Add defines for the architectural memory types that can be shoved into
various MSRs and registers, e.g. MTRRs, PAT, VMX capabilities MSRs, EPTPs,
etc. While most MSRs/registers support only a subset of all memory types,
the values themselves are architectural and identical across all users.
Leave the goofy MTRR_TYPE_* definitions as-is since they are in a uapi
header, but add compile-time assertions to connect the dots (and sanity
check that the msr-index.h values didn't get fat-fingered).
Keep the VMX_EPTP_MT_* defines so that it's slightly more obvious that the
EPTP holds a single memory type in 3 of its 64 bits; those bits just
happen to be 2:0, i.e. don't need to be shifted.
Opportunistically use X86_MEMTYPE_WB instead of an open coded '6' in
setup_vmcs_config().
No functional change intended.
Signed-off-by: Sean Christopherson <[email protected]>
---
arch/x86/include/asm/msr-index.h | 15 ++++++++++++++-
arch/x86/include/asm/vmx.h | 5 +++--
arch/x86/kernel/cpu/mtrr/mtrr.c | 6 ++++++
arch/x86/kvm/vmx/nested.c | 2 +-
arch/x86/kvm/vmx/vmx.c | 2 +-
arch/x86/mm/pat/memtype.c | 33 ++++++++++++--------------------
6 files changed, 37 insertions(+), 26 deletions(-)
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index f1bd7b91b3c6..29f0ea78e41c 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -36,6 +36,20 @@
#define EFER_FFXSR (1<<_EFER_FFXSR)
#define EFER_AUTOIBRS (1<<_EFER_AUTOIBRS)
+/*
+ * Architectural memory types that are common to MTRRs, PAT, VMX MSRs, etc.
+ * Most MSRs support/allow only a subset of memory types, but the values
+ * themselves are common across all relevant MSRs.
+ */
+#define X86_MEMTYPE_UC 0ull /* Uncacheable, a.k.a. Strong Uncacheable */
+#define X86_MEMTYPE_WC 1ull /* Write Combining */
+/* RESERVED 2 */
+/* RESERVED 3 */
+#define X86_MEMTYPE_WT 4ull /* Write Through */
+#define X86_MEMTYPE_WP 5ull /* Write Protected */
+#define X86_MEMTYPE_WB 6ull /* Write Back */
+#define X86_MEMTYPE_UC_MINUS 7ull /* Weak Uncacheabled (PAT only) */
+
/* Intel MSRs. Some also available on other CPUs */
#define MSR_TEST_CTRL 0x00000033
@@ -1108,7 +1122,6 @@
#define VMX_BASIC_64 0x0001000000000000LLU
#define VMX_BASIC_MEM_TYPE_SHIFT 50
#define VMX_BASIC_MEM_TYPE_MASK 0x003c000000000000LLU
-#define VMX_BASIC_MEM_TYPE_WB 6LLU
#define VMX_BASIC_INOUT 0x0040000000000000LLU
/* Resctrl MSRs: */
diff --git a/arch/x86/include/asm/vmx.h b/arch/x86/include/asm/vmx.h
index 0e73616b82f3..4fdc76263066 100644
--- a/arch/x86/include/asm/vmx.h
+++ b/arch/x86/include/asm/vmx.h
@@ -504,9 +504,10 @@ enum vmcs_field {
#define VMX_EPTP_PWL_4 0x18ull
#define VMX_EPTP_PWL_5 0x20ull
#define VMX_EPTP_AD_ENABLE_BIT (1ull << 6)
+/* The EPTP memtype is encoded in bits 2:0, i.e. doesn't need to be shifted. */
#define VMX_EPTP_MT_MASK 0x7ull
-#define VMX_EPTP_MT_WB 0x6ull
-#define VMX_EPTP_MT_UC 0x0ull
+#define VMX_EPTP_MT_WB X86_MEMTYPE_WB
+#define VMX_EPTP_MT_UC X86_MEMTYPE_UC
#define VMX_EPT_READABLE_MASK 0x1ull
#define VMX_EPT_WRITABLE_MASK 0x2ull
#define VMX_EPT_EXECUTABLE_MASK 0x4ull
diff --git a/arch/x86/kernel/cpu/mtrr/mtrr.c b/arch/x86/kernel/cpu/mtrr/mtrr.c
index 767bf1c71aad..125e36010b82 100644
--- a/arch/x86/kernel/cpu/mtrr/mtrr.c
+++ b/arch/x86/kernel/cpu/mtrr/mtrr.c
@@ -55,6 +55,12 @@
#include "mtrr.h"
+static_assert(X86_MEMTYPE_UC == MTRR_TYPE_UNCACHABLE);
+static_assert(X86_MEMTYPE_WC == MTRR_TYPE_WRCOMB);
+static_assert(X86_MEMTYPE_WT == MTRR_TYPE_WRTHROUGH);
+static_assert(X86_MEMTYPE_WP == MTRR_TYPE_WRPROT);
+static_assert(X86_MEMTYPE_WB == MTRR_TYPE_WRBACK);
+
/* arch_phys_wc_add returns an MTRR register index plus this offset. */
#define MTRR_TO_PHYS_WC_OFFSET 1000
diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
index d05ddf751491..82a35aba7d2b 100644
--- a/arch/x86/kvm/vmx/nested.c
+++ b/arch/x86/kvm/vmx/nested.c
@@ -7006,7 +7006,7 @@ static void nested_vmx_setup_basic(struct nested_vmx_msrs *msrs)
VMCS12_REVISION |
VMX_BASIC_TRUE_CTLS |
((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
- (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
+ (X86_MEMTYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
if (cpu_has_vmx_basic_inout())
msrs->basic |= VMX_BASIC_INOUT;
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 7a74388f9ecf..71cc6e3b3221 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -2692,7 +2692,7 @@ static int setup_vmcs_config(struct vmcs_config *vmcs_conf,
#endif
/* Require Write-Back (WB) memory type for VMCS accesses. */
- if (((vmx_msr_high >> 18) & 15) != 6)
+ if (((vmx_msr_high >> 18) & 15) != X86_MEMTYPE_WB)
return -EIO;
rdmsrl(MSR_IA32_VMX_MISC, misc_msr);
diff --git a/arch/x86/mm/pat/memtype.c b/arch/x86/mm/pat/memtype.c
index 0904d7e8e126..3e0ba044925f 100644
--- a/arch/x86/mm/pat/memtype.c
+++ b/arch/x86/mm/pat/memtype.c
@@ -175,15 +175,6 @@ static inline void set_page_memtype(struct page *pg,
}
#endif
-enum {
- PAT_UC = 0, /* uncached */
- PAT_WC = 1, /* Write combining */
- PAT_WT = 4, /* Write Through */
- PAT_WP = 5, /* Write Protected */
- PAT_WB = 6, /* Write Back (default) */
- PAT_UC_MINUS = 7, /* UC, but can be overridden by MTRR */
-};
-
#define CM(c) (_PAGE_CACHE_MODE_ ## c)
static enum page_cache_mode __init pat_get_cache_mode(unsigned int pat_val,
@@ -193,13 +184,13 @@ static enum page_cache_mode __init pat_get_cache_mode(unsigned int pat_val,
char *cache_mode;
switch (pat_val) {
- case PAT_UC: cache = CM(UC); cache_mode = "UC "; break;
- case PAT_WC: cache = CM(WC); cache_mode = "WC "; break;
- case PAT_WT: cache = CM(WT); cache_mode = "WT "; break;
- case PAT_WP: cache = CM(WP); cache_mode = "WP "; break;
- case PAT_WB: cache = CM(WB); cache_mode = "WB "; break;
- case PAT_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
- default: cache = CM(WB); cache_mode = "WB "; break;
+ case X86_MEMTYPE_UC: cache = CM(UC); cache_mode = "UC "; break;
+ case X86_MEMTYPE_WC: cache = CM(WC); cache_mode = "WC "; break;
+ case X86_MEMTYPE_WT: cache = CM(WT); cache_mode = "WT "; break;
+ case X86_MEMTYPE_WP: cache = CM(WP); cache_mode = "WP "; break;
+ case X86_MEMTYPE_WB: cache = CM(WB); cache_mode = "WB "; break;
+ case X86_MEMTYPE_UC_MINUS: cache = CM(UC_MINUS); cache_mode = "UC- "; break;
+ default: cache = CM(WB); cache_mode = "WB "; break;
}
memcpy(msg, cache_mode, 4);
@@ -254,11 +245,11 @@ void pat_cpu_init(void)
void __init pat_bp_init(void)
{
struct cpuinfo_x86 *c = &boot_cpu_data;
-#define PAT(p0, p1, p2, p3, p4, p5, p6, p7) \
- (((u64)PAT_ ## p0) | ((u64)PAT_ ## p1 << 8) | \
- ((u64)PAT_ ## p2 << 16) | ((u64)PAT_ ## p3 << 24) | \
- ((u64)PAT_ ## p4 << 32) | ((u64)PAT_ ## p5 << 40) | \
- ((u64)PAT_ ## p6 << 48) | ((u64)PAT_ ## p7 << 56))
+#define PAT(p0, p1, p2, p3, p4, p5, p6, p7) \
+ ((X86_MEMTYPE_ ## p0) | (X86_MEMTYPE_ ## p1 << 8) | \
+ (X86_MEMTYPE_ ## p2 << 16) | (X86_MEMTYPE_ ## p3 << 24) | \
+ (X86_MEMTYPE_ ## p4 << 32) | (X86_MEMTYPE_ ## p5 << 40) | \
+ (X86_MEMTYPE_ ## p6 << 48) | (X86_MEMTYPE_ ## p7 << 56))
if (!IS_ENABLED(CONFIG_X86_PAT))
--
2.44.0.278.ge034bb2e1d-goog
On Fri, 2024-03-08 at 17:27 -0800, Sean Christopherson wrote:
> Add defines for the architectural memory types that can be shoved into
> various MSRs and registers, e.g. MTRRs, PAT, VMX capabilities MSRs, EPTPs,
> etc. While most MSRs/registers support only a subset of all memory types,
> the values themselves are architectural and identical across all users.
>
> Leave the goofy MTRR_TYPE_* definitions as-is since they are in a uapi
> header, but add compile-time assertions to connect the dots (and sanity
> check that the msr-index.h values didn't get fat-fingered).
>
> Keep the VMX_EPTP_MT_* defines so that it's slightly more obvious that the
> EPTP holds a single memory type in 3 of its 64 bits; those bits just
> happen to be 2:0, i.e. don't need to be shifted.
>
> Opportunistically use X86_MEMTYPE_WB instead of an open coded '6' in
> setup_vmcs_config().
>
> No functional change intended.
>
> Signed-off-by: Sean Christopherson <[email protected]>
>
[...]
>
> #include "mtrr.h"
>
> +static_assert(X86_MEMTYPE_UC == MTRR_TYPE_UNCACHABLE);
> +static_assert(X86_MEMTYPE_WC == MTRR_TYPE_WRCOMB);
> +static_assert(X86_MEMTYPE_WT == MTRR_TYPE_WRTHROUGH);
> +static_assert(X86_MEMTYPE_WP == MTRR_TYPE_WRPROT);
> +static_assert(X86_MEMTYPE_WB == MTRR_TYPE_WRBACK);
> +
>
Hi Sean,
IIUC, the purpose of this patch is for the kernel to use X86_MEMTYPE_xx, which
are architectural values, where applicable?
Yeah we need to keep MTRR_TYPE_xx in the uapi header, but in the kernel, should
we change all places that use MTRR_TYPE_xx to X86_MEMTYPE_xx? The
static_assert()s above have guaranteed the two are the same, so there's nothing
wrong for the kernel to use X86_MEMTYPE_xx instead.
Both PAT_xx and VMX_BASIC_MEM_TYPE_xx to X86_MEMTYPE_xx, it seems a little bit
odd if we don't switch for MTRR_TYPE_xx.
However by simple search MEM_TYPE_xx are intensively used in many files, so...
On Wed, Mar 27, 2024, Kai Huang wrote:
> On Fri, 2024-03-08 at 17:27 -0800, Sean Christopherson wrote:
> > Add defines for the architectural memory types that can be shoved into
> > various MSRs and registers, e.g. MTRRs, PAT, VMX capabilities MSRs, EPTPs,
> > etc. While most MSRs/registers support only a subset of all memory types,
> > the values themselves are architectural and identical across all users.
> >
> > Leave the goofy MTRR_TYPE_* definitions as-is since they are in a uapi
> > header, but add compile-time assertions to connect the dots (and sanity
> > check that the msr-index.h values didn't get fat-fingered).
> >
> > Keep the VMX_EPTP_MT_* defines so that it's slightly more obvious that the
> > EPTP holds a single memory type in 3 of its 64 bits; those bits just
> > happen to be 2:0, i.e. don't need to be shifted.
> >
> > Opportunistically use X86_MEMTYPE_WB instead of an open coded '6' in
> > setup_vmcs_config().
> >
> > No functional change intended.
> >
> > Signed-off-by: Sean Christopherson <[email protected]>
> >
>
> [...]
>
> >
> > #include "mtrr.h"
> >
> > +static_assert(X86_MEMTYPE_UC == MTRR_TYPE_UNCACHABLE);
> > +static_assert(X86_MEMTYPE_WC == MTRR_TYPE_WRCOMB);
> > +static_assert(X86_MEMTYPE_WT == MTRR_TYPE_WRTHROUGH);
> > +static_assert(X86_MEMTYPE_WP == MTRR_TYPE_WRPROT);
> > +static_assert(X86_MEMTYPE_WB == MTRR_TYPE_WRBACK);
> > +
> >
>
> Hi Sean,
>
> IIUC, the purpose of this patch is for the kernel to use X86_MEMTYPE_xx, which
> are architectural values, where applicable?
Maybe? Probably?
> Yeah we need to keep MTRR_TYPE_xx in the uapi header, but in the kernel, should
> we change all places that use MTRR_TYPE_xx to X86_MEMTYPE_xx? The
> static_assert()s above have guaranteed the two are the same, so there's nothing
> wrong for the kernel to use X86_MEMTYPE_xx instead.
>
> Both PAT_xx and VMX_BASIC_MEM_TYPE_xx to X86_MEMTYPE_xx, it seems a little bit
> odd if we don't switch for MTRR_TYPE_xx.
>
> However by simple search MEM_TYPE_xx are intensively used in many files, so...
Yeah, I definitely don't want to do it in this series due to the amount of churn
that would be required.
$ git grep MTRR_TYPE_ | wc -l
100
I'm not even entirely convinced that it would be a net positive. Much of the KVM
usage that's being cleaned up is flat out wrong, e.g. using "MTRR" enums in places
that having nothing to do with MTRRs. But the majority of the remaining usage is
in MTRR code, i.e. isn't wrong, and is arguably better off using the MTRR specific
#defines.
On 4/04/2024 7:57 am, Sean Christopherson wrote:
> On Wed, Mar 27, 2024, Kai Huang wrote:
>> On Fri, 2024-03-08 at 17:27 -0800, Sean Christopherson wrote:
>>> Add defines for the architectural memory types that can be shoved into
>>> various MSRs and registers, e.g. MTRRs, PAT, VMX capabilities MSRs, EPTPs,
>>> etc. While most MSRs/registers support only a subset of all memory types,
>>> the values themselves are architectural and identical across all users.
>>>
>>> Leave the goofy MTRR_TYPE_* definitions as-is since they are in a uapi
>>> header, but add compile-time assertions to connect the dots (and sanity
>>> check that the msr-index.h values didn't get fat-fingered).
>>>
>>> Keep the VMX_EPTP_MT_* defines so that it's slightly more obvious that the
>>> EPTP holds a single memory type in 3 of its 64 bits; those bits just
>>> happen to be 2:0, i.e. don't need to be shifted.
>>>
>>> Opportunistically use X86_MEMTYPE_WB instead of an open coded '6' in
>>> setup_vmcs_config().
>>>
>>> No functional change intended.
>>>
>>> Signed-off-by: Sean Christopherson <[email protected]>
>>>
>>
>> [...]
>>
>>>
>>> #include "mtrr.h"
>>>
>>> +static_assert(X86_MEMTYPE_UC == MTRR_TYPE_UNCACHABLE);
>>> +static_assert(X86_MEMTYPE_WC == MTRR_TYPE_WRCOMB);
>>> +static_assert(X86_MEMTYPE_WT == MTRR_TYPE_WRTHROUGH);
>>> +static_assert(X86_MEMTYPE_WP == MTRR_TYPE_WRPROT);
>>> +static_assert(X86_MEMTYPE_WB == MTRR_TYPE_WRBACK);
>>> +
>>>
>>
>> Hi Sean,
>>
>> IIUC, the purpose of this patch is for the kernel to use X86_MEMTYPE_xx, which
>> are architectural values, where applicable?
>
> Maybe? Probably?
>
>> Yeah we need to keep MTRR_TYPE_xx in the uapi header, but in the kernel, should
>> we change all places that use MTRR_TYPE_xx to X86_MEMTYPE_xx? The
>> static_assert()s above have guaranteed the two are the same, so there's nothing
>> wrong for the kernel to use X86_MEMTYPE_xx instead.
>>
>> Both PAT_xx and VMX_BASIC_MEM_TYPE_xx to X86_MEMTYPE_xx, it seems a little bit
>> odd if we don't switch for MTRR_TYPE_xx.
>>
>> However by simple search MEM_TYPE_xx are intensively used in many files, so...
>
> Yeah, I definitely don't want to do it in this series due to the amount of churn
> that would be required.
>
> $ git grep MTRR_TYPE_ | wc -l
> 100
>
> I'm not even entirely convinced that it would be a net positive. Much of the KVM
> usage that's being cleaned up is flat out wrong, e.g. using "MTRR" enums in places
> that having nothing to do with MTRRs. But the majority of the remaining usage is
> in MTRR code, i.e. isn't wrong, and is arguably better off using the MTRR specific
> #defines.
Yeah understood.
But the patch title says we also "add common defines for ... MTRRs", so
to me looks we should get rid of MTRR_TPYE_xx and use the common ones
instead. And it also looks a little bit inconsistent if we remove the
PAT_xx but keep the MTRR_TYPE_xx.
Perhaps we can keep PAT_xx but add macros?
#define PAT_UC X86_MEMTYPE_UC
...
But looks not nice either because the only purpose is to keep the PAT_xx..
So up to you :-)
On Thu, Apr 04, 2024, Kai Huang wrote:
> On 4/04/2024 7:57 am, Sean Christopherson wrote:
> > On Wed, Mar 27, 2024, Kai Huang wrote:
> > > IIUC, the purpose of this patch is for the kernel to use X86_MEMTYPE_xx, which
> > > are architectural values, where applicable?
> >
> > Maybe? Probably?
> >
> > > Yeah we need to keep MTRR_TYPE_xx in the uapi header, but in the kernel, should
> > > we change all places that use MTRR_TYPE_xx to X86_MEMTYPE_xx? The
> > > static_assert()s above have guaranteed the two are the same, so there's nothing
> > > wrong for the kernel to use X86_MEMTYPE_xx instead.
> > >
> > > Both PAT_xx and VMX_BASIC_MEM_TYPE_xx to X86_MEMTYPE_xx, it seems a little bit
> > > odd if we don't switch for MTRR_TYPE_xx.
> > >
> > > However by simple search MEM_TYPE_xx are intensively used in many files, so...
> >
> > Yeah, I definitely don't want to do it in this series due to the amount of churn
> > that would be required.
> >
> > $ git grep MTRR_TYPE_ | wc -l
> > 100
> >
> > I'm not even entirely convinced that it would be a net positive. Much of the KVM
> > usage that's being cleaned up is flat out wrong, e.g. using "MTRR" enums in places
> > that having nothing to do with MTRRs. But the majority of the remaining usage is
> > in MTRR code, i.e. isn't wrong, and is arguably better off using the MTRR specific
> > #defines.
>
> Yeah understood.
>
> But the patch title says we also "add common defines for ... MTRRs", so to
> me looks we should get rid of MTRR_TPYE_xx and use the common ones instead.
> And it also looks a little bit inconsistent if we remove the PAT_xx but keep
> the MTRR_TYPE_xx.
>
> Perhaps we can keep PAT_xx but add macros?
>
> #define PAT_UC X86_MEMTYPE_UC
> ...
>
> But looks not nice either because the only purpose is to keep the PAT_xx..
Ya, keeping PAT_* is the only option I strongly object to. I have no problem
replacing MTRR_TPYE_* usage, I would simply prefer not to do it in this series.
And I obviously have no problem leaving the MTRR stuff as-is.