Skip to contentSkip to conversation

xen: CONFIG_PV_SHIM_EXCLUSIVE and CONFIG_HVM are mutually exclusive

PATCH In review Juergen Gross 3 participants 5 messages Dec 8, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xen: CONFIG_PV_SHIM_EXCLUSIVE and CONFIG_HVM are mutually exclusive

With CONFIG_PV_SHIM_EXCLUSIVE some sources required for CONFIG_HVM are not built, so let CONFIG_HVM depend on !CONFIG_PV_SHIM_EXCLUSIVE.

Let CONFIG_HVM default to !CONFIG_PV_SHIM instead.

---
 xen/arch/x86/Kconfig | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
xen/arch/x86/Kconfig +2 -1
index 24868aa6ad..0107cfa12f 100644
@@ -90,7 +90,8 @@ config PV_LINEAR_PT
          If unsure, say Y.
 
 config HVM
-	def_bool !PV_SHIM_EXCLUSIVE
+	depends on !PV_SHIM_EXCLUSIVE
+	def_bool !PV_SHIM
 	prompt "HVM support"
 	---help---
 	  Interfaces to support HVM domains.  HVM domains require hardware
Signature · 1 line
2.26.2
#1 lore raw
Quoted 9 lines · Juergen Gross

With CONFIG_PV_SHIM_EXCLUSIVE some sources required for CONFIG_HVM are not built, so let CONFIG_HVM depend on !CONFIG_PV_SHIM_EXCLUSIVE.

Let CONFIG_HVM default to !CONFIG_PV_SHIM instead.

---
 xen/arch/x86/Kconfig | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

See "x86/shim: don't permit HVM and PV_SHIM_EXCLUSIVE at the same time" posted on Oct 19. I'd be fine switching to the !PV_SHIM default you have here. But Andrew looks to be objecting to a change like this, sadly without pointing out a good alternative so far.

Jan

Quoted 15 lines
xen/arch/x86/Kconfig +2 -1
index 24868aa6ad..0107cfa12f 100644
@@ -90,7 +90,8 @@ config PV_LINEAR_PT
          If unsure, say Y.
 
 config HVM
-	def_bool !PV_SHIM_EXCLUSIVE
+	depends on !PV_SHIM_EXCLUSIVE
+	def_bool !PV_SHIM
 	prompt "HVM support"
 	---help---
 	  Interfaces to support HVM domains.  HVM domains require hardware
#2 lore raw
In reply to Juergen Gross
Quoted 6 lines · Juergen Gross

With CONFIG_PV_SHIM_EXCLUSIVE some sources required for CONFIG_HVM are not built, so let CONFIG_HVM depend on !CONFIG_PV_SHIM_EXCLUSIVE.

Let CONFIG_HVM default to !CONFIG_PV_SHIM instead.

So while this will fix the randconfig failure, the statement isn't true.  There are HVM codepaths which aren't even dead in shim-exclusive mode.

The problem here is the way CONFIG_PV_SHIM_EXCLUSIVE abuses the Kconfig system.  What is currently happening is that this option is trying to enforce the pv shim defconfig in the dependency system.

We already have a defconfig, which is used in appropriate locations.  We should not have two different things fighting over control.

This is the fault of c/s 8b5b49ceb3d which went in despite my objections.  The change is not related to PV_SHIM_EXCLUSIVE - it is to do with not supporting a control domain, which a) better describes what it is actually doing, and b) has wider utility than PV Shim.

~Andrew

#3 lore raw
Quoted 11 lines · Andrew Cooper
Quoted 6 lines · Juergen Gross

With CONFIG_PV_SHIM_EXCLUSIVE some sources required for CONFIG_HVM are not built, so let CONFIG_HVM depend on !CONFIG_PV_SHIM_EXCLUSIVE.

Let CONFIG_HVM default to !CONFIG_PV_SHIM instead.

So while this will fix the randconfig failure, the statement isn't true.  There are HVM codepaths which aren't even dead in shim-exclusive mode.

I only said that CONFIG_PV_SHIM_EXCLUSIVE disables building some sources which are required for CONFIG_HVM, and this is certainly true.

Quoted 11 lines

The problem here is the way CONFIG_PV_SHIM_EXCLUSIVE abuses the Kconfig system.  What is currently happening is that this option is trying to enforce the pv shim defconfig in the dependency system.

We already have a defconfig, which is used in appropriate locations.  We should not have two different things fighting over control.

This is the fault of c/s 8b5b49ceb3d which went in despite my objections.  The change is not related to PV_SHIM_EXCLUSIVE - it is to do with not supporting a control domain, which a) better describes what it is actually doing, and b) has wider utility than PV Shim.

Yes, maybe.

Random build failures are not nice, so in case there is no agreement how to proceed I'd be in favor for fixing the fallout and then discuss a proper solution.

Juergen

#4 lore raw
In reply to Andrew Cooper
Quoted 16 lines · Andrew Cooper
Quoted 6 lines · Juergen Gross

With CONFIG_PV_SHIM_EXCLUSIVE some sources required for CONFIG_HVM are not built, so let CONFIG_HVM depend on !CONFIG_PV_SHIM_EXCLUSIVE.

Let CONFIG_HVM default to !CONFIG_PV_SHIM instead.

So while this will fix the randconfig failure, the statement isn't true.  There are HVM codepaths which aren't even dead in shim-exclusive mode.

The problem here is the way CONFIG_PV_SHIM_EXCLUSIVE abuses the Kconfig system.  What is currently happening is that this option is trying to enforce the pv shim defconfig in the dependency system.

Quoted 7 lines

We already have a defconfig, which is used in appropriate locations.  We should not have two different things fighting over control.

This is the fault of c/s 8b5b49ceb3d which went in despite my objections.  The change is not related to PV_SHIM_EXCLUSIVE - it is to do with not supporting a control domain, which a) better describes what it is actually doing, and b) has wider utility than PV Shim.

Would you mind pointing me at where you had voiced objections to that change? I've just searched both my inbox and the list archives, without finding any. I only recall your objections to the patch I sent later which is similar to Jürgen's. And I'm quite certain I'd have stayed away from committing anything while aware of unresolved objections, even if - more often than not - this means waiting almost indefinitely, which I don't appreciate as a way to deal with disagreement.

From what you further state, I derive that you'd like to see e.g. !PV_SHIM_EXCLUSIVE be a dependency of a new CONTROL_DOMAIN Kconfig setting? I'm not sure though I see how this would help the situation (I'm not even sure what scope this control would
have: just domctl, or also sysctl, or additionally platform-op). Nor do I see what's wrong with forcing HVM off in shim-exclusive builds - there can't be HVM domains in such a configuration (but I'm pretty sure I said so somewhere else already, without ever hearing back, albeit it apparently wasn't on either of the patches' threads according to my outbox).

Jan