Skip to contentSkip to conversation

xen: Fix PV-only build

PATCH In review Jason Andryuk 3 participants 10 messages Aug 6

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover xen: Fix PV-only build

Allow disabling XEN_PVHVM for a PV-only. A stub in the event channel code needs to be fixed first.

Jason Andryuk (2):
xen/events: Fix xen_set_upcall_vector stub
xen/Kconfig: select XEN_PVHVM

 arch/x86/xen/Kconfig             | 8 +++++---
 drivers/xen/events/events_base.c | 2 +-
 2 files changed, 6 insertions(+), 4 deletions(-)
Signature · 1 line
2.55.0
#1 lore raw

PATCH 1/2 xen/events: Fix xen_set_upcall_vector stub

Building the xen_set_upcall_vector stub fails with
error: control reaches end of non-void function.

Return -EINVAL, which matches the hypercall's return for a non-HVM domain.

This is needed to allow disabling CONFIG_XEN_PVHVM.

Fixes b1c3497e604d ("x86/xen: Add support for HVMOP_set_evtchn_upcall_vector")
---
 drivers/xen/events/events_base.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/xen/events/events_base.c +1 -1
index 6ea945508a89..3a5ae96e73cc 100644
@@ -2245,7 +2245,7 @@ static __init void xen_alloc_callback_vector(void)
 #else
 void xen_setup_callback_vector(void) {}
 static inline void xen_init_setup_upcall_vector(void) {}
-int xen_set_upcall_vector(unsigned int cpu) {}
+int xen_set_upcall_vector(unsigned int cpu) { return -EINVAL; }
 static inline void xen_alloc_callback_vector(void) {}
 #endif /* CONFIG_XEN_PVHVM */
 #endif /* CONFIG_X86 */
Signature · 1 line
2.55.0
#2 lore raw

PATCH 2/2 xen/Kconfig: select XEN_PVHVM

In reply to Jason Andryuk

XEN_PVHVM cannot be disabled as it is a hidden variable with def_bool y. Switch XEN_PVHVM to a plain bool, and make XEN_PVH and XEN_PVHVM_GUEST select it. It will be pulled in as needed, and drop the code from PV-only builds.

Fixes 34aff14580d1 ("xen: Remove Xen PVH/PVHVM dependency on PCI")
---
 arch/x86/xen/Kconfig | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
arch/x86/xen/Kconfig +5 -3
index 99b06f5c47cd..7bdaa28ff232 100644
@@ -51,7 +51,7 @@ config XEN_PV_DOM0
 	depends on XEN_PV && XEN_DOM0
 
 config XEN_PVHVM
-	def_bool y
+	bool
 	depends on XEN && X86_LOCAL_APIC
 
 config XEN_PVHVM_SMP
@@ -61,13 +61,15 @@ config XEN_PVHVM_SMP
 config XEN_PVHVM_GUEST
 	bool "Xen PVHVM guest support"
 	default y
-	depends on XEN_PVHVM && PCI
+	depends on XEN && PCI
+	select XEN_PVHVM
 	help
 	  Support running as a Xen PVHVM guest.
 
 config XEN_PVH
 	bool "Xen PVH guest support"
-	depends on XEN && XEN_PVHVM && ACPI
+	depends on XEN && ACPI
+	select XEN_PVHVM
 	select PVH
 	help
 	  Support for running as a Xen PVH guest.
Signature · 1 line
2.55.0
#3 lore raw

Re: [PATCH 2/2] xen/Kconfig: select XEN_PVHVM

arch/x86/xen/Kconfig config XEN_PV_DOM0 · line 55

Quoted 9 lines · Jason Andryuk
arch/x86/xen/Kconfig +1 -1
@@ -51,7 +51,7 @@ config XEN_PV_DOM0
 	depends on XEN_PV && XEN_DOM0
 
 config XEN_PVHVM
-	def_bool y
+	bool
 	depends on XEN && X86_LOCAL_APIC

Imo "depends on" on prompt-less, default-off options are at best unhelpful in the common case. (Aiui it can be helpful when "imply" is used instead of "select".) Hence I think this dependency wants dropping right in this patch, justified further by ...

Quoted 8 lines
arch/x86/xen/Kconfig +2 -1
@@ -61,13 +61,15 @@ config XEN_PVHVM_SMP
 config XEN_PVHVM_GUEST
 	bool "Xen PVHVM guest support"
 	default y
-	depends on XEN_PVHVM && PCI
+	depends on XEN && PCI
+	select XEN_PVHVM
 	help

arch/x86/xen/Kconfig config XEN_PVHVM_SMP · line 75

Quoted 10 lines
arch/x86/xen/Kconfig +2 -1
 	  Support running as a Xen PVHVM guest.
 
 config XEN_PVH
 	bool "Xen PVH guest support"
-	depends on XEN && XEN_PVHVM && ACPI
+	depends on XEN && ACPI
+	select XEN_PVHVM
 	select PVH
 	help
 	  Support for running as a Xen PVH guest.

... the XEN dependency being in both of the options selecting XEN_PVHVM and, as per Jürgen's patch, X86_LOCAL_APIC being redundant anyway.

Jan

maintainer
#4 lore raw
In reply to Jason Andryuk
Quoted 11 lines · Jason Andryuk

Allow disabling XEN_PVHVM for a PV-only. A stub in the event channel code needs to be fixed first.

Jason Andryuk (2):
xen/events: Fix xen_set_upcall_vector stub
xen/Kconfig: select XEN_PVHVM

  arch/x86/xen/Kconfig             | 8 +++++---
  drivers/xen/events/events_base.c | 2 +-
  2 files changed, 6 insertions(+), 4 deletions(-)

I did a comparison of a kernel built with your patches disabling XEN_PVHVM and my patches with XEN_PVHVM_GUEST disabled.

The kernel built with my patches is 6 bytes smaller than the one with your patches.

So I don't see any reason to take your patches, which conflict with mine, especially as my patches have a negative diffstat on source level, too.

Juergen

#5 lore raw
Quoted 13 lines · Juergen Gross
Quoted 11 lines · Jason Andryuk

Allow disabling XEN_PVHVM for a PV-only. A stub in the event channel code needs to be fixed first.

Jason Andryuk (2):
xen/events: Fix xen_set_upcall_vector stub
xen/Kconfig: select XEN_PVHVM

  arch/x86/xen/Kconfig             | 8 +++++---
  drivers/xen/events/events_base.c | 2 +-
  2 files changed, 6 insertions(+), 4 deletions(-)
Quoted 5 lines

I did a comparison of a kernel built with your patches disabling XEN_PVHVM and my patches with XEN_PVHVM_GUEST disabled.

The kernel built with my patches is 6 bytes smaller than the one with your patches.

Isn't this a sign of something else needing tweaking, somewhere?

Quoted 2 lines

So I don't see any reason to take your patches, which conflict with mine, especially as my patches have a negative diffstat on source level, too.

Hmm, Jason's patches look to move things into a more adequate direction, though. In which case I think a negative diffstat becomes an irrelevant argument?

Jan

maintainer
#6 lore raw
Quoted 14 lines · Jan Beulich
Quoted 13 lines · Juergen Gross
Quoted 11 lines · Jason Andryuk

Allow disabling XEN_PVHVM for a PV-only. A stub in the event channel code needs to be fixed first.

Jason Andryuk (2):
xen/events: Fix xen_set_upcall_vector stub
xen/Kconfig: select XEN_PVHVM

   arch/x86/xen/Kconfig             | 8 +++++---
   drivers/xen/events/events_base.c | 2 +-
   2 files changed, 6 insertions(+), 4 deletions(-)
Quoted 7 lines
Quoted 5 lines

I did a comparison of a kernel built with your patches disabling XEN_PVHVM and my patches with XEN_PVHVM_GUEST disabled.

The kernel built with my patches is 6 bytes smaller than the one with your patches.

Isn't this a sign of something else needing tweaking, somewhere?

This is a sign that there are probably only very few really HVM specific paths (in the sense of: explicitly not marked as irrelevant for PV) in the kernel. Yes, I'm sure you can find some more, but I'm really not sure this is relevant for more than a handful of users.

Quoted 6 lines
Quoted 2 lines

So I don't see any reason to take your patches, which conflict with mine, especially as my patches have a negative diffstat on source level, too.

Hmm, Jason's patches look to move things into a more adequate direction, though. In which case I think a negative diffstat becomes an irrelevant argument?

Depends on what you are looking for.

My take from this is that a PV-only kernel with Jason's patches is not really adding any value, while my simplification is at least making things simpler in terms of code volume and number of Xen related config options.

Of course it would be possible to have a smaller PV-only kernel, but as I said already, there has been no public demand for that in the last years and the downsides IMHO far outweigh the potential gain.

IMO the "adequate direction" regarding Xen specific kernel code is towards PVH and not towards more PV specific tweaking. And I'm very sure the kernel community outside of the Xen community is agreeing with me here.

Juergen

#7 lore raw
Quoted 21 lines · Jürgen Groß
Quoted 20 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 11 lines · Jason Andryuk

Allow disabling XEN_PVHVM for a PV-only.  A stub in the event channel code needs to be fixed first.

Jason Andryuk (2):
    xen/events: Fix xen_set_upcall_vector stub
    xen/Kconfig: select XEN_PVHVM

   arch/x86/xen/Kconfig             | 8 +++++---
   drivers/xen/events/events_base.c | 2 +-

   2 files changed, 6 insertions(+), 4 deletions(-)

I did a comparison of a kernel built with your patches disabling XEN_PVHVM and my patches with XEN_PVHVM_GUEST disabled.

The kernel built with my patches is 6 bytes smaller than the one with your patches.

Quoted 6 lines
Quoted 1 line

Isn't this a sign of something else needing tweaking, somewhere?

This is a sign that there are probably only very few really HVM specific paths (in the sense of: explicitly not marked as irrelevant for PV) in the kernel. Yes, I'm sure you can find some more, but I'm really not sure this is relevant for more than a handful of users.

I see more reduction:
15029248 - arch/x86/boot/bzImage
15021056 - arch/x86/boot/bzImage.after

~8k

52542480 - vmlinux
52523184 - vmlinux.after

~18k

$ ../linux/scripts/bloat-o-meter vmlinux vmlinux.after
add/remove: 1/136 grow/shrink: 14/48 up/down: 15902/-22430 (-6528)
Quoted 13 lines
Quoted 6 lines
Quoted 2 lines

So I don't see any reason to take your patches, which conflict with mine, especially as my patches have a negative diffstat on source level, too.

Hmm, Jason's patches look to move things into a more adequate direction, though. In which case I think a negative diffstat becomes an irrelevant argument?

Depends on what you are looking for.

My take from this is that a PV-only kernel with Jason's patches is not really adding any value, while my simplification is at least making things simpler in terms of code volume and number of Xen related config options.

Quoted 7 lines

Of course it would be possible to have a smaller PV-only kernel, but as I said already, there has been no public demand for that in the last years and the downsides IMHO far outweigh the potential gain.

IMO the "adequate direction" regarding Xen specific kernel code is towards PVH and not towards more PV specific tweaking. And I'm very sure the kernel community outside of the Xen community is agreeing with me here.

I don't need PV-only kernels, so I am fine with not pursuing this patch set.

Mainly I wanted to post this alternative since restoring PV-only is possible (and I inadvertently broke it). Converting CONFIG_XEN_PVHVM to CONFIG_XEN looked wrong when it didn't apply to PV.

Pursuing your patches for the reasons you give also makes sense.

Regards,
Jason

maintainer
#8 lore raw
Quoted 35 lines · Jason Andryuk
Quoted 21 lines · Jürgen Groß
Quoted 20 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 11 lines · Jason Andryuk

Allow disabling XEN_PVHVM for a PV-only.  A stub in the event channel code needs to be fixed first.

Jason Andryuk (2):
    xen/events: Fix xen_set_upcall_vector stub
    xen/Kconfig: select XEN_PVHVM

   arch/x86/xen/Kconfig             | 8 +++++---
   drivers/xen/events/events_base.c | 2 +-

   2 files changed, 6 insertions(+), 4 deletions(-)

I did a comparison of a kernel built with your patches disabling XEN_PVHVM and my patches with XEN_PVHVM_GUEST disabled.

The kernel built with my patches is 6 bytes smaller than the one with your patches.

Quoted 6 lines
Quoted 1 line

Isn't this a sign of something else needing tweaking, somewhere?

This is a sign that there are probably only very few really HVM specific paths (in the sense of: explicitly not marked as irrelevant for PV) in the kernel. Yes, I'm sure you can find some more, but I'm really not sure this is relevant for more than a handful of users.

I see more reduction:
15029248 - arch/x86/boot/bzImage
15021056 - arch/x86/boot/bzImage.after

~8k

Quoted 7 lines

52542480 - vmlinux
52523184 - vmlinux.after

~18k

$ ../linux/scripts/bloat-o-meter vmlinux vmlinux.after
add/remove: 1/136 grow/shrink: 14/48 up/down: 15902/-22430 (-6528)

Did you disable CONFIG_XEN_PVHVM_GUEST in the before kernel, too?

Quoted 18 lines
Quoted 18 lines
Quoted 7 lines
Quoted 3 lines

So I don't see any reason to take your patches, which conflict with mine, especially as my patches have a negative diffstat on source level, too.

Hmm, Jason's patches look to move things into a more adequate direction, though. In which case I think a negative diffstat becomes an irrelevant argument?

Depends on what you are looking for.

My take from this is that a PV-only kernel with Jason's patches is not really adding any value, while my simplification is at least making things simpler in terms of code volume and number of Xen related config options.

Of course it would be possible to have a smaller PV-only kernel, but as I said already, there has been no public demand for that in the last years and the downsides IMHO far outweigh the potential gain.

Quoted 8 lines
Quoted 3 lines

IMO the "adequate direction" regarding Xen specific kernel code is towards PVH and not towards more PV specific tweaking. And I'm very sure the kernel community outside of the Xen community is agreeing with me here.

I don't need PV-only kernels, so I am fine with not pursuing this patch set.

Mainly I wanted to post this alternative since restoring PV-only is possible (and I inadvertently broke it).  Converting CONFIG_XEN_PVHVM to CONFIG_XEN looked wrong when it didn't apply to PV.

Pursuing your patches for the reasons you give also makes sense.

Thanks,

Juergen

#9 lore raw
Quoted 39 lines · Juergen Gross
Quoted 38 lines · Jason Andryuk
Quoted 21 lines · Jürgen Groß
Quoted 20 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 11 lines · Jason Andryuk

Allow disabling XEN_PVHVM for a PV-only.  A stub in the event channel code needs to be fixed first.

Jason Andryuk (2):
    xen/events: Fix xen_set_upcall_vector stub
    xen/Kconfig: select XEN_PVHVM

   arch/x86/xen/Kconfig             | 8 +++++---
   drivers/xen/events/events_base.c | 2 +-

   2 files changed, 6 insertions(+), 4 deletions(-)

I did a comparison of a kernel built with your patches disabling XEN_PVHVM and my patches with XEN_PVHVM_GUEST disabled.

The kernel built with my patches is 6 bytes smaller than the one with your patches.

Quoted 6 lines
Quoted 1 line

Isn't this a sign of something else needing tweaking, somewhere?

This is a sign that there are probably only very few really HVM specific paths (in the sense of: explicitly not marked as irrelevant for PV) in the kernel. Yes, I'm sure you can find some more, but I'm really not sure this is relevant for more than a handful of users.

I see more reduction:
15029248 - arch/x86/boot/bzImage
15021056 - arch/x86/boot/bzImage.after

~8k

52542480 - vmlinux
52523184 - vmlinux.after

Quoted 6 lines
Quoted 4 lines

~18k

$ ../linux/scripts/bloat-o-meter vmlinux vmlinux.after
add/remove: 1/136 grow/shrink: 14/48 up/down: 15902/-22430 (-6528)

Did you disable CONFIG_XEN_PVHVM_GUEST in the before kernel, too?

Yes

$ grep PVHVM config-before config-after config-pvh-before:CONFIG_XEN_PVHVM=y
config-pvh-before:CONFIG_XEN_PVHVM_SMP=y
config-pvh-before:# CONFIG_XEN_PVHVM_GUEST is not set
config-pvh-after:# CONFIG_XEN_PVHVM_GUEST is not set

These three options change:

$ diff -u config-before config-after
-CONFIG_XEN_PVHVM=y
-CONFIG_XEN_PVHVM_SMP=y
-CONFIG_XEN_AUTO_XLATE=y

Regards,
Jason