Re: [kvm-unit-tests PATCH 1/1] s390x: stsi: Define vm_is_kvm to be used in different tests

Pierre Morel <pmorel@xxxxxxxxxxxxx> · Wed, 16 Feb 2022 13:26:21 +0100

On 2/16/22 09:13, Janosch Frank wrote:
On 2/15/22 18:30, Pierre Morel wrote:


On 2/15/22 16:21, Claudio Imbrenda wrote:
On Tue, 15 Feb 2022 16:08:16 +0100
Janosch Frank <frankja@xxxxxxxxxxxxx> wrote:

On 2/15/22 13:06, Claudio Imbrenda wrote:
On Tue, 15 Feb 2022 11:46:32 +0100
Pierre Morel <pmorel@xxxxxxxxxxxxx> wrote:
Several tests are in need of a way to check on which hypervisor
and virtualization level they are running on to be able to fence
certain tests. This patch adds functions that return true if a
vm is running under KVM, LPAR or generally as a level 2 guest.

To check if we're running under KVM we use the STSI 3.2.2
instruction, let's define it's response structure in a central
header.

sorry, I had replied to the old series, let me reply here too


I think it would look cleaner if there was only one
"detect_environment" function, that would call stsi once and detect 
the
environment, then the various vm_is_* would become something like

bool vm_is_*(void)
{
    return detect_environment() == VM_IS_*;
}

of course detect_environment would also cache the result with static
variables.

bonus, we could make that function public, so a testcase could just
switch over the type of hypervisor it's being run on, instead of 
having
to use a series of ifs.

and then maybe the various vm_is_* could become static inlines to 
be put
in the header.

please note that "detect_environment" is just the first thing that 
came
to my mind, I have no preference regarding the name.

I'd like to keep this patch as simple as possible because there are
multiple patch sets which are gated by it.

The vm.h code and the skey.c z/VM 6 check is a thorn in my side anyway
and I'd rather have it fixed properly which will likely result in a lot
of opinions being voiced.

So I'd propose to rename vm_is_vm() to vm_is_guest2() and pick this 
patch.

ok for me

I'll rename the function and queue the patch


Not OK for me, in the POP PTF do not do any difference between guest 2
and guest 3.

If we're running with HW virtualization then every guest >= 2 is a guest 
2 at the end. And most of the time we don't want to know the HW level 
anyway, we want to know who our hypervisor is and vm_is_vm() doesn't 
tell you one bit about that.

It tells us that we are running under a VM, the POP defines 1 as the 
basic machine, 2 as the LPAR and 3 as the Virtual Machine

I find this definition clear, much more clear than guest 2 that is why I 
used it.


At this point I would be happier if we remove the function and use 
stsi_get_fc() == 3 directly. There's no arguing about what we're 
checking when using that.

Speaking of function name, I do not understand the name of this function 
stsi_get_fc() : returning the function code ?


We're currently arguing about a function that's only used in this patch, 
no?

It is absolutely unimportant for me, if you prefer this we do this.
I send the changes.



Or do you want to implement a VSIE specificity ?



Signed-off-by: Pierre Morel <pmorel@xxxxxxxxxxxxx>
---
    lib/s390x/stsi.h | 32 ++++++++++++++++++++++++++++
    lib/s390x/vm.c   | 55 
++++++++++++++++++++++++++++++++++++++++++++++--
    lib/s390x/vm.h   |  3 +++
    s390x/stsi.c     | 23 ++------------------
    4 files changed, 90 insertions(+), 23 deletions(-)
    create mode 100644 lib/s390x/stsi.h

diff --git a/lib/s390x/stsi.h b/lib/s390x/stsi.h
new file mode 100644
index 00000000..bebc492d
--- /dev/null
+++ b/lib/s390x/stsi.h
@@ -0,0 +1,32 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Structures used to Store System Information
+ *
+ * Copyright IBM Corp. 2022
+ */
+
+#ifndef _S390X_STSI_H_
+#define _S390X_STSI_H_
+
+struct sysinfo_3_2_2 {
+    uint8_t reserved[31];
+    uint8_t count;
+    struct {
+        uint8_t reserved2[4];
+        uint16_t total_cpus;
+        uint16_t conf_cpus;
+        uint16_t standby_cpus;
+        uint16_t reserved_cpus;
+        uint8_t name[8];
+        uint32_t caf;
+        uint8_t cpi[16];
+        uint8_t reserved5[3];
+        uint8_t ext_name_encoding;
+        uint32_t reserved3;
+        uint8_t uuid[16];
+    } vm[8];
+    uint8_t reserved4[1504];
+    uint8_t ext_names[8][256];
+};
+
+#endif  /* _S390X_STSI_H_ */
diff --git a/lib/s390x/vm.c b/lib/s390x/vm.c
index a5b92863..91acd05b 100644
--- a/lib/s390x/vm.c
+++ b/lib/s390x/vm.c
@@ -12,6 +12,7 @@
    #include <alloc_page.h>
    #include <asm/arch_def.h>
    #include "vm.h"
+#include "stsi.h"
    /**
     * Detect whether we are running with TCG (instead of KVM)
@@ -26,9 +27,13 @@ bool vm_is_tcg(void)
        if (initialized)
            return is_tcg;
+    if (!vm_is_vm()) {
+        initialized = true;
+        return is_tcg;
+    }
+
        buf = alloc_page();
-    if (!buf)
-        return false;
+    assert(buf);
        if (stsi(buf, 1, 1, 1))
            goto out;
@@ -43,3 +48,49 @@ out:
        free_page(buf);
        return is_tcg;
    }
+
+/**
+ * Detect whether we are running with KVM
+ */
+bool vm_is_kvm(void)
+{
+    /* EBCDIC for "KVM/" */
+    const uint8_t kvm_ebcdic[] = { 0xd2, 0xe5, 0xd4, 0x61 };
+    static bool initialized;
+    static bool is_kvm;
+    struct sysinfo_3_2_2 *stsi_322;
+
+    if (initialized)
+        return is_kvm;
+
+    if (!vm_is_vm() || vm_is_tcg()) {
+        initialized = true;
+        return is_kvm;
+    }
+
+    stsi_322 = alloc_page();
+    assert(stsi_322);
+
+    if (stsi(stsi_322, 3, 2, 2))
+        goto out;
+
+    /*
+     * If the manufacturer string is "KVM/" in EBCDIC, then we
+     * are on KVM.
+     */
+    is_kvm = !memcmp(&stsi_322->vm[0].cpi, kvm_ebcdic, 
sizeof(kvm_ebcdic));
+    initialized = true;
+out:
+    free_page(stsi_322);
+    return is_kvm;
+}
+
+bool vm_is_lpar(void)
+{
+    return stsi_get_fc() == 2;
+}
+
+bool vm_is_vm(void)
+{
+    return stsi_get_fc() == 3;
+}
diff --git a/lib/s390x/vm.h b/lib/s390x/vm.h
index 7abba0cc..3aaf76af 100644
--- a/lib/s390x/vm.h
+++ b/lib/s390x/vm.h
@@ -9,5 +9,8 @@
    #define _S390X_VM_H_
    bool vm_is_tcg(void);
+bool vm_is_kvm(void);
+bool vm_is_vm(void);
+bool vm_is_lpar(void);
    #endif  /* _S390X_VM_H_ */
diff --git a/s390x/stsi.c b/s390x/stsi.c
index 391f8849..dccc53e7 100644
--- a/s390x/stsi.c
+++ b/s390x/stsi.c
@@ -13,27 +13,8 @@
    #include <asm/asm-offsets.h>
    #include <asm/interrupt.h>
    #include <smp.h>
+#include <stsi.h>
-struct stsi_322 {
-    uint8_t reserved[31];
-    uint8_t count;
-    struct {
-        uint8_t reserved2[4];
-        uint16_t total_cpus;
-        uint16_t conf_cpus;
-        uint16_t standby_cpus;
-        uint16_t reserved_cpus;
-        uint8_t name[8];
-        uint32_t caf;
-        uint8_t cpi[16];
-        uint8_t reserved5[3];
-        uint8_t ext_name_encoding;
-        uint32_t reserved3;
-        uint8_t uuid[16];
-    } vm[8];
-    uint8_t reserved4[1504];
-    uint8_t ext_names[8][256];
-};
    static uint8_t pagebuf[PAGE_SIZE * 2] 
__attribute__((aligned(PAGE_SIZE * 2)));
    static void test_specs(void)
@@ -91,7 +72,7 @@ static void test_3_2_2(void)
        /* EBCDIC for "KVM/" */
        const uint8_t cpi_kvm[] = { 0xd2, 0xe5, 0xd4, 0x61 };
        const char vm_name_ext[] = "kvm-unit-test";
-    struct stsi_322 *data = (void *)pagebuf;
+    struct sysinfo_3_2_2 *data = (void *)pagebuf;
        report_prefix_push("3.2.2");





--
Pierre Morel
IBM Lab Boeblingen






