Skip to contentSkip to conversation

x86: Drop opt_pku entirely

PATCH In review Andrew Cooper 3 participants 6 messages Jun 20, 2023 – Aug 7, 2023

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86: Drop opt_pku entirely

This option is particularly dubious as Xen does not use Protection Keys, owing to the sharing of pagetables with PV guests. All this option does is hide PKU by default from HVM guests, and is therefore redundant with the more generic cpuid=no-pku.

The variable ought to be in __initdata given it's single user, but deleting it entirely looks to be a better course of action.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>
CC: Henry Wang <[email protected]>

---
 CHANGELOG.md                      |  3 +++
 docs/misc/xen-command-line.pandoc | 10 ----------
 xen/arch/x86/cpu/common.c         |  7 -------
 3 files changed, 3 insertions(+), 17 deletions(-)
CHANGELOG.md +3 -0
index 7d7e0590f8c6..43f15dc34cbf 100644
@@ -25,6 +25,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/)
  - Add support for AVX512-FP16 on x86.
  - On Arm, Xen supports guests running SVE/SVE2 instructions. (Tech Preview)
 
+### Removed
+ - On x86, the "pku" command line option has been removed.  It has never
+   behaved precisely as described, and redundant with "cpuid=no-pku".
 
 ## [4.17.0](https://xenbits.xen.org/gitweb/?p=xen.git;a=shortlog;h=RELEASE-4.17.0) - 2022-12-12
 
docs/misc/xen-command-line.pandoc +0 -10
index 4060ebdc5d76..9d66688bd1ff 100644
@@ -1950,16 +1950,6 @@ for all of them (`true`), only for those subject to XPTI (`xpti`) or for
 those not subject to XPTI (`no-xpti`). The feature is used only in case
 INVPCID is supported and not disabled via `invpcid=false`.
 
-### pku (x86)
-> `= <boolean>`
-
-> Default: `true`
-
-Flag to enable Memory Protection Keys.
-
-The protection-key feature provides an additional mechanism by which IA-32e
-paging controls access to usermode addresses.
-
 ### ple_gap
 > `= <integer>`
 
xen/arch/x86/cpu/common.c +0 -7
index cfcdaace125b..14021ffc66d8 100644
@@ -29,10 +29,6 @@ bool __read_mostly opt_dom0_cpuid_faulting = true;
 bool_t opt_arat = 1;
 boolean_param("arat", opt_arat);
 
-/* pku: Flag to enable Memory Protection Keys (default on). */
-static bool_t opt_pku = 1;
-boolean_param("pku", opt_pku);
-
 unsigned int opt_cpuid_mask_ecx = ~0u;
 integer_param("cpuid_mask_ecx", opt_cpuid_mask_ecx);
 unsigned int opt_cpuid_mask_edx = ~0u;
@@ -522,9 +518,6 @@ void identify_cpu(struct cpuinfo_x86 *c)
 		this_cpu->c_init(c);
 
 
-   	if (c == &boot_cpu_data && !opt_pku)
-		setup_clear_cpu_cap(X86_FEATURE_PKU);
-
 	/*
 	 * The vendor-specific functions might have changed features.  Now
 	 * we do "generic changes."
Signature · 1 line
2.30.2
#1 lore raw

CHANGELOG.md The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) · line 30

Quoted 9 lines · Andrew Cooper
CHANGELOG.md +3 -0
@@ -25,6 +25,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/)
  - Add support for AVX512-FP16 on x86.
  - On Arm, Xen supports guests running SVE/SVE2 instructions. (Tech Preview)
 
+### Removed
+ - On x86, the "pku" command line option has been removed.  It has never
+   behaved precisely as described, and redundant with "cpuid=no-pku".

Nit: Missing "was"?

Quoted 8 lines
docs/misc/xen-command-line.pandoc +0 -2
@@ -1950,16 +1950,6 @@ for all of them (`true`), only for those subject to XPTI (`xpti`) or for
 those not subject to XPTI (`no-xpti`). The feature is used only in case
 INVPCID is supported and not disabled via `invpcid=false`.
 
-### pku (x86)
-> `= <boolean>`

docs/misc/xen-command-line.pandoc for all of them (`true`), only for those subject to XPTI (`xpti`) or for · line 1954

Quoted 10 lines
docs/misc/xen-command-line.pandoc +0 -8
-
-> Default: `true`
-
-Flag to enable Memory Protection Keys.
-
-The protection-key feature provides an additional mechanism by which IA-32e
-paging controls access to usermode addresses.
-
 ### ple_gap
 > `= <integer>`

Elsewhere you said that we kind of imply that only the explicitly named sub-options of cpuid= are supported. If that's the case (which could do with saying more explicitly), you will want to add "pku" there in order to not regress what is (deemed) supported.

It also looks as if the speculation control related enumeration there has gone stale / is now incomplete.

Jan

#2 lore raw
Quoted 11 lines · Jan Beulich

CHANGELOG.md The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) · line 30

Quoted 9 lines · Andrew Cooper
CHANGELOG.md +3 -0
@@ -25,6 +25,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/)
  - Add support for AVX512-FP16 on x86.
  - On Arm, Xen supports guests running SVE/SVE2 instructions. (Tech Preview)
 
+### Removed
+ - On x86, the "pku" command line option has been removed.  It has never
+   behaved precisely as described, and redundant with "cpuid=no-pku".

Nit: Missing "was"?

Fixed

Quoted 7 lines
Quoted 7 lines
docs/misc/xen-command-line.pandoc +0 -0
@@ -1950,16 +1950,6 @@ for all of them (`true`), only for those subject to XPTI (`xpti`) or for
 those not subject to XPTI (`no-xpti`). The feature is used only in case
 INVPCID is supported and not disabled via `invpcid=false`.
 
Quoted 16 lines

docs/misc/xen-command-line.pandoc for all of them (`true`), only for those subject to XPTI (`xpti`) or for · line 1954

Quoted 12 lines
docs/misc/xen-command-line.pandoc +0 -10
-### pku (x86)
-> `= <boolean>`
-
-> Default: `true`
-
-Flag to enable Memory Protection Keys.
-
-The protection-key feature provides an additional mechanism by which IA-32e
-paging controls access to usermode addresses.
-
 ### ple_gap
 > `= <integer>`

Elsewhere you said that we kind of imply that only the explicitly named sub-options of cpuid= are supported. If that's the case (which could do with saying more explicitly), you will want to add "pku" there in order to not regress what is (deemed) supported.

I disagree.  I can say it was equivalent to X without X being an explicitly supported option.

PKU shouldn't be adjusted by either of these options; it should be controlled in the VM config file (if at all).  I'm unwilling to make any suggestion that this is supported.

~Andrew

Acked-by
#3 lore raw
Quoted 39 lines · Andrew Cooper
Quoted 11 lines · Jan Beulich

CHANGELOG.md The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) · line 30

Quoted 9 lines · Andrew Cooper
CHANGELOG.md +3 -0
@@ -25,6 +25,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/)
  - Add support for AVX512-FP16 on x86.
  - On Arm, Xen supports guests running SVE/SVE2 instructions. (Tech Preview)
 
+### Removed
+ - On x86, the "pku" command line option has been removed.  It has never
+   behaved precisely as described, and redundant with "cpuid=no-pku".

Nit: Missing "was"?

Fixed

Quoted 7 lines
Quoted 7 lines
docs/misc/xen-command-line.pandoc +0 -0
@@ -1950,16 +1950,6 @@ for all of them (`true`), only for those subject to XPTI (`xpti`) or for
 those not subject to XPTI (`no-xpti`). The feature is used only in case
 INVPCID is supported and not disabled via `invpcid=false`.
 
Quoted 16 lines

docs/misc/xen-command-line.pandoc for all of them (`true`), only for those subject to XPTI (`xpti`) or for · line 1954

Quoted 12 lines
docs/misc/xen-command-line.pandoc +0 -10
-### pku (x86)
-> `= <boolean>`
-
-> Default: `true`
-
-Flag to enable Memory Protection Keys.
-
-The protection-key feature provides an additional mechanism by which IA-32e
-paging controls access to usermode addresses.
-
 ### ple_gap
 > `= <integer>`

Elsewhere you said that we kind of imply that only the explicitly named sub-options of cpuid= are supported. If that's the case (which could do with saying more explicitly), you will want to add "pku" there in order to not regress what is (deemed) supported.

Quoted 6 lines

I disagree.  I can say it was equivalent to X without X being an explicitly supported option.

PKU shouldn't be adjusted by either of these options; it should be controlled in the VM config file (if at all).  I'm unwilling to make any suggestion that this is supported.

Hmm, so you're suggesting "pku=" use was unsupported, too? If so, perhaps worth adding the word "unsupported" to the CHANGELOG.md entry. At which point I'm fine with no adjustment to cpuid= doc. And then also

Jan

#4 lore raw
Quoted 25 lines · Jan Beulich
Quoted 24 lines · Andrew Cooper
Quoted 23 lines · Jan Beulich
Quoted 8 lines · Andrew Cooper
docs/misc/xen-command-line.pandoc +0 -2
@@ -1950,16 +1950,6 @@ for all of them (`true`), only for those subject to XPTI (`xpti`) or for
 those not subject to XPTI (`no-xpti`). The feature is used only in case
 INVPCID is supported and not disabled via `invpcid=false`.
 
-### pku (x86)
-> `= <boolean>`

docs/misc/xen-command-line.pandoc for all of them (`true`), only for those subject to XPTI (`xpti`) or for · line 1954

Quoted 10 lines
docs/misc/xen-command-line.pandoc +0 -8
-
-> Default: `true`
-
-Flag to enable Memory Protection Keys.
-
-The protection-key feature provides an additional mechanism by which IA-32e
-paging controls access to usermode addresses.
-
 ### ple_gap
 > `= <integer>`

Elsewhere you said that we kind of imply that only the explicitly named sub-options of cpuid= are supported. If that's the case (which could do with saying more explicitly), you will want to add "pku" there in order to not regress what is (deemed) supported.

Quoted 8 lines
Quoted 6 lines

I disagree.  I can say it was equivalent to X without X being an explicitly supported option.

PKU shouldn't be adjusted by either of these options; it should be controlled in the VM config file (if at all).  I'm unwilling to make any suggestion that this is supported.

Hmm, so you're suggesting "pku=" use was unsupported, too? If so, perhaps worth adding the word "unsupported" to the CHANGELOG.md entry.

Well - it's just hiding a feature so nothing (interesting) is going to break.  But it also isn't an option that anyone ought to have been using.

Quoted 3 lines

At which
point I'm fine with no adjustment to cpuid= doc. And then also

Thanks.  I'll tweak.

~Andrew