x86: more flexible alignment handling for xen binary
cover x86: more flexible alignment handling for xen binary
Wei Liu (2):
x86: decouple xen alignment setting from EFI/non-EFI build x86/pvshim: use 2M alignment
xen/arch/x86/Kconfig | 26 ++++++++++++++++++++++++++ xen/arch/x86/configs/pvshim_defconfig | 1 + xen/arch/x86/xen.lds.S | 16 ++++++++++++++-- 3 files changed, 41 insertions(+), 2 deletions(-)
Signature · 7 lines
2.20.1 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH 1/2 x86: decouple xen alignment setting from EFI/non-EFI build
Introduce a new Kconfig option to pick the alignment for xen binary. To retain original behaviour, the default pick for EFI build is 2M and non-EFI build 4K.
--- xen/arch/x86/Kconfig | 26 ++++++++++++++++++++++++++ xen/arch/x86/xen.lds.S | 16 ++++++++++++++-- 2 files changed, 40 insertions(+), 2 deletions(-)
xen/arch/x86/Kconfig +26 -0
@@ -138,6 +138,32 @@ config TBOOT If unsure, say Y. +choice + prompt "Alignment of Xen binary" + depends on X86 + default XEN_ALIGN_DEFAULT + ---help--- + Specify alignment for Xen binary. + + If unsure, choose "default". + +config XEN_ALIGN_DEFAULT + bool "Default alignment" + ---help--- + Pick alignment according to build variants. + + For EFI build the default alignment is 2M. For non-EFI build + the default alignment is 4K due to syslinux failing to handle + 2M alignment. + +config XEN_ALIGN_4K + bool "4K alignment" + +config XEN_ALIGN_2M + bool "2M alignment" + +endchoice + config XEN_GUEST def_bool n prompt "Xen Guest"
xen/arch/x86/xen.lds.S +14 -2
@@ -12,7 +12,6 @@ #define FORMAT "pei-x86-64" #undef __XEN_VIRT_START #define __XEN_VIRT_START __image_base__ -#define SECTION_ALIGN MB(2) #define DECL_SECTION(x) x : ENTRY(efi_start)
@@ -20,13 +19,26 @@ ENTRY(efi_start) #else /* !EFI */ #define FORMAT "elf64-x86-64" -#define SECTION_ALIGN PAGE_SIZE #define DECL_SECTION(x) x : AT(ADDR(x) - __XEN_VIRT_START) ENTRY(start_pa) #endif /* EFI */ +#if defined CONFIG_XEN_ALIGN_2M +#define SECTION_ALIGN MB(2) +#elif defined CONFIG_XEN_ALIGN_4K +#define SECTION_ALIGN PAGE_SIZE +#elif defined CONFIG_XEN_ALIGN_DEFAULT + #ifdef EFI + #define SECTION_ALIGN MB(2) + #else + #define SECTION_ALIGN PAGE_SIZE + #endif +#else +#error "Section alignment undefined" +#endif + OUTPUT_FORMAT(FORMAT, FORMAT, FORMAT) OUTPUT_ARCH(i386:x86-64)
Signature · 7 lines
2.20.1 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 1/2] x86: decouple xen alignment setting from EFI/non-EFI build
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 18.03.19 at 15:57, <[email protected]> wrote:
Introduce a new
Kconfigoption to pick the alignment for xen binary. To retain original behaviour, the default pick for EFI build is 2M and non-EFI build 4K.
Is this a worthwhile step to take, considering that we mean to switch to 2M main section boundaries anyway? I realize it's still not necessarily clear how to get there without re-introducing the boot regression with syslinux that was observed back then, but perhaps time would better be spent there than into introducing configurability? (To be honest I don't recall whether it was determined that there was no way out of this dilemma, but it seems to me there would have been a plan.)
Quoted 13 lines
xen/arch/x86/Kconfig +7
-0@@ -138,6 +138,32 @@ config TBOOT If unsure, say Y. +choice + prompt "Alignment of Xen binary" + depends on X86 + default XEN_ALIGN_DEFAULT + ---help--- + Specify alignment for Xen binary. +
xen/arch/x86/Kconfig config TBOOT · line 157
Quoted 10 lines
xen/arch/x86/Kconfig +10
-0+ If unsure, choose "default". + +config XEN_ALIGN_DEFAULT + bool "Default alignment" + ---help--- + Pick alignment according to build variants. + + For EFI build the default alignment is 2M. For non-EFI build + the default alignment is 4K due to syslinux failing to handle + 2M alignment.
Wasn't it the resulting image size rather than the alignment itself which did cause the problem? Has the issue been fixed in newer syslinux?
xen/arch/x86/Kconfig config TBOOT · line 163
Quoted 5 lines
xen/arch/x86/Kconfig +5
-0+config XEN_ALIGN_4K + bool "4K alignment" + +config XEN_ALIGN_2M + bool "2M alignment"
In patch 2 you only need the latter - is there really much value in allowing explicitly requesting 4k?
Quoted 14 lines
xen/arch/x86/Kconfig +3
-1@@ -20,13 +19,26 @@ ENTRY(efi_start) #else /* !EFI */ #define FORMAT "elf64-x86-64" -#define SECTION_ALIGN PAGE_SIZE #define DECL_SECTION(x) x : AT(ADDR(x) - __XEN_VIRT_START) ENTRY(start_pa) #endif /* EFI */ +#if defined CONFIG_XEN_ALIGN_2M +#define SECTION_ALIGN MB(2) +#elif defined CONFIG_XEN_ALIGN_4K
xen/arch/x86/Kconfig ENTRY(efi_start) · line 40
Quoted 10 lines
xen/arch/x86/Kconfig +10
-0+#define SECTION_ALIGN PAGE_SIZE +#elif defined CONFIG_XEN_ALIGN_DEFAULT + #ifdef EFI + #define SECTION_ALIGN MB(2) + #else + #define SECTION_ALIGN PAGE_SIZE + #endif +#else +#error "Section alignment undefined" +#endif
How about converting the last #elif to #else and omitting the #error? Also would you mind using the defined() style (i.e. with parentheses), as we commonly do elsewhere? And please use uniform indentation (or not) of directives. I also find the indentation itself quite unusual - we have almost no examples of it being like this outside of files we've imported from elsewhere. Typically we retain the # in column 1 and indent only the actual directive keyword.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 1/2] x86: decouple xen alignment setting from EFI/non-EFI build
Quoted 5 lines · Jan Beulich
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 18.03.19 at 15:57, <[email protected]> wrote:
Introduce a new
Kconfigoption to pick the alignment for xen binary. To retain original behaviour, the default pick for EFI build is 2M and non-EFI build 4K.
Quoted 8 lines
Is this a worthwhile step to take, considering that we mean to switch to 2M main section boundaries anyway? I realize it's still not necessarily clear how to get there without re-introducing the boot regression with syslinux that was observed back then, but perhaps time would better be spent there than into introducing configurability? (To be honest I don't recall whether it was determined that there was no way out of this dilemma, but it seems to me there would have been a plan.)
I think it would definitely be sensible to fix it there. But we still have older syslinux to deal with. I don't think they're going away any time soon.
Quoted 20 lines
Quoted 20 lines
xen/arch/x86/Kconfig +13
-0@@ -138,6 +138,32 @@ config TBOOT If unsure, say Y. +choice + prompt "Alignment of Xen binary" + depends on X86 + default XEN_ALIGN_DEFAULT + ---help--- + Specify alignment for Xen binary. + + If unsure, choose "default". + +config XEN_ALIGN_DEFAULT + bool "Default alignment" + ---help--- + Pick alignment according to build variants.
Quoted 9 lines
xen/arch/x86/Kconfig config TBOOT · line 157
Quoted 4 lines
xen/arch/x86/Kconfig +4
-0+ + For EFI build the default alignment is 2M. For non-EFI build + the default alignment is 4K due to syslinux failing to handle + 2M alignment.Wasn't it the resulting image size rather than the alignment itself which did cause the problem? Has the issue been fixed in newer syslinux?
Right. I can make this clearer.
"due to syslinux failing to handle the image size increment induced by 2M alignment".
Also, I think s/binary/image/ in the description would be better.
Looking at syslinux.git I don't immediately see patches to fix the issue (if there is such issue in the first place).
Quoted 9 lines
xen/arch/x86/Kconfig config TBOOT · line 163
Quoted 5 lines
xen/arch/x86/Kconfig +5
-0+config XEN_ALIGN_4K + bool "4K alignment" + +config XEN_ALIGN_2M + bool "2M alignment"In patch 2 you only need the latter - is there really much value in allowing explicitly requesting 4k?
I thought there would be a case in which users want 4K explicitly? I'm certainly fine with dropping the 4K alignment option.
Quoted 25 lines
Quoted 14 lines
xen/arch/x86/Kconfig +3
-1@@ -20,13 +19,26 @@ ENTRY(efi_start) #else /* !EFI */ #define FORMAT "elf64-x86-64" -#define SECTION_ALIGN PAGE_SIZE #define DECL_SECTION(x) x : AT(ADDR(x) - __XEN_VIRT_START) ENTRY(start_pa) #endif /* EFI */ +#if defined CONFIG_XEN_ALIGN_2M +#define SECTION_ALIGN MB(2) +#elif defined CONFIG_XEN_ALIGN_4Kxen/arch/x86/Kconfig ENTRY(efi_start) · line 40
Quoted 10 lines
xen/arch/x86/Kconfig +10
-0+#define SECTION_ALIGN PAGE_SIZE +#elif defined CONFIG_XEN_ALIGN_DEFAULT + #ifdef EFI + #define SECTION_ALIGN MB(2) + #else + #define SECTION_ALIGN PAGE_SIZE + #endif +#else +#error "Section alignment undefined" +#endif
Quoted 8 lines
How about converting the last #elif to #else and omitting the #error? Also would you mind using the
defined()style (i.e. with parentheses), as we commonly do elsewhere? And please use uniform indentation (or not) of directives. I also find the indentation itself quite unusual - we have almost no examples of it being like this outside of files we've imported from elsewhere. Typically we retain the # in column 1 and indent only the actual directive keyword.
Will fix (after we determine what to do with 4K option).
Wei.
Quoted 4 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH 1/2] x86: decouple xen alignment setting from EFI/non-EFI build
Quoted 32 lines
Quoted 1 line
Quoted 1 line · On 19.03.19 at 12:24, <[email protected]> wrote:
Quoted 20 lines · Jan Beulich
Quoted 20 lines
Quoted 1 line
Quoted 1 line · On 18.03.19 at 15:57, <[email protected]> wrote:
xen/arch/x86/Kconfig +13
-0@@ -138,6 +138,32 @@ config TBOOT If unsure, say Y. +choice + prompt "Alignment of Xen binary" + depends on X86 + default XEN_ALIGN_DEFAULT + ---help--- + Specify alignment for Xen binary. + + If unsure, choose "default". + +config XEN_ALIGN_DEFAULT + bool "Default alignment" + ---help--- + Pick alignment according to build variants.Quoted 9 lines
xen/arch/x86/Kconfig config TBOOT · line 157
Quoted 4 lines
xen/arch/x86/Kconfig +4
-0+ + For EFI build the default alignment is 2M. For non-EFI build + the default alignment is 4K due to syslinux failing to handle + 2M alignment.Wasn't it the resulting image size rather than the alignment itself which did cause the problem? Has the issue been fixed in newer syslinux?
Quoted 6 lines
Right. I can make this clearer.
"due to syslinux failing to handle the image size increment induced by 2M alignment".
Also, I think s/binary/image/ in the description would be better.
Agreed.
Quoted 11 lines
Quoted 8 lines
xen/arch/x86/Kconfig config TBOOT · line 163
Quoted 5 lines
xen/arch/x86/Kconfig +5
-0+config XEN_ALIGN_4K + bool "4K alignment" + +config XEN_ALIGN_2M + bool "2M alignment"In patch 2 you only need the latter - is there really much value in allowing explicitly requesting 4k?
I thought there would be a case in which users want 4K explicitly? I'm certainly fine with dropping the 4K alignment option.
Well, without a specific need or request I'd prefer not to see extra options introduced.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH 2/2 x86/pvshim: use 2M alignment
In reply to Wei LiuThe pvshim is loaded directly by toolstack. Use 2M alignment for potential better performance.
--- xen/arch/x86/configs/pvshim_defconfig | 1 + 1 file changed, 1 insertion(+)
xen/arch/x86/configs/pvshim_defconfig +1 -0
@@ -5,6 +5,7 @@ CONFIG_PVH_GUEST=y CONFIG_PV_SHIM=y CONFIG_PV_SHIM_EXCLUSIVE=y CONFIG_NR_CPUS=32 +CONFIG_XEN_ALIGN_2M=y # Disable features not used by the PV shim # CONFIG_SHADOW_PAGING is not set # CONFIG_BIGMEM is not set
Signature · 7 lines
2.20.1 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel