Skip to contentSkip to conversation

s390x: Add snippet support

PATCH Superseded Janosch Frank 2 participants 9 messages Jun 24, 2021 – Jun 25, 2021

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

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.

RFC -> V1:
* Replaced handle_validity() with assert()
* Added clearing of registers before sigp in cstart.S
* Added a nice way to compile the snippets

Janosch Frank (2):
s390x: snippets: Add gitignore as well as linker script and start
assembly
s390x: mvpg: Add SIE mvpg test

Steffen Eiden (1):
s390x: snippets: Add snippet compilation

 .gitignore                      |   1 +
 s390x/Makefile                  |  29 ++++--
 s390x/mvpg-sie.c                | 150 ++++++++++++++++++++++++++++++++
 s390x/snippets/c/cstart.S       |  15 ++++
 s390x/snippets/c/flat.lds       |  51 +++++++++++
 s390x/snippets/c/mvpg-snippet.c |  33 +++++++
 s390x/unittests.cfg             |   3 +
 7 files changed, 276 insertions(+), 6 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
2.27.0
#1 lore raw

PATCH 1/3 s390x: snippets: Add gitignore as well as linker script and start assembly

Snippets are small guests That can be run under a unit test as the hypervisor. They can be written in C or assembly. The C code needs a linker script and a start assembly file that jumps to main to work properly. So let's add that as well as a gitignore entry for the new files.

---
 .gitignore                |  1 +
 s390x/snippets/c/cstart.S | 15 ++++++++++++
 s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++
 3 files changed, 67 insertions(+)
 create mode 100644 s390x/snippets/c/cstart.S
 create mode 100644 s390x/snippets/c/flat.lds
.gitignore +1 -0
index 8534fb7..b3cf2cb 100644
@@ -23,3 +23,4 @@ cscope.*
 /api/dirty-log
 /api/dirty-log-perf
 /s390x/*.bin
+/s390x/snippets/*/*.gbin
s390x/snippets/c/cstart.S new file +15 -0
new file mode 100644
index 0000000..d7f6525
@@ -0,0 +1,15 @@
+#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 - 160
+	brasl	%r14, main
+	/* For now let's only use cpu 0 in snippets so this will always work. */
+	xgr	%r0, %r0
+	sigp    %r2, %r0, SIGP_STOP
s390x/snippets/c/flat.lds new file +51 -0

51 lines not on this page · raw message

Signature · 1 line
2.27.0
#2 lore raw

Re: [PATCH 1/3] s390x: snippets: Add gitignore as well as linker script and start assembly

Quoted 51 lines · Janosch Frank

Snippets are small guests That can be run under a unit test as the hypervisor. They can be written in C or assembly. The C code needs a linker script and a start assembly file that jumps to main to work properly. So let's add that as well as a gitignore entry for the new files.

---
  .gitignore                |  1 +
  s390x/snippets/c/cstart.S | 15 ++++++++++++
  s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++
  3 files changed, 67 insertions(+)

create mode 100644 s390x/snippets/c/cstart.S
create mode 100644 s390x/snippets/c/flat.lds

.gitignore +1 -0
index 8534fb7..b3cf2cb 100644
@@ -23,3 +23,4 @@ cscope.*
  /api/dirty-log
  /api/dirty-log-perf
  /s390x/*.bin
+/s390x/snippets/*/*.gbin
s390x/snippets/c/cstart.S new file +15 -0
new file mode 100644
index 0000000..d7f6525
@@ -0,0 +1,15 @@
+#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 - 160
+	brasl	%r14, main
+	/* For now let's only use cpu 0 in snippets so this will always work. */
+	xgr	%r0, %r0
+	sigp    %r2, %r0, SIGP_STOP
s390x/snippets/c/flat.lds new file +0 -0
new file mode 100644
index 0000000..5e70732

s390x/snippets/c/flat.lds line 7

Quoted 7 lines
s390x/snippets/c/flat.lds new file +7 -0
@@ -0,0 +1,51 @@
+SECTIONS
+{
+	.lowcore : {
+		/*
+		 * Initial short psw for disk boot, with 31 bit addressing for
+		 * non z/Arch environment compatibility and the instruction
+		 * address 0x10000 (cstart64.S .init).

I think this comment needs some adjustments (0x10000 => 0x4000 and do not talk about cstart64.S)?

Also, what about switching to 64-bit mode in the snippets?

Thomas

Quoted 45 lines
s390x/snippets/c/flat.lds new file +44 -0
+		 */
+		. = 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);
+}
#3 lore raw

Re: [PATCH 1/3] s390x: snippets: Add gitignore as well as linker script and start assembly

Quoted 54 lines · Thomas Huth
Quoted 53 lines · Janosch Frank

Snippets are small guests That can be run under a unit test as the hypervisor. They can be written in C or assembly. The C code needs a linker script and a start assembly file that jumps to main to work properly. So let's add that as well as a gitignore entry for the new files.

---
  .gitignore                |  1 +
  s390x/snippets/c/cstart.S | 15 ++++++++++++
  s390x/snippets/c/flat.lds | 51 +++++++++++++++++++++++++++++++++++++++
  3 files changed, 67 insertions(+)

create mode 100644 s390x/snippets/c/cstart.S
create mode 100644 s390x/snippets/c/flat.lds

.gitignore +1 -0
index 8534fb7..b3cf2cb 100644
@@ -23,3 +23,4 @@ cscope.*
  /api/dirty-log
  /api/dirty-log-perf
  /s390x/*.bin
+/s390x/snippets/*/*.gbin
s390x/snippets/c/cstart.S new file +15 -0
new file mode 100644
index 0000000..d7f6525
@@ -0,0 +1,15 @@
+#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 - 160
+	brasl	%r14, main
+	/* For now let's only use cpu 0 in snippets so this will always work. */
+	xgr	%r0, %r0
+	sigp    %r2, %r0, SIGP_STOP
s390x/snippets/c/flat.lds new file +2 -0
new file mode 100644
index 0000000..5e70732
@@ -0,0 +1,51 @@
+SECTIONS
+{
Quoted 8 lines

s390x/snippets/c/flat.lds line 7

Quoted 5 lines
s390x/snippets/c/flat.lds new file +5 -0
+	.lowcore : {
+		/*
+		 * Initial short psw for disk boot, with 31 bit addressing for
+		 * non z/Arch environment compatibility and the instruction
+		 * address 0x10000 (cstart64.S .init).

I think this comment needs some adjustments (0x10000 => 0x4000 and do not talk about cstart64.S)?

Right, will do

Quoted 2 lines

Also, what about switching to 64-bit mode in the snippets?

I thought about that for some time a while ago.
It's not really necessary since the host test can also decide which guest PSW is used to start the snippet but it also doesn't hurt so I just added a sam64.

Quoted 50 lines

Thomas

Quoted 45 lines
s390x/snippets/c/flat.lds new file +44 -0
+		 */
+		. = 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);
+}
#4 lore raw

PATCH 2/3 s390x: snippets: Add snippet compilation

In reply to Janosch Frank

From: Steffen Eiden <[email protected]>

This patch provides the necessary make targets to properly compile snippets and link them into their associated host.

To define the guest-host definition, we use the make-feature SECONDEXPANSION in combination with Target specific Variable Values. The variable snippets has different values depending on the current target. This enables us to define which snippets (=guests) belong to which hosts. Furthermore, using the second-expansion, we can use snippets in the perquisites of the host's %.elf rule, which otherwise would be not set.

---
 s390x/Makefile | 27 +++++++++++++++++++++------
 1 file changed, 21 insertions(+), 6 deletions(-)
s390x/Makefile +21 -6
index 8820e99..ba32f4c 100644
@@ -76,11 +76,26 @@ OBJDIRS += lib/s390x
 asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.o
 
 FLATLIBS = $(libcflat)
-%.elf: %.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)
+
+SNIPPET_DIR = $(TEST_DIR)/snippets
+snippet_asmlib = $(SNIPPET_DIR)/c/cstart.o
+
+# perquisites (=guests) for the snippet hosts.
+# $(TEST_DIR)/<snippet-host>.elf: snippets = $(SNIPPET_DIR)/<c/asm>/<snippet>.gbin
+
+$(SNIPPET_DIR)/asm/%.gbin: $(SNIPPET_DIR)/asm/%.o $(FLATLIBS)
+	$(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@
+	$(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@
+
+$(SNIPPET_DIR)/c/%.gbin: $(SNIPPET_DIR)/c/%.o $(snippet_asmlib) $(FLATLIBS)
+	$(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds $(patsubst %.gbin,%.o,$@) $(snippet_asmlib) $(FLATLIBS)
+	$(OBJCOPY) -O binary $@ $@
+	$(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@
+
+.SECONDEXPANSION:
+%.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) $(snippets) $(@:.elf=.aux.o)
 	$(RM) $(@:.elf=.aux.o)
 	@chmod a-x $@
 
@@ -94,7 +109,7 @@ 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} $(TEST_DIR)/.*.d $(SNIPPET_DIR)/c/*.{o,gbin} $(SNIPPET_DIR)/c/.*.d lib/s390x/.*.d
 
 generated-files = $(asm-offsets)
 $(tests:.elf=.o) $(asmlib) $(cflatobjs): $(generated-files)
Signature · 1 line
2.27.0
Acked-by on v1 2/3
#5 lore raw

Re: [PATCH 2/3] s390x: snippets: Add snippet compilation

Quoted 41 lines · Janosch Frank

From: Steffen Eiden <[email protected]>

This patch provides the necessary make targets to properly compile snippets and link them into their associated host.

To define the guest-host definition, we use the make-feature SECONDEXPANSION in combination with Target specific Variable Values. The variable snippets has different values depending on the current target. This enables us to define which snippets (=guests) belong to which hosts. Furthermore, using the second-expansion, we can use snippets in the perquisites of the host's %.elf rule, which otherwise would be not set.

---
  s390x/Makefile | 27 +++++++++++++++++++++------
  1 file changed, 21 insertions(+), 6 deletions(-)
s390x/Makefile +10 -5
index 8820e99..ba32f4c 100644
@@ -76,11 +76,26 @@ OBJDIRS += lib/s390x
  asmlib = $(TEST_DIR)/cstart64.o $(TEST_DIR)/cpu.o
 
  FLATLIBS = $(libcflat)
-%.elf: %.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)
+
+SNIPPET_DIR = $(TEST_DIR)/snippets
+snippet_asmlib = $(SNIPPET_DIR)/c/cstart.o
+
+# perquisites (=guests) for the snippet hosts.
+# $(TEST_DIR)/<snippet-host>.elf: snippets = $(SNIPPET_DIR)/<c/asm>/<snippet>.gbin
+
+$(SNIPPET_DIR)/asm/%.gbin: $(SNIPPET_DIR)/asm/%.o $(FLATLIBS)
+	$(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@
+	$(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@

s390x/Makefile OBJDIRS += lib/s390x · line 98

Quoted 10 lines
s390x/Makefile +10 -0
+
+$(SNIPPET_DIR)/c/%.gbin: $(SNIPPET_DIR)/c/%.o $(snippet_asmlib) $(FLATLIBS)
+	$(CC) $(LDFLAGS) -o $@ -T $(SNIPPET_DIR)/c/flat.lds $(patsubst %.gbin,%.o,$@) $(snippet_asmlib) $(FLATLIBS)
+	$(OBJCOPY) -O binary $@ $@
+	$(OBJCOPY) -I binary -O elf64-s390 -B "s390:64-bit" $@ $@
+
+.SECONDEXPANSION:
+%.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) $(snippets) $(@:.elf=.aux.o)

Wow, weird Makefile magic, never seend that before ... but it sounds reasonable, so:

#6 lore raw

PATCH 3/3 s390x: mvpg: Add SIE mvpg test

In reply to Janosch Frank

Let's also check the PEI values to make sure our VSIE implementation is correct.

---
 s390x/Makefile                  |   2 +
 s390x/mvpg-sie.c                | 150 ++++++++++++++++++++++++++++++++
 s390x/snippets/c/mvpg-snippet.c |  33 +++++++
 s390x/unittests.cfg             |   3 +
 4 files changed, 188 insertions(+)
 create mode 100644 s390x/mvpg-sie.c
 create mode 100644 s390x/snippets/c/mvpg-snippet.c
s390x/Makefile +2 -0
index ba32f4c..07af26d 100644
@@ -23,6 +23,7 @@ tests += $(TEST_DIR)/sie.elf
 tests += $(TEST_DIR)/mvpg.elf
 tests += $(TEST_DIR)/uv-host.elf
 tests += $(TEST_DIR)/edat.elf
+tests += $(TEST_DIR)/mvpg-sie.elf
 
 tests_binary = $(patsubst %.elf,%.bin,$(tests))
 ifneq ($(HOST_KEY_DOCUMENT),)
@@ -82,6 +83,7 @@ snippet_asmlib = $(SNIPPET_DIR)/c/cstart.o
 
 # perquisites (=guests) for the snippet hosts.
 # $(TEST_DIR)/<snippet-host>.elf: snippets = $(SNIPPET_DIR)/<c/asm>/<snippet>.gbin
+$(TEST_DIR)/mvpg-sie.elf: snippets = $(SNIPPET_DIR)/c/mvpg-snippet.gbin
 
 $(SNIPPET_DIR)/asm/%.gbin: $(SNIPPET_DIR)/asm/%.o $(FLATLIBS)
 	$(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@
s390x/mvpg-sie.c new file +150 -0

150 lines not on this page · raw message

s390x/snippets/c/mvpg-snippet.c new file +33 -0
new file mode 100644
index 0000000..96b70c9
@@ -0,0 +1,33 @@
+#include <libcflat.h>
+
+static inline void force_exit(void)
+{
+	asm volatile("	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;
+
+	asm volatile("	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
index a0ec886..9e1802f 100644
@@ -106,3 +106,6 @@ timeout = 10
 
 [edat]
 file = edat.elf
+
+[mvpg-sie]
+file = mvpg-sie.elf
Signature · 1 line
2.27.0
#7 lore raw

Re: [PATCH 3/3] s390x: mvpg: Add SIE mvpg test

Quoted 45 lines · Janosch Frank

Let's also check the PEI values to make sure our VSIE implementation is correct.

---
  s390x/Makefile                  |   2 +
  s390x/mvpg-sie.c                | 150 ++++++++++++++++++++++++++++++++
  s390x/snippets/c/mvpg-snippet.c |  33 +++++++
  s390x/unittests.cfg             |   3 +
  4 files changed, 188 insertions(+)

create mode 100644 s390x/mvpg-sie.c
create mode 100644 s390x/snippets/c/mvpg-snippet.c

s390x/Makefile +2 -0
index ba32f4c..07af26d 100644
@@ -23,6 +23,7 @@ tests += $(TEST_DIR)/sie.elf
  tests += $(TEST_DIR)/mvpg.elf
  tests += $(TEST_DIR)/uv-host.elf
  tests += $(TEST_DIR)/edat.elf
+tests += $(TEST_DIR)/mvpg-sie.elf
 
  tests_binary = $(patsubst %.elf,%.bin,$(tests))
  ifneq ($(HOST_KEY_DOCUMENT),)
@@ -82,6 +83,7 @@ snippet_asmlib = $(SNIPPET_DIR)/c/cstart.o
 
  # perquisites (=guests) for the snippet hosts.
  # $(TEST_DIR)/<snippet-host>.elf: snippets = $(SNIPPET_DIR)/<c/asm>/<snippet>.gbin
+$(TEST_DIR)/mvpg-sie.elf: snippets = $(SNIPPET_DIR)/c/mvpg-snippet.gbin
 
  $(SNIPPET_DIR)/asm/%.gbin: $(SNIPPET_DIR)/asm/%.o $(FLATLIBS)
  	$(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@
s390x/mvpg-sie.c new file +6 -0
new file mode 100644
index 0000000..a18c1b0
@@ -0,0 +1,150 @@
+#include <libcflat.h>
+#include <asm/asm-offsets.h>
+#include <asm-generic/barrier.h>
+#include <asm/interrupt.h>
+#include <asm/pgtable.h>
+#include <mmu.h>

s390x/mvpg-sie.c line 16

Quoted 10 lines
s390x/mvpg-sie.c new file +10 -0
+#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>

The list of headers that get included here is rather long for this file that is rather short ... e.g. do we really need such headers like sigp.h and sclp.h here?

Quoted 11 lines
s390x/mvpg-sie.c new file +11 -0
+static u8 *guest;
+static u8 *guest_instr;
+static struct vm vm;
+
+static uint8_t *src;
+static uint8_t *dst;
+static uint8_t *cmp;
+
+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;

s390x/mvpg-sie.c line 38

Quoted 10 lines
s390x/mvpg-sie.c new file +10 -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)
+			assert(0);
                assert(vm->sblk->icptcode != ICPT_VALIDITY)

?

s390x/mvpg-sie.c line 47

Quoted 9 lines
s390x/mvpg-sie.c new file +9 -0
+	}
+	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);
+	uint64_t **pei_src = (uint64_t **)((uintptr_t) vm.sblk + 0xc8);

Hmm, magic values ... according to the public SA22-7095-0 from 1984, the array at offset 0xc0 is called "Interruption parameters" ... could we maybe at least use a similar name in our kvm_s390_sie_block?

Quoted 5 lines
s390x/mvpg-sie.c new file +5 -0
+	report_prefix_push("pei");
+
+	report_prefix_push("src");
+	memset(dst, 0, PAGE_SIZE);
+	protect_page(src, PAGE_ENTRY_I);

s390x/mvpg-sie.c line 63

Quoted 10 lines
s390x/mvpg-sie.c new file +10 -0
+	sie(&vm);
+	report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partial execution");
+	report((uintptr_t)**pei_src == ((uintptr_t)vm.sblk->mso) + 0x6000 + PAGE_ENTRY_I, "PEI_SRC correct");
+	report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000, "PEI_DST correct");
+	unprotect_page(src, PAGE_ENTRY_I);
+	report(!memcmp(cmp, dst, PAGE_SIZE), "Destination intact");
+	/* Jump over the diag44 */
+	sie(&vm);
+	assert(vm.sblk->icptcode == ICPT_INST &&
+	       vm.sblk->ipa == 0x8300 && vm.sblk->ipb == 0x440000);

Maybe add a comment about the magic values?

Quoted 99 lines
s390x/mvpg-sie.c new file +87 -0
+	report_prefix_pop();
+
+	/* Clear PEI data for next check */
+	report_prefix_push("dst");
+	memset((uint64_t *)((uintptr_t) vm.sblk + 0xc0), 0, 16);
+	memset(dst, 0, PAGE_SIZE);
+	protect_page(dst, PAGE_ENTRY_I);
+	sie(&vm);
+	report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partial execution");
+	report((uintptr_t)**pei_src == vm.sblk->mso + 0x6000, "PEI_SRC correct");
+	report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000 + PAGE_ENTRY_I, "PEI_DST correct");
+	/* Needed for the memcmp and general cleanup */
+	unprotect_page(dst, PAGE_ENTRY_I);
+	report(!memcmp(cmp, dst, PAGE_SIZE), "Destination intact");
+	report_prefix_pop();
+
+	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);
+
+	memcpy(guest, _binary_s390x_snippets_c_mvpg_snippet_gbin_start, binary_size);
+	memset(src, 0x42, PAGE_SIZE);
+	memset(dst, 0x43, PAGE_SIZE);
+	sie(&vm);
+	mb();
+	report(!memcmp(src, dst, PAGE_SIZE) && *dst == 0x42, "Page moved");
+}
+
+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
+	 * 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.
+	 */
+	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 */
+	vm.sblk->eca = ECA_MVPGI;
+
+	src = guest + PAGE_SIZE * 6;
+	dst = guest + PAGE_SIZE * 5;
+	cmp = alloc_page();
+	memset(cmp, 0, PAGE_SIZE);
+}
+
+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();
+
+}
s390x/snippets/c/mvpg-snippet.c new file +6 -0
new file mode 100644
index 0000000..96b70c9
@@ -0,0 +1,33 @@
+#include <libcflat.h>
+
+static inline void force_exit(void)
+{
+	asm volatile("	diag	0,0,0x44\n");
+}

s390x/snippets/c/mvpg-snippet.c line 16

Quoted 10 lines
s390x/snippets/c/mvpg-snippet.c new file +10 -0
+
+static inline int mvpg(unsigned long r0, void *dest, void *src)
+{
+	register unsigned long reg0 asm ("0") = r0;
+	int cc;
+
+	asm volatile("	mvpg    %1,%2\n"
+		     "	ipm     %0\n"
+		     "	srl     %0,28"
+		     : "=&d" (cc) : "a" (dest), "a" (src), "d" (reg0)

Why is cc marked with "&" here?

Thomas

#8 lore raw

Re: [PATCH 3/3] s390x: mvpg: Add SIE mvpg test

Quoted 52 lines · Thomas Huth
Quoted 51 lines · Janosch Frank

Let's also check the PEI values to make sure our VSIE implementation is correct.

---
  s390x/Makefile                  |   2 +
  s390x/mvpg-sie.c                | 150 ++++++++++++++++++++++++++++++++
  s390x/snippets/c/mvpg-snippet.c |  33 +++++++
  s390x/unittests.cfg             |   3 +
  4 files changed, 188 insertions(+)

create mode 100644 s390x/mvpg-sie.c
create mode 100644 s390x/snippets/c/mvpg-snippet.c

s390x/Makefile +2 -0
index ba32f4c..07af26d 100644
@@ -23,6 +23,7 @@ tests += $(TEST_DIR)/sie.elf
  tests += $(TEST_DIR)/mvpg.elf
  tests += $(TEST_DIR)/uv-host.elf
  tests += $(TEST_DIR)/edat.elf
+tests += $(TEST_DIR)/mvpg-sie.elf
 
  tests_binary = $(patsubst %.elf,%.bin,$(tests))
  ifneq ($(HOST_KEY_DOCUMENT),)
@@ -82,6 +83,7 @@ snippet_asmlib = $(SNIPPET_DIR)/c/cstart.o
 
  # perquisites (=guests) for the snippet hosts.
  # $(TEST_DIR)/<snippet-host>.elf: snippets = $(SNIPPET_DIR)/<c/asm>/<snippet>.gbin
+$(TEST_DIR)/mvpg-sie.elf: snippets = $(SNIPPET_DIR)/c/mvpg-snippet.gbin
 
  $(SNIPPET_DIR)/asm/%.gbin: $(SNIPPET_DIR)/asm/%.o $(FLATLIBS)
  	$(OBJCOPY) -O binary $(patsubst %.gbin,%.o,$@) $@
s390x/mvpg-sie.c new file +12 -0
new file mode 100644
index 0000000..a18c1b0
@@ -0,0 +1,150 @@
+#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>
Quoted 8 lines

s390x/mvpg-sie.c line 16

Quoted 4 lines
s390x/mvpg-sie.c new file +4 -0
+#include <bitops.h>
+#include <vm.h>
+#include <sclp.h>
+#include <sie.h>

The list of headers that get included here is rather long for this file that is rather short ... e.g. do we really need such headers like sigp.h and sclp.h here?

sclp.h is used for the sie facility check.
I've removed asm/sigp.h, smp.h and bitops.h

Quoted 28 lines
Quoted 12 lines
s390x/mvpg-sie.c new file +11 -0
+static u8 *guest;
+static u8 *guest_instr;
+static struct vm vm;
+
+static uint8_t *src;
+static uint8_t *dst;
+static uint8_t *cmp;
+
+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;

s390x/mvpg-sie.c line 38

Quoted 10 lines
s390x/mvpg-sie.c new file +10 -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)
+			assert(0);
                 assert(vm->sblk->icptcode != ICPT_VALIDITY)

?

Quoted 14 lines

s390x/mvpg-sie.c line 47

Quoted 9 lines
s390x/mvpg-sie.c new file +9 -0
+	}
+	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);
+	uint64_t **pei_src = (uint64_t **)((uintptr_t) vm.sblk + 0xc8);

Hmm, magic values ... according to the public SA22-7095-0 from 1984, the array at offset 0xc0 is called "Interruption parameters" ... could we maybe at least use a similar name in our
kvm_s390_sie_block?

That's very much a thing that moved over from KVM and the reason for that is that the interruption parameters contain a lot of things especially when PV is added into the mix.

We can now either choose to make the interruption parameters an interesing looking union or we define the offsets properly. I'm also not too happy about the VSIE PEI code using "u64 *pei_block = &vsie_page->scb_o->mcic;" for that.

#define SBLK_OFF_PEI_DST 0xc0
#define SBLK_OFF_PEI_SRC 0xc8

Quoted 13 lines

s390x/mvpg-sie.c line 60

Quoted 13 lines
s390x/mvpg-sie.c new file +12 -0
+	report_prefix_push("pei");
+
+	report_prefix_push("src");
+	memset(dst, 0, PAGE_SIZE);
+	protect_page(src, PAGE_ENTRY_I);
+	sie(&vm);
+	report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partial execution");
+	report((uintptr_t)**pei_src == ((uintptr_t)vm.sblk->mso) + 0x6000 + PAGE_ENTRY_I, "PEI_SRC correct");
+	report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000, "PEI_DST correct");
+	unprotect_page(src, PAGE_ENTRY_I);
+	report(!memcmp(cmp, dst, PAGE_SIZE), "Destination intact");
+	/* Jump over the diag44 */

How about:

/*
* We need to execute the diag44 which is used as a blocker behind the
* mvpg. It makes sure we fail the tests above if the mvpg wouldn't
* have intercepted.
*/

Quoted 1 line

s390x/mvpg-sie.c line 59

Quoted 1 line
s390x/mvpg-sie.c new file +1 -0
+	sie(&vm);

And:

/* Make sure we intercepted for the diag44 and noting else */

Quoted 4 lines

s390x/mvpg-sie.c line 63

Quoted 2 lines
s390x/mvpg-sie.c new file +2 -0
+	assert(vm.sblk->icptcode == ICPT_INST &&
+	       vm.sblk->ipa == 0x8300 && vm.sblk->ipb == 0x440000);

Maybe add a comment about the magic values?

Since these will be used extensively in the future I think it might make sense to introduce constants.

#define SIE_ICPT_IPA_DIAG 0x8300

Quoted 104 lines
Quoted 104 lines
s390x/mvpg-sie.c new file +87 -0
+	report_prefix_pop();
+
+	/* Clear PEI data for next check */
+	report_prefix_push("dst");
+	memset((uint64_t *)((uintptr_t) vm.sblk + 0xc0), 0, 16);
+	memset(dst, 0, PAGE_SIZE);
+	protect_page(dst, PAGE_ENTRY_I);
+	sie(&vm);
+	report(vm.sblk->icptcode == ICPT_PARTEXEC, "Partial execution");
+	report((uintptr_t)**pei_src == vm.sblk->mso + 0x6000, "PEI_SRC correct");
+	report((uintptr_t)**pei_dst == vm.sblk->mso + 0x5000 + PAGE_ENTRY_I, "PEI_DST correct");
+	/* Needed for the memcmp and general cleanup */
+	unprotect_page(dst, PAGE_ENTRY_I);
+	report(!memcmp(cmp, dst, PAGE_SIZE), "Destination intact");
+	report_prefix_pop();
+
+	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);
+
+	memcpy(guest, _binary_s390x_snippets_c_mvpg_snippet_gbin_start, binary_size);
+	memset(src, 0x42, PAGE_SIZE);
+	memset(dst, 0x43, PAGE_SIZE);
+	sie(&vm);
+	mb();
+	report(!memcmp(src, dst, PAGE_SIZE) && *dst == 0x42, "Page moved");
+}
+
+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
+	 * 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.
+	 */
+	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 */
+	vm.sblk->eca = ECA_MVPGI;
+
+	src = guest + PAGE_SIZE * 6;
+	dst = guest + PAGE_SIZE * 5;
+	cmp = alloc_page();
+	memset(cmp, 0, PAGE_SIZE);
+}
+
+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();
+
+}
s390x/snippets/c/mvpg-snippet.c new file +10 -0
new file mode 100644
index 0000000..96b70c9
@@ -0,0 +1,33 @@
+#include <libcflat.h>
+
+static inline void force_exit(void)
+{
+	asm volatile("	diag	0,0,0x44\n");
+}
+
+static inline int mvpg(unsigned long r0, void *dest, void *src)
+{
+	register unsigned long reg0 asm ("0") = r0;
Quoted 8 lines

s390x/snippets/c/mvpg-snippet.c line 16

Quoted 6 lines
s390x/snippets/c/mvpg-snippet.c new file +6 -0
+	int cc;
+
+	asm volatile("	mvpg    %1,%2\n"
+		     "	ipm     %0\n"
+		     "	srl     %0,28"
+		     : "=&d" (cc) : "a" (dest), "a" (src), "d" (reg0)

Why is cc marked with "&" here?

I copied that from s390x/mvpg.c

Quoted 3 lines

Thomas