Skip to contentSkip to conversation

x86/shim: Fix defconfig selection and trim the build further

PATCH In review Andrew Cooper 3 participants 4 messages Jun 3, 2020 – Jun 4, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/shim: Fix defconfig selection and trim the build further

Several options (TBOOT, XENOPROF, Scheduler) depend on EXPERT to be able to deselect/configure.

Enabling EXPERT now causes the request of the Credit1 scheduler to be honoured (rather than giving us Credit2), but take this opportunity to switch to Null, as the previously problematic issues are now believed to be fixed.

Enabling EXPERT also allows XEN_SHSTK to be selected, and we don't want this being built for shim. We also don't want TRACEBUFFER or GDBSX either.

Take this oppotunity to swap the disable of HVM_FEP for a general disable of HVM (likely to have wider impliciations in the future), and disable ARGO (will necesserily need plumbing work to function in shim).

---
CC: Jan Beulich <[email protected]>
CC: Wei Liu <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Juergen Gross <[email protected]>
CC: Paul Durrant <[email protected]>
CC: Dario Faggioli <[email protected]>

---
 xen/arch/x86/configs/pvshim_defconfig | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
xen/arch/x86/configs/pvshim_defconfig +8 -2
index 830660e022..3af48d6c06 100644
@@ -5,19 +5,25 @@ CONFIG_PVH_GUEST=y
 CONFIG_PV_SHIM=y
 CONFIG_PV_SHIM_EXCLUSIVE=y
 CONFIG_NR_CPUS=32
+CONFIG_EXPERT=y
+CONFIG_SCHED_NULL=y
 # Disable features not used by the PV shim
+# CONFIG_HVM is not set
+# CONFIG_XEN_SHSTK is not set
 # CONFIG_HYPFS is not set
 # CONFIG_SHADOW_PAGING is not set
 # CONFIG_BIGMEM is not set
-# CONFIG_HVM_FEP is not set
 # CONFIG_TBOOT is not set
 # CONFIG_KEXEC is not set
 # CONFIG_XENOPROF is not set
 # CONFIG_XSM is not set
+# CONFIG_ARGO is not set
+# CONFIG_SCHED_CREDIT is not set
 # CONFIG_SCHED_CREDIT2 is not set
 # CONFIG_SCHED_RTDS is not set
 # CONFIG_SCHED_ARINC653 is not set
-# CONFIG_SCHED_NULL is not set
 # CONFIG_LIVEPATCH is not set
 # CONFIG_SUPPRESS_DUPLICATE_SYMBOL_WARNINGS is not set
+# CONFIG_TRACEBUFFER is not set
 # CONFIG_DEBUG is not set
+# CONFIG_GDBSX is not set
Signature · 1 line
2.11.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 7 lines · Andrew Cooper

Several options (TBOOT, XENOPROF, Scheduler) depend on EXPERT to be able to deselect/configure.

Enabling EXPERT now causes the request of the Credit1 scheduler to be honoured (rather than giving us Credit2), but take this opportunity to switch to Null, as the previously problematic issues are now believed to be fixed.

Quoted 6 lines

Enabling EXPERT also allows XEN_SHSTK to be selected, and we don't want this being built for shim. We also don't want TRACEBUFFER or GDBSX either.

Take this oppotunity to swap the disable of HVM_FEP for a general disable of HVM (likely to have wider impliciations in the future), and disable ARGO (will necesserily need plumbing work to function in shim).

Odd. I was quite sure this is the case already; in particular my own build test of a shim config has this already.

Quoted 1 line

I have a question though (without implying the patch here needs adjusting, but rather with a look towards after 4.14):

Quoted 20 lines
xen/arch/x86/configs/pvshim_defconfig +5 -1
@@ -5,19 +5,25 @@ CONFIG_PVH_GUEST=y
 CONFIG_PV_SHIM=y
 CONFIG_PV_SHIM_EXCLUSIVE=y
 CONFIG_NR_CPUS=32
+CONFIG_EXPERT=y
+CONFIG_SCHED_NULL=y
 # Disable features not used by the PV shim
+# CONFIG_HVM is not set
+# CONFIG_XEN_SHSTK is not set
 # CONFIG_HYPFS is not set
 # CONFIG_SHADOW_PAGING is not set
 # CONFIG_BIGMEM is not set
-# CONFIG_HVM_FEP is not set
 # CONFIG_TBOOT is not set
 # CONFIG_KEXEC is not set
 # CONFIG_XENOPROF is not set
 # CONFIG_XSM is not set
+# CONFIG_ARGO is not set

xen/arch/x86/configs/pvshim_defconfig CONFIG_PVH_GUEST=y · line 29

Quoted 10 lines
xen/arch/x86/configs/pvshim_defconfig +3 -1
+# CONFIG_SCHED_CREDIT is not set
 # CONFIG_SCHED_CREDIT2 is not set
 # CONFIG_SCHED_RTDS is not set
 # CONFIG_SCHED_ARINC653 is not set
-# CONFIG_SCHED_NULL is not set
 # CONFIG_LIVEPATCH is not set
 # CONFIG_SUPPRESS_DUPLICATE_SYMBOL_WARNINGS is not set
+# CONFIG_TRACEBUFFER is not set
 # CONFIG_DEBUG is not set
+# CONFIG_GDBSX is not set

I assume both the "enable" and "disable" sections here are ordered like they would be in a resulting full .config. But this being two separate sections, doing so doesn't help e.g. diff-ing. How about we sort both sections alphabetically (short of other good sorting criteria, yet better than entirely unsorted)?

Jan

#2 lore raw
Quoted 24 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 04 June 2020 07:43
To: Andrew Cooper <[email protected]>
Cc: Xen-devel <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; Juergen Gross <[email protected]>; Paul Durrant <[email protected]>; Dario Faggioli <[email protected]>
Subject: Re: [PATCH for-4.14] x86/shim: Fix defconfig selection and trim the build further

Quoted 7 lines · Andrew Cooper

Several options (TBOOT, XENOPROF, Scheduler) depend on EXPERT to be able to deselect/configure.

Enabling EXPERT now causes the request of the Credit1 scheduler to be honoured (rather than giving us Credit2), but take this opportunity to switch to Null, as the previously problematic issues are now believed to be fixed.

Quoted 6 lines

Enabling EXPERT also allows XEN_SHSTK to be selected, and we don't want this being built for shim. We also don't want TRACEBUFFER or GDBSX either.

Take this oppotunity to swap the disable of HVM_FEP for a general disable of HVM (likely to have wider impliciations in the future), and disable ARGO (will necesserily need plumbing work to function in shim).

Quoted 6 lines

Odd. I was quite sure this is the case already; in particular my own build test of a shim config has this already.

Quoted 1 line
Quoted 42 lines

I have a question though (without implying the patch here needs adjusting, but rather with a look towards after 4.14):

Quoted 20 lines
xen/arch/x86/configs/pvshim_defconfig +5 -1
@@ -5,19 +5,25 @@ CONFIG_PVH_GUEST=y
 CONFIG_PV_SHIM=y
 CONFIG_PV_SHIM_EXCLUSIVE=y
 CONFIG_NR_CPUS=32
+CONFIG_EXPERT=y
+CONFIG_SCHED_NULL=y
 # Disable features not used by the PV shim
+# CONFIG_HVM is not set
+# CONFIG_XEN_SHSTK is not set
 # CONFIG_HYPFS is not set
 # CONFIG_SHADOW_PAGING is not set
 # CONFIG_BIGMEM is not set
-# CONFIG_HVM_FEP is not set
 # CONFIG_TBOOT is not set
 # CONFIG_KEXEC is not set
 # CONFIG_XENOPROF is not set
 # CONFIG_XSM is not set
+# CONFIG_ARGO is not set

xen/arch/x86/configs/pvshim_defconfig CONFIG_PVH_GUEST=y · line 29

Quoted 10 lines
xen/arch/x86/configs/pvshim_defconfig +3 -1
+# CONFIG_SCHED_CREDIT is not set
 # CONFIG_SCHED_CREDIT2 is not set
 # CONFIG_SCHED_RTDS is not set
 # CONFIG_SCHED_ARINC653 is not set
-# CONFIG_SCHED_NULL is not set
 # CONFIG_LIVEPATCH is not set
 # CONFIG_SUPPRESS_DUPLICATE_SYMBOL_WARNINGS is not set
+# CONFIG_TRACEBUFFER is not set
 # CONFIG_DEBUG is not set
+# CONFIG_GDBSX is not set

I assume both the "enable" and "disable" sections here are ordered like they would be in a resulting full .config. But this being two separate sections, doing so doesn't help e.g. diff-ing. How about we sort both sections alphabetically (short of other good sorting criteria, yet better than entirely unsorted)?

Jan

#3 lore raw
In reply to Jan Beulich
Quoted 8 lines · Jan Beulich
Quoted 7 lines · Andrew Cooper

Several options (TBOOT, XENOPROF, Scheduler) depend on EXPERT to be able to deselect/configure.

Enabling EXPERT now causes the request of the Credit1 scheduler to be honoured (rather than giving us Credit2), but take this opportunity to switch to Null, as the previously problematic issues are now believed to be fixed.

Quoted 8 lines
Quoted 6 lines

Enabling EXPERT also allows XEN_SHSTK to be selected, and we don't want this being built for shim. We also don't want TRACEBUFFER or GDBSX either.

Take this oppotunity to swap the disable of HVM_FEP for a general disable of HVM (likely to have wider impliciations in the future), and disable ARGO (will necesserily need plumbing work to function in shim).

Odd. I was quite sure this is the case already; in particular my own build test of a shim config has this already.

It is currently off because of its default in Xen, but even if that were to change, its not usable in shim without further development work.

Quoted 3 lines
Quoted 2 lines

Thanks.

Quoted 28 lines

I have a question though (without implying the patch here needs adjusting, but rather with a look towards after 4.14):

Quoted 24 lines
xen/arch/x86/configs/pvshim_defconfig +6 -1
@@ -5,19 +5,25 @@ CONFIG_PVH_GUEST=y
 CONFIG_PV_SHIM=y
 CONFIG_PV_SHIM_EXCLUSIVE=y
 CONFIG_NR_CPUS=32
+CONFIG_EXPERT=y
+CONFIG_SCHED_NULL=y
 # Disable features not used by the PV shim
+# CONFIG_HVM is not set
+# CONFIG_XEN_SHSTK is not set
 # CONFIG_HYPFS is not set
 # CONFIG_SHADOW_PAGING is not set
 # CONFIG_BIGMEM is not set
-# CONFIG_HVM_FEP is not set
 # CONFIG_TBOOT is not set
 # CONFIG_KEXEC is not set
 # CONFIG_XENOPROF is not set
 # CONFIG_XSM is not set
+# CONFIG_ARGO is not set
+# CONFIG_SCHED_CREDIT is not set
 # CONFIG_SCHED_CREDIT2 is not set
 # CONFIG_SCHED_RTDS is not set
 # CONFIG_SCHED_ARINC653 is not set
Quoted 8 lines

xen/arch/x86/configs/pvshim_defconfig CONFIG_PVH_GUEST=y · line 29

Quoted 6 lines
xen/arch/x86/configs/pvshim_defconfig +2 -1
-# CONFIG_SCHED_NULL is not set
 # CONFIG_LIVEPATCH is not set
 # CONFIG_SUPPRESS_DUPLICATE_SYMBOL_WARNINGS is not set
+# CONFIG_TRACEBUFFER is not set
 # CONFIG_DEBUG is not set
+# CONFIG_GDBSX is not set

I assume both the "enable" and "disable" sections here are ordered like they would be in a resulting full .config.

They are, yes.

Quoted 2 lines

But this being two
separate sections, doing so doesn't help e.g. diff-ing.

Having them in order helps massively with scanning through the two files together.

I'm not entirely certain why the two sections are separate to begin with.  Merging them would probably make things even easier, but I think the file does want to stay in .config order.

~Andrew