_enc_region_locked(kvm, region);
>  
> @@ -2899,12 +2905,7 @@ void sev_vm_destroy(struct kvm *kvm)
>  		return;
>  	}
>  
> -	/*
> -	 * Ensure that all guest tagged cache entries are flushed before
> -	 * releasing the pages back to the system for use. CLFLUSH will
> -	 * not do this, so issue a WBINVD.
> -	 */
> -	wbinvd_on_all_cpus();
> +	sev_writeback_caches();
>  
>  	/*
>  	 * if userspace was terminated before unregistering the memory regions
> @@ -3126,16 +3127,16 @@ static void sev_flush_encrypted_page(struct kvm_vcpu *vcpu, void *va)
>  
>  	/*
>  	 * VM Page Flush takes a host virtual address and a guest ASID.  Fall
> -	 * back to WBINVD if this faults so as not to make any problems worse
> +	 * back to WBNOINVD if this faults so as not to make any problems worse

I don't like replacing WBINVD with WBNOINVD everywhere.  It incorrectly implies
that WBNOINVD is guaranteed.  Easiest thing is just to avoid mnemonics and describe
the behavior in conversational language (well, as conversational as cache flushin
gets :-D).
 
> @@ -3858,7 +3859,7 @@ static int __sev_snp_update_protected_guest_state(struct kvm_vcpu *vcpu)
>  		 * guest-mapped page rather than the initial one allocated
>  		 * by KVM in svm->sev_es.vmsa. In theory, svm->sev_es.vmsa
>  		 * could be free'd and cleaned up here, but that involves
> -		 * cleanups like wbinvd_on_all_cpus() which would ideally
> +		 * cleanups like sev_writeback_caches() which would ideally

Similarly, avoid function names in comments, because they too become stale.

>  		 * be handled during teardown rather than guest boot.
>  		 * Deferring that also allows the existing logic for SEV-ES
>  		 * VMSAs to be re-used with minimal SNP-specific changes.