s390x: Add snippet support
cover s390x: Add snippet support
The SIE support allows us to run guests and test the hypervisor's (V)SIE implementation. However it requires that the guest instructions are binary which limits the complexity of the guest code.
The snippet support provides a way to write guest code as ASM or C and simply memcpy it into guest memory. Some of the KVM-unit-test library can be re-used which further speeds up guest code development.
The included mvpg-sie test helped us to deliver the KVM mvpg fixes which Claudio posted a short while ago. In the future I'll post Secure Execution snippet support patches which was my initial goal with this series anyway.
I heard you liked tests so I put tests inside tests so you can test while you test.
Janosch Frank (2):
s390x: Add guest snippet support
s390x: mvpg: Add SIE mvpg test
.gitignore | 2 + s390x/Makefile | 29 ++++++- s390x/mvpg-sie.c | 139 ++++++++++++++++++++++++++++++++ s390x/snippets/c/cstart.S | 13 +++ s390x/snippets/c/flat.lds | 51 ++++++++++++ s390x/snippets/c/mvpg-snippet.c | 33 ++++++++ s390x/unittests.cfg | 3 + 7 files changed, 267 insertions(+), 3 deletions(-) create mode 100644 s390x/mvpg-sie.c create mode 100644 s390x/snippets/c/cstart.S create mode 100644 s390x/snippets/c/flat.lds create mode 100644 s390x/snippets/c/mvpg-snippet.c
Signature · 1 line
2.30.2
RFC 1/2 s390x: Add guest snippet support
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-) create mode 100644 s390x/snippets/c/cstart.S create mode 100644 s390x/snippets/c/flat.lds
.gitignore +2 -0
@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbin
s390x/Makefile +25 -3
@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.o FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a = + +snippets = $(snippets-a)$(snippets-c) +snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) $(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets) $(@:.elf=.aux.o) $(RM) $(@:.elf=.aux.o) @chmod a-x $@
@@ -93,7 +115,7 @@ FLATLIBS = $(libcflat) $(GENPROTIMG) --host-key-document $(HOST_KEY_DOCUMENT) --no-verify --image $< -o $@ arch_clean: asm_offsets_clean - $(RM) $(TEST_DIR)/*.{o,elf,bin} $(TEST_DIR)/.*.d lib/s390x/.*.d + $(RM) $(TEST_DIR)/*.{o,elf,bin} $(SNIPPET_DIR)/c/*.{o,elf,bin,gbin} $(SNIPPET_DIR)/.*.d $(TEST_DIR)/.*.d lib/s390x/.*.d generated-files = $(asm-offsets) $(tests:.elf=.o) $(asmlib) $(cflatobjs): $(generated-files)
s390x/snippets/c/cstart.S new file +13 -0
@@ -0,0 +1,13 @@ +#include <asm/sigp.h> + +.section .init + .globl start +start: + /* XOR all registers with themselves to clear them fully. */ + .irp i, 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 + xgr \i,\i + .endr + /* 0x3000 is the stack page for now */ + lghi %r15, 0x4000 + brasl %r14, main + sigp %r1, %r0, SIGP_STOP
s390x/snippets/c/flat.lds new file +51 -0
51 lines not on this page · raw message
Signature · 1 line
2.30.2
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
Quoted 33 lines · On Thu, 20 May 2021 09:47:29 +0000 Janosch Frank <[email protected]> wrote:
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51+++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-) create mode 100644
s390x/snippets/c/cstart.Screate mode 100644s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-1@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.o FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)
s390x/Makefile OBJDIRS += lib/s390x · line 87
Quoted 10 lines
s390x/Makefile +10
-0+ +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a = + +snippets = $(snippets-a)$(snippets-c)
↑↑
I'm not a Makefile expert, but, don't you need a space between the two variable expansions?
s390x/Makefile OBJDIRS += lib/s390x · line 100
Quoted 13 lines
s390x/Makefile +13
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)
I would keep the %.o as the first in the list
s390x/Makefile OBJDIRS += lib/s390x
Quoted 5 lines
s390x/Makefile +1
-1$(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets)
so all the snippets are always baked in every test?
Quoted 91 lines
s390x/Makefile +0
-2$(@:.elf=.aux.o) $(RM) $(@:.elf=.aux.o) @chmod a-x $@@@ -93,7 +115,7 @@ FLATLIBS = $(libcflat) $(GENPROTIMG) --host-key-document $(HOST_KEY_DOCUMENT) --no-verify --image $< -o $@ arch_clean: asm_offsets_clean - $(RM) $(TEST_DIR)/*.{o,elf,bin} $(TEST_DIR)/.*.d
lib/s390x/.*.ds390x/Makefile +1
-0+ $(RM) $(TEST_DIR)/*.{o,elf,bin}$(SNIPPET_DIR)/c/*.{o,elf,bin,gbin} $(SNIPPET_DIR)/.*.d $(TEST_DIR)/.*.d
lib/s390x/.*.dgenerated-files = $(asm-offsets)
$(tests:.elf=.o) $(asmlib) $(cflatobjs): $(generated-files)s390x/snippets/c/cstart.S new file +13
-0@@ -0,0 +1,13 @@ +#include <asm/sigp.h> + +.section .init + .globl start +start: + /* XOR all registers with themselves to clear them fully. */ + .irp i, 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 + xgr \i,\i + .endr + /* 0x3000 is the stack page for now */ + lghi %r15, 0x4000 + brasl %r14, main + sigp %r1, %r0, SIGP_STOPs390x/snippets/c/flat.lds new file +5
-0@@ -0,0 +1,51 @@ +SECTIONS +{ + .lowcore : { + /* + * Initial short psw for disk boot, with 31 bitaddressing for
s390x/snippets/c/flat.lds new file +1
-0+ * non z/Arch environment compatibility and theinstruction
s390x/snippets/c/flat.lds new file +45
-0+ * address 0x10000 (cstart64.S .init). + */ + . = 0; + LONG(0x00080000) + LONG(0x80004000) + /* Restart new PSW for booting via PSW restart. */ + . = 0x1a0; + QUAD(0x0000000180000000) + QUAD(0x0000000000004000) + } + . = 0x4000; + .text : { + *(.init) + *(.text) + *(.text.*) + } + . = ALIGN(64K); + etext = .; + .opd : { *(.opd) } + . = ALIGN(16); + .dynamic : { + dynamic_start = .; + *(.dynamic) + } + .dynsym : { + dynsym_start = .; + *(.dynsym) + } + .rela.dyn : { *(.rela*) } + . = ALIGN(16); + .data : { + *(.data) + *(.data.rel*) + } + . = ALIGN(16); + .rodata : { *(.rodata) *(.rodata.*) } + . = ALIGN(16); + __bss_start = .; + .bss : { *(.bss) } + __bss_end = .; + . = ALIGN(64K); + edata = .; + . += 64K; + . = ALIGN(64K); +}
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
Quoted 34 lines · Claudio Imbrenda
Quoted 31 lines · On Thu, 20 May 2021 09:47:29 +0000 Janosch Frank <[email protected]> wrote:
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51+++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-) create mode 100644
s390x/snippets/c/cstart.Screate mode 100644s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-0@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.o
Quoted 16 lines
s390x/Makefile OBJDIRS += lib/s390x · line 87
Quoted 12 lines
s390x/Makefile +10
-1FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a = + +snippets = $(snippets-a)$(snippets-c)↑↑
I'm not aMakefileexpert, but, don't you need a space between the two variable expansions?
Yup, I already fixed that one
Quoted 15 lines
s390x/Makefile OBJDIRS += lib/s390x · line 100
Quoted 13 lines
s390x/Makefile +13
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)I would keep the %.o as the first in the list>
Quoted 7 lines
s390x/Makefile OBJDIRS += lib/s390x
Quoted 5 lines
s390x/Makefile +1
-1$(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets)so all the snippets are always baked in every test?
Currently yes.
Do you have better ideas?
Quoted 93 lines
Quoted 92 lines
s390x/Makefile +0
-2$(@:.elf=.aux.o) $(RM) $(@:.elf=.aux.o) @chmod a-x $@@@ -93,7 +115,7 @@ FLATLIBS = $(libcflat) $(GENPROTIMG) --host-key-document $(HOST_KEY_DOCUMENT) --no-verify --image $< -o $@ arch_clean: asm_offsets_clean - $(RM) $(TEST_DIR)/*.{o,elf,bin} $(TEST_DIR)/.*.d
lib/s390x/.*.ds390x/Makefile +1
-0+ $(RM) $(TEST_DIR)/*.{o,elf,bin}$(SNIPPET_DIR)/c/*.{o,elf,bin,gbin} $(SNIPPET_DIR)/.*.d $(TEST_DIR)/.*.d
lib/s390x/.*.dgenerated-files = $(asm-offsets)
$(tests:.elf=.o) $(asmlib) $(cflatobjs): $(generated-files)s390x/snippets/c/cstart.S new file +13
-0@@ -0,0 +1,13 @@ +#include <asm/sigp.h> + +.section .init + .globl start +start: + /* XOR all registers with themselves to clear them fully. */ + .irp i, 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 + xgr \i,\i + .endr + /* 0x3000 is the stack page for now */ + lghi %r15, 0x4000 + brasl %r14, main + sigp %r1, %r0, SIGP_STOPs390x/snippets/c/flat.lds new file +5
-0@@ -0,0 +1,51 @@ +SECTIONS +{ + .lowcore : { + /* + * Initial short psw for disk boot, with 31 bitaddressing for
s390x/snippets/c/flat.lds new file +1
-0+ * non z/Arch environment compatibility and theinstruction
s390x/snippets/c/flat.lds new file +45
-0+ * address 0x10000 (cstart64.S .init). + */ + . = 0; + LONG(0x00080000) + LONG(0x80004000) + /* Restart new PSW for booting via PSW restart. */ + . = 0x1a0; + QUAD(0x0000000180000000) + QUAD(0x0000000000004000) + } + . = 0x4000; + .text : { + *(.init) + *(.text) + *(.text.*) + } + . = ALIGN(64K); + etext = .; + .opd : { *(.opd) } + . = ALIGN(16); + .dynamic : { + dynamic_start = .; + *(.dynamic) + } + .dynsym : { + dynsym_start = .; + *(.dynsym) + } + .rela.dyn : { *(.rela*) } + . = ALIGN(16); + .data : { + *(.data) + *(.data.rel*) + } + . = ALIGN(16); + .rodata : { *(.rodata) *(.rodata.*) } + . = ALIGN(16); + __bss_start = .; + .bss : { *(.bss) } + __bss_end = .; + . = ALIGN(64K); + edata = .; + . += 64K; + . = ALIGN(64K); +}
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
In reply to Janosch FrankQuoted 31 lines · Janosch Frank
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-)create mode 100644
s390x/snippets/c/cstart.S
create mode 100644s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-0@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.o
s390x/Makefile OBJDIRS += lib/s390x · line 85
Quoted 10 lines
s390x/Makefile +8
-1FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a =
Could you please call this snippets-s instead of ...-a ? The -a suffix looks like an archive to me otherwise.
s390x/Makefile OBJDIRS += lib/s390x · line 87
Quoted 1 line
s390x/Makefile +1
-0+snippets = $(snippets-a)$(snippets-c)
Shouldn't there be a space between the two?
Quoted 8 lines
s390x/Makefile +8
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \
s390x/Makefile OBJDIRS += lib/s390x · line 104
Quoted 10 lines
s390x/Makefile +6
-1+ $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) $(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets) $(@:.elf=.aux.o)
Does this link the snippets into all elf files? ... wouldn't it be better to restrict it somehow to the files that really need them?
Quoted 21 lines
s390x/Makefile +1
-1$(RM) $(@:.elf=.aux.o) @chmod a-x $@@@ -93,7 +115,7 @@ FLATLIBS = $(libcflat) $(GENPROTIMG) --host-key-document $(HOST_KEY_DOCUMENT) --no-verify --image $< -o $@ arch_clean: asm_offsets_clean - $(RM) $(TEST_DIR)/*.{o,elf,bin} $(TEST_DIR)/.*.d lib/s390x/.*.d + $(RM) $(TEST_DIR)/*.{o,elf,bin} $(SNIPPET_DIR)/c/*.{o,elf,bin,gbin} $(SNIPPET_DIR)/.*.d $(TEST_DIR)/.*.d lib/s390x/.*.d generated-files = $(asm-offsets) $(tests:.elf=.o) $(asmlib) $(cflatobjs): $(generated-files)s390x/snippets/c/cstart.S new file +3
-0@@ -0,0 +1,13 @@ +#include <asm/sigp.h> + +.section .init
s390x/snippets/c/cstart.S line 13
Quoted 10 lines
s390x/snippets/c/cstart.S new file +10
-0+ .globl start +start: + /* XOR all registers with themselves to clear them fully. */ + .irp i, 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 + xgr \i,\i + .endr + /* 0x3000 is the stack page for now */ + lghi %r15, 0x4000 + brasl %r14, main + sigp %r1, %r0, SIGP_STOP
I think you should clear r0 before using it here?
Thomas
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
Quoted 32 lines · Thomas Huth
Quoted 31 lines · Janosch Frank
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-)create mode 100644
s390x/snippets/c/cstart.S
create mode 100644s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-0@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.o
Quoted 13 lines
s390x/Makefile OBJDIRS += lib/s390x · line 85
Quoted 10 lines
s390x/Makefile +8
-1FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a =Could you please call this snippets-s instead of ...-a ? The -a suffix looks like an archive to me otherwise.
Sure
Quoted 4 lines
s390x/Makefile OBJDIRS += lib/s390x · line 87
Quoted 2 lines
s390x/Makefile +1
-0+snippets = $(snippets-a)$(snippets-c)Shouldn't there be a space between the two?
Yes, already fixed that a long while ago
I thought I had sent out a new version already, maybe that was an illusion as I can't seem to find it right now.
Quoted 14 lines
Quoted 14 lines
s390x/Makefile +13
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)
Quoted 8 lines
s390x/Makefile OBJDIRS += lib/s390x · line 104
Quoted 5 lines
s390x/Makefile +1
-1$(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets) $(@:.elf=.aux.o)Does this link the snippets into all elf files? ... wouldn't it be better to restrict it somehow to the files that really need them?
Yes it does.
I'd like to avoid having to specify a makefile rule for every test that uses snippets as we already have more than the mvpg one in the queue.
So I'm having Steffen looking into a solution for this problem. My first idea was to bring the used snippets into the unittests.cfg but I disliked that we then would have compile instructions in another file. Maybe there's a way to include that into the makefile in a clever way?
Quoted 26 lines
Quoted 26 lines
s390x/Makefile +1
-1$(RM) $(@:.elf=.aux.o) @chmod a-x $@@@ -93,7 +115,7 @@ FLATLIBS = $(libcflat) $(GENPROTIMG) --host-key-document $(HOST_KEY_DOCUMENT) --no-verify --image $< -o $@ arch_clean: asm_offsets_clean - $(RM) $(TEST_DIR)/*.{o,elf,bin} $(TEST_DIR)/.*.d lib/s390x/.*.d + $(RM) $(TEST_DIR)/*.{o,elf,bin} $(SNIPPET_DIR)/c/*.{o,elf,bin,gbin} $(SNIPPET_DIR)/.*.d $(TEST_DIR)/.*.d lib/s390x/.*.d generated-files = $(asm-offsets) $(tests:.elf=.o) $(asmlib) $(cflatobjs): $(generated-files)s390x/snippets/c/cstart.S new file +7
-0@@ -0,0 +1,13 @@ +#include <asm/sigp.h> + +.section .init + .globl start +start: + /* XOR all registers with themselves to clear them fully. */ + .irp i, 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
Quoted 8 lines
s390x/snippets/c/cstart.S line 13
Quoted 6 lines
s390x/snippets/c/cstart.S new file +6
-0+ xgr \i,\i + .endr + /* 0x3000 is the stack page for now */ + lghi %r15, 0x4000 + brasl %r14, main + sigp %r1, %r0, SIGP_STOPI think you should clear r0 before using it here?
Right. will fix
Quoted 3 lines
Thomas
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
Quoted 81 lines · Janosch Frank
Quoted 32 lines · Thomas Huth
Quoted 31 lines · Janosch Frank
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-)create mode 100644
s390x/snippets/c/cstart.S
create mode 100644s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-0@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.oQuoted 13 lines
s390x/Makefile OBJDIRS += lib/s390x · line 85
Quoted 10 lines
s390x/Makefile +8
-1FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a =Could you please call this snippets-s instead of ...-a ? The -a suffix looks like an archive to me otherwise.
Sure
Quoted 4 lines
s390x/Makefile OBJDIRS += lib/s390x · line 87
Quoted 2 lines
s390x/Makefile +1
-0+snippets = $(snippets-a)$(snippets-c)Shouldn't there be a space between the two?
Yes, already fixed that a long while ago
I thought I had sent out a new version already, maybe that was an illusion as I can't seem to find it right now.Quoted 14 lines
Quoted 14 lines
s390x/Makefile +13
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)Quoted 8 lines
s390x/Makefile OBJDIRS += lib/s390x · line 104
Quoted 5 lines
s390x/Makefile +1
-1$(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets) $(@:.elf=.aux.o)Does this link the snippets into all elf files? ... wouldn't it be better to restrict it somehow to the files that really need them?
Quoted 8 lines
Yes it does.
I'd like to avoid having to specify a makefile rule for every test that uses snippets as we already have more than the mvpg one in the queue.So I'm having Steffen looking into a solution for this problem. My first idea was to bring the used snippets into the unittests.cfg but I disliked that we then would have compile instructions in another file. Maybe there's a way to include that into the makefile in a clever way?
I haven't tried, but maybe you could replace the $(snippets) in the last line with
$(wildcard snippets/[email protected])
or something similar?
Thomas
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
Quoted 91 lines · Thomas Huth
Quoted 81 lines · Janosch Frank
Quoted 32 lines · Thomas Huth
Quoted 31 lines · Janosch Frank
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-)create mode 100644
s390x/snippets/c/cstart.S
create mode 100644s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-0@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.oQuoted 13 lines
s390x/Makefile OBJDIRS += lib/s390x · line 85
Quoted 10 lines
s390x/Makefile +8
-1FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a =Could you please call this snippets-s instead of ...-a ? The -a suffix looks like an archive to me otherwise.
Sure
Quoted 4 lines
s390x/Makefile OBJDIRS += lib/s390x · line 87
Quoted 2 lines
s390x/Makefile +1
-0+snippets = $(snippets-a)$(snippets-c)Shouldn't there be a space between the two?
Yes, already fixed that a long while ago
I thought I had sent out a new version already, maybe that was an illusion as I can't seem to find it right now.Quoted 14 lines
Quoted 14 lines
s390x/Makefile +13
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)Quoted 8 lines
s390x/Makefile OBJDIRS += lib/s390x · line 104
Quoted 5 lines
s390x/Makefile +1
-1$(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets) $(@:.elf=.aux.o)Does this link the snippets into all elf files? ... wouldn't it be better to restrict it somehow to the files that really need them?
Quoted 8 lines
Yes it does.
I'd like to avoid having to specify a makefile rule for every test that uses snippets as we already have more than the mvpg one in the queue.So I'm having Steffen looking into a solution for this problem. My first idea was to bring the used snippets into the unittests.cfg but I disliked that we then would have compile instructions in another file. Maybe there's a way to include that into the makefile in a clever way?
Quoted 6 lines
I haven't tried, but maybe you could replace the $(snippets) in the last line with
$(wildcard snippets/[email protected])or something similar?
That starts falling apart when multiple tests use the same snippet, no?
Quoted 3 lines
Thomas
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
Quoted 92 lines · Janosch Frank
Quoted 91 lines · Thomas Huth
Quoted 89 lines · Janosch Frank
Quoted 32 lines · Thomas Huth
Quoted 31 lines · Janosch Frank
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-) create mode 100644 s390x/snippets/c/cstart.S create mode 100644 s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-0@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.oQuoted 13 lines
s390x/Makefile OBJDIRS += lib/s390x · line 85
Quoted 10 lines
s390x/Makefile +8
-1FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a =Could you please call this snippets-s instead of ...-a ? The -a suffix looks like an archive to me otherwise.
Sure
Quoted 4 lines
s390x/Makefile OBJDIRS += lib/s390x · line 87
Quoted 2 lines
s390x/Makefile +1
-0+snippets = $(snippets-a)$(snippets-c)Shouldn't there be a space between the two?
Yes, already fixed that a long while ago
I thought I had sent out a new version already, maybe that was an illusion as I can't seem to find it right now.Quoted 14 lines
Quoted 14 lines
s390x/Makefile +13
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)Quoted 8 lines
s390x/Makefile OBJDIRS += lib/s390x · line 104
Quoted 5 lines
s390x/Makefile +1
-1$(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets) $(@:.elf=.aux.o)Does this link the snippets into all elf files? ... wouldn't it be better to restrict it somehow to the files that really need them?
Yes it does.
I'd like to avoid having to specify a makefile rule for every test that uses snippets as we already have more than the mvpg one in the queue.So I'm having Steffen looking into a solution for this problem. My first idea was to bring the used snippets into the unittests.cfg but I disliked that we then would have compile instructions in another file. Maybe there's a way to include that into the makefile in a clever way?
Quoted 8 lines
Quoted 6 lines
I haven't tried, but maybe you could replace the $(snippets) in the last line with
$(wildcard snippets/[email protected])or something similar?
That starts falling apart when multiple tests use the same snippet, no?
That's true ... Maybe something like:
$(filter %.gbin,$^)
?
Thomas
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
Quoted 98 lines · Thomas Huth
Quoted 97 lines · Janosch Frank
Quoted 96 lines · Thomas Huth
Quoted 81 lines · Janosch Frank
Quoted 32 lines · Thomas Huth
Quoted 31 lines · Janosch Frank
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-) create mode 100644 s390x/snippets/c/cstart.S create mode 100644 s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-0@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.oQuoted 13 lines
s390x/Makefile OBJDIRS += lib/s390x · line 85
Quoted 10 lines
s390x/Makefile +8
-1FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a =Could you please call this snippets-s instead of ...-a ? The -a suffix looks like an archive to me otherwise.
Sure
Quoted 4 lines
s390x/Makefile OBJDIRS += lib/s390x · line 87
Quoted 2 lines
s390x/Makefile +1
-0+snippets = $(snippets-a)$(snippets-c)Shouldn't there be a space between the two?
Yes, already fixed that a long while ago
I thought I had sent out a new version already, maybe that was an illusion as I can't seem to find it right now.Quoted 14 lines
Quoted 14 lines
s390x/Makefile +13
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)Quoted 8 lines
s390x/Makefile OBJDIRS += lib/s390x · line 104
Quoted 5 lines
s390x/Makefile +1
-1$(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets) $(@:.elf=.aux.o)Does this link the snippets into all elf files? ... wouldn't it be better to restrict it somehow to the files that really need them?
Quoted 8 lines
Yes it does.
I'd like to avoid having to specify a makefile rule for every test that uses snippets as we already have more than the mvpg one in the queue.So I'm having Steffen looking into a solution for this problem. My first idea was to bring the used snippets into the unittests.cfg but I disliked that we then would have compile instructions in another file. Maybe there's a way to include that into the makefile in a clever way?
I haven't tried, but maybe you could replace the $(snippets) in the last line with
$(wildcard snippets/[email protected])
Quoted 7 lines
Quoted 3 lines
Quoted 1 line
or something similar?
That starts falling apart when multiple tests use the same snippet, no?
That's true ... Maybe something like:
$(filter %.gbin,$^)
That filters all files that are not gbins from the prereqs, right? In what way is that different to linking all the snippets? After all they are currently a prereq for %.elf or am I missing something here?
Just to expand our use-case a bit more with an example since the code is still too rough to post right now:
We test diagnose handling for Secure Execution / PV in one host test where we have a gbin each for most diagnose codes 0x44, 0x9c, 0x288, 0x500 (0x308 needs a whole test and possible multiple gbins on its own due to its complexity).
Other test bundles are PV IO, misc (stsi, sthyi, ...) and of course other VSIE instruction handling tests in the future.
Quoted 5 lines
?
Thomas
Re: [kvm-unit-tests RFC 1/2] s390x: Add guest snippet support
Quoted 103 lines · Janosch Frank
Quoted 102 lines · Thomas Huth
Quoted 100 lines · Janosch Frank
Quoted 91 lines · Thomas Huth
Quoted 81 lines · Janosch Frank
Quoted 32 lines · Thomas Huth
Quoted 31 lines · Janosch Frank
Snippets can be used to easily write and run guest (SIE) tests. The snippet is linked into the test binaries and can therefore be accessed via a ptr.
Signed-off-by Janosch Frank <[email protected]>--- .gitignore | 2 ++ s390x/Makefile | 28 ++++++++++++++++++--- s390x/snippets/c/cstart.S | 13 ++++++++++ s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 91 insertions(+), 3 deletions(-) create mode 100644 s390x/snippets/c/cstart.S create mode 100644 s390x/snippets/c/flat.lds.gitignore +2
-0@@ -22,3 +22,5 @@ cscope.* /api/dirty-log /api/dirty-log-perf /s390x/*.bin +/s390x/snippets/*/*.bin +/s390x/snippets/*/*.gbins390x/Makefile +0
-0@@ -75,11 +75,33 @@ OBJDIRS += lib/s390x asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.oQuoted 13 lines
s390x/Makefile OBJDIRS += lib/s390x · line 85
Quoted 10 lines
s390x/Makefile +8
-1FLATLIBS = $(libcflat) -%.elf: %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib) + +SNIPPET_DIR = $(TEST_DIR)/snippets + +# C snippets that need to be linked +snippets-c = + +# ASM snippets that are directly compiled and converted to a *.gbin +snippets-a =Could you please call this snippets-s instead of ...-a ? The -a suffix looks like an archive to me otherwise.
Sure
Quoted 4 lines
s390x/Makefile OBJDIRS += lib/s390x · line 86
Quoted 2 lines
s390x/Makefile +1
-0+snippets = $(snippets-a)$(snippets-c)Shouldn't there be a space between the two?
Yes, already fixed that a long while ago
I thought I had sent out a new version already, maybe that was an illusion as I can't seem to find it right now.Quoted 14 lines
Quoted 14 lines
s390x/Makefile +13
-0+snippets-o += $(patsubst %.gbin,%.o,$(snippets)) + +$(snippets-a): $(snippets-o) $(FLATLIBS) + $(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +$(snippets-c): $(snippets-o) $(SNIPPET_DIR)/c/cstart.o $(FLATLIBS) + $(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds \ + $(filter %.o, $^) $(FLATLIBS) + $(OBJCOPY) -O binary $@ $@ + $(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@ + +%.elf: $(snippets) %.o $(FLATLIBS) $(SRCDIR)/s390x/flat.lds $(asmlib)Quoted 8 lines
s390x/Makefile OBJDIRS += lib/s390x · line 103
Quoted 5 lines
s390x/Makefile +1
-1$(CC) $(CFLAGS) -c -o $(@:.elf=.aux.o) \ $(SRCDIR)/lib/auxinfo.c -DPROGNAME=\"$@\" $(CC) $(LDFLAGS) -o $@ -T $(SRCDIR)/s390x/flat.lds \ - $(filter %.o, $^) $(FLATLIBS) $(@:.elf=.aux.o) + $(filter %.o, $^) $(FLATLIBS) $(snippets) $(@:.elf=.aux.o)Does this link the snippets into all elf files? ... wouldn't it be better to restrict it somehow to the files that really need them?
Quoted 8 lines
Yes it does.
I'd like to avoid having to specify a makefile rule for every test that uses snippets as we already have more than the mvpg one in the queue.So I'm having Steffen looking into a solution for this problem. My first idea was to bring the used snippets into the unittests.cfg but I disliked that we then would have compile instructions in another file. Maybe there's a way to include that into the makefile in a clever way?
Quoted 6 lines
I haven't tried, but maybe you could replace the $(snippets) in the last line with
$(wildcard snippets/[email protected])or something similar?
That starts falling apart when multiple tests use the same snippet, no?
Quoted 7 lines
Quoted 3 lines
That's true ... Maybe something like:
$(filter %.gbin,$^)That filters all files that are not gbins from the prereqs, right? In what way is that different to linking all the snippets? After all they are currently a prereq for %.elf or am I missing something here?
Sorry, I meant we could then add the prereqs manually for each tests that require it, e.g. add a single line like:
mvpg-sie.elf: snippets/mvpg.gbin
(without specifying a rule, just the dependency)
... not sure whether that works, though...
Thomas
RFC 2/2 s390x: mvpg: Add SIE mvpg test
In reply to Janosch FrankLet's also check the PEI values to make sure our VSIE implementation is correct.
--- s390x/Makefile | 3 +- s390x/mvpg-sie.c | 139 ++++++++++++++++++++++++++++++++ s390x/snippets/c/mvpg-snippet.c | 33 ++++++++ s390x/unittests.cfg | 3 + 4 files changed, 177 insertions(+), 1 deletion(-) create mode 100644 s390x/mvpg-sie.c create mode 100644 s390x/snippets/c/mvpg-snippet.c
s390x/Makefile +2 -1
@@ -22,6 +22,7 @@ tests += $(TEST_DIR)/uv-guest.elf tests += $(TEST_DIR)/sie.elf tests += $(TEST_DIR)/mvpg.elf tests += $(TEST_DIR)/uv-host.elf +tests += $(TEST_DIR)/mvpg-sie.elf tests_binary = $(patsubst %.elf,%.bin,$(tests)) ifneq ($(HOST_KEY_DOCUMENT),)
@@ -79,7 +80,7 @@ FLATLIBS = $(libcflat) SNIPPET_DIR = $(TEST_DIR)/snippets # C snippets that need to be linked -snippets-c = +snippets-c = $(SNIPPET_DIR)/c/mvpg-snippet.gbin # ASM snippets that are directly compiled and converted to a *.gbin snippets-a =
s390x/mvpg-sie.c new file +139 -0
139 lines not on this page · raw message
s390x/snippets/c/mvpg-snippet.c new file +33 -0
@@ -0,0 +1,33 @@ +#include <libcflat.h> + +static inline void force_exit (void ) +{ + asmvolatile (" diag 0,0,0x44\n" ); +} + +static inline int mvpg (unsigned long r0,void *dest,void *src) +{ + register unsigned long reg0 asm ("0") = r0; +int cc; + + asmvolatile (" mvpg %1,%2\n" +" ipm %0\n" +" srl %0,28" + :"=&d" (cc) :"a" (dest),"a" (src),"d" (reg0) + :"memory" ,"cc" ); +return cc; +} + +static void test_mvpg_real (void ) +{ +mvpg (0 , (void *)0x5000 , (void *)0x6000 ); +force_exit (); +} + +__attribute__((section (".text" )))int main (void ) +{ +test_mvpg_real (); +test_mvpg_real (); +test_mvpg_real (); +return 0 ; +}
s390x/unittests.cfg +3 -0
@@ -103,3 +103,6 @@ file = sie.elf [mvpg] file = mvpg.elf timeout = 10 + +[mvpg-sie] +file = mvpg-sie.elf
Signature · 1 line
2.30.2
Re: [kvm-unit-tests RFC 2/2] s390x: mvpg: Add SIE mvpg test
Quoted 61 lines · On Thu, 20 May 2021 09:47:30 +0000 Janosch Frank <[email protected]> wrote:
Let's also check the PEI values to make sure our VSIE implementation is correct.
Signed-off-by Janosch Frank <[email protected]>--- s390x/Makefile | 3 +- s390x/mvpg-sie.c | 139 ++++++++++++++++++++++++++++++++ s390x/snippets/c/mvpg-snippet.c | 33 ++++++++ s390x/unittests.cfg | 3 + 4 files changed, 177 insertions(+), 1 deletion(-) create mode 100644 s390x/mvpg-sie.c create mode 100644 s390x/snippets/c/mvpg-snippet.cs390x/Makefile +2
-1@@ -22,6 +22,7 @@ tests += $(TEST_DIR)/uv-guest.elf tests += $(TEST_DIR)/sie.elf tests += $(TEST_DIR)/mvpg.elf tests += $(TEST_DIR)/uv-host.elf +tests += $(TEST_DIR)/mvpg-sie.elf tests_binary = $(patsubst %.elf,%.bin,$(tests)) ifneq ($(HOST_KEY_DOCUMENT),)@@ -79,7 +80,7 @@ FLATLIBS = $(libcflat) SNIPPET_DIR = $(TEST_DIR)/snippets # C snippets that need to be linked -snippets-c = +snippets-c = $(SNIPPET_DIR)/c/mvpg-snippet.gbin # ASM snippets that are directly compiled and converted to a *.gbin snippets-a =s390x/mvpg-sie.c new file +21
-0@@ -0,0 +1,139 @@ +#include <libcflat.h> +#include <asm/asm-offsets.h> +#include <asm-generic/barrier.h> +#include <asm/interrupt.h> +#include <asm/pgtable.h> +#include <mmu.h> +#include <asm/page.h> +#include <asm/facility.h> +#include <asm/mem.h> +#include <asm/sigp.h> +#include <smp.h> +#include <alloc_page.h> +#include <bitops.h> +#include <vm.h> +#include <sclp.h> +#include <sie.h> + +static u8 *guest; +static u8 *guest_instr; +static struct vm vm; +
s390x/mvpg-sie.c line 31
Quoted 10 lines
s390x/mvpg-sie.c new file +10
-0+static uint8_t *src; +static uint8_t *dst; + +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_start[]; +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_end[]; +int binary_size; + +static void handle_validity (struct vm *vm) +{ +report (0 ,"VALIDITY: %x" , vm->sblk->ipb >>16 );
I think an assert would be better. This should not happen, and if it happens something went very wrong and we have no guarantee that we will be able to continue
Quoted 22 lines
s390x/mvpg-sie.c new file +19
-0+} + +static void sie (struct vm *vm) +{ +/* Reset icptcode so we don't trip below */ + vm->sblk->icptcode =0 ; + +while (vm->sblk->icptcode ==0 ) { +sie64a (vm->sblk, &vm->save_area); +if (vm->sblk->icptcode ==ICPT_VALIDITY ) +handle_validity (vm); + } + vm->save_area.guest.grs[14 ] = vm->sblk->gg14; + vm->save_area.guest.grs[15 ] = vm->sblk->gg15; +} + +static void test_mvpg_pei (void ) +{ + uint64_t **pei_dst = (uint64_t **)((uintptr_t) vm.sblk +
0xc0);s390x/mvpg-sie.c new file +1
-0+ uint64_t **pei_src = (uint64_t **)((uintptr_t) vm.sblk +
0xc8); +
s390x/mvpg-sie.c line 53
Quoted 1 line
s390x/mvpg-sie.c new file +1
-0+report_prefix_push ("pei" );
maybe clear the destination buffer...
Quoted 8 lines
s390x/mvpg-sie.c new file +3
-0+protect_page (guest +0x6000 ,PAGE_ENTRY_I ); +sie (&vm); + report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partialexecution");
s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_src == ((uintptr_t)vm.sblk->mso) +
0x6000+PAGE_ENTRY_I, "PEI_SRCcorrect");s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000,"
PEI_DSTcorrect");
... and check that the page was not copied
s390x/mvpg-sie.c
Quoted 2 lines
s390x/mvpg-sie.c new file +2
-0+/* Jump over the diag44 */ +sie (&vm);
I would check if you really got a diag44
Quoted 22 lines
s390x/mvpg-sie.c new file +6
-0+/* Clear PEI data for next check */ +memset ((uint64_t *)((uintptr_t ) vm.sblk +0xc0 ),0 ,16 ); +unprotect_page (guest +0x6000 ,PAGE_ENTRY_I ); +protect_page (guest +0x5000 ,PAGE_ENTRY_I ); +sie (&vm); + report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partialexecution");
s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_src == vm.sblk->mso + 0x6000,"
PEI_SRCcorrect");s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000 +
PAGE_ENTRY_I, "PEI_DSTcorrect"); +s390x/mvpg-sie.c new file +6
-0+report_prefix_pop (); +} + +static void test_mvpg (void ) +{ + int binary_size =((uintptr_t)_binary_s390x_snippets_c_mvpg_snippet_gbin_end - + (uintptr_t)_binary_s390x_snippets_c_mvpg_snippet_gbin_start); +
s390x/mvpg-sie.c new file +1
-0+ memcpy(guest,
_binary_s390x_snippets_c_mvpg_snippet_gbin_start,binary_size);
s390x/mvpg-sie.c
Quoted 5 lines
s390x/mvpg-sie.c new file +5
-0+memset (src,0x42 ,PAGE_SIZE ); +memset (dst,0x43 ,PAGE_SIZE ); +sie (&vm); +mb (); + report(!memcmp(src, dst, PAGE_SIZE) && *dst == 0x42, "Page
or maybe you can clear the destination buffer here, if you prefer
Quoted 108 lines
moved"); +}
s390x/mvpg-sie.c new file +16
-0+ +static void setup_guest (void ) +{ +setup_vm (); + +/* Allocate 1MB as guest memory */ + guest =alloc_pages (8 ); +/* The first two pages are the lowcore */ + guest_instr = guest +PAGE_SIZE *2 ; + + vm.sblk =alloc_page (); + + vm.sblk->cpuflags =CPUSTAT_ZARCH |CPUSTAT_RUNNING ; + vm.sblk->prefix =0 ; +/* +* Pageable guest with the same ASCE as the test programm, but
s390x/mvpg-sie.c new file +4
-0+* the guest memory 0x0 is offset to start at the allocated +* guest pages and end after 1MB. +* +* It's not pretty but faster and easier than managing guest ASCEs.
s390x/mvpg-sie.c new file +11
-0+*/ + vm.sblk->mso = (u64 )guest; + vm.sblk->msl = (u64 )guest; + vm.sblk->ihcpu =0xffff ; + + vm.sblk->crycbd = (uint64_t )alloc_page (); + + vm.sblk->gpsw.addr =PAGE_SIZE *4 ; + vm.sblk->gpsw.mask =0x0000000180000000ULL ; + vm.sblk->ictl =ICTL_OPEREXC |ICTL_PINT ; +/* Enable MVPG interpretation as we want to test KVM and not ourselves */
s390x/mvpg-sie.c new file +23
-0+ vm.sblk->eca =ECA_MVPGI ; + + src = guest +PAGE_SIZE *6 ; + dst = guest +PAGE_SIZE *5 ; +} + +int main (void ) +{ +report_prefix_push ("mvpg-sie" ); +if (!sclp_facilities.has_sief2) { +report_skip ("SIEF2 facility unavailable" ); + goto done; + } + +setup_guest (); +test_mvpg (); +test_mvpg_pei (); + +done: +report_prefix_pop (); +return report_summary (); + +}(quoted diff) +0
-0
b/s390x/snippets/c/mvpg-snippet.cnew file mode 100644 index 00000000..96b70c9cs390x/snippets/c/mvpg-snippet.c new file +33
-0@@ -0,0 +1,33 @@ +#include <libcflat.h> + +static inline void force_exit (void ) +{ + asmvolatile (" diag 0,0,0x44\n" ); +} + +static inline int mvpg (unsigned long r0,void *dest,void *src) +{ + register unsigned long reg0 asm ("0") = r0; +int cc; + + asmvolatile (" mvpg %1,%2\n" +" ipm %0\n" +" srl %0,28" + :"=&d" (cc) :"a" (dest),"a" (src),"d" (reg0) + :"memory" ,"cc" ); +return cc; +} + +static void test_mvpg_real (void ) +{ +mvpg (0 , (void *)0x5000 , (void *)0x6000 ); +force_exit (); +} + +__attribute__((section (".text" )))int main (void ) +{ +test_mvpg_real (); +test_mvpg_real (); +test_mvpg_real (); +return 0 ; +}s390x/unittests.cfg +3
-0@@ -103,3 +103,6 @@ file = sie.elf [mvpg] file = mvpg.elf timeout = 10 + +[mvpg-sie] +file = mvpg-sie.elf
Re: [kvm-unit-tests RFC 2/2] s390x: mvpg: Add SIE mvpg test
Quoted 70 lines · Claudio Imbrenda
Quoted 67 lines · On Thu, 20 May 2021 09:47:30 +0000 Janosch Frank <[email protected]> wrote:
Let's also check the PEI values to make sure our VSIE implementation is correct.
Signed-off-by Janosch Frank <[email protected]>--- s390x/Makefile | 3 +- s390x/mvpg-sie.c | 139 ++++++++++++++++++++++++++++++++ s390x/snippets/c/mvpg-snippet.c | 33 ++++++++ s390x/unittests.cfg | 3 + 4 files changed, 177 insertions(+), 1 deletion(-) create mode 100644 s390x/mvpg-sie.c create mode 100644 s390x/snippets/c/mvpg-snippet.cs390x/Makefile +2
-1@@ -22,6 +22,7 @@ tests += $(TEST_DIR)/uv-guest.elf tests += $(TEST_DIR)/sie.elf tests += $(TEST_DIR)/mvpg.elf tests += $(TEST_DIR)/uv-host.elf +tests += $(TEST_DIR)/mvpg-sie.elf tests_binary = $(patsubst %.elf,%.bin,$(tests)) ifneq ($(HOST_KEY_DOCUMENT),)@@ -79,7 +80,7 @@ FLATLIBS = $(libcflat) SNIPPET_DIR = $(TEST_DIR)/snippets # C snippets that need to be linked -snippets-c = +snippets-c = $(SNIPPET_DIR)/c/mvpg-snippet.gbin # ASM snippets that are directly compiled and converted to a *.gbin snippets-a =s390x/mvpg-sie.c new file +27
-0@@ -0,0 +1,139 @@ +#include <libcflat.h> +#include <asm/asm-offsets.h> +#include <asm-generic/barrier.h> +#include <asm/interrupt.h> +#include <asm/pgtable.h> +#include <mmu.h> +#include <asm/page.h> +#include <asm/facility.h> +#include <asm/mem.h> +#include <asm/sigp.h> +#include <smp.h> +#include <alloc_page.h> +#include <bitops.h> +#include <vm.h> +#include <sclp.h> +#include <sie.h> + +static u8 *guest; +static u8 *guest_instr; +static struct vm vm; + +static uint8_t *src; +static uint8_t *dst; + +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_start[]; +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_end[]; +int binary_size;
Quoted 8 lines
s390x/mvpg-sie.c line 31
Quoted 4 lines
s390x/mvpg-sie.c new file +4
-0+ +static void handle_validity (struct vm *vm) +{ +report (0 ,"VALIDITY: %x" , vm->sblk->ipb >>16 );I think an assert would be better. This should not happen, and if it happens something went very wrong and we have no guarantee that we will be able to continue
That depends on the test. If you use multiple gbins/snippets in a test, then a validity for one of those does not necessarily break the others. Especially when you think about PV tests where the destroy will clean out all the state anyway.
Quoted 27 lines
Quoted 23 lines
s390x/mvpg-sie.c new file +19
-0+} + +static void sie (struct vm *vm) +{ +/* Reset icptcode so we don't trip below */ + vm->sblk->icptcode =0 ; + +while (vm->sblk->icptcode ==0 ) { +sie64a (vm->sblk, &vm->save_area); +if (vm->sblk->icptcode ==ICPT_VALIDITY ) +handle_validity (vm); + } + vm->save_area.guest.grs[14 ] = vm->sblk->gg14; + vm->save_area.guest.grs[15 ] = vm->sblk->gg15; +} + +static void test_mvpg_pei (void ) +{ + uint64_t **pei_dst = (uint64_t **)((uintptr_t) vm.sblk +
0xc0);s390x/mvpg-sie.c new file +1
-0+ uint64_t **pei_src = (uint64_t **)((uintptr_t) vm.sblk +
0xc8); +s390x/mvpg-sie.c line 53
Quoted 1 line
s390x/mvpg-sie.c new file +1
-0+report_prefix_push ("pei" );maybe clear the destination buffer...
Quoted 10 lines
Quoted 8 lines
s390x/mvpg-sie.c new file +3
-0+protect_page (guest +0x6000 ,PAGE_ENTRY_I ); +sie (&vm); + report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partialexecution");
s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_src == ((uintptr_t)vm.sblk->mso) +
0x6000+PAGE_ENTRY_I, "PEI_SRCcorrect");s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000,"
PEI_DSTcorrect");... and check that the page was not copied
Sure
Quoted 5 lines
s390x/mvpg-sie.c
Quoted 3 lines
s390x/mvpg-sie.c new file +2
-0+/* Jump over the diag44 */ +sie (&vm);I would check if you really got a diag44
Sure, reporting doesn't make sense to me but an assert should be fine
Quoted 22 lines
Quoted 22 lines
s390x/mvpg-sie.c new file +6
-0+/* Clear PEI data for next check */ +memset ((uint64_t *)((uintptr_t ) vm.sblk +0xc0 ),0 ,16 ); +unprotect_page (guest +0x6000 ,PAGE_ENTRY_I ); +protect_page (guest +0x5000 ,PAGE_ENTRY_I ); +sie (&vm); + report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partialexecution");
s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_src == vm.sblk->mso + 0x6000,"
PEI_SRCcorrect");s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000 +
PAGE_ENTRY_I, "PEI_DSTcorrect"); +s390x/mvpg-sie.c new file +6
-0+report_prefix_pop (); +} + +static void test_mvpg (void ) +{ + int binary_size =((uintptr_t)_binary_s390x_snippets_c_mvpg_snippet_gbin_end - + (uintptr_t)_binary_s390x_snippets_c_mvpg_snippet_gbin_start); +
s390x/mvpg-sie.c new file +1
-0+ memcpy(guest,
Quoted 8 lines
s390x/mvpg-sie.c
Quoted 6 lines
_binary_s390x_snippets_c_mvpg_snippet_gbin_start,binary_size);s390x/mvpg-sie.c new file +5
-0+memset (src,0x42 ,PAGE_SIZE ); +memset (dst,0x43 ,PAGE_SIZE ); +sie (&vm); +mb (); + report(!memcmp(src, dst, PAGE_SIZE) && *dst == 0x42, "Pageor maybe you can clear the destination buffer here, if you prefer
I'll have a look, thanks for the review!
Quoted 110 lines
Quoted 109 lines
moved"); +}
s390x/mvpg-sie.c new file +16
-0+ +static void setup_guest (void ) +{ +setup_vm (); + +/* Allocate 1MB as guest memory */ + guest =alloc_pages (8 ); +/* The first two pages are the lowcore */ + guest_instr = guest +PAGE_SIZE *2 ; + + vm.sblk =alloc_page (); + + vm.sblk->cpuflags =CPUSTAT_ZARCH |CPUSTAT_RUNNING ; + vm.sblk->prefix =0 ; +/* +* Pageable guest with the same ASCE as the test programm, but
s390x/mvpg-sie.c new file +4
-0+* the guest memory 0x0 is offset to start at the allocated +* guest pages and end after 1MB. +* +* It's not pretty but faster and easier than managing guest ASCEs.
s390x/mvpg-sie.c new file +11
-0+*/ + vm.sblk->mso = (u64 )guest; + vm.sblk->msl = (u64 )guest; + vm.sblk->ihcpu =0xffff ; + + vm.sblk->crycbd = (uint64_t )alloc_page (); + + vm.sblk->gpsw.addr =PAGE_SIZE *4 ; + vm.sblk->gpsw.mask =0x0000000180000000ULL ; + vm.sblk->ictl =ICTL_OPEREXC |ICTL_PINT ; +/* Enable MVPG interpretation as we want to test KVM and not ourselves */
s390x/mvpg-sie.c new file +23
-0+ vm.sblk->eca =ECA_MVPGI ; + + src = guest +PAGE_SIZE *6 ; + dst = guest +PAGE_SIZE *5 ; +} + +int main (void ) +{ +report_prefix_push ("mvpg-sie" ); +if (!sclp_facilities.has_sief2) { +report_skip ("SIEF2 facility unavailable" ); + goto done; + } + +setup_guest (); +test_mvpg (); +test_mvpg_pei (); + +done: +report_prefix_pop (); +return report_summary (); + +}(quoted diff) +0
-0
b/s390x/snippets/c/mvpg-snippet.cnew file mode 100644 index 00000000..96b70c9cs390x/snippets/c/mvpg-snippet.c new file +33
-0@@ -0,0 +1,33 @@ +#include <libcflat.h> + +static inline void force_exit (void ) +{ + asmvolatile (" diag 0,0,0x44\n" ); +} + +static inline int mvpg (unsigned long r0,void *dest,void *src) +{ + register unsigned long reg0 asm ("0") = r0; +int cc; + + asmvolatile (" mvpg %1,%2\n" +" ipm %0\n" +" srl %0,28" + :"=&d" (cc) :"a" (dest),"a" (src),"d" (reg0) + :"memory" ,"cc" ); +return cc; +} + +static void test_mvpg_real (void ) +{ +mvpg (0 , (void *)0x5000 , (void *)0x6000 ); +force_exit (); +} + +__attribute__((section (".text" )))int main (void ) +{ +test_mvpg_real (); +test_mvpg_real (); +test_mvpg_real (); +return 0 ; +}s390x/unittests.cfg +3
-0@@ -103,3 +103,6 @@ file = sie.elf [mvpg] file = mvpg.elf timeout = 10 + +[mvpg-sie] +file = mvpg-sie.elf
Re: [kvm-unit-tests RFC 2/2] s390x: mvpg: Add SIE mvpg test
In reply to Claudio ImbrendaQuoted 116 lines · Claudio Imbrenda
Quoted 61 lines · On Thu, 20 May 2021 09:47:30 +0000 Janosch Frank <[email protected]> wrote:
Let's also check the PEI values to make sure our VSIE implementation is correct.
Signed-off-by Janosch Frank <[email protected]>--- s390x/Makefile | 3 +- s390x/mvpg-sie.c | 139 ++++++++++++++++++++++++++++++++ s390x/snippets/c/mvpg-snippet.c | 33 ++++++++ s390x/unittests.cfg | 3 + 4 files changed, 177 insertions(+), 1 deletion(-) create mode 100644 s390x/mvpg-sie.c create mode 100644 s390x/snippets/c/mvpg-snippet.cs390x/Makefile +2
-1@@ -22,6 +22,7 @@ tests += $(TEST_DIR)/uv-guest.elf tests += $(TEST_DIR)/sie.elf tests += $(TEST_DIR)/mvpg.elf tests += $(TEST_DIR)/uv-host.elf +tests += $(TEST_DIR)/mvpg-sie.elf tests_binary = $(patsubst %.elf,%.bin,$(tests)) ifneq ($(HOST_KEY_DOCUMENT),)@@ -79,7 +80,7 @@ FLATLIBS = $(libcflat) SNIPPET_DIR = $(TEST_DIR)/snippets # C snippets that need to be linked -snippets-c = +snippets-c = $(SNIPPET_DIR)/c/mvpg-snippet.gbin # ASM snippets that are directly compiled and converted to a *.gbin snippets-a =s390x/mvpg-sie.c new file +21
-0@@ -0,0 +1,139 @@ +#include <libcflat.h> +#include <asm/asm-offsets.h> +#include <asm-generic/barrier.h> +#include <asm/interrupt.h> +#include <asm/pgtable.h> +#include <mmu.h> +#include <asm/page.h> +#include <asm/facility.h> +#include <asm/mem.h> +#include <asm/sigp.h> +#include <smp.h> +#include <alloc_page.h> +#include <bitops.h> +#include <vm.h> +#include <sclp.h> +#include <sie.h> + +static u8 *guest; +static u8 *guest_instr; +static struct vm vm; +s390x/mvpg-sie.c line 31
Quoted 10 lines
s390x/mvpg-sie.c new file +10
-0+static uint8_t *src; +static uint8_t *dst; + +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_start[]; +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_end[]; +int binary_size; + +static void handle_validity (struct vm *vm) +{ +report (0 ,"VALIDITY: %x" , vm->sblk->ipb >>16 );I think an assert would be better. This should not happen, and if it happens something went very wrong and we have no guarantee that we will be able to continue
Quoted 22 lines
s390x/mvpg-sie.c new file +19
-0+} + +static void sie (struct vm *vm) +{ +/* Reset icptcode so we don't trip below */ + vm->sblk->icptcode =0 ; + +while (vm->sblk->icptcode ==0 ) { +sie64a (vm->sblk, &vm->save_area); +if (vm->sblk->icptcode ==ICPT_VALIDITY ) +handle_validity (vm); + } + vm->save_area.guest.grs[14 ] = vm->sblk->gg14; + vm->save_area.guest.grs[15 ] = vm->sblk->gg15; +} + +static void test_mvpg_pei (void ) +{ + uint64_t **pei_dst = (uint64_t **)((uintptr_t) vm.sblk +
0xc0);s390x/mvpg-sie.c new file +1
-0+ uint64_t **pei_src = (uint64_t **)((uintptr_t) vm.sblk +
0xc8); +s390x/mvpg-sie.c line 53
Quoted 1 line
s390x/mvpg-sie.c new file +1
-0+report_prefix_push ("pei" );maybe clear the destination buffer...
Quoted 8 lines
s390x/mvpg-sie.c new file +3
-0+protect_page (guest +0x6000 ,PAGE_ENTRY_I ); +sie (&vm); + report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partialexecution");
s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_src == ((uintptr_t)vm.sblk->mso) +
0x6000+PAGE_ENTRY_I, "PEI_SRCcorrect");s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000,"
PEI_DSTcorrect");... and check that the page was not copied
Quoted 8 lines
s390x/mvpg-sie.c
Quoted 2 lines
s390x/mvpg-sie.c new file +2
-0+/* Jump over the diag44 */ +sie (&vm);I would check if you really got a diag44
s390x/mvpg-sie.c line 64
Quoted 3 lines
s390x/mvpg-sie.c new file +3
-0+/* Clear PEI data for next check */ +memset ((uint64_t *)((uintptr_t ) vm.sblk +0xc0 ),0 ,16 ); +unprotect_page (guest +0x6000 ,PAGE_ENTRY_I );
I'll also move these over to src/dst so we only use the magic constants once.
Quoted 136 lines
Quoted 19 lines
s390x/mvpg-sie.c new file +3
-0+protect_page (guest +0x5000 ,PAGE_ENTRY_I ); +sie (&vm); + report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partialexecution");
s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_src == vm.sblk->mso + 0x6000,"
PEI_SRCcorrect");s390x/mvpg-sie.c new file +1
-0+ report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000 +
PAGE_ENTRY_I, "PEI_DSTcorrect"); +s390x/mvpg-sie.c new file +6
-0+report_prefix_pop (); +} + +static void test_mvpg (void ) +{ + int binary_size =((uintptr_t)_binary_s390x_snippets_c_mvpg_snippet_gbin_end - + (uintptr_t)_binary_s390x_snippets_c_mvpg_snippet_gbin_start); +
s390x/mvpg-sie.c new file +1
-0+ memcpy(guest,
_binary_s390x_snippets_c_mvpg_snippet_gbin_start,binary_size);s390x/mvpg-sie.c
Quoted 5 lines
s390x/mvpg-sie.c new file +5
-0+memset (src,0x42 ,PAGE_SIZE ); +memset (dst,0x43 ,PAGE_SIZE ); +sie (&vm); +mb (); + report(!memcmp(src, dst, PAGE_SIZE) && *dst == 0x42, "Pageor maybe you can clear the destination buffer here, if you prefer
Quoted 108 lines
moved"); +}
s390x/mvpg-sie.c new file +16
-0+ +static void setup_guest (void ) +{ +setup_vm (); + +/* Allocate 1MB as guest memory */ + guest =alloc_pages (8 ); +/* The first two pages are the lowcore */ + guest_instr = guest +PAGE_SIZE *2 ; + + vm.sblk =alloc_page (); + + vm.sblk->cpuflags =CPUSTAT_ZARCH |CPUSTAT_RUNNING ; + vm.sblk->prefix =0 ; +/* +* Pageable guest with the same ASCE as the test programm, but
s390x/mvpg-sie.c new file +4
-0+* the guest memory 0x0 is offset to start at the allocated +* guest pages and end after 1MB. +* +* It's not pretty but faster and easier than managing guest ASCEs.
s390x/mvpg-sie.c new file +11
-0+*/ + vm.sblk->mso = (u64 )guest; + vm.sblk->msl = (u64 )guest; + vm.sblk->ihcpu =0xffff ; + + vm.sblk->crycbd = (uint64_t )alloc_page (); + + vm.sblk->gpsw.addr =PAGE_SIZE *4 ; + vm.sblk->gpsw.mask =0x0000000180000000ULL ; + vm.sblk->ictl =ICTL_OPEREXC |ICTL_PINT ; +/* Enable MVPG interpretation as we want to test KVM and not ourselves */
s390x/mvpg-sie.c new file +23
-0+ vm.sblk->eca =ECA_MVPGI ; + + src = guest +PAGE_SIZE *6 ; + dst = guest +PAGE_SIZE *5 ; +} + +int main (void ) +{ +report_prefix_push ("mvpg-sie" ); +if (!sclp_facilities.has_sief2) { +report_skip ("SIEF2 facility unavailable" ); + goto done; + } + +setup_guest (); +test_mvpg (); +test_mvpg_pei (); + +done: +report_prefix_pop (); +return report_summary (); + +}(quoted diff) +0
-0
b/s390x/snippets/c/mvpg-snippet.cnew file mode 100644 index 00000000..96b70c9cs390x/snippets/c/mvpg-snippet.c new file +33
-0@@ -0,0 +1,33 @@ +#include <libcflat.h> + +static inline void force_exit (void ) +{ + asmvolatile (" diag 0,0,0x44\n" ); +} + +static inline int mvpg (unsigned long r0,void *dest,void *src) +{ + register unsigned long reg0 asm ("0") = r0; +int cc; + + asmvolatile (" mvpg %1,%2\n" +" ipm %0\n" +" srl %0,28" + :"=&d" (cc) :"a" (dest),"a" (src),"d" (reg0) + :"memory" ,"cc" ); +return cc; +} + +static void test_mvpg_real (void ) +{ +mvpg (0 , (void *)0x5000 , (void *)0x6000 ); +force_exit (); +} + +__attribute__((section (".text" )))int main (void ) +{ +test_mvpg_real (); +test_mvpg_real (); +test_mvpg_real (); +return 0 ; +}s390x/unittests.cfg +3
-0@@ -103,3 +103,6 @@ file = sie.elf [mvpg] file = mvpg.elf timeout = 10 + +[mvpg-sie] +file = mvpg-sie.elf
Re: [kvm-unit-tests RFC 2/2] s390x: mvpg: Add SIE mvpg test
In reply to Janosch FrankQuoted 62 lines · Janosch Frank
Let's also check the PEI values to make sure our VSIE implementation is correct.
Signed-off-by Janosch Frank <[email protected]>--- s390x/Makefile | 3 +- s390x/mvpg-sie.c | 139 ++++++++++++++++++++++++++++++++ s390x/snippets/c/mvpg-snippet.c | 33 ++++++++ s390x/unittests.cfg | 3 + 4 files changed, 177 insertions(+), 1 deletion(-)create mode 100644
s390x/mvpg-sie.c
create mode 100644s390x/snippets/c/mvpg-snippet.cs390x/Makefile +2
-1@@ -22,6 +22,7 @@ tests += $(TEST_DIR)/uv-guest.elf tests += $(TEST_DIR)/sie.elf tests += $(TEST_DIR)/mvpg.elf tests += $(TEST_DIR)/uv-host.elf +tests += $(TEST_DIR)/mvpg-sie.elf tests_binary = $(patsubst %.elf,%.bin,$(tests)) ifneq ($(HOST_KEY_DOCUMENT),)@@ -79,7 +80,7 @@ FLATLIBS = $(libcflat) SNIPPET_DIR = $(TEST_DIR)/snippets report_abort # C snippets that need to be linked -snippets-c = +snippets-c = $(SNIPPET_DIR)/c/mvpg-snippet.gbin # ASM snippets that are directly compiled and converted to a *.gbin snippets-a =s390x/mvpg-sie.c renamed from dev/nullreport_abort +22
-0@@ -0,0 +1,139 @@ +#include <libcflat.h> +#include <asm/asm-offsets.h> +#include <asm-generic/barrier.h> +#include <asm/interrupt.h> +#include <asm/pgtable.h> +#include <mmu.h> +#include <asm/page.h> +#include <asm/facility.h> +#include <asm/mem.h> +#include <asm/sigp.h> +#include <smp.h> +#include <alloc_page.h> +#include <bitops.h> +#include <vm.h> +#include <sclp.h> +#include <sie.h> + +static u8 *guest; +static u8 *guest_instr; +static struct vm vm; + +static uint8_t *src;
s390x/mvpg-sie.c line 32
Quoted 10 lines
s390x/mvpg-sie.c renamed from dev/nullreport_abort +10
-0+static uint8_t *dst; + +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_start[]; +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_end[]; +int binary_size; + +static void handle_validity (struct vm *vm) +{ +report (0 ,"VALIDITY: %x" , vm->sblk->ipb >>16 ); +}
Maybe rather use report_abort() in that case? Or does it make sense to continue running the other tests afterwards?
Thomas
Re: [kvm-unit-tests RFC 2/2] s390x: mvpg: Add SIE mvpg test
Quoted 77 lines · Thomas Huth
Quoted 62 lines · Janosch Frank
Let's also check the PEI values to make sure our VSIE implementation is correct.
Signed-off-by Janosch Frank <[email protected]>--- s390x/Makefile | 3 +- s390x/mvpg-sie.c | 139 ++++++++++++++++++++++++++++++++ s390x/snippets/c/mvpg-snippet.c | 33 ++++++++ s390x/unittests.cfg | 3 + 4 files changed, 177 insertions(+), 1 deletion(-)create mode 100644
s390x/mvpg-sie.c
create mode 100644s390x/snippets/c/mvpg-snippet.cs390x/Makefile +2
-1@@ -22,6 +22,7 @@ tests += $(TEST_DIR)/uv-guest.elf tests += $(TEST_DIR)/sie.elf tests += $(TEST_DIR)/mvpg.elf tests += $(TEST_DIR)/uv-host.elf +tests += $(TEST_DIR)/mvpg-sie.elf tests_binary = $(patsubst %.elf,%.bin,$(tests)) ifneq ($(HOST_KEY_DOCUMENT),)@@ -79,7 +80,7 @@ FLATLIBS = $(libcflat) SNIPPET_DIR = $(TEST_DIR)/snippets report_abort # C snippets that need to be linked -snippets-c = +snippets-c = $(SNIPPET_DIR)/c/mvpg-snippet.gbin # ASM snippets that are directly compiled and converted to a *.gbin snippets-a =s390x/mvpg-sie.c renamed from dev/nullreport_abort +22
-0@@ -0,0 +1,139 @@ +#include <libcflat.h> +#include <asm/asm-offsets.h> +#include <asm-generic/barrier.h> +#include <asm/interrupt.h> +#include <asm/pgtable.h> +#include <mmu.h> +#include <asm/page.h> +#include <asm/facility.h> +#include <asm/mem.h> +#include <asm/sigp.h> +#include <smp.h> +#include <alloc_page.h> +#include <bitops.h> +#include <vm.h> +#include <sclp.h> +#include <sie.h> + +static u8 *guest; +static u8 *guest_instr; +static struct vm vm; + +static uint8_t *src;s390x/mvpg-sie.c line 32
Quoted 10 lines
s390x/mvpg-sie.c renamed from dev/nullreport_abort +10
-0+static uint8_t *dst; + +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_start[]; +extern const char _binary_s390x_snippets_c_mvpg_snippet_gbin_end[]; +int binary_size; + +static void handle_validity (struct vm *vm) +{ +report (0 ,"VALIDITY: %x" , vm->sblk->ipb >>16 ); +}Maybe rather use
report_abort()in that case? Or does it make sense to continue running the other tests afterwards?
Quoted 2 lines
Thomas
Hmm, right this is not yet in the SIE lib.
Quoted 8 lines · Janosch Frank
The SIE support allows us to run guests and test the hypervisor's (V)SIE implementation. However it requires that the guest instructions are binary which limits the complexity of the guest code.
The snippet support provides a way to write guest code as ASM or C and simply memcpy it into guest memory. Some of the KVM-unit-test library can be re-used which further speeds up guest code development.
Quoted 7 lines
The included mvpg-sie test helped us to deliver the KVM mvpg fixes which Claudio posted a short while ago. In the future I'll post Secure Execution snippet support patches which was my initial goal with this series anyway.
I heard you liked tests so I put tests inside tests so you can test while you test.
The idea sounds sane to me.
Signature · 3 lines
Thanks, David / dhildenb