Skip to contentSkip to conversation

s390x: Protected Virtualization support

PATCH Superseded Janosch Frank 7 participants 62 messages Feb 26, 2020 – Mar 31, 2020

v5 of 11 Superseded Latest is v11 →

lore
Share & tools
lore
#0 lore raw

v5 cover s390x: Protected Virtualization support

Most of the QEMU changes for PV are related to the new IPL type with subcodes 8 - 10 and the execution of the necessary Ultravisor calls to IPL secure guests. Note that we can only boot into secure mode from normal mode, i.e. stfle 161 is not active in secure mode.

The other changes related to data gathering for emulation and disabling addressing checks in secure mode, as well as CPU resets.

v5:
* Added balloon inhibition
* Moved docs into docs/system
* Some more enable/disable changes
* Moved enablement/disablement of pv in separate functions
* Some review fixes

v4:
* Sync with KVM changes
* Review changes

V3:
* Use dedicated functions to access SIDA
* Smaller cleanups and segfault fixes
* Error reporting for Ultravisor calls
* Inject of RC of diag308 subcode 10 fails

V2:
* Split out cleanups
* Internal PV state tracking
* Review feedback

Christian Borntraeger (1):
s390x: Add unpack facility feature to GA1

Janosch Frank (17):
s390x: Use constant for ESA PSW address
Sync pv
s390x: protvirt: Add diag308 subcodes 8 - 10
s390x: protvirt: Support unpack facility
s390x: protvirt: Add migration blocker
s390x: protvirt: Handle diag 308 subcodes 0,1,3,4
s390x: protvirt: Inhibit balloon when switching to protected mode
s390x: protvirt: KVM intercept changes
s390x: Add SIDA memory ops
s390x: protvirt: Move STSI data over SIDAD
s390x: protvirt: SCLP interpretation
s390x: protvirt: Set guest IPL PSW
s390x: protvirt: Move diag 308 data over SIDAD
s390x: protvirt: Disable address checks for PV guest IO emulation
s390x: protvirt: Move IO control structures over SIDA
s390x: protvirt: Handle SIGP store status correctly
docs: Add protvirt docs

 docs/system/index.rst               |   1 +
 docs/system/protvirt.rst            |  57 +++++++++++
 hw/s390x/Makefile.objs              |   1 +
 hw/s390x/ipl.c                      |  82 +++++++++++++++-
 hw/s390x/ipl.h                      |  33 +++++++
 hw/s390x/pv.c                       | 106 +++++++++++++++++++++
 hw/s390x/pv.h                       |  34 +++++++
 hw/s390x/s390-virtio-ccw.c          | 143 +++++++++++++++++++++++++++-
 hw/s390x/sclp.c                     |  17 ++++
 include/hw/s390x/s390-virtio-ccw.h  |   1 +
 include/hw/s390x/sclp.h             |   2 +
 linux-headers/linux/kvm.h           |  45 ++++++++-
 target/s390x/cpu.c                  |  27 ++++--
 target/s390x/cpu.h                  |   9 +-
 target/s390x/cpu_features_def.inc.h |   1 +
 target/s390x/diag.c                 |  61 ++++++++++--
 target/s390x/gen-features.c         |   1 +
 target/s390x/helper.c               |   4 +
 target/s390x/ioinst.c               | 113 +++++++++++++++-------
 target/s390x/kvm.c                  |  54 ++++++++++-
 target/s390x/kvm_s390x.h            |   2 +
 target/s390x/mmu_helper.c           |  14 +++
 target/s390x/sigp.c                 |   1 +
 23 files changed, 744 insertions(+), 65 deletions(-)
 create mode 100644 docs/system/protvirt.rst
 create mode 100644 hw/s390x/pv.c
 create mode 100644 hw/s390x/pv.h
Signature · 1 line
2.20.1
#1 lore raw

PATCH v5 1/18 s390x: Use constant for ESA PSW address

Lets make it a bit more clear that we're extracting the 31 bit address from the short psw.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +1 -0
index 8a557fd8d1..74e66fe0c2 100644
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL
 
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
Signature · 1 line
2.20.1
Reviewed-by on v5 1/18
#2 lore raw

Re: [PATCH v5 01/18] s390x: Use constant for ESA PSW address

Quoted 4 lines · Janosch Frank

Lets make it a bit more clear that we're extracting the 31 bit address from the short psw.

PSW_MASK_ESA_MASK and PSW_MASK_ESA_ADDR look pretty similar, but I cant find a good name. We could use ~PSW_MASK_ESA_ADDR as an alternative.

Either way, this makes sense.

Quoted 54 lines
---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
 
 
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);

}

target/s390x/cpu.h +1 -0
index 8a557fd8d1..74e66fe0c2 100644
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL
 
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
#3 lore raw

Re: [PATCH v5 01/18] s390x: Use constant for ESA PSW address

Quoted 6 lines · Christian Borntraeger
Quoted 3 lines · Janosch Frank

Lets make it a bit more clear that we're extracting the 31 bit address from the short psw.

Quoted 6 lines
Quoted 1 line

PSW_MASK_ESA_MASK and PSW_MASK_ESA_ADDR look pretty similar, but I cant find a good name. We could use ~PSW_MASK_ESA_ADDR as an alternative.

I'm also not too happy about it, if there's a better suggestion I'd take it any day.

Quoted 4 lines

Either way, this makes sense.

Thanks!

Quoted 55 lines
Quoted 55 lines
---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
 
 
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);

}

target/s390x/cpu.h +1 -0
index 8a557fd8d1..74e66fe0c2 100644
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL
 
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
#4 lore raw

Re: [PATCH v5 01/18] s390x: Use constant for ESA PSW address

In reply to Janosch Frank
Quoted 48 lines · Janosch Frank

Lets make it a bit more clear that we're extracting the 31 bit address from the short psw.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +0 -0
index 8a557fd8d1..74e66fe0c2 100644

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu; · line 280

Quoted 4 lines
target/s390x/cpu.h +1 -0
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL

..._MASK_..._MASK

Isn't there a better name for all the bits in the PSW that are not an address?

PSW_MASK_ESA_BITS
PSW_MASK_ESA_FLAGS
...

Quoted 4 lines
target/s390x/cpu.h +0 -0
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
 
Signature · 3 lines
Thanks,

David / dhildenb
#5 lore raw

Re: [PATCH v5 01/18] s390x: Use constant for ESA PSW address

Quoted 2 lines · On Wed, 26 Feb 2020 15:27:52 +0100 David Hildenbrand <[email protected]> wrote:
Quoted 1 line · Janosch Frank

Lets make it a bit more clear that we're extracting the 31 bit address

s/Lets/Let's/ :)

Quoted 54 lines
Quoted 47 lines

from the short psw.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +0 -0
index 8a557fd8d1..74e66fe0c2 100644

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu; · line 280

Quoted 4 lines
target/s390x/cpu.h +1 -0
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL  

..._MASK_..._MASK

Quoted 6 lines

Isn't there a better name for all the bits in the PSW that are not an address?

PSW_MASK_ESA_BITS
PSW_MASK_ESA_FLAGS
...

Hm, the PoP says that the PSW "includes the instruction address, condition code, and other control fields"; it also talks about the 'short' PSW as being distinct from the 'ESA' PSW (bit 31 may be 0 or 1 in the short PSW). Maybe

PSW_MASK_SHORT_ADDR
PSW_MASK_SHORT_CTRL

(Or keep _ESA_ if renaming creates too much churn.)

Quoted 7 lines

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu;

Quoted 5 lines
target/s390x/cpu.h +0 -0
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
 

This patch is also independent of the protected virtualization support... I plan to send a pull request tomorrow, so I can include this patch, if we agree on a name for the constant :)

#6 lore raw

Re: [PATCH v5 01/18] s390x: Use constant for ESA PSW address

Quoted 69 lines · Cornelia Huck
Quoted 2 lines · On Wed, 26 Feb 2020 15:27:52 +0100 David Hildenbrand <[email protected]> wrote:
Quoted 1 line · Janosch Frank

Lets make it a bit more clear that we're extracting the 31 bit address

s/Lets/Let's/ :)

Quoted 54 lines
Quoted 47 lines

from the short psw.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +0 -0
index 8a557fd8d1..74e66fe0c2 100644

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu; · line 280

Quoted 4 lines
target/s390x/cpu.h +1 -0
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL  

..._MASK_..._MASK

Quoted 6 lines

Isn't there a better name for all the bits in the PSW that are not an address?

PSW_MASK_ESA_BITS
PSW_MASK_ESA_FLAGS
...

Quoted 7 lines

Hm, the PoP says that the PSW "includes the instruction address, condition code, and other control fields"; it also talks about the 'short' PSW as being distinct from the 'ESA' PSW (bit 31 may be 0 or 1 in the short PSW). Maybe

PSW_MASK_SHORT_ADDR
PSW_MASK_SHORT_CTRL

+1

Signature · 3 lines
Thanks,

David / dhildenb
#7 lore raw

Re: [PATCH v5 01/18] s390x: Use constant for ESA PSW address

In reply to Cornelia Huck
Quoted 7 lines · Cornelia Huck
Quoted 2 lines · On Wed, 26 Feb 2020 15:27:52 +0100 David Hildenbrand <[email protected]> wrote:
Quoted 1 line · Janosch Frank

Lets make it a bit more clear that we're extracting the 31 bit address

s/Lets/Let's/ :)

Ack

Quoted 62 lines
Quoted 55 lines
Quoted 48 lines

from the short psw.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +0 -0
index 8a557fd8d1..74e66fe0c2 100644

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu; · line 280

Quoted 4 lines
target/s390x/cpu.h +1 -0
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL  

..._MASK_..._MASK

Quoted 6 lines

Isn't there a better name for all the bits in the PSW that are not an address?

PSW_MASK_ESA_BITS
PSW_MASK_ESA_FLAGS
...

Quoted 7 lines

Hm, the PoP says that the PSW "includes the instruction address, condition code, and other control fields"; it also talks about the 'short' PSW as being distinct from the 'ESA' PSW (bit 31 may be 0 or 1 in the short PSW). Maybe

PSW_MASK_SHORT_ADDR
PSW_MASK_SHORT_CTRL

Sure, why not

Quoted 11 lines

(Or keep _ESA_ if renaming creates too much churn.)

Quoted 7 lines

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu;

Quoted 5 lines
target/s390x/cpu.h +0 -0
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
 
Quoted 3 lines

This patch is also independent of the protected virtualization support... I plan to send a pull request tomorrow, so I can include this patch, if we agree on a name for the constant :)

Well, you would also need to rename all users of PSW_MASK_ESA_ADDR Let me split that up into two patches, the rename for the ADDR and this one. I'll send it out once I'm more or less awake.

#8 lore raw

Re: [PATCH v5 01/18] s390x: Use constant for ESA PSW address

Quoted 94 lines · Janosch Frank
Quoted 7 lines · Cornelia Huck
Quoted 2 lines · On Wed, 26 Feb 2020 15:27:52 +0100 David Hildenbrand <[email protected]> wrote:
Quoted 1 line · Janosch Frank

Lets make it a bit more clear that we're extracting the 31 bit address

s/Lets/Let's/ :)

Ack

Quoted 62 lines
Quoted 55 lines
Quoted 48 lines

from the short psw.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +0 -0
index 8a557fd8d1..74e66fe0c2 100644

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu; · line 280

Quoted 4 lines
target/s390x/cpu.h +1 -0
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL  

..._MASK_..._MASK

Quoted 6 lines

Isn't there a better name for all the bits in the PSW that are not an address?

PSW_MASK_ESA_BITS
PSW_MASK_ESA_FLAGS
...

Quoted 7 lines

Hm, the PoP says that the PSW "includes the instruction address, condition code, and other control fields"; it also talks about the 'short' PSW as being distinct from the 'ESA' PSW (bit 31 may be 0 or 1 in the short PSW). Maybe

PSW_MASK_SHORT_ADDR
PSW_MASK_SHORT_CTRL

Sure, why not

Quoted 11 lines

(Or keep _ESA_ if renaming creates too much churn.)

Quoted 7 lines
Quoted 5 lines
target/s390x/cpu.h +0 -0
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
 
Quoted 7 lines
Quoted 3 lines

This patch is also independent of the protected virtualization support... I plan to send a pull request tomorrow, so I can include this patch, if we agree on a name for the constant :)

Well, you would also need to rename all users of PSW_MASK_ESA_ADDR Let me split that up into two patches, the rename for the ADDR and this one. I'll send it out once I'm more or less awake.

Seems like the ADDR constant has never been used anyway... Ok, I renounce everything I said before, if you want to fix this up yourself that would be wonderful, if not I'd be happy to provide you with a patch.

#9 lore raw

Re: [PATCH v5 01/18] s390x: Use constant for ESA PSW address

Quoted 99 lines · On Thu, 27 Feb 2020 09:09:47 +0100 Janosch Frank <[email protected]> wrote:
Quoted 98 lines · Janosch Frank
Quoted 7 lines · Cornelia Huck
Quoted 2 lines · On Wed, 26 Feb 2020 15:27:52 +0100 David Hildenbrand <[email protected]> wrote:
Quoted 1 line · Janosch Frank

Lets make it a bit more clear that we're extracting the 31 bit address

s/Lets/Let's/ :)

Ack

Quoted 62 lines
Quoted 55 lines
Quoted 48 lines

from the short psw.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 1 +
 3 files changed, 4 insertions(+), 3 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..42e21e7a6a 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_ESA_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..43360912a0 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +0 -0
index 8a557fd8d1..74e66fe0c2 100644

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu; · line 280

Quoted 4 lines
target/s390x/cpu.h +1 -0
@@ -277,6 +277,7 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
 #define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_ESA_MASK       0xffffffff80000000ULL    

..._MASK_..._MASK

Quoted 6 lines

Isn't there a better name for all the bits in the PSW that are not an address?

PSW_MASK_ESA_BITS
PSW_MASK_ESA_FLAGS
...

Quoted 7 lines

Hm, the PoP says that the PSW "includes the instruction address, condition code, and other control fields"; it also talks about the 'short' PSW as being distinct from the 'ESA' PSW (bit 31 may be 0 or 1 in the short PSW). Maybe

PSW_MASK_SHORT_ADDR
PSW_MASK_SHORT_CTRL

Sure, why not

Quoted 14 lines

(Or keep _ESA_ if renaming creates too much churn.)

Quoted 7 lines

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu;

Quoted 5 lines
target/s390x/cpu.h +0 -0
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
 

This patch is also independent of the protected virtualization support... I plan to send a pull request tomorrow, so I can include this patch, if we agree on a name for the constant :)

Quoted 8 lines
Quoted 3 lines

Well, you would also need to rename all users of PSW_MASK_ESA_ADDR Let me split that up into two patches, the rename for the ADDR and this one. I'll send it out once I'm more or less awake.

Seems like the ADDR constant has never been used anyway... Ok, I renounce everything I said before, if you want to fix this up yourself that would be wonderful, if not I'd be happy to provide you with a patch.

A quick respin of this patch would be easiest for me.

#10 lore raw

PATCH v6 1/1 s390x: Rename and use constants for short PSW address and mask

Let's rename PSW_MASK_ESA_ADDR to PSW_MASK_SHORT_ADDR because we're not working with a ESA PSW which would not support the extended addressing bit. Also let's actually use it.

Additionally we introduce PSW_MASK_SHORT_CTRL and use it throughout the codebase.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 3 ++-
 3 files changed, 5 insertions(+), 4 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..e36f770a72 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_SHORT_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..3dd396e870 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_SHORT_CTRL;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_SHORT_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +2 -1
index 8a557fd8d1..1d17709d6e 100644
@@ -276,7 +276,8 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_RI             0x0000008000000000ULL
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
-#define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_SHORT_ADDR     0x000000007fffffffULL
+#define PSW_MASK_SHORT_CTRL     0xffffffff80000000ULL
 
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
Signature · 1 line
2.20.1
Reviewed-by on v6 1/1
#11 lore raw

Re: [PATCH v6] s390x: Rename and use constants for short PSW address and mask

Quoted 54 lines · Janosch Frank

Let's rename PSW_MASK_ESA_ADDR to PSW_MASK_SHORT_ADDR because we're not working with a ESA PSW which would not support the extended addressing bit. Also let's actually use it.

Additionally we introduce PSW_MASK_SHORT_CTRL and use it throughout the codebase.

---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 3 ++-
 3 files changed, 5 insertions(+), 4 deletions(-)
hw/s390x/ipl.c +1 -1
index 7773499d7f..e36f770a72 100644
@@ -179,7 +179,7 @@ static void s390_ipl_realize(DeviceState *dev, Error **errp)
                 /* if not Linux load the address of the (short) IPL PSW */
                 ipl_psw = rom_ptr(4, 4);
                 if (ipl_psw) {
-                    pentry = be32_to_cpu(*ipl_psw) & 0x7fffffffUL;
+                    pentry = be32_to_cpu(*ipl_psw) & PSW_MASK_SHORT_ADDR;
                 } else {
                     error_setg(&err, "Could not get IPL PSW");
                     goto error;
target/s390x/cpu.c +2 -2
index 8da1905485..3dd396e870 100644
@@ -78,13 +78,13 @@ static void s390_cpu_load_normal(CPUState *s)
     S390CPU *cpu = S390_CPU(s);
     uint64_t spsw = ldq_phys(s->as, 0);
 
-    cpu->env.psw.mask = spsw & 0xffffffff80000000ULL;
+    cpu->env.psw.mask = spsw & PSW_MASK_SHORT_CTRL;
     /*
      * Invert short psw indication, so SIE will report a specification
      * exception if it was not set.
      */
     cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & 0x7fffffffULL;
+    cpu->env.psw.addr = spsw & PSW_MASK_SHORT_ADDR;
 
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
target/s390x/cpu.h +0 -0
index 8a557fd8d1..1d17709d6e 100644

target/s390x/cpu.h extern const VMStateDescription vmstate_s390_cpu; · line 283

Quoted 9 lines
target/s390x/cpu.h +2 -1
@@ -276,7 +276,8 @@ extern const VMStateDescription vmstate_s390_cpu;
 #define PSW_MASK_RI             0x0000008000000000ULL
 #define PSW_MASK_64             0x0000000100000000ULL
 #define PSW_MASK_32             0x0000000080000000ULL
-#define PSW_MASK_ESA_ADDR       0x000000007fffffffULL
+#define PSW_MASK_SHORT_ADDR     0x000000007fffffffULL
+#define PSW_MASK_SHORT_CTRL     0xffffffff80000000ULL
 
 #undef PSW_ASC_PRIMARY
 #undef PSW_ASC_ACCREG
Signature · 3 lines
Thanks,

David / dhildenb
#12 lore raw

Re: [PATCH v6] s390x: Rename and use constants for short PSW address and mask

In reply to Janosch Frank
Quoted 7 lines · On Thu, 27 Feb 2020 04:23:41 -0500 Janosch Frank <[email protected]> wrote:

Let's rename PSW_MASK_ESA_ADDR to PSW_MASK_SHORT_ADDR because we're not working with a ESA PSW which would not support the extended addressing bit. Also let's actually use it.

Additionally we introduce PSW_MASK_SHORT_CTRL and use it throughout the codebase.

Quoted 7 lines
---
 hw/s390x/ipl.c     | 2 +-
 target/s390x/cpu.c | 4 ++--
 target/s390x/cpu.h | 3 ++-
 3 files changed, 5 insertions(+), 4 deletions(-)

Thanks, applied.

#13 lore raw

PATCH v5 2/18 Sync pv

In reply to Janosch Frank
---
 linux-headers/linux/kvm.h | 43 +++++++++++++++++++++++++++++++++++++--
 1 file changed, 41 insertions(+), 2 deletions(-)
linux-headers/linux/kvm.h +41 -2
index 265099100e..e36f761194 100644
@@ -474,8 +474,11 @@ struct kvm_s390_mem_op {
 	__u32 size;		/* amount of bytes */
 	__u32 op;		/* type of operation */
 	__u64 buf;		/* buffer in userspace */
-	__u8 ar;		/* the access register number */
-	__u8 reserved[31];	/* should be set to 0 */
+	union {
+		__u8 ar;	/* the access register number */
+		__u32 sida_offset; /* offset into the sida */
+		__u8 reserved[32]; /* should be set to 0 */
+	};
 };
 /* types for kvm_s390_mem_op->op */
 #define KVM_S390_MEMOP_LOGICAL_READ	0
@@ -1010,6 +1013,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_ARM_NISV_TO_USER 177
 #define KVM_CAP_ARM_INJECT_EXT_DABT 178
 #define KVM_CAP_S390_VCPU_RESETS 179
+#define KVM_CAP_S390_PROTECTED 180
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
@@ -1478,6 +1482,41 @@ struct kvm_enc_region {
 #define KVM_S390_NORMAL_RESET	_IO(KVMIO,   0xc3)
 #define KVM_S390_CLEAR_RESET	_IO(KVMIO,   0xc4)
 
+struct kvm_s390_pv_sec_parm {
+	__u64 origin;
+	__u64 length;
+};
+
+struct kvm_s390_pv_unp {
+	__u64 addr;
+	__u64 size;
+	__u64 tweak;
+};
+
+enum pv_cmd_id {
+	KVM_PV_ENABLE,
+	KVM_PV_DISABLE,
+	KVM_PV_VM_SET_SEC_PARMS,
+	KVM_PV_VM_UNPACK,
+	KVM_PV_VM_VERIFY,
+	KVM_PV_VM_PREP_RESET,
+	KVM_PV_VM_UNSHARE_ALL,
+	KVM_PV_VCPU_CREATE,
+	KVM_PV_VCPU_DESTROY,
+};
+
+struct kvm_pv_cmd {
+	__u32 cmd;	/* Command to be executed */
+	__u16 rc;	/* Ultravisor return code */
+	__u16 rrc;	/* Ultravisor return reason code */
+	__u64 data;	/* Data or address */
+	__u32 flags;    /* flags for future extensions. Must be 0 for now */
+	__u32 reserved[3];
+};
+
+/* Available with KVM_CAP_S390_PROTECTED */
+#define KVM_S390_PV_COMMAND		_IOWR(KVMIO, 0xc5, struct kvm_pv_cmd)
+
 /* Secure Encrypted Virtualization command */
 enum sev_cmd_id {
 	/* Guest initialization commands */
Signature · 1 line
2.20.1
#14 lore raw

PATCH v5 3/18 s390x: protvirt: Add diag308 subcodes 8 - 10

In reply to Janosch Frank

For diag308 subcodes 8 - 10 we have a new ipib of type 5. The ipib holds the address and length of the secure execution header, as well as a list of guest components.

Each component is a block of memory, for example kernel or initrd, which needs to be decrypted by the Ultravisor in order to run a protected VM. The secure execution header instructs the Ultravisor on how to handle the protected VM and its components.

Subcodes 8 and 9 are similiar to 5 and 6 and subcode 10 will finally start the protected guest.

Subcodes 8-10 are not valid in protected mode, we have to do a subcode 3 and then the 8 and 10 combination for a protected reboot.

---
 hw/s390x/ipl.c      | 47 ++++++++++++++++++++++++++++++++++++++++++---
 hw/s390x/ipl.h      | 31 ++++++++++++++++++++++++++++++
 target/s390x/diag.c | 27 +++++++++++++++++++++++---
 3 files changed, 99 insertions(+), 6 deletions(-)
hw/s390x/ipl.c +44 -3
index 42e21e7a6a..a017b3edda 100644
@@ -538,15 +538,55 @@ static bool is_virtio_scsi_device(IplParameterBlock *iplb)
     return is_virtio_ccw_device_of_type(iplb, VIRTIO_ID_SCSI);
 }
 
+int s390_ipl_pv_check_components(IplParameterBlock *iplb)
+{
+    int i;
+    IPLBlockPV *ipib_pv = &iplb->pv;
+
+    if (ipib_pv->num_comp == 0) {
+        return -EINVAL;
+    }
+
+    for (i = 0; i < ipib_pv->num_comp; i++) {
+        /* Addr must be 4k aligned */
+        if (ipib_pv->components[i].addr & ~TARGET_PAGE_MASK) {
+            return -EINVAL;
+        }
+
+        /* Tweak prefix is monotonously increasing with each component */
+        if (i < ipib_pv->num_comp - 1 &&
+            ipib_pv->components[i].tweak_pref >
+            ipib_pv->components[i + 1].tweak_pref) {
+            return -EINVAL;
+        }
+    }
+    return 0;
+}
+
 void s390_ipl_update_diag308(IplParameterBlock *iplb)
 {
     S390IPLState *ipl = get_ipl_device();
 
-    ipl->iplb = *iplb;
-    ipl->iplb_valid = true;
+    if (iplb->pbt == S390_IPL_TYPE_PV) {
+        ipl->iplb_pv = *iplb;
+        ipl->iplb_valid_pv = true;
+    } else {
+        ipl->iplb = *iplb;
+        ipl->iplb_valid = true;
+    }
     ipl->netboot = is_virtio_net_device(iplb);
 }
 
+IplParameterBlock *s390_ipl_get_iplb_secure(void)
+{
+    S390IPLState *ipl = get_ipl_device();
+
+    if (!ipl->iplb_valid_pv) {
+        return NULL;
+    }
+    return &ipl->iplb_pv;
+}
+
 IplParameterBlock *s390_ipl_get_iplb(void)
 {
     S390IPLState *ipl = get_ipl_device();
@@ -561,7 +601,8 @@ void s390_ipl_reset_request(CPUState *cs, enum s390_reset reset_type)
 {
     S390IPLState *ipl = get_ipl_device();
 
-    if (reset_type == S390_RESET_EXTERNAL || reset_type == S390_RESET_REIPL) {
+    if (reset_type == S390_RESET_EXTERNAL || reset_type == S390_RESET_REIPL ||
+        reset_type == S390_RESET_PV) {
         /* use CPU 0 for full resets */
         ipl->reset_cpu_index = 0;
     } else {
hw/s390x/ipl.h +31 -0
index d4813105db..3c4bb66eda 100644
@@ -15,6 +15,23 @@
 #include "cpu.h"
 #include "hw/qdev-core.h"
 
+struct IPLBlockPVComp {
+    uint64_t tweak_pref;
+    uint64_t addr;
+    uint64_t size;
+} QEMU_PACKED;
+typedef struct IPLBlockPVComp IPLBlockPVComp;
+
+struct IPLBlockPV {
+    uint8_t  reserved[87];
+    uint8_t  version;
+    uint32_t num_comp;
+    uint64_t pv_header_addr;
+    uint64_t pv_header_len;
+    struct IPLBlockPVComp components[];
+} QEMU_PACKED;
+typedef struct IPLBlockPV IPLBlockPV;
+
 struct IplBlockCcw {
     uint8_t  reserved0[85];
     uint8_t  ssid;
@@ -71,6 +88,7 @@ union IplParameterBlock {
         union {
             IplBlockCcw ccw;
             IplBlockFcp fcp;
+            IPLBlockPV pv;
             IplBlockQemuScsi scsi;
         };
     } QEMU_PACKED;
@@ -84,9 +102,11 @@ union IplParameterBlock {
 typedef union IplParameterBlock IplParameterBlock;
 
 int s390_ipl_set_loadparm(uint8_t *loadparm);
+int s390_ipl_pv_check_components(IplParameterBlock *iplb);
 void s390_ipl_update_diag308(IplParameterBlock *iplb);
 void s390_ipl_prepare_cpu(S390CPU *cpu);
 IplParameterBlock *s390_ipl_get_iplb(void);
+IplParameterBlock *s390_ipl_get_iplb_secure(void);
 
 enum s390_reset {
     /* default is a reset not triggered by a CPU e.g. issued by QMP */
@@ -94,6 +114,7 @@ enum s390_reset {
     S390_RESET_REIPL,
     S390_RESET_MODIFIED_CLEAR,
     S390_RESET_LOAD_NORMAL,
+    S390_RESET_PV,
 };
 void s390_ipl_reset_request(CPUState *cs, enum s390_reset reset_type);
 void s390_ipl_get_reset_request(CPUState **cs, enum s390_reset *reset_type);
@@ -133,6 +154,7 @@ struct S390IPLState {
     /*< private >*/
     DeviceState parent_obj;
     IplParameterBlock iplb;
+    IplParameterBlock iplb_pv;
     QemuIplParameters qipl;
     uint64_t start_addr;
     uint64_t compat_start_addr;
@@ -140,6 +162,7 @@ struct S390IPLState {
     uint64_t compat_bios_start_addr;
     bool enforce_bios;
     bool iplb_valid;
+    bool iplb_valid_pv;
     bool netboot;
     /* reset related properties don't have to be migrated or reset */
     enum s390_reset reset_type;
@@ -161,9 +184,11 @@ QEMU_BUILD_BUG_MSG(offsetof(S390IPLState, iplb) & 3, "alignment of iplb wrong");
 
 #define S390_IPL_TYPE_FCP 0x00
 #define S390_IPL_TYPE_CCW 0x02
+#define S390_IPL_TYPE_PV 0x05
 #define S390_IPL_TYPE_QEMU_SCSI 0xff
 
 #define S390_IPLB_HEADER_LEN 8
+#define S390_IPLB_MIN_PV_LEN 148
 #define S390_IPLB_MIN_CCW_LEN 200
 #define S390_IPLB_MIN_FCP_LEN 384
 #define S390_IPLB_MIN_QEMU_SCSI_LEN 200
@@ -185,4 +210,10 @@ static inline bool iplb_valid_fcp(IplParameterBlock *iplb)
            iplb->pbt == S390_IPL_TYPE_FCP;
 }
 
+static inline bool iplb_valid_pv(IplParameterBlock *iplb)
+{
+    return be32_to_cpu(iplb->len) >= S390_IPLB_MIN_PV_LEN &&
+           iplb->pbt == S390_IPL_TYPE_PV;
+}
+
 #endif
target/s390x/diag.c +24 -3
index b5aec06d6b..d6ceb1f75d 100644
@@ -52,6 +52,8 @@ int handle_diag_288(CPUS390XState *env, uint64_t r1, uint64_t r3)
 #define DIAG_308_RC_OK              0x0001
 #define DIAG_308_RC_NO_CONF         0x0102
 #define DIAG_308_RC_INVALID         0x0402
+#define DIAG_308_RC_NO_PV_CONF      0x0a02
+#define DIAG_308_RC_INVAL_FOR_PV    0x0b02
 
 #define DIAG308_RESET_MOD_CLR       0
 #define DIAG308_RESET_LOAD_NORM     1
@@ -59,6 +61,9 @@ int handle_diag_288(CPUS390XState *env, uint64_t r1, uint64_t r3)
 #define DIAG308_LOAD_NORMAL_DUMP    4
 #define DIAG308_SET                 5
 #define DIAG308_STORE               6
+#define DIAG308_PV_SET              8
+#define DIAG308_PV_STORE            9
+#define DIAG308_PV_START            10
 
 static int diag308_parm_check(CPUS390XState *env, uint64_t r1, uint64_t addr,
                               uintptr_t ra, bool write)
@@ -105,6 +110,7 @@ void handle_diag_308(CPUS390XState *env, uint64_t r1, uint64_t r3, uintptr_t ra)
         s390_ipl_reset_request(cs, S390_RESET_REIPL);
         break;
     case DIAG308_SET:
+    case DIAG308_PV_SET:
         if (diag308_parm_check(env, r1, addr, ra, false)) {
             return;
         }
@@ -117,7 +123,8 @@ void handle_diag_308(CPUS390XState *env, uint64_t r1, uint64_t r3, uintptr_t ra)
 
         cpu_physical_memory_read(addr, iplb, be32_to_cpu(iplb->len));
 
-        if (!iplb_valid_ccw(iplb) && !iplb_valid_fcp(iplb)) {
+        if (!iplb_valid_ccw(iplb) && !iplb_valid_fcp(iplb) &&
+            !(iplb_valid_pv(iplb) && !s390_ipl_pv_check_components(iplb))) {
             env->regs[r1 + 1] = DIAG_308_RC_INVALID;
             goto out;
         }
@@ -128,17 +135,31 @@ out:
         g_free(iplb);
         return;
     case DIAG308_STORE:
+    case DIAG308_PV_STORE:
         if (diag308_parm_check(env, r1, addr, ra, true)) {
             return;
         }
-        iplb = s390_ipl_get_iplb();
+        if (subcode == DIAG308_PV_STORE) {
+            iplb = s390_ipl_get_iplb_secure();
+        } else {
+            iplb = s390_ipl_get_iplb();
+        }
         if (iplb) {
             cpu_physical_memory_write(addr, iplb, be32_to_cpu(iplb->len));
             env->regs[r1 + 1] = DIAG_308_RC_OK;
         } else {
             env->regs[r1 + 1] = DIAG_308_RC_NO_CONF;
         }
-        return;
+        break;
+    case DIAG308_PV_START:
+        iplb = s390_ipl_get_iplb_secure();
+        if (!iplb || !iplb_valid_pv(iplb)) {
+            env->regs[r1 + 1] = DIAG_308_RC_NO_PV_CONF;
+            return;
+        }
+
+        s390_ipl_reset_request(cs, S390_RESET_PV);
+        break;
     default:
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         break;
Signature · 1 line
2.20.1
#15 lore raw

PATCH v5 4/18 s390x: protvirt: Support unpack facility

In reply to Janosch Frank

When a guest has saved a ipib of type 5 and calls diagnose308 with subcode 10, we have to setup the protected processing environment via Ultravisor calls. The calls are done by KVM and are exposed via an API.

The following steps are necessary:
1. Enable protected mode for the VM (register it and its cpus with the Ultravisor)
2. Forward the secure header to the Ultravisor (has all information on how to decrypt the image and VM information)
3. Protect image pages from the host and decrypt them
4. Verify the image integrity

Only after step 4 a protected VM is allowed to run.

to machine]

---
 hw/s390x/Makefile.objs              |   1 +
 hw/s390x/ipl.c                      |  33 +++++++++
 hw/s390x/ipl.h                      |   2 +
 hw/s390x/pv.c                       | 106 ++++++++++++++++++++++++++++
 hw/s390x/pv.h                       |  34 +++++++++
 hw/s390x/s390-virtio-ccw.c          |  87 +++++++++++++++++++++++
 include/hw/s390x/s390-virtio-ccw.h  |   1 +
 target/s390x/cpu.c                  |   4 ++
 target/s390x/cpu.h                  |   1 +
 target/s390x/cpu_features_def.inc.h |   1 +
 10 files changed, 270 insertions(+)
 create mode 100644 hw/s390x/pv.c
 create mode 100644 hw/s390x/pv.h
hw/s390x/Makefile.objs +1 -0
index e02ed80b68..a46a1c7894 100644
@@ -31,6 +31,7 @@ obj-y += tod-qemu.o
 obj-$(CONFIG_KVM) += tod-kvm.o
 obj-$(CONFIG_KVM) += s390-skeys-kvm.o
 obj-$(CONFIG_KVM) += s390-stattrib-kvm.o
+obj-$(CONFIG_KVM) += pv.o
 obj-y += s390-ccw.o
 obj-y += ap-device.o
 obj-y += ap-bridge.o
hw/s390x/ipl.c +33 -0
index a017b3edda..53876637ee 100644
@@ -33,6 +33,7 @@
 #include "qemu/cutils.h"
 #include "qemu/option.h"
 #include "exec/exec-all.h"
+#include "pv.h"
 
 #define KERN_IMAGE_START                0x010000UL
 #define LINUX_MAGIC_ADDR                0x010008UL
@@ -676,6 +677,38 @@ static void s390_ipl_prepare_qipl(S390CPU *cpu)
     cpu_physical_memory_unmap(addr, len, 1, len);
 }
 
+int s390_ipl_prepare_pv_header(void)
+{
+    S390IPLState *ipl = get_ipl_device();
+    IPLBlockPV *ipib_pv = &ipl->iplb_pv.pv;
+    void *hdr = g_malloc(ipib_pv->pv_header_len);
+    int rc;
+
+    cpu_physical_memory_read(ipib_pv->pv_header_addr, hdr,
+                             ipib_pv->pv_header_len);
+    rc = s390_pv_set_sec_parms((uint64_t)hdr,
+                          ipib_pv->pv_header_len);
+    g_free(hdr);
+    return rc;
+}
+
+int s390_ipl_pv_unpack(void)
+{
+    int i, rc = 0;
+    S390IPLState *ipl = get_ipl_device();
+    IPLBlockPV *ipib_pv = &ipl->iplb_pv.pv;
+
+    for (i = 0; i < ipib_pv->num_comp; i++) {
+        rc = s390_pv_unpack(ipib_pv->components[i].addr,
+                            TARGET_PAGE_ALIGN(ipib_pv->components[i].size),
+                            ipib_pv->components[i].tweak_pref);
+        if (rc) {
+            break;
+        }
+    }
+    return rc;
+}
+
 void s390_ipl_prepare_cpu(S390CPU *cpu)
 {
     S390IPLState *ipl = get_ipl_device();
hw/s390x/ipl.h +2 -0
index 3c4bb66eda..92467738a7 100644
@@ -104,6 +104,8 @@ typedef union IplParameterBlock IplParameterBlock;
 int s390_ipl_set_loadparm(uint8_t *loadparm);
 int s390_ipl_pv_check_components(IplParameterBlock *iplb);
 void s390_ipl_update_diag308(IplParameterBlock *iplb);
+int s390_ipl_prepare_pv_header(void);
+int s390_ipl_pv_unpack(void);
 void s390_ipl_prepare_cpu(S390CPU *cpu);
 IplParameterBlock *s390_ipl_get_iplb(void);
 IplParameterBlock *s390_ipl_get_iplb_secure(void);
hw/s390x/pv.c new file +106 -0

106 lines not on this page · raw message

hw/s390x/pv.h new file +34 -0
new file mode 100644
index 0000000000..936262889f
@@ -0,0 +1,34 @@
+/*
+ * Protected Virtualization header
+ *
+ * Copyright IBM Corp. 2020
+ * Author(s):
+ *  Janosch Frank <[email protected]>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or (at
+ * your option) any later version. See the COPYING file in the top-level
+ * directory.
+ */
+
+#ifndef HW_S390_PV_H
+#define HW_S390_PV_H
+
+#ifdef CONFIG_KVM
+int s390_pv_vm_enable(void);
+void s390_pv_vm_disable(void);
+int s390_pv_set_sec_parms(uint64_t origin, uint64_t length);
+int s390_pv_unpack(uint64_t addr, uint64_t size, uint64_t tweak);
+void s390_pv_perf_clear_reset(void);
+int s390_pv_verify(void);
+void s390_pv_unshare(void);
+#else
+int s390_pv_vm_enable(void) { return 0; }
+void s390_pv_vm_disable(void) {}
+int s390_pv_set_sec_parms(uint64_t origin, uint64_t length) { return 0; }
+int s390_pv_unpack(uint64_t addr, uint64_t size, uint64_t tweak) { return 0; }
+void s390_pv_perf_clear_reset(void) {}
+int s390_pv_verify(void) { return 0; }
+void s390_pv_unshare(void) {}
+#endif
+
+#endif /* HW_S390_PV_H */
hw/s390x/s390-virtio-ccw.c +87 -0

124 lines not on this page · raw message

include/hw/s390x/s390-virtio-ccw.h +1 -0
index 8aa27199c9..cd1dccc6e3 100644
@@ -28,6 +28,7 @@ typedef struct S390CcwMachineState {
     /*< public >*/
     bool aes_key_wrap;
     bool dea_key_wrap;
+    bool pv;
     uint8_t loadparm[8];
 } S390CcwMachineState;
 
target/s390x/cpu.c +4 -0
index 43360912a0..c4d937e35a 100644
@@ -37,6 +37,8 @@
 #include "sysemu/hw_accel.h"
 #include "hw/qdev-properties.h"
 #ifndef CONFIG_USER_ONLY
+#include "hw/s390x/s390-virtio-ccw.h"
+#include "hw/s390x/pv.h"
 #include "hw/boards.h"
 #include "sysemu/arch_init.h"
 #include "sysemu/sysemu.h"
@@ -191,6 +193,7 @@ static void s390_cpu_realizefn(DeviceState *dev, Error **errp)
 
 #if !defined(CONFIG_USER_ONLY)
     MachineState *ms = MACHINE(qdev_get_machine());
+    S390CcwMachineState *ccw = S390_CCW_MACHINE(ms);
     unsigned int max_cpus = ms->smp.max_cpus;
     if (cpu->env.core_id >= max_cpus) {
         error_setg(&err, "Unable to add CPU with core-id: %" PRIu32
@@ -205,6 +208,7 @@ static void s390_cpu_realizefn(DeviceState *dev, Error **errp)
         goto out;
     }
 
+    cpu->env.pv = ccw->pv;
     /* sync cs->cpu_index and env->core_id. The latter is needed for TCG. */
     cs->cpu_index = cpu->env.core_id;
 #endif
target/s390x/cpu.h +1 -0
index 74e66fe0c2..acff5c9a65 100644
@@ -114,6 +114,7 @@ struct CPUS390XState {
 
     /* Fields up to this point are cleared by a CPU reset */
     struct {} end_reset_fields;
+    bool pv; /* protected virtualization */
 
 #if !defined(CONFIG_USER_ONLY)
     uint32_t core_id; /* PoP "CPU address", same as cpu_index */
target/s390x/cpu_features_def.inc.h +1 -0
index 31dff0d84e..60db28351d 100644
@@ -107,6 +107,7 @@ DEF_FEAT(DEFLATE_BASE, "deflate-base", STFL, 151, "Deflate-conversion facility (
 DEF_FEAT(VECTOR_PACKED_DECIMAL_ENH, "vxpdeh", STFL, 152, "Vector-Packed-Decimal-Enhancement Facility")
 DEF_FEAT(MSA_EXT_9, "msa9-base", STFL, 155, "Message-security-assist-extension-9 facility (excluding subfunctions)")
 DEF_FEAT(ETOKEN, "etoken", STFL, 156, "Etoken facility")
+DEF_FEAT(UNPACK, "unpack", STFL, 161, "Unpack facility")
 
 /* Features exposed via SCLP SCCB Byte 80 - 98  (bit numbers relative to byte-80) */
 DEF_FEAT(SIE_GSLS, "gsls", SCLP_CONF_CHAR, 40, "SIE: Guest-storage-limit-suppression facility")
Signature · 1 line
2.20.1
#16 lore raw

PATCH v5 5/18 s390x: protvirt: Add migration blocker

In reply to Janosch Frank

Migration is not yet supported.

---
 hw/s390x/s390-virtio-ccw.c | 28 ++++++++++++++++++++++------
 1 file changed, 22 insertions(+), 6 deletions(-)
hw/s390x/s390-virtio-ccw.c +22 -6
index e506dd65ed..79f472c309 100644
@@ -42,6 +42,9 @@
 #include "hw/s390x/tod.h"
 #include "sysemu/sysemu.h"
 #include "hw/s390x/pv.h"
+#include "migration/blocker.h"
+
+static Error *pv_mig_blocker;
 
 S390CPU *s390_cpu_addr2state(uint16_t cpu_addr)
 {
@@ -325,18 +328,30 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
 {
     CPUState *t;
 
-    s390_pv_vm_disable();
-    CPU_FOREACH(t) {
-        S390_CPU(t)->env.pv = false;
+    if (ms->pv) {
+        s390_pv_vm_disable();
+        CPU_FOREACH(t) {
+            S390_CPU(t)->env.pv = false;
+        }
+        ms->pv = false;
     }
-    ms->pv = false;
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
 {
+    static Error *local_err;
     CPUState *t;
     int rc;
 
+    if (!pv_mig_blocker) {
+        error_setg(&pv_mig_blocker,
+                   "protected VMs are currently not migrateable.");
+    }
+    migrate_add_blocker(pv_mig_blocker, &local_err);
+    if (local_err) {
+        goto out_err;
+    }
+
     /* Create SE VM */
     rc = s390_pv_vm_enable();
     if (rc) {
@@ -438,11 +453,12 @@ static void s390_machine_reset(MachineState *machine)
 
         if (s390_machine_protect(ms)) {
             s390_machine_inject_pv_error(cs);
-            s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
-            return;
+            goto pv_err;
         }
 
         run_on_cpu(cs, s390_do_cpu_load_normal, RUN_ON_CPU_NULL);
+pv_err:
+        s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
         break;
     default:
         g_assert_not_reached();
Signature · 1 line
2.20.1
#17 lore raw

Re: [PATCH v5 05/18] s390x: protvirt: Add migration blocker

Quoted 27 lines · Janosch Frank

Migration is not yet supported.

---
 hw/s390x/s390-virtio-ccw.c | 28 ++++++++++++++++++++++------
 1 file changed, 22 insertions(+), 6 deletions(-)
hw/s390x/s390-virtio-ccw.c +3 -2
index e506dd65ed..79f472c309 100644
@@ -42,6 +42,9 @@
 #include "hw/s390x/tod.h"
 #include "sysemu/sysemu.h"
 #include "hw/s390x/pv.h"
+#include "migration/blocker.h"
+
+static Error *pv_mig_blocker;
 
 S390CPU *s390_cpu_addr2state(uint16_t cpu_addr)
 {
@@ -325,18 +328,30 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
 {
     CPUState *t;
 
-    s390_pv_vm_disable();
-    CPU_FOREACH(t) {

hw/s390x/s390-virtio-ccw.c static void s390_machine_unprotect(S390CcwMachineState *ms) · line 338

Quoted 10 lines
hw/s390x/s390-virtio-ccw.c +6 -2
-        S390_CPU(t)->env.pv = false;
+    if (ms->pv) {
+        s390_pv_vm_disable();
+        CPU_FOREACH(t) {
+            S390_CPU(t)->env.pv = false;
+        }
+        ms->pv = false;
     }
-    ms->pv = false;
 }

Shouldnt that hunk go into the previous patch?

#18 lore raw

Re: [PATCH v5 05/18] s390x: protvirt: Add migration blocker

Quoted 28 lines · Christian Borntraeger
Quoted 25 lines · Janosch Frank

Migration is not yet supported.

---
 hw/s390x/s390-virtio-ccw.c | 28 ++++++++++++++++++++++------
 1 file changed, 22 insertions(+), 6 deletions(-)
hw/s390x/s390-virtio-ccw.c +3 -0
index e506dd65ed..79f472c309 100644
@@ -42,6 +42,9 @@
 #include "hw/s390x/tod.h"
 #include "sysemu/sysemu.h"
 #include "hw/s390x/pv.h"
+#include "migration/blocker.h"
+
+static Error *pv_mig_blocker;
 
 S390CPU *s390_cpu_addr2state(uint16_t cpu_addr)
 {
@@ -325,18 +328,30 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
 {
     CPUState *t;
 
Quoted 16 lines

hw/s390x/s390-virtio-ccw.c static void s390_machine_unprotect(S390CcwMachineState *ms) · line 338

Quoted 12 lines
hw/s390x/s390-virtio-ccw.c +6 -4
-    s390_pv_vm_disable();
-    CPU_FOREACH(t) {
-        S390_CPU(t)->env.pv = false;
+    if (ms->pv) {
+        s390_pv_vm_disable();
+        CPU_FOREACH(t) {
+            S390_CPU(t)->env.pv = false;
+        }
+        ms->pv = false;
     }
-    ms->pv = false;
 }

Shouldnt that hunk go into the previous patch?

Yes, something went wrong

#19 lore raw

PATCH v5 6/18 s390x: protvirt: Handle diag 308 subcodes 0,1,3,4

In reply to Janosch Frank

As we now have access to the protection state of the cpus, we can implement special handling of diag 308 subcodes for cpus in the protected state.

For subcodes 0 and 1 we need to unshare all pages before continuing, so the guest doesn't accidentally expose data when dumping.

For subcode 3/4 we tear down the protected VM and reboot into unprotected mode. We do not provide a secure reboot.

Before we can do the unshare calls, we need to mark all cpus as stopped.

---
 hw/s390x/s390-virtio-ccw.c | 37 ++++++++++++++++++++++++++++++++++---
 target/s390x/diag.c        |  4 ++++
 2 files changed, 38 insertions(+), 3 deletions(-)
hw/s390x/s390-virtio-ccw.c +34 -3
index 79f472c309..9983165b05 100644
@@ -335,6 +335,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         }
         ms->pv = false;
     }
+    migrate_del_blocker(pv_mig_blocker);
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
@@ -396,12 +397,27 @@ static void s390_machine_inject_pv_error(CPUState *cs)
     env->regs[r1 + 1] = 0xa02;
 }
 
+static void s390_pv_prepare_reset(CPUS390XState *env)
+{
+    CPUState *cs;
+
+    if (!env->pv) {
+        return;
+    }
+    CPU_FOREACH(cs) {
+        s390_cpu_set_state(S390_CPU_STATE_STOPPED, S390_CPU(cs));
+    }
+    s390_pv_unshare();
+    s390_pv_perf_clear_reset();
+}
+
 static void s390_machine_reset(MachineState *machine)
 {
     enum s390_reset reset_type;
     CPUState *cs, *t;
     S390CPU *cpu;
     S390CcwMachineState *ms = S390_CCW_MACHINE(machine);
+    CPUS390XState *env;
 
     /* get the reset parameters, reset them once done */
     s390_ipl_get_reset_request(&cs, &reset_type);
@@ -410,10 +426,15 @@ static void s390_machine_reset(MachineState *machine)
     s390_cmma_reset();
 
     cpu = S390_CPU(cs);
+    env = &cpu->env;
 
     switch (reset_type) {
     case S390_RESET_EXTERNAL:
     case S390_RESET_REIPL:
+        if (ms->pv) {
+            s390_machine_unprotect(ms);
+        }
+
         qemu_devices_reset();
         s390_crypto_reset();
 
@@ -421,21 +442,31 @@ static void s390_machine_reset(MachineState *machine)
         run_on_cpu(cs, s390_do_cpu_ipl, RUN_ON_CPU_NULL);
         break;
     case S390_RESET_MODIFIED_CLEAR:
+        /*
+         * Susbsystem reset needs to be done before we unshare memory
+         * and loose access to VIRTIO structures in guest memory.
+         */
+        subsystem_reset();
+        s390_crypto_reset();
+        s390_pv_prepare_reset(env);
         CPU_FOREACH(t) {
             run_on_cpu(t, s390_do_cpu_full_reset, RUN_ON_CPU_NULL);
         }
-        subsystem_reset();
-        s390_crypto_reset();
         run_on_cpu(cs, s390_do_cpu_load_normal, RUN_ON_CPU_NULL);
         break;
     case S390_RESET_LOAD_NORMAL:
+        /*
+         * Susbsystem reset needs to be done before we unshare memory
+         * and loose access to VIRTIO structures in guest memory.
+         */
+        subsystem_reset();
+        s390_pv_prepare_reset(env);
         CPU_FOREACH(t) {
             if (t == cs) {
                 continue;
             }
             run_on_cpu(t, s390_do_cpu_reset, RUN_ON_CPU_NULL);
         }
-        subsystem_reset();
         run_on_cpu(cs, s390_do_cpu_initial_reset, RUN_ON_CPU_NULL);
         run_on_cpu(cs, s390_do_cpu_load_normal, RUN_ON_CPU_NULL);
         break;
target/s390x/diag.c +4 -0
index d6ceb1f75d..840335d40a 100644
@@ -68,6 +68,10 @@ int handle_diag_288(CPUS390XState *env, uint64_t r1, uint64_t r3)
 static int diag308_parm_check(CPUS390XState *env, uint64_t r1, uint64_t addr,
                               uintptr_t ra, bool write)
 {
+    /* Handled by the Ultravisor */
+    if (env->pv) {
+        return 0;
+    }
     if ((r1 & 1) || (addr & ~TARGET_PAGE_MASK)) {
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         return -1;
Signature · 1 line
2.20.1
#20 lore raw

Re: [PATCH v5 06/18] s390x: protvirt: Handle diag 308 subcodes 0,1,3,4

Quoted 24 lines · Janosch Frank

As we now have access to the protection state of the cpus, we can implement special handling of diag 308 subcodes for cpus in the protected state.

For subcodes 0 and 1 we need to unshare all pages before continuing, so the guest doesn't accidentally expose data when dumping.

For subcode 3/4 we tear down the protected VM and reboot into unprotected mode. We do not provide a secure reboot.

Before we can do the unshare calls, we need to mark all cpus as stopped.

---
 hw/s390x/s390-virtio-ccw.c | 37 ++++++++++++++++++++++++++++++++++---
 target/s390x/diag.c        |  4 ++++
 2 files changed, 38 insertions(+), 3 deletions(-)
hw/s390x/s390-virtio-ccw.c +0 -0
index 79f472c309..9983165b05 100644

hw/s390x/s390-virtio-ccw.c static void s390_machine_unprotect(S390CcwMachineState *ms) · line 340

Quoted 6 lines
hw/s390x/s390-virtio-ccw.c +1 -0
@@ -335,6 +335,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         }
         ms->pv = false;
     }
+    migrate_del_blocker(pv_mig_blocker);
 }
 

and that part into patch 5?

#21 lore raw

Re: [PATCH v5 06/18] s390x: protvirt: Handle diag 308 subcodes 0,1,3,4

Quoted 22 lines · Christian Borntraeger
Quoted 19 lines · Janosch Frank

As we now have access to the protection state of the cpus, we can implement special handling of diag 308 subcodes for cpus in the protected state.

For subcodes 0 and 1 we need to unshare all pages before continuing, so the guest doesn't accidentally expose data when dumping.

For subcode 3/4 we tear down the protected VM and reboot into unprotected mode. We do not provide a secure reboot.

Before we can do the unshare calls, we need to mark all cpus as stopped.

---
 hw/s390x/s390-virtio-ccw.c | 37 ++++++++++++++++++++++++++++++++++---
 target/s390x/diag.c        |  4 ++++
 2 files changed, 38 insertions(+), 3 deletions(-)
Quoted 15 lines

hw/s390x/s390-virtio-ccw.c static void s390_machine_unprotect(S390CcwMachineState *ms) · line 340

Quoted 11 lines
hw/s390x/s390-virtio-ccw.c +1 -0
index 79f472c309..9983165b05 100644
@@ -335,6 +335,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         }
         ms->pv = false;
     }
+    migrate_del_blocker(pv_mig_blocker);
 }
 

and that part into patch 5?

Already fixed in the branch :-)

#22 lore raw

PATCH v5 7/18 s390x: protvirt: Inhibit balloon when switching to protected mode

In reply to Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

---
 hw/s390x/s390-virtio-ccw.c | 3 +++
 1 file changed, 3 insertions(+)
hw/s390x/s390-virtio-ccw.c +3 -0
index 9983165b05..0f4455d1df 100644
@@ -41,6 +41,7 @@
 #include "hw/qdev-properties.h"
 #include "hw/s390x/tod.h"
 #include "sysemu/sysemu.h"
+#include "sysemu/balloon.h"
 #include "hw/s390x/pv.h"
 #include "migration/blocker.h"
 
@@ -336,6 +337,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         ms->pv = false;
     }
     migrate_del_blocker(pv_mig_blocker);
+    qemu_balloon_inhibit(false);
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
@@ -344,6 +346,7 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     CPUState *t;
     int rc;
 
+    qemu_balloon_inhibit(true);
     if (!pv_mig_blocker) {
         error_setg(&pv_mig_blocker,
                    "protected VMs are currently not migrateable.");
Signature · 1 line
2.20.1
#23 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Is that more like the UV will detect this as an integrity issue and crash the VM?

Quoted 35 lines
---
 hw/s390x/s390-virtio-ccw.c | 3 +++
 1 file changed, 3 insertions(+)
hw/s390x/s390-virtio-ccw.c +3 -0
index 9983165b05..0f4455d1df 100644
@@ -41,6 +41,7 @@
 #include "hw/qdev-properties.h"
 #include "hw/s390x/tod.h"
 #include "sysemu/sysemu.h"
+#include "sysemu/balloon.h"
 #include "hw/s390x/pv.h"
 #include "migration/blocker.h"
 
@@ -336,6 +337,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         ms->pv = false;
     }
     migrate_del_blocker(pv_mig_blocker);
+    qemu_balloon_inhibit(false);
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
@@ -344,6 +346,7 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     CPUState *t;
     int rc;
 
+    qemu_balloon_inhibit(true);
     if (!pv_mig_blocker) {
         error_setg(&pv_mig_blocker,
                    "protected VMs are currently not migrateable.");
Signature · 3 lines
Thanks,

David / dhildenb
#24 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 6 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

Quoted 6 lines

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Is that more like the UV will detect this as an integrity issue and crash the VM?

yes, the UV will detect a fresh page as an integrity issue. Only if the page was defined to be shared by the guest, we would avoid the integrity check.

Reviewed-by
#25 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 13 lines · Christian Borntraeger
Quoted 10 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Quoted 7 lines
Quoted 2 lines

Is that more like the UV will detect this as an integrity issue and crash the VM?

yes, the UV will detect a fresh page as an integrity issue. Only if the page was defined to be shared by the guest, we would avoid the integrity check.

Please make that clearer in the patch description. With that

Signature · 3 lines
Thanks,

David / dhildenb
#26 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 17 lines · David Hildenbrand
Quoted 16 lines · Christian Borntraeger
Quoted 12 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Is that more like the UV will detect this as an integrity issue and crash the VM?

Quoted 9 lines
Quoted 4 lines

yes, the UV will detect a fresh page as an integrity issue. Only if the page was defined to be shared by the guest, we would avoid the integrity check.

Please make that clearer in the patch description. With that

How about:
s390x: protvirt: Inhibit balloon when switching to protected mode

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. If pages are not shared, the integrity checks will fail once those pages have been altered and are given back to the guest.

Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

#27 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 31 lines · Janosch Frank
Quoted 17 lines · David Hildenbrand
Quoted 16 lines · Christian Borntraeger
Quoted 12 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Is that more like the UV will detect this as an integrity issue and crash the VM?

Quoted 9 lines
Quoted 4 lines

yes, the UV will detect a fresh page as an integrity issue. Only if the page was defined to be shared by the guest, we would avoid the integrity check.

Please make that clearer in the patch description. With that

How about:
s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 9 lines

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. If pages are not shared, the integrity checks will fail once those pages have been altered and are given back to the guest.

Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

Yep, sounds good!

Signature · 3 lines
Thanks,

David / dhildenb
#28 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

In reply to Janosch Frank
Quoted 28 lines · On Wed, 26 Feb 2020 16:30:32 +0100 Janosch Frank <[email protected]> wrote:
Quoted 17 lines · David Hildenbrand
Quoted 16 lines · Christian Borntraeger
Quoted 12 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Is that more like the UV will detect this as an integrity issue and crash the VM?

Quoted 9 lines
Quoted 4 lines

yes, the UV will detect a fresh page as an integrity issue. Only if the page was defined to be shared by the guest, we would avoid the integrity check.

Please make that clearer in the patch description. With that

Quoted 7 lines

How about:
s390x: protvirt: Inhibit balloon when switching to protected mode

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. If pages are not shared, the integrity checks will fail once those pages have been altered and are given back to the guest.

This makes sense to me...

Quoted 4 lines

Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

...however, I'm scratching my head here.

If we have a future guest that knows how to handle this, how do we know? We know that the current Linux driver clears VIRTIO_F_IOMMU_PLATFORM during feature negotiation, and presumably a guest that knows how to handle this will not do that. But it still won't work as expected, as we inhibit ballooning...

So, either
- we don't inhibit ballooning now; any guest that clears the (required) virtio feature bit won't be able to drive the virtio-balloon device anyway, but a future guest that negotiates the bit would work; or
- we inhibit ballooning now; no guest can therefore use ballooning, regardless what they are doing or not doing (this includes guests that negotiate the feature bit, but fail to handle pages properly).

Or is there some other reason why we need to inhibit ballooning for protected vms?

#29 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 62 lines · Cornelia Huck
Quoted 28 lines · On Wed, 26 Feb 2020 16:30:32 +0100 Janosch Frank <[email protected]> wrote:
Quoted 17 lines · David Hildenbrand
Quoted 16 lines · Christian Borntraeger
Quoted 12 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Is that more like the UV will detect this as an integrity issue and crash the VM?

Quoted 9 lines
Quoted 4 lines

yes, the UV will detect a fresh page as an integrity issue. Only if the page was defined to be shared by the guest, we would avoid the integrity check.

Please make that clearer in the patch description. With that

Quoted 7 lines

How about:
s390x: protvirt: Inhibit balloon when switching to protected mode

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. If pages are not shared, the integrity checks will fail once those pages have been altered and are given back to the guest.

This makes sense to me...

Quoted 4 lines

Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

...however, I'm scratching my head here.

If we have a future guest that knows how to handle this, how do we know? We know that the current Linux driver clears VIRTIO_F_IOMMU_PLATFORM during feature negotiation, and presumably a guest that knows how to handle this will not do that. But it still won't work as expected, as we inhibit ballooning...

So, either
- we don't inhibit ballooning now; any guest that clears the (required) virtio feature bit won't be able to drive the virtio-balloon device anyway, but a future guest that negotiates the bit would work; or
- we inhibit ballooning now; no guest can therefore use ballooning, regardless what they are doing or not doing (this includes guests that negotiate the feature bit, but fail to handle pages properly).

Quoted 2 lines

Or is there some other reason why we need to inhibit ballooning for protected vms?

So here is my proposal.
1. we block ballooning now in QEMU (take this patch now)
2. Later Halil will provide a change to virtio that removes the blocker and adds VIRTIO_F_IOMMU_PLATFORM automatically by QEMU when doing the protvirt switch. This is ok as the guest balloon driver will reject to work with the IOMMU change (see
https://lore.kernel.org/qemu-devel/[email protected]/)
3. later we can fix the guest balloon driver to do the right thing for this case (e.g. do the make shared call)

#30 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

In reply to David Hildenbrand
Quoted 9 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Yes, as soon as the host alters non-shared memory we'll run into integrity issues.

I've been talking to Halil after I sent this out and it looks like we'll rather try to automatically enable the IOMMU for all devices when switching into protected mode. He said that if the IOMMU is set the balloon code will do an early exit on feature negotiation.

Quoted 41 lines

Is that more like the UV will detect this as an integrity issue and crash the VM?

Quoted 35 lines
---
 hw/s390x/s390-virtio-ccw.c | 3 +++
 1 file changed, 3 insertions(+)
hw/s390x/s390-virtio-ccw.c +3 -0
index 9983165b05..0f4455d1df 100644
@@ -41,6 +41,7 @@
 #include "hw/qdev-properties.h"
 #include "hw/s390x/tod.h"
 #include "sysemu/sysemu.h"
+#include "sysemu/balloon.h"
 #include "hw/s390x/pv.h"
 #include "migration/blocker.h"
 
@@ -336,6 +337,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         ms->pv = false;
     }
     migrate_del_blocker(pv_mig_blocker);
+    qemu_balloon_inhibit(false);
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
@@ -344,6 +346,7 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     CPUState *t;
     int rc;
 
+    qemu_balloon_inhibit(true);
     if (!pv_mig_blocker) {
         error_setg(&pv_mig_blocker,
                    "protected VMs are currently not migrateable.");
#31 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 11 lines · Janosch Frank
Quoted 9 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Quoted 8 lines

Yes, as soon as the host alters non-shared memory we'll run into integrity issues.

I've been talking to Halil after I sent this out and it looks like we'll rather try to automatically enable the IOMMU for all devices when switching into protected mode. He said that if the IOMMU is set the balloon code will do an early exit on feature negotiation.

I think we should fence the balloon here nevertheless, so the patch in itself is probably fine.

#32 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 18 lines · Christian Borntraeger
Quoted 15 lines · Janosch Frank
Quoted 9 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Yes, as soon as the host alters non-shared memory we'll run into integrity issues.

Quoted 7 lines
Quoted 4 lines

I've been talking to Halil after I sent this out and it looks like we'll rather try to automatically enable the IOMMU for all devices when switching into protected mode. He said that if the IOMMU is set the balloon code will do an early exit on feature negotiation.

I think we should fence the balloon here nevertheless, so the patch in itself is probably fine.

+1, this is a global "don't use ram_block_discard" trigger.

Signature · 3 lines
Thanks,

David / dhildenb
#33 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

In reply to Janosch Frank
Quoted 11 lines · On Wed, 26 Feb 2020 16:11:03 +0100 Janosch Frank <[email protected]> wrote:
Quoted 9 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Quoted 9 lines

Yes, as soon as the host alters non-shared memory we'll run into integrity issues.

I've been talking to Halil after I sent this out and it looks like we'll rather try to automatically enable the IOMMU for all devices when switching into protected mode. He said that if the IOMMU is set the balloon code will do an early exit on feature negotiation.

I have a discussion starter RFC for you.

--------------------------8<----------------------------------------------
From: Halil Pasic <[email protected]>
Date: Wed, 26 Feb 2020 16:48:21 +0100
Subject: [RFC PATCH 1/1] virtio-ccw: auto-manage VIRTIO_F_IOMMU_PLATFORM

The virtio specification tells that the device is to present VIRTIO_F_ACCESS_PLATFORM (a.k.a. VIRTIO_F_IOMMU_PLATFORM) when the device "can only access certain memory addresses with said access specified and/or granted by the platform". This is the case for a protected VM, as the device can access only memory addresses that are in pages that are currently shared (only the guest can share/unsare its page).

No VM however starts out as a protected VM, but some VMs may be converted to protected VMs if the guest decides so.

Making the end user explicitly manage the VIRTIO_F_ACCESS_PLATFORM via the property iommu_on is a minor disaster. If the correctness of the paravirtualized virtio devices depends (and thus in a sense the correctness of the hypervisor), then the hypervisor should have the last word about whether VIRTIO_F_ACCESS_PLATFORM is to be presented or not.

Let's manage the VIRTIO_F_ACCESS_PLATFORM virtio feature automatically for virtio-ccw devices, so that we set it before we start the protected configuration, and unset it when our VM is not protected.

---
NOTES:
* I wanted to have a discussion starter fast, there are multiple open questions.

* Doing more than one system_resets() is hackish. We should look into this.

* The user interface implications of this patch are also an ugly can of worms. Needs to be discussed.

* We should consider keeping the original value if !pv and restoring it on pv --> !pv, instead of forcing to unset when leaving pv, and actually not forcing unset when !pv.

* It might make sense to do something like this in virtio core, but I decided start the discussion with a ccw-only change.

* Maybe we need a machine option that enables this sort of behavior, especially if we decide not to do the conserving/restoring.

* Lightly tested.

---
 hw/s390x/s390-virtio-ccw.c |  4 ++--
 hw/s390x/virtio-ccw.c      | 13 +++++++++++++
 2 files changed, 15 insertions(+), 2 deletions(-)
hw/s390x/s390-virtio-ccw.c +2 -2
index 0f4455d1df..996124f152 100644
@@ -337,7 +337,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         ms->pv = false;
     }
     migrate_del_blocker(pv_mig_blocker);
-    qemu_balloon_inhibit(false);
+    subsystem_reset();
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
@@ -346,7 +346,6 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     CPUState *t;
     int rc;
 
-    qemu_balloon_inhibit(true);
     if (!pv_mig_blocker) {
         error_setg(&pv_mig_blocker,
                    "protected VMs are currently not migrateable.");
@@ -384,6 +383,7 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     if (rc) {
         goto out_err;
     }
+    subsystem_reset();
     return rc;
 
 out_err:
hw/s390x/virtio-ccw.c +13 -0
index 13f57e7b67..48bb54f68e 100644
@@ -869,12 +869,24 @@ static void virtio_ccw_notify(DeviceState *d, uint16_t vector)
     }
 }
 
+static inline void virtio_ccw_pv_enforce_features(VirtIODevice *vdev)
+{
+    S390CcwMachineState *ms = S390_CCW_MACHINE(qdev_get_machine());
+
+    if (ms->pv) {
+        virtio_add_feature(&vdev->host_features, VIRTIO_F_IOMMU_PLATFORM);
+    } else {
+        virtio_clear_feature(&vdev->host_features, VIRTIO_F_IOMMU_PLATFORM);
+    }
+}
+
 static void virtio_ccw_reset(DeviceState *d)
 {
     VirtioCcwDevice *dev = VIRTIO_CCW_DEVICE(d);
     VirtIODevice *vdev = virtio_bus_get_device(&dev->bus);
     VirtIOCCWDeviceClass *vdc = VIRTIO_CCW_DEVICE_GET_CLASS(dev);
 
+    virtio_ccw_pv_enforce_features(vdev);
     virtio_ccw_reset_virtio(dev, vdev);
     if (vdc->parent_reset) {
         vdc->parent_reset(d);
@@ -1103,6 +1115,7 @@ static void virtio_ccw_pre_plugged(DeviceState *d, Error **errp)
     if (dev->max_rev >= 1) {
         virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1);
     }
+    virtio_ccw_pv_enforce_features(vdev);
 }
 
 /* This is called by virtio-bus just after the device is plugged. */

base-commit: 8665f2475f5999d4c9f33598f1360f0b0797c489

Signature · 1 line
2.17.1
#34 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 68 lines · On Thu, 27 Feb 2020 13:24:02 +0100 Halil Pasic <[email protected]> wrote:
Quoted 11 lines · On Wed, 26 Feb 2020 16:11:03 +0100 Janosch Frank <[email protected]> wrote:
Quoted 9 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Quoted 9 lines

Yes, as soon as the host alters non-shared memory we'll run into integrity issues.

I've been talking to Halil after I sent this out and it looks like we'll rather try to automatically enable the IOMMU for all devices when switching into protected mode. He said that if the IOMMU is set the balloon code will do an early exit on feature negotiation.

I have a discussion starter RFC for you.

--------------------------8<----------------------------------------------
From: Halil Pasic <[email protected]>
Date: Wed, 26 Feb 2020 16:48:21 +0100
Subject: [RFC PATCH 1/1] virtio-ccw: auto-manage VIRTIO_F_IOMMU_PLATFORM

The virtio specification tells that the device is to present VIRTIO_F_ACCESS_PLATFORM (a.k.a. VIRTIO_F_IOMMU_PLATFORM) when the device "can only access certain memory addresses with said access specified and/or granted by the platform". This is the case for a protected VM, as the device can access only memory addresses that are in pages that are currently shared (only the guest can share/unsare its page).

No VM however starts out as a protected VM, but some VMs may be converted to protected VMs if the guest decides so.

Making the end user explicitly manage the VIRTIO_F_ACCESS_PLATFORM via the property iommu_on is a minor disaster. If the correctness of the paravirtualized virtio devices depends (and thus in a sense the correctness of the hypervisor), then the hypervisor should have the last word about whether VIRTIO_F_ACCESS_PLATFORM is to be presented or not.

Let's manage the VIRTIO_F_ACCESS_PLATFORM virtio feature automatically for virtio-ccw devices, so that we set it before we start the protected configuration, and unset it when our VM is not protected.

---
NOTES:
* I wanted to have a discussion starter fast, there are multiple open questions.

* Doing more than one system_resets() is hackish. We should look into this.

* The user interface implications of this patch are also an ugly can of worms. Needs to be discussed.

* We should consider keeping the original value if !pv and restoring it on pv --> !pv, instead of forcing to unset when leaving pv, and actually not forcing unset when !pv.

Quoted 7 lines

* It might make sense to do something like this in virtio core, but I decided start the discussion with a ccw-only change.

* Maybe we need a machine option that enables this sort of behavior, especially if we decide not to do the conserving/restoring.

* Lightly tested.

ping

Any interest in this?

Quoted 73 lines
---
 hw/s390x/s390-virtio-ccw.c |  4 ++--
 hw/s390x/virtio-ccw.c      | 13 +++++++++++++
 2 files changed, 15 insertions(+), 2 deletions(-)
hw/s390x/s390-virtio-ccw.c +2 -2
index 0f4455d1df..996124f152 100644
@@ -337,7 +337,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         ms->pv = false;
     }
     migrate_del_blocker(pv_mig_blocker);
-    qemu_balloon_inhibit(false);
+    subsystem_reset();
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
@@ -346,7 +346,6 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     CPUState *t;
     int rc;
 
-    qemu_balloon_inhibit(true);
     if (!pv_mig_blocker) {
         error_setg(&pv_mig_blocker,
                    "protected VMs are currently not migrateable.");
@@ -384,6 +383,7 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     if (rc) {
         goto out_err;
     }
+    subsystem_reset();
     return rc;
 
 out_err:
hw/s390x/virtio-ccw.c +13 -0
index 13f57e7b67..48bb54f68e 100644
@@ -869,12 +869,24 @@ static void virtio_ccw_notify(DeviceState *d, uint16_t vector)
     }
 }
 
+static inline void virtio_ccw_pv_enforce_features(VirtIODevice *vdev)
+{
+    S390CcwMachineState *ms = S390_CCW_MACHINE(qdev_get_machine());
+
+    if (ms->pv) {
+        virtio_add_feature(&vdev->host_features, VIRTIO_F_IOMMU_PLATFORM);
+    } else {
+        virtio_clear_feature(&vdev->host_features, VIRTIO_F_IOMMU_PLATFORM);
+    }
+}
+
 static void virtio_ccw_reset(DeviceState *d)
 {
     VirtioCcwDevice *dev = VIRTIO_CCW_DEVICE(d);
     VirtIODevice *vdev = virtio_bus_get_device(&dev->bus);
     VirtIOCCWDeviceClass *vdc = VIRTIO_CCW_DEVICE_GET_CLASS(dev);
 
+    virtio_ccw_pv_enforce_features(vdev);
     virtio_ccw_reset_virtio(dev, vdev);
     if (vdc->parent_reset) {
         vdc->parent_reset(d);
@@ -1103,6 +1115,7 @@ static void virtio_ccw_pre_plugged(DeviceState *d, Error **errp)
     if (dev->max_rev >= 1) {
         virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1);
     }
+    virtio_ccw_pv_enforce_features(vdev);
 }
 
 /* This is called by virtio-bus just after the device is plugged. */

base-commit: 8665f2475f5999d4c9f33598f1360f0b0797c489

#35 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

In reply to Halil Pasic
Quoted 147 lines · Halil Pasic
Quoted 11 lines · On Wed, 26 Feb 2020 16:11:03 +0100 Janosch Frank <[email protected]> wrote:
Quoted 9 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Quoted 9 lines

Yes, as soon as the host alters non-shared memory we'll run into integrity issues.

I've been talking to Halil after I sent this out and it looks like we'll rather try to automatically enable the IOMMU for all devices when switching into protected mode. He said that if the IOMMU is set the balloon code will do an early exit on feature negotiation.

I have a discussion starter RFC for you.

--------------------------8<----------------------------------------------
From: Halil Pasic <[email protected]>
Date: Wed, 26 Feb 2020 16:48:21 +0100
Subject: [RFC PATCH 1/1] virtio-ccw: auto-manage VIRTIO_F_IOMMU_PLATFORM

The virtio specification tells that the device is to present VIRTIO_F_ACCESS_PLATFORM (a.k.a. VIRTIO_F_IOMMU_PLATFORM) when the device "can only access certain memory addresses with said access specified and/or granted by the platform". This is the case for a protected VM, as the device can access only memory addresses that are in pages that are currently shared (only the guest can share/unsare its page).

No VM however starts out as a protected VM, but some VMs may be converted to protected VMs if the guest decides so.

Making the end user explicitly manage the VIRTIO_F_ACCESS_PLATFORM via the property iommu_on is a minor disaster. If the correctness of the paravirtualized virtio devices depends (and thus in a sense the correctness of the hypervisor), then the hypervisor should have the last word about whether VIRTIO_F_ACCESS_PLATFORM is to be presented or not.

Let's manage the VIRTIO_F_ACCESS_PLATFORM virtio feature automatically for virtio-ccw devices, so that we set it before we start the protected configuration, and unset it when our VM is not protected.

---
NOTES:
* I wanted to have a discussion starter fast, there are multiple open questions.

* Doing more than one system_resets() is hackish. We should look into this.

* The user interface implications of this patch are also an ugly can of worms. Needs to be discussed.

* We should consider keeping the original value if !pv and restoring it on pv --> !pv, instead of forcing to unset when leaving pv, and actually not forcing unset when !pv.

* It might make sense to do something like this in virtio core, but I decided start the discussion with a ccw-only change.

* Maybe we need a machine option that enables this sort of behavior, especially if we decide not to do the conserving/restoring.

* Lightly tested.

---
 hw/s390x/s390-virtio-ccw.c |  4 ++--
 hw/s390x/virtio-ccw.c      | 13 +++++++++++++
 2 files changed, 15 insertions(+), 2 deletions(-)
hw/s390x/s390-virtio-ccw.c +2 -2
index 0f4455d1df..996124f152 100644
@@ -337,7 +337,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         ms->pv = false;
     }
     migrate_del_blocker(pv_mig_blocker);
-    qemu_balloon_inhibit(false);
+    subsystem_reset();
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
@@ -346,7 +346,6 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     CPUState *t;
     int rc;
 
-    qemu_balloon_inhibit(true);
     if (!pv_mig_blocker) {
         error_setg(&pv_mig_blocker,
                    "protected VMs are currently not migrateable.");
@@ -384,6 +383,7 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     if (rc) {
         goto out_err;
     }
+    subsystem_reset();
     return rc;
 
 out_err:
hw/s390x/virtio-ccw.c +13 -0
index 13f57e7b67..48bb54f68e 100644
@@ -869,12 +869,24 @@ static void virtio_ccw_notify(DeviceState *d, uint16_t vector)
     }
 }
 
+static inline void virtio_ccw_pv_enforce_features(VirtIODevice *vdev)
+{
+    S390CcwMachineState *ms = S390_CCW_MACHINE(qdev_get_machine());
+
+    if (ms->pv) {
+        virtio_add_feature(&vdev->host_features, VIRTIO_F_IOMMU_PLATFORM);
+    } else {
+        virtio_clear_feature(&vdev->host_features, VIRTIO_F_IOMMU_PLATFORM);
+    }
+}
+
 static void virtio_ccw_reset(DeviceState *d)
 {
     VirtioCcwDevice *dev = VIRTIO_CCW_DEVICE(d);
     VirtIODevice *vdev = virtio_bus_get_device(&dev->bus);
     VirtIOCCWDeviceClass *vdc = VIRTIO_CCW_DEVICE_GET_CLASS(dev);
 
+    virtio_ccw_pv_enforce_features(vdev);
     virtio_ccw_reset_virtio(dev, vdev);
     if (vdc->parent_reset) {
         vdc->parent_reset(d);
@@ -1103,6 +1115,7 @@ static void virtio_ccw_pre_plugged(DeviceState *d, Error **errp)
     if (dev->max_rev >= 1) {
         virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1);
     }
+    virtio_ccw_pv_enforce_features(vdev);
 }
 
 /* This is called by virtio-bus just after the device is plugged. */
Quoted 2 lines

base-commit: 8665f2475f5999d4c9f33598f1360f0b0797c489

I asked this question already to Michael (cc) via a different channel, but hare is it again:

Why does the balloon driver not support VIRTIO_F_IOMMU_PLATFORM? It is absolutely not clear to me. The introducing commit mentioned that it "bypasses DMA". I fail to see that.

At least the communication via the SG mechanism should work perfectly fine with an IOMMU enabled. So I assume it boils down to the pages that we inflate/deflate not being referenced via IOVA?

I don't think they have to be IOVA addresses. We're neither reading nor writing these pages. We really speak about "physical memory in the system" when ballooning. Everything else doesn't really make sense. There is no need to map/unmap pages we inflate/deflate AFAIKs.

IMHO, we should not try to piggy-back on VIRTIO_F_IOMMU_PLATFORM here, but instead explicitly disable it either in the hypervisor or the guest.

I hope someone can clarify what the real issue with an IOMMU and ballooning is, because I'll be having the same "issue" with virtio-mem.

Signature · 3 lines
Thanks,

David / dhildenb
#36 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 154 lines · On Thu, 19 Mar 2020 14:54:11 +0100 David Hildenbrand <[email protected]> wrote:
Quoted 147 lines · Halil Pasic
Quoted 11 lines · On Wed, 26 Feb 2020 16:11:03 +0100 Janosch Frank <[email protected]> wrote:
Quoted 9 lines · David Hildenbrand
Quoted 4 lines · Janosch Frank

Ballooning in protected VMs can only be done when the guest shares the pages it gives to the host. Hence, until we have a solution for this in the guest kernel, we inhibit ballooning when switching into protected mode and reverse that once we move out of it.

I don't understand what you mean here, sorry. zapping a page will mean that a fresh one will be faulted in when accessed. And AFAIK, that means it will be encrypted again when needed.

Quoted 9 lines

Yes, as soon as the host alters non-shared memory we'll run into integrity issues.

I've been talking to Halil after I sent this out and it looks like we'll rather try to automatically enable the IOMMU for all devices when switching into protected mode. He said that if the IOMMU is set the balloon code will do an early exit on feature negotiation.

I have a discussion starter RFC for you.

--------------------------8<----------------------------------------------
From: Halil Pasic <[email protected]>
Date: Wed, 26 Feb 2020 16:48:21 +0100
Subject: [RFC PATCH 1/1] virtio-ccw: auto-manage VIRTIO_F_IOMMU_PLATFORM

The virtio specification tells that the device is to present VIRTIO_F_ACCESS_PLATFORM (a.k.a. VIRTIO_F_IOMMU_PLATFORM) when the device "can only access certain memory addresses with said access specified and/or granted by the platform". This is the case for a protected VM, as the device can access only memory addresses that are in pages that are currently shared (only the guest can share/unsare its page).

No VM however starts out as a protected VM, but some VMs may be converted to protected VMs if the guest decides so.

Making the end user explicitly manage the VIRTIO_F_ACCESS_PLATFORM via the property iommu_on is a minor disaster. If the correctness of the paravirtualized virtio devices depends (and thus in a sense the correctness of the hypervisor), then the hypervisor should have the last word about whether VIRTIO_F_ACCESS_PLATFORM is to be presented or not.

Let's manage the VIRTIO_F_ACCESS_PLATFORM virtio feature automatically for virtio-ccw devices, so that we set it before we start the protected configuration, and unset it when our VM is not protected.

---
NOTES:
* I wanted to have a discussion starter fast, there are multiple open questions.

* Doing more than one system_resets() is hackish. We should look into this.

* The user interface implications of this patch are also an ugly can of worms. Needs to be discussed.

* We should consider keeping the original value if !pv and restoring it on pv --> !pv, instead of forcing to unset when leaving pv, and actually not forcing unset when !pv.

* It might make sense to do something like this in virtio core, but I decided start the discussion with a ccw-only change.

* Maybe we need a machine option that enables this sort of behavior, especially if we decide not to do the conserving/restoring.

* Lightly tested.

---
 hw/s390x/s390-virtio-ccw.c |  4 ++--
 hw/s390x/virtio-ccw.c      | 13 +++++++++++++
 2 files changed, 15 insertions(+), 2 deletions(-)
hw/s390x/s390-virtio-ccw.c +2 -2
index 0f4455d1df..996124f152 100644
@@ -337,7 +337,7 @@ static void s390_machine_unprotect(S390CcwMachineState *ms)
         ms->pv = false;
     }
     migrate_del_blocker(pv_mig_blocker);
-    qemu_balloon_inhibit(false);
+    subsystem_reset();
 }
 
 static int s390_machine_protect(S390CcwMachineState *ms)
@@ -346,7 +346,6 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     CPUState *t;
     int rc;
 
-    qemu_balloon_inhibit(true);
     if (!pv_mig_blocker) {
         error_setg(&pv_mig_blocker,
                    "protected VMs are currently not migrateable.");
@@ -384,6 +383,7 @@ static int s390_machine_protect(S390CcwMachineState *ms)
     if (rc) {
         goto out_err;
     }
+    subsystem_reset();
     return rc;
 
 out_err:
hw/s390x/virtio-ccw.c +13 -0
index 13f57e7b67..48bb54f68e 100644
@@ -869,12 +869,24 @@ static void virtio_ccw_notify(DeviceState *d, uint16_t vector)
     }
 }
 
+static inline void virtio_ccw_pv_enforce_features(VirtIODevice *vdev)
+{
+    S390CcwMachineState *ms = S390_CCW_MACHINE(qdev_get_machine());
+
+    if (ms->pv) {
+        virtio_add_feature(&vdev->host_features, VIRTIO_F_IOMMU_PLATFORM);
+    } else {
+        virtio_clear_feature(&vdev->host_features, VIRTIO_F_IOMMU_PLATFORM);
+    }
+}
+
 static void virtio_ccw_reset(DeviceState *d)
 {
     VirtioCcwDevice *dev = VIRTIO_CCW_DEVICE(d);
     VirtIODevice *vdev = virtio_bus_get_device(&dev->bus);
     VirtIOCCWDeviceClass *vdc = VIRTIO_CCW_DEVICE_GET_CLASS(dev);
 
+    virtio_ccw_pv_enforce_features(vdev);
     virtio_ccw_reset_virtio(dev, vdev);
     if (vdc->parent_reset) {
         vdc->parent_reset(d);
@@ -1103,6 +1115,7 @@ static void virtio_ccw_pre_plugged(DeviceState *d, Error **errp)
     if (dev->max_rev >= 1) {
         virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1);
     }
+    virtio_ccw_pv_enforce_features(vdev);
 }
 
 /* This is called by virtio-bus just after the device is plugged. */
Quoted 2 lines

base-commit: 8665f2475f5999d4c9f33598f1360f0b0797c489

I asked this question already to Michael (cc) via a different channel, but hare is it again:

Quoted 7 lines

Why does the balloon driver not support VIRTIO_F_IOMMU_PLATFORM? It is absolutely not clear to me. The introducing commit mentioned that it "bypasses DMA". I fail to see that.

At least the communication via the SG mechanism should work perfectly fine with an IOMMU enabled. So I assume it boils down to the pages that we inflate/deflate not being referenced via IOVA?

AFAIU the IOVA/GPA stuff is not the problem here. You have said it yourself, the SG mechanism would work for balloon out of the box, as it does for the other virtio devices.

But VIRTIO_F_ACCESS_PLATFORM (aka VIRTIO_F_IOMMU_PLATFORM) not presented means according to Michael that the device has full access to the entire guest RAM. If VIRTIO_F_ACCESS_PLATFORM is negotiated this may or may not be the case.

The actual problem is that the pages denoted by the buffer transmitted via the virtqueue are normally not shared pages. I.e. the hypervisor can not reuse them (what is the point of balloon inflate). To make this work, the guest would need to share the pages before saying 'host these are in my balloon, so you can use them'. This is a piece of logic we need only if the host/the device does not have full access to the guest RAM. That is in my opinion why the balloon driver fences VIRTIO_F_ACCESS_PLATFORM.

Does that make sense?

Quoted 9 lines

I don't think they have to be IOVA addresses. We're neither reading nor writing these pages. We really speak about "physical memory in the system" when ballooning. Everything else doesn't really make sense. There is no need to map/unmap pages we inflate/deflate AFAIKs.

IMHO, we should not try to piggy-back on VIRTIO_F_IOMMU_PLATFORM here, but instead explicitly disable it either in the hypervisor or the guest.

We need a feature bit here. We can say fencing VIRTIO_F_ACCESS_PLATFORM was a bug, fix that bug, and then invent another 'the guest RAM is somehow different' feature bit specific to the balloon, and then create arch hooks in the driver that get active if this feature is negotiated.

I assumed the fact that the balloon driver fences
VIRTIO_F_ACCESS_PLATFORM is not a bug.

Quoted 2 lines

I hope someone can clarify what the real issue with an IOMMU and ballooning is, because I'll be having the same "issue" with

virtio-mem.

Quoted 1 line

The issue is not with the IOMMU, the issue is with restricted access to guest RAM. The definition of VIRTIO_F_ACCESS_PLATFORM is such that we pretty much know what's up when VIRTIO_F_ACCESS_PLATFORM is not presented, but VIRTIO_F_ACCESS_PLATFORM presented can mean a couple of things.

Regards,
Halil

#37 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

[...]

Quoted 12 lines
Quoted 11 lines

I asked this question already to Michael (cc) via a different channel, but hare is it again:

Why does the balloon driver not support VIRTIO_F_IOMMU_PLATFORM? It is absolutely not clear to me. The introducing commit mentioned that it "bypasses DMA". I fail to see that.

At least the communication via the SG mechanism should work perfectly fine with an IOMMU enabled. So I assume it boils down to the pages that we inflate/deflate not being referenced via IOVA?

Quoted 8 lines

AFAIU the IOVA/GPA stuff is not the problem here. You have said it yourself, the SG mechanism would work for balloon out of the box, as it does for the other virtio devices.

But VIRTIO_F_ACCESS_PLATFORM (aka VIRTIO_F_IOMMU_PLATFORM) not presented means according to Michael that the device has full access to the entire guest RAM. If VIRTIO_F_ACCESS_PLATFORM is negotiated this may or may not be the case.

So you say

"The virtio specification tells that the device is to present VIRTIO_F_ACCESS_PLATFORM (a.k.a. VIRTIO_F_IOMMU_PLATFORM) when the device "can only access certain memory addresses with said access specified and/or granted by the platform"."

So, AFAIU, *any* virtio device (hypervisor side) has to present this flag when PV is enabled. In that regard, your patch makes perfect sense (although I am not sure it's a good idea to overwrite these feature bits
- maybe they should be activated on the cmdline permanently instead when PV is to be used? (or enable )).

Quoted 6 lines

The actual problem is that the pages denoted by the buffer transmitted via the virtqueue are normally not shared pages. I.e. the hypervisor can not reuse them (what is the point of balloon inflate). To make this work, the guest would need to share the pages before saying 'host these are in my balloon, so you can use them'. This is a piece of logic we

What exactly would have to be done in the hypervisor to support it?

Assume we have to trigger sharing/unsharing - this sounds like a very architecture specific thing? Or is this e.g., doing a map/unmap operation like mapping/unmapping the SG?

Right now it sounds to me "we have to do $ARCHSPECIFIC when inflating/deflating in the guest", which feels wrong.

Quoted 3 lines

need only if the host/the device does not have full access to the guest RAM. That is in my opinion why the balloon driver fences VIRTIO_F_ACCESS_PLATFORM.> Does that make sense?

Yeah, I understood the "device has to set VIRTIO_F_ACCESS_PLATFORM" part. Struggling with the "what can the guest driver actually do" part.

Quoted 21 lines
Quoted 10 lines

I don't think they have to be IOVA addresses. We're neither reading nor writing these pages. We really speak about "physical memory in the system" when ballooning. Everything else doesn't really make sense. There is no need to map/unmap pages we inflate/deflate AFAIKs.

IMHO, we should not try to piggy-back on VIRTIO_F_IOMMU_PLATFORM here, but instead explicitly disable it either in the hypervisor or the guest.

We need a feature bit here. We can say fencing VIRTIO_F_ACCESS_PLATFORM was a bug, fix that bug, and then invent another 'the guest RAM is somehow different' feature bit specific to the balloon, and then create arch hooks in the driver that get active if this feature is negotiated.

I assumed the fact that the balloon driver fences
VIRTIO_F_ACCESS_PLATFORM is not a bug.

Quoted 2 lines

I hope someone can clarify what the real issue with an IOMMU and ballooning is, because I'll be having the same "issue" with

Quoted 8 lines

virtio-mem.

Quoted 1 line

The issue is not with the IOMMU, the issue is with restricted access to guest RAM. The definition of VIRTIO_F_ACCESS_PLATFORM is such that we pretty much know what's up when VIRTIO_F_ACCESS_PLATFORM is not presented, but VIRTIO_F_ACCESS_PLATFORM presented can mean a couple of things.

Understood.

Signature · 3 lines
Thanks,

David / dhildenb
#38 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 25 lines · On Thu, 19 Mar 2020 18:31:11 +0100 David Hildenbrand <[email protected]> wrote:

[...]

Quoted 12 lines
Quoted 11 lines

I asked this question already to Michael (cc) via a different channel, but hare is it again:

Why does the balloon driver not support VIRTIO_F_IOMMU_PLATFORM? It is absolutely not clear to me. The introducing commit mentioned that it "bypasses DMA". I fail to see that.

At least the communication via the SG mechanism should work perfectly fine with an IOMMU enabled. So I assume it boils down to the pages that we inflate/deflate not being referenced via IOVA?

Quoted 8 lines

AFAIU the IOVA/GPA stuff is not the problem here. You have said it yourself, the SG mechanism would work for balloon out of the box, as it does for the other virtio devices.

But VIRTIO_F_ACCESS_PLATFORM (aka VIRTIO_F_IOMMU_PLATFORM) not presented means according to Michael that the device has full access to the entire guest RAM. If VIRTIO_F_ACCESS_PLATFORM is negotiated this may or may not be the case.

So you say

Quoted 7 lines

"The virtio specification tells that the device is to present VIRTIO_F_ACCESS_PLATFORM (a.k.a. VIRTIO_F_IOMMU_PLATFORM) when the device "can only access certain memory addresses with said access specified and/or granted by the platform"."

So, AFAIU, *any* virtio device (hypervisor side) has to present this flag when PV is enabled.

Yes, and balloon says bye bye when running in PV mode is only a secondary objective. I've compiled some references:

"To summarize, the necessary conditions for a hack along these lines (using DMA API without VIRTIO_F_ACCESS_PLATFORM) are that we detect that:

- secure guest mode is enabled - so we know that since we don't share most memory regular virtio code won't
work, even though the buggy hypervisor didn't set VIRTIO_F_ACCESS_PLATFORM"
(Michael Tsirkin, https://lkml.org/lkml/2020/2/20/1021)
I.e.: PV but !VIRTIO_F_ACCESS_PLATFORM \implies bugy hypervisor

"If VIRTIO_F_ACCESS_PLATFORM is set then things just work. If VIRTIO_F_ACCESS_PLATFORM is clear device is supposed to have access to all of memory. You can argue in various ways but it's easier to just declare a behaviour that violates this a bug."
(Michael Tsirkin, https://lkml.org/lkml/2020/2/21/1626)
This one is about all memory guest, and not just the buffers transfered via the virtqueue, which surprised me a bit at the beginning. But balloon actually needs this.

"A device SHOULD offer VIRTIO_F_ACCESS_PLATFORM if its access to memory is through bus addresses distinct from and translated by the platform to physical addresses used by the driver, and/or if it can only access certain memory addresses with said access specified and/or granted by the platform. A device MAY fail to operate further if VIRTIO_F_ACCESS_PLATFORM is not accepted. "
(https://docs.oasis-open.org/virtio/virtio/v1.1/cs01/virtio-v1.1-cs01.html#x1-4120002)

Quoted 5 lines

In that regard, your patch makes perfect sense
(although I am not sure it's a good idea to overwrite these feature bits
- maybe they should be activated on the cmdline permanently instead when PV is to be used? (or enable )).

I didn't understand the last part. I believe conserving the user specified value when not running in PV mode is better than the hard overwrite I did here. I wanted a discussion starter.

I think the other option (with respect to let QEMU manage this for user, i.e. what I try to do here) is to fence the conversion if virtio devices that do not offer VIRTIO_F_ACCESS_PLATFORM are attached; and disallow hotplug of such devices at some point during the conversion.

I believe that alternative is even uglier.

IMHO we don't want the end user to fiddle with iommu_platform, because all the 'benefit' he gets from that is possibility to make a mistake. For example, I got an internal bug report saying virtio is broken with PV, which boiled down to an overlooked auto generated NIC, which of course had iommu_platform (VIRTIO_F_ACCESS_PLATFORM) not set.

Quoted 10 lines
Quoted 8 lines

The actual problem is that the pages denoted by the buffer transmitted via the virtqueue are normally not shared pages. I.e. the hypervisor can not reuse them (what is the point of balloon inflate). To make this work, the guest would need to share the pages before saying 'host these are in my balloon, so you can use them'. This is a piece of logic we

What exactly would have to be done in the hypervisor to support it?

AFAIK nothing. The guest needs to share the pages, and everything works. Janosch, can you help me with this one?

Quoted 3 lines

Assume we have to trigger sharing/unsharing - this sounds like a very architecture specific thing?

It is, but any guest having sovereignty about its memory may need something similar.

Quoted 2 lines

Or is this e.g., doing a map/unmap
operation like mapping/unmapping the SG?

No this is something different. We need stronger guarantees than the streaming portion of the DMA API provides. And what we actually want is not DMA but something very different.

Quoted 4 lines

Right now it sounds to me "we have to do $ARCHSPECIFIC when inflating/deflating in the guest", which feels wrong.

It is wrong in a sense. Drivers are mostly supposed to be portable. But balloon is not a run of the mill device. I don't see any other way to make this work.

Quoted 7 lines
Quoted 3 lines

need only if the host/the device does not have full access to the guest RAM. That is in my opinion why the balloon driver fences VIRTIO_F_ACCESS_PLATFORM.> Does that make sense?

Yeah, I understood the "device has to set VIRTIO_F_ACCESS_PLATFORM" part. Struggling with the "what can the guest driver actually do" part.

Let me try to reword this. The point of PV is that the guest has exclusive access to his pages unless the guest decides to share some of the using a dedicated ultravisor call.

The point of the memballoon is, as far as I understand, to effectively dynamically manage the guests memory size within given boundaries, and without requiring memory hotplug. The basic idea is that the pages in the balloon belong to the host. The host attempting to re-use a non-shared page of a guest leads to problems. AFAIR the main problem was that shall we ever want to deflate such a page (make it again available for guest use) we would need to do an import, and that can only work if we have the exact same content as when it was exported. Otherwise integrity check fails as if we had a malicious hypervisor, that is trying to inject stuff into the guest.

I'm sure Janosch can provide a better explanation.

I really don't see another way, how memory ballooning could work with something like PV, without the balloon driver relinquishing the guests ownership of the pages that are going to leave the guest via the balloon.

On that note ccing the AMD SEV people. Balloon is at this point dysfunctional for them as well. @Tom: Right? If yes what problems need to be solved so virtio-balloon can work under SEV?

Regards,
Halil

#39 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 136 lines · Halil Pasic
Quoted 22 lines · On Thu, 19 Mar 2020 18:31:11 +0100 David Hildenbrand <[email protected]> wrote:

[...]

Quoted 10 lines
Quoted 10 lines

I asked this question already to Michael (cc) via a different channel, but hare is it again:

Why does the balloon driver not support VIRTIO_F_IOMMU_PLATFORM? It is absolutely not clear to me. The introducing commit mentioned that it "bypasses DMA". I fail to see that.

At least the communication via the SG mechanism should work perfectly fine with an IOMMU enabled. So I assume it boils down to the pages that we inflate/deflate not being referenced via IOVA?

Quoted 8 lines

AFAIU the IOVA/GPA stuff is not the problem here. You have said it yourself, the SG mechanism would work for balloon out of the box, as it does for the other virtio devices.

But VIRTIO_F_ACCESS_PLATFORM (aka VIRTIO_F_IOMMU_PLATFORM) not presented means according to Michael that the device has full access to the entire guest RAM. If VIRTIO_F_ACCESS_PLATFORM is negotiated this may or may not be the case.

So you say

Quoted 7 lines

"The virtio specification tells that the device is to present VIRTIO_F_ACCESS_PLATFORM (a.k.a. VIRTIO_F_IOMMU_PLATFORM) when the device "can only access certain memory addresses with said access specified and/or granted by the platform"."

So, AFAIU, *any* virtio device (hypervisor side) has to present this flag when PV is enabled.

Yes, and balloon says bye bye when running in PV mode is only a secondary objective. I've compiled some references:

"To summarize, the necessary conditions for a hack along these lines (using DMA API without VIRTIO_F_ACCESS_PLATFORM) are that we detect that:

- secure guest mode is enabled - so we know that since we don't share
most memory regular virtio code won't
work, even though the buggy hypervisor didn't set VIRTIO_F_ACCESS_PLATFORM"
(Michael Tsirkin, https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flkml.org%2Flkml%2F2020%2F2%2F20%2F1021&amp;data=02%7C01%7Cbrijesh.singh%40amd.com%7C52b79b5c9e894dd968c508d7ccfe9479%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637203266090844487&amp;sdata=aNS%2FW2nL27mPSl1Xz3iXUY31qtrzmVHYhzVHEILAaQQ%3D&amp;reserved=0)
I.e.: PV but !VIRTIO_F_ACCESS_PLATFORM \implies bugy hypervisor

"If VIRTIO_F_ACCESS_PLATFORM is set then things just work. If
VIRTIO_F_ACCESS_PLATFORM is clear device is supposed to have access to
all of memory. You can argue in various ways but it's easier to just
declare a behaviour that violates this a bug."
(Michael Tsirkin, https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flkml.org%2Flkml%2F2020%2F2%2F21%2F1626&amp;data=02%7C01%7Cbrijesh.singh%40amd.com%7C52b79b5c9e894dd968c508d7ccfe9479%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637203266090854439&amp;sdata=d3knybBUZ5NL0Lv1C2JS040A3toiCxXVYLkBlzXSrqc%3D&amp;reserved=0)
This one is about all memory guest, and not just the buffers transfered
via the virtqueue, which surprised me a bit at the beginning. But balloon
actually needs this.

"A device SHOULD offer VIRTIO_F_ACCESS_PLATFORM if its access to memory is through bus addresses distinct from and translated by the platform to physical addresses used by the driver, and/or if it can only access certain memory addresses with said access specified and/or granted by the platform. A device MAY fail to operate further if VIRTIO_F_ACCESS_PLATFORM is not accepted. "
(https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdocs.oasis-open.org%2Fvirtio%2Fvirtio%2Fv1.1%2Fcs01%2Fvirtio-v1.1-cs01.html%23x1-4120002&amp;data=02%7C01%7Cbrijesh.singh%40amd.com%7C52b79b5c9e894dd968c508d7ccfe9479%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637203266090854439&amp;sdata=RBx8cBr8I%2FWFChtVFTjBygRiHIXMmsjT8W%2BwLaTNQ24%3D&amp;reserved=0)

Quoted 5 lines

In that regard, your patch makes perfect sense
(although I am not sure it's a good idea to overwrite these feature bits
- maybe they should be activated on the cmdline permanently instead when PV is to be used? (or enable )).

I didn't understand the last part. I believe conserving the user specified value when not running in PV mode is better than the hard overwrite I did here. I wanted a discussion starter.

I think the other option (with respect to let QEMU manage this for user, i.e. what I try to do here) is to fence the conversion if virtio devices that do not offer VIRTIO_F_ACCESS_PLATFORM are attached; and disallow hotplug of such devices at some point during the conversion.

I believe that alternative is even uglier.

IMHO we don't want the end user to fiddle with iommu_platform, because all the 'benefit' he gets from that is possibility to make a mistake. For example, I got an internal bug report saying virtio is broken with PV, which boiled down to an overlooked auto generated NIC, which of course had iommu_platform (VIRTIO_F_ACCESS_PLATFORM) not set.

Quoted 7 lines
Quoted 6 lines

The actual problem is that the pages denoted by the buffer transmitted via the virtqueue are normally not shared pages. I.e. the hypervisor can not reuse them (what is the point of balloon inflate). To make this work, the guest would need to share the pages before saying 'host these are in my balloon, so you can use them'. This is a piece of logic we

What exactly would have to be done in the hypervisor to support it?

AFAIK nothing. The guest needs to share the pages, and everything works. Janosch, can you help me with this one?

Quoted 2 lines

Assume we have to trigger sharing/unsharing - this sounds like a very architecture specific thing?

It is, but any guest having sovereignty about its memory may need something similar.

Quoted 2 lines

Or is this e.g., doing a map/unmap
operation like mapping/unmapping the SG?

No this is something different. We need stronger guarantees than the streaming portion of the DMA API provides. And what we actually want is not DMA but something very different.

Quoted 3 lines

Right now it sounds to me "we have to do $ARCHSPECIFIC when inflating/deflating in the guest", which feels wrong.

It is wrong in a sense. Drivers are mostly supposed to be portable. But balloon is not a run of the mill device. I don't see any other way to make this work.

Quoted 6 lines
Quoted 3 lines

need only if the host/the device does not have full access to the guest RAM. That is in my opinion why the balloon driver fences VIRTIO_F_ACCESS_PLATFORM.> Does that make sense?

Yeah, I understood the "device has to set VIRTIO_F_ACCESS_PLATFORM" part. Struggling with the "what can the guest driver actually do" part.

Let me try to reword this. The point of PV is that the guest has exclusive access to his pages unless the guest decides to share some of the using a dedicated ultravisor call.

The point of the memballoon is, as far as I understand, to effectively dynamically manage the guests memory size within given boundaries, and without requiring memory hotplug. The basic idea is that the pages in the balloon belong to the host. The host attempting to re-use a non-shared page of a guest leads to problems. AFAIR the main problem was that shall we ever want to deflate such a page (make it again available for guest use) we would need to do an import, and that can only work if we have the exact same content as when it was exported. Otherwise integrity check fails as if we had a malicious hypervisor, that is trying to inject stuff into the guest.

I'm sure Janosch can provide a better explanation.

Quoted 7 lines

I really don't see another way, how memory ballooning could work with something like PV, without the balloon driver relinquishing the guests ownership of the pages that are going to leave the guest via the balloon.

On that note ccing the AMD SEV people. Balloon is at this point dysfunctional for them as well. @Tom: Right? If yes what problems need to be solved so virtio-balloon can work under SEV?

Yes, Balloon does not work for SEV as well. I did not investigated in the great detail on what it will take to make it work. At that time the main issue was that the balloon virtio device did not had VIRTIO_F_ACCESS_PLATFORM flag set. So, we were failing to create a shared virtio ring. The current SEV does not have integrity protection support so we should be okay to release the buffer from guest and import the buffer into the guest. We probably need to deal with some cache flushes because cache may still contain the data with different C-bit etc. AMD has recently published the SNP architecture, which supports the integrity protection and it appears that this architecture provides some instruction which guest can used by the guest to validate and invalidate the pages as it leaves and enter the guest memory space. Overall, i think its doable but devil in the detail.

Quoted 4 lines

Regards,
Halil

#40 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

In reply to Halil Pasic
Quoted 5 lines
Quoted 2 lines

So, AFAIU, *any* virtio device (hypervisor side) has to present this flag when PV is enabled.

Yes, and balloon says bye bye when running in PV mode is only a secondary objective. I've compiled some references:

Thanks!

Quoted 62 lines

"To summarize, the necessary conditions for a hack along these lines (using DMA API without VIRTIO_F_ACCESS_PLATFORM) are that we detect that:

- secure guest mode is enabled - so we know that since we don't share most memory regular virtio code won't
work, even though the buggy hypervisor didn't set VIRTIO_F_ACCESS_PLATFORM"
(Michael Tsirkin, https://lkml.org/lkml/2020/2/20/1021)
I.e.: PV but !VIRTIO_F_ACCESS_PLATFORM \implies bugy hypervisor

"If VIRTIO_F_ACCESS_PLATFORM is set then things just work. If VIRTIO_F_ACCESS_PLATFORM is clear device is supposed to have access to all of memory. You can argue in various ways but it's easier to just declare a behaviour that violates this a bug."
(Michael Tsirkin, https://lkml.org/lkml/2020/2/21/1626)
This one is about all memory guest, and not just the buffers transfered via the virtqueue, which surprised me a bit at the beginning. But balloon actually needs this.

"A device SHOULD offer VIRTIO_F_ACCESS_PLATFORM if its access to memory is through bus addresses distinct from and translated by the platform to physical addresses used by the driver, and/or if it can only access certain memory addresses with said access specified and/or granted by the platform. A device MAY fail to operate further if VIRTIO_F_ACCESS_PLATFORM is not accepted. "
(https://docs.oasis-open.org/virtio/virtio/v1.1/cs01/virtio-v1.1-cs01.html#x1-4120002)

Quoted 5 lines

In that regard, your patch makes perfect sense
(although I am not sure it's a good idea to overwrite these feature bits
- maybe they should be activated on the cmdline permanently instead when PV is to be used? (or enable )).

I didn't understand the last part. I believe conserving the user specified value when not running in PV mode is better than the hard overwrite I did here. I wanted a discussion starter.

I think the other option (with respect to let QEMU manage this for user, i.e. what I try to do here) is to fence the conversion if virtio devices that do not offer VIRTIO_F_ACCESS_PLATFORM are attached; and disallow hotplug of such devices at some point during the conversion.

I believe that alternative is even uglier.

IMHO we don't want the end user to fiddle with iommu_platform, because all the 'benefit' he gets from that is possibility to make a mistake. For example, I got an internal bug report saying virtio is broken with PV, which boiled down to an overlooked auto generated NIC, which of course had iommu_platform (VIRTIO_F_ACCESS_PLATFORM) not set.

Quoted 9 lines
Quoted 8 lines

The actual problem is that the pages denoted by the buffer transmitted via the virtqueue are normally not shared pages. I.e. the hypervisor can not reuse them (what is the point of balloon inflate). To make this work, the guest would need to share the pages before saying 'host these are in my balloon, so you can use them'. This is a piece of logic we

Quoted 5 lines
Quoted 1 line

What exactly would have to be done in the hypervisor to support it?

AFAIK nothing. The guest needs to share the pages, and everything works. Janosch, can you help me with this one?

See below, making this work on the hypervisor side would be much cleaner IMHO, but most probably not possible due to guest integrity.

FWIW, "Free page reporting" will (never) work with PV, where there is basically no manual "deflation" step anymore.

Quoted 7 lines
Quoted 3 lines

Assume we have to trigger sharing/unsharing - this sounds like a very architecture specific thing?

It is, but any guest having sovereignty about its memory may need something similar.

Quoted 6 lines
Quoted 2 lines

Or is this e.g., doing a map/unmap
operation like mapping/unmapping the SG?

No this is something different. We need stronger guarantees than the streaming portion of the DMA API provides. And what we actually want is not DMA but something very different.

Right, that's what I was expecting ...

Quoted 9 lines
Quoted 5 lines

Right now it sounds to me "we have to do $ARCHSPECIFIC when inflating/deflating in the guest", which feels wrong.

It is wrong in a sense. Drivers are mostly supposed to be portable. But balloon is not a run of the mill device. I don't see any other way to make this work.

Well, it is mostly architecture independent until now ...

Quoted 24 lines
Quoted 8 lines
Quoted 4 lines

need only if the host/the device does not have full access to the guest RAM. That is in my opinion why the balloon driver fences VIRTIO_F_ACCESS_PLATFORM.> Does that make sense?

Yeah, I understood the "device has to set VIRTIO_F_ACCESS_PLATFORM" part. Struggling with the "what can the guest driver actually do" part.

Let me try to reword this. The point of PV is that the guest has exclusive access to his pages unless the guest decides to share some of the using a dedicated ultravisor call.

The point of the memballoon is, as far as I understand, to effectively dynamically manage the guests memory size within given boundaries, and without requiring memory hotplug. The basic idea is that the pages in the balloon belong to the host. The host attempting to re-use a non-shared page of a guest leads to problems. AFAIR the main problem was that shall we ever want to deflate such a page (make it again available for guest use) we would need to do an import, and that can only work if we have the exact same content as when it was exported. Otherwise integrity check fails as if we had a malicious hypervisor, that is trying to inject stuff into the guest.

Quoted 8 lines

I'm sure Janosch can provide a better explanation.

I really don't see another way, how memory ballooning could work with something like PV, without the balloon driver relinquishing the guests ownership of the pages that are going to leave the guest via the balloon.> On that note ccing the AMD SEV people. Balloon is at this point dysfunctional for them as well. @Tom: Right? If yes what problems need to be solved so virtio-balloon can work under SEV?

SEV even pins all guest memory, so it's useless and would be even dangerous to use.

Some thoughts:

1. I would really prefer if there is a way to zero-out+share a page and zero-out+unshare a page triggered by the hypervisor. Then only the hypervisor has to do "the right thing" when inflating/deflating/rebooting etc. I know we can "unshare all" via the UV - we e.g., have to do that on reboots. But I assume this might mess with "guest integrity" (being able to zero out random guest pages technically) and is therefore not possible.

2. Have some other way to communicate "careful, ballooning won't work". E.g., the guest detecting *itself* that it is running inside a PV environment and not loading virtio-balloon until it properly shares/unshares. Again, piggy-backing on IOMMU/VIRTIO_F_ACCESS_PLATFORM somehow feels wrong.

E.g., once you would support inflation/deflation in virtio-balloon, free page reporting could not be supported. So it's more than just a single arch specific inflation/deflation callback.

And virtio-mem [1] will have similar issues once we want to use that on s390x. But there, an arch-specific share/unshare callback should be sufficient most probably. Still, there would have to be a way to block it on s390x PV until implemented. Ideally it will be the same as for virtio-balloon.

Again, being able to do that in the hypervisor instead of in the guest would be much cleaner.

[1] https://lkml.kernel.org/r/[email protected]

Signature · 3 lines
Thanks,

David / dhildenb
#41 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

In reply to David Hildenbrand
Quoted 3 lines · David Hildenbrand

Why does the balloon driver not support VIRTIO_F_IOMMU_PLATFORM? It is absolutely not clear to me. The introducing commit mentioned that it "bypasses DMA". I fail to see that.

Well sure one can put the balloon behind an IOMMU. If will shuffle PFN lists through a shared page. Problem is, you can't run an untrusted driver with it since if you do it can corrupt guest memory. And VIRTIO_F_IOMMU_PLATFORM so far meant that you can run
a userspace driver.

Maybe we need a separate feature bit for this kind of thing where you assume the driver is trusted? Such a bit - unlike VIRTIO_F_IOMMU_PLATFORM - would allow legacy guests ...

Signature · 1 line
MST
#42 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 10 lines · Michael S. Tsirkin
Quoted 3 lines · David Hildenbrand

Why does the balloon driver not support VIRTIO_F_IOMMU_PLATFORM? It is absolutely not clear to me. The introducing commit mentioned that it "bypasses DMA". I fail to see that.

Well sure one can put the balloon behind an IOMMU. If will shuffle PFN lists through a shared page. Problem is, you can't run an untrusted driver with it since if you do it can corrupt guest memory. And VIRTIO_F_IOMMU_PLATFORM so far meant that you can run
a userspace driver.

Just to clarify: Is it sufficient to clear VIRTIO_F_IOMMU_PLATFORM in the *guest kernel driver* to prohibit *guest userspace drivers*? I would have thought we would have to disallow on the hypervisor/device side. (no expert on user space drivers, especially how they detect/enable/access virtio devices)

Quoted 4 lines

Maybe we need a separate feature bit for this kind of thing where you assume the driver is trusted? Such a bit - unlike VIRTIO_F_IOMMU_PLATFORM - would allow legacy guests ...

Let's take virtio-mem as an example. You cannot zap memory outside of the scope of a virtio-mem device. So I assume having a user space driver would be ok (although most probably of limited use :) )?

Still, for virtio-mem, special s390x handling, similar to virtio-balloon
- (un)sharing of pages - would have to be performed.

So some feature bits to cleanly separate the different limitations would be great. At least in regard to s390x, I guess we don't have to worry too much about legacy guests.

Signature · 3 lines
Thanks,

David / dhildenb
#43 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 6 lines · David Hildenbrand
Quoted 5 lines · Michael S. Tsirkin
Quoted 3 lines · David Hildenbrand

Why does the balloon driver not support VIRTIO_F_IOMMU_PLATFORM? It is absolutely not clear to me. The introducing commit mentioned that it "bypasses DMA". I fail to see that.

Quoted 8 lines
Quoted 5 lines

Well sure one can put the balloon behind an IOMMU. If will shuffle PFN lists through a shared page. Problem is, you can't run an untrusted driver with it since if you do it can corrupt guest memory. And VIRTIO_F_IOMMU_PLATFORM so far meant that you can run
a userspace driver.

Just to clarify: Is it sufficient to clear VIRTIO_F_IOMMU_PLATFORM in the *guest kernel driver* to prohibit *guest userspace drivers*?

No it isn't sufficient.

Quoted 3 lines

I would have thought we would have to disallow on the hypervisor/device side. (no expert on user space drivers, especially how they detect/enable/access virtio devices)

QEMU does exactly this:

static int virtio_validate_features(VirtIODevice *vdev)

{
    VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);

    if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
        !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
        return -EFAULT;
    }

...
}

Quoted 9 lines
Quoted 4 lines

Maybe we need a separate feature bit for this kind of thing where you assume the driver is trusted? Such a bit - unlike VIRTIO_F_IOMMU_PLATFORM - would allow legacy guests ...

Let's take virtio-mem as an example. You cannot zap memory outside of the scope of a virtio-mem device. So I assume having a user space driver would be ok (although most probably of limited use :) )?

Quoted 6 lines

Still, for virtio-mem, special s390x handling, similar to virtio-balloon
- (un)sharing of pages - would have to be performed.

So some feature bits to cleanly separate the different limitations would be great. At least in regard to s390x, I guess we don't have to worry too much about legacy guests.

So if you have the cycles to think through and document how balloon interacts with different access limitations, that would be great!

Quoted 4 lines
Signature · 3 lines
Thanks,

David / dhildenb
#44 lore raw

Re: [PATCH v5 07/18] s390x: protvirt: Inhibit balloon when switching to protected mode

Quoted 14 lines
Quoted 3 lines

I would have thought we would have to disallow on the hypervisor/device side. (no expert on user space drivers, especially how they detect/enable/access virtio devices)

QEMU does exactly this:

static int virtio_validate_features(VirtIODevice *vdev)

{
    VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);

    if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
        !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
        return -EFAULT;
    }
Quoted 2 lines

...
}

Okay, that makes sense. Thanks!

Quoted 14 lines
Quoted 14 lines
Quoted 6 lines

Maybe we need a separate feature bit for this kind of thing where you assume the driver is trusted? Such a bit - unlike VIRTIO_F_IOMMU_PLATFORM - would allow legacy guests ...

Let's take virtio-mem as an example. You cannot zap memory outside of the scope of a virtio-mem device. So I assume having a user space driver would be ok (although most probably of limited use :) )?

Still, for virtio-mem, special s390x handling, similar to virtio-balloon
- (un)sharing of pages - would have to be performed.

Quoted 6 lines
Quoted 3 lines

So some feature bits to cleanly separate the different limitations would be great. At least in regard to s390x, I guess we don't have to worry too much about legacy guests.

So if you have the cycles to think through and document how balloon interacts with different access limitations, that would be great!

I'll add it to my ever-growing todo list. Would be great if Halil could help out thinking how to express the semantics so we can handle PV properly (both, virtio-balloon, but also virtio-mem).

Cheers!

Signature · 3 lines
Thanks,

David / dhildenb
#45 lore raw

PATCH v5 8/18 s390x: protvirt: KVM intercept changes

In reply to Janosch Frank

Secure guests no longer intercept with code 4 for an instruction interception. Instead they have codes 104 and 108 for secure instruction interception and secure instruction notification respectively.

The 104 mirrors the 4 interception.

The 108 is a notification interception to let KVM and QEMU know that something changed and we need to update tracking information or perform specific tasks. It's currently taken for the following instructions:

* stpx (To inform about the changed prefix location)
* sclp (On incorrect SCCB values, so we can inject a IRQ)
* sigp (All but "stop and store status")
* diag308 (Subcodes 0/1)

---
 target/s390x/kvm.c | 4 ++++
 1 file changed, 4 insertions(+)
target/s390x/kvm.c +4 -0
index 1d6fd6a27b..eec0b92479 100644
@@ -115,6 +115,8 @@
 #define ICPT_CPU_STOP                   0x28
 #define ICPT_OPEREXC                    0x2c
 #define ICPT_IO                         0x40
+#define ICPT_PV_INSTR                   0x68
+#define ICPT_PV_INSTR_NOTIFICATION      0x6c
 
 #define NR_LOCAL_IRQS 32
 /*
@@ -1693,6 +1695,8 @@ static int handle_intercept(S390CPU *cpu)
             (long)cs->kvm_run->psw_addr);
     switch (icpt_code) {
         case ICPT_INSTRUCTION:
+        case ICPT_PV_INSTR:
+        case ICPT_PV_INSTR_NOTIFICATION:
             r = handle_instruction(cpu, run);
             break;
         case ICPT_PROGRAM:
Signature · 1 line
2.20.1
#46 lore raw

PATCH v5 9/18 s390x: Add SIDA memory ops

In reply to Janosch Frank

Protected guests save the instruction control blocks in the SIDA instead of QEMU/KVM directly accessing the guest's memory.

Let's introduce new functions to access the SIDA.

---
 linux-headers/linux/kvm.h |  2 ++
 target/s390x/cpu.h        |  7 ++++++-
 target/s390x/kvm.c        | 25 +++++++++++++++++++++++++
 target/s390x/kvm_s390x.h  |  2 ++
 target/s390x/mmu_helper.c | 14 ++++++++++++++
 5 files changed, 49 insertions(+), 1 deletion(-)
linux-headers/linux/kvm.h +2 -0
index e36f761194..c30344ab00 100644
@@ -483,6 +483,8 @@ struct kvm_s390_mem_op {
 /* types for kvm_s390_mem_op->op */
 #define KVM_S390_MEMOP_LOGICAL_READ	0
 #define KVM_S390_MEMOP_LOGICAL_WRITE	1
+#define KVM_S390_MEMOP_SIDA_READ	2
+#define KVM_S390_MEMOP_SIDA_WRITE	3
 /* flags for kvm_s390_mem_op->flags */
 #define KVM_S390_MEMOP_F_CHECK_ONLY		(1ULL << 0)
 #define KVM_S390_MEMOP_F_INJECT_EXCEPTION	(1ULL << 1)
target/s390x/cpu.h +6 -1
index acff5c9a65..c9bbb51214 100644
@@ -824,7 +824,12 @@ int s390_cpu_virt_mem_rw(S390CPU *cpu, vaddr laddr, uint8_t ar, void *hostbuf,
 #define s390_cpu_virt_mem_check_write(cpu, laddr, ar, len)   \
         s390_cpu_virt_mem_rw(cpu, laddr, ar, NULL, len, true)
 void s390_cpu_virt_mem_handle_exc(S390CPU *cpu, uintptr_t ra);
-
+int s390_cpu_pv_mem_rw(S390CPU *cpu, unsigned int offset,  void *hostbuf,
+                       int len, bool is_write);
+#define s390_cpu_pv_mem_read(cpu, offset, dest, len)    \
+        s390_cpu_pv_mem_rw(cpu, offset, dest, len, false)
+#define s390_cpu_pv_mem_write(cpu, offset, dest, len)       \
+        s390_cpu_pv_mem_rw(cpu, offset, dest, len, true)
 
 /* sigp.c */
 int s390_cpu_restart(S390CPU *cpu);
target/s390x/kvm.c +25 -0
index eec0b92479..cdcd538b4f 100644
@@ -154,6 +154,7 @@ static int cap_ri;
 static int cap_gs;
 static int cap_hpage_1m;
 static int cap_vcpu_resets;
+static int cap_protected;
 
 static int active_cmma;
 
@@ -346,6 +347,7 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
     cap_mem_op = kvm_check_extension(s, KVM_CAP_S390_MEM_OP);
     cap_s390_irq = kvm_check_extension(s, KVM_CAP_S390_INJECT_IRQ);
     cap_vcpu_resets = kvm_check_extension(s, KVM_CAP_S390_VCPU_RESETS);
+    cap_protected = kvm_check_extension(s, KVM_CAP_S390_PROTECTED);
 
     if (!kvm_check_extension(s, KVM_CAP_S390_GMAP)
         || !kvm_check_extension(s, KVM_CAP_S390_COW)) {
@@ -846,6 +848,29 @@ int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
     return ret;
 }
 
+int kvm_s390_mem_op_pv(S390CPU *cpu, uint64_t offset, void *hostbuf,
+                       int len, bool is_write)
+{
+    struct kvm_s390_mem_op mem_op = {
+        .sida_offset = offset,
+        .size = len,
+        .op = is_write ? KVM_S390_MEMOP_SIDA_WRITE
+                       : KVM_S390_MEMOP_SIDA_READ,
+        .buf = (uint64_t)hostbuf,
+    };
+    int ret;
+
+    if (!cap_mem_op || !cap_protected) {
+        return -ENOSYS;
+    }
+
+    ret = kvm_vcpu_ioctl(CPU(cpu), KVM_S390_MEM_OP, &mem_op);
+    if (ret < 0) {
+        error_report("KVM_S390_MEM_OP failed: %s", strerror(-ret));
+    }
+    return ret;
+}
+
 /*
  * Legacy layout for s390:
  * Older S390 KVM requires the topmost vma of the RAM to be
target/s390x/kvm_s390x.h +2 -0
index 0b21789796..9c38f6ccce 100644
@@ -19,6 +19,8 @@ void kvm_s390_vcpu_interrupt(S390CPU *cpu, struct kvm_s390_irq *irq);
 void kvm_s390_access_exception(S390CPU *cpu, uint16_t code, uint64_t te_code);
 int kvm_s390_mem_op(S390CPU *cpu, vaddr addr, uint8_t ar, void *hostbuf,
                     int len, bool is_write);
+int kvm_s390_mem_op_pv(S390CPU *cpu, vaddr addr, void *hostbuf, int len,
+                       bool is_write);
 void kvm_s390_program_interrupt(S390CPU *cpu, uint16_t code);
 int kvm_s390_set_cpu_state(S390CPU *cpu, uint8_t cpu_state);
 void kvm_s390_vcpu_interrupt_pre_save(S390CPU *cpu);
target/s390x/mmu_helper.c +14 -0
index c9f3f34750..ec8befbdc8 100644
@@ -474,6 +474,20 @@ static int translate_pages(S390CPU *cpu, vaddr addr, int nr_pages,
     return 0;
 }
 
+int s390_cpu_pv_mem_rw(S390CPU *cpu, unsigned int offset, void *hostbuf,
+                       int len, bool is_write)
+{
+    int ret;
+
+    if (kvm_enabled()) {
+        ret = kvm_s390_mem_op_pv(cpu, offset, hostbuf, len, is_write);
+    } else {
+        /* Protected Virtualization is a KVM/Hardware only feature */
+        g_assert_not_reached();
+    }
+    return ret;
+}
+
 /**
  * s390_cpu_virt_mem_rw:
  * @laddr:     the logical start address
Signature · 1 line
2.20.1
#47 lore raw

PATCH v5 10/18 s390x: protvirt: Move STSI data over SIDAD

In reply to Janosch Frank

For protected guests, we need to put the STSI emulation results into the SIDA, so SIE will write them into the guest at the next entry.

---
 target/s390x/kvm.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)
target/s390x/kvm.c +12 -3
index cdcd538b4f..43fc0c088b 100644
@@ -1797,11 +1797,16 @@ static int handle_tsch(S390CPU *cpu)
 
 static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
 {
+    CPUS390XState *env = &cpu->env;
     SysIB_322 sysib;
     int del;
 
-    if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
-        return;
+    if (env->pv) {
+        s390_cpu_pv_mem_read(cpu, 0, &sysib, sizeof(sysib));
+    } else {
+        if (s390_cpu_virt_mem_read(cpu, addr, ar, &sysib, sizeof(sysib))) {
+            return;
+        }
     }
     /* Shift the stack of Extended Names to prepare for our own data */
     memmove(&sysib.ext_names[1], &sysib.ext_names[0],
@@ -1840,7 +1845,11 @@ static void insert_stsi_3_2_2(S390CPU *cpu, __u64 addr, uint8_t ar)
     /* Insert UUID */
     memcpy(sysib.vm[0].uuid, &qemu_uuid, sizeof(sysib.vm[0].uuid));
 
-    s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
+    if (env->pv) {
+        s390_cpu_pv_mem_write(cpu, 0, &sysib, sizeof(sysib));
+    } else {
+        s390_cpu_virt_mem_write(cpu, addr, ar, &sysib, sizeof(sysib));
+    }
 }
 
 static int handle_stsi(S390CPU *cpu)
Signature · 1 line
2.20.1
#48 lore raw

PATCH v5 11/18 s390x: protvirt: SCLP interpretation

In reply to Janosch Frank

SCLP for a protected guest is done over the SIDAD, so we need to use the s390_cpu_virt_mem_* functions to access the SIDAD instead of guest memory when reading/writing SCBs.

To not confuse the sclp emulation, we set 0x4000 as the SCCB address, since the function that injects the sclp external interrupt would reject a zero sccb address.

---
 hw/s390x/sclp.c         | 17 +++++++++++++++++
 include/hw/s390x/sclp.h |  2 ++
 target/s390x/kvm.c      |  5 +++++
 3 files changed, 24 insertions(+)
hw/s390x/sclp.c +17 -0
index af0bfbc2ec..5136f5fcbe 100644
@@ -193,6 +193,23 @@ static void sclp_execute(SCLPDevice *sclp, SCCB *sccb, uint32_t code)
     }
 }
 
+#define SCLP_PV_DUMMY_ADDR 0x4000
+int sclp_service_call_protected(CPUS390XState *env, uint64_t sccb,
+                                uint32_t code)
+{
+    SCLPDevice *sclp = get_sclp_device();
+    SCLPDeviceClass *sclp_c = SCLP_GET_CLASS(sclp);
+    SCCB work_sccb;
+    hwaddr sccb_len = sizeof(SCCB);
+
+    s390_cpu_pv_mem_read(env_archcpu(env), 0, &work_sccb, sccb_len);
+    sclp_c->execute(sclp, &work_sccb, code);
+    s390_cpu_pv_mem_write(env_archcpu(env), 0, &work_sccb,
+                          be16_to_cpu(work_sccb.h.length));
+    sclp_c->service_interrupt(sclp, SCLP_PV_DUMMY_ADDR);
+    return 0;
+}
+
 int sclp_service_call(CPUS390XState *env, uint64_t sccb, uint32_t code)
 {
     SCLPDevice *sclp = get_sclp_device();
include/hw/s390x/sclp.h +2 -0
index c54413b78c..c0a3faa37d 100644
@@ -217,5 +217,7 @@ void s390_sclp_init(void);
 void sclp_service_interrupt(uint32_t sccb);
 void raise_irq_cpu_hotplug(void);
 int sclp_service_call(CPUS390XState *env, uint64_t sccb, uint32_t code);
+int sclp_service_call_protected(CPUS390XState *env, uint64_t sccb,
+                                uint32_t code);
 
 #endif
target/s390x/kvm.c +5 -0
index 43fc0c088b..a4cbdc5fc6 100644
@@ -1226,6 +1226,11 @@ static void kvm_sclp_service_call(S390CPU *cpu, struct kvm_run *run,
     sccb = env->regs[ipbh0 & 0xf];
     code = env->regs[(ipbh0 & 0xf0) >> 4];
 
+    if (run->s390_sieic.icptcode == ICPT_PV_INSTR) {
+        sclp_service_call_protected(env, sccb, code);
+        return;
+    }
+
     r = sclp_service_call(env, sccb, code);
     if (r < 0) {
         kvm_s390_program_interrupt(cpu, -r);
Signature · 1 line
2.20.1
#49 lore raw

PATCH v5 12/18 s390x: protvirt: Set guest IPL PSW

In reply to Janosch Frank

Handling of CPU reset and setting of the IPL psw from guest storage at offset 0 is done by a Ultravisor call. Let's only fetch it if necessary.

---
 target/s390x/cpu.c | 23 ++++++++++++++---------
 1 file changed, 14 insertions(+), 9 deletions(-)
target/s390x/cpu.c +14 -9
index c4d937e35a..cd53e123f3 100644
@@ -78,16 +78,21 @@ static bool s390_cpu_has_work(CPUState *cs)
 static void s390_cpu_load_normal(CPUState *s)
 {
     S390CPU *cpu = S390_CPU(s);
-    uint64_t spsw = ldq_phys(s->as, 0);
-
-    cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
-    /*
-     * Invert short psw indication, so SIE will report a specification
-     * exception if it was not set.
-     */
-    cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
-    cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
+    CPUS390XState *env = &cpu->env;
+    uint64_t spsw;
 
+    if (!env->pv) {
+        spsw = ldq_phys(s->as, 0);
+        cpu->env.psw.mask = spsw & PSW_MASK_ESA_MASK;
+        /*
+         * Invert short psw indication, so SIE will report a specification
+         * exception if it was not set.
+         */
+        cpu->env.psw.mask ^= PSW_MASK_SHORTPSW;
+        cpu->env.psw.addr = spsw & PSW_MASK_ESA_ADDR;
+    } else {
+        s390_cpu_set_state(S390_CPU_STATE_LOAD, cpu);
+    }
     s390_cpu_set_state(S390_CPU_STATE_OPERATING, cpu);
 }
 #endif
Signature · 1 line
2.20.1
#50 lore raw

PATCH v5 13/18 s390x: protvirt: Move diag 308 data over SIDAD

In reply to Janosch Frank

For protected guests the IPIB is written/read to/from the satellite block, so we need those accesses to go through s390_cpu_pv_mem_read/write().

---
 target/s390x/diag.c | 30 +++++++++++++++++++++++-------
 1 file changed, 23 insertions(+), 7 deletions(-)
target/s390x/diag.c +23 -7
index 840335d40a..266ac78b83 100644
@@ -88,6 +88,7 @@ static int diag308_parm_check(CPUS390XState *env, uint64_t r1, uint64_t addr,
 void handle_diag_308(CPUS390XState *env, uint64_t r1, uint64_t r3, uintptr_t ra)
 {
     CPUState *cs = env_cpu(env);
+    S390CPU *cpu = S390_CPU(cs);
     uint64_t addr =  env->regs[r1];
     uint64_t subcode = env->regs[r3];
     IplParameterBlock *iplb;
@@ -119,13 +120,22 @@ void handle_diag_308(CPUS390XState *env, uint64_t r1, uint64_t r3, uintptr_t ra)
             return;
         }
         iplb = g_new0(IplParameterBlock, 1);
-        cpu_physical_memory_read(addr, iplb, sizeof(iplb->len));
+        if (!env->pv) {
+            cpu_physical_memory_read(addr, iplb, sizeof(iplb->len));
+        } else {
+            s390_cpu_pv_mem_read(cpu, 0, iplb, sizeof(iplb->len));
+        }
+
         if (!iplb_valid_len(iplb)) {
             env->regs[r1 + 1] = DIAG_308_RC_INVALID;
             goto out;
         }
 
-        cpu_physical_memory_read(addr, iplb, be32_to_cpu(iplb->len));
+        if (!env->pv) {
+            cpu_physical_memory_read(addr, iplb, be32_to_cpu(iplb->len));
+        } else {
+            s390_cpu_pv_mem_read(cpu, 0, iplb, be32_to_cpu(iplb->len));
+        }
 
         if (!iplb_valid_ccw(iplb) && !iplb_valid_fcp(iplb) &&
             !(iplb_valid_pv(iplb) && !s390_ipl_pv_check_components(iplb))) {
@@ -137,7 +147,7 @@ void handle_diag_308(CPUS390XState *env, uint64_t r1, uint64_t r3, uintptr_t ra)
         env->regs[r1 + 1] = DIAG_308_RC_OK;
 out:
         g_free(iplb);
-        return;
+        break;
     case DIAG308_STORE:
     case DIAG308_PV_STORE:
         if (diag308_parm_check(env, r1, addr, ra, true)) {
@@ -148,12 +158,18 @@ out:
         } else {
             iplb = s390_ipl_get_iplb();
         }
-        if (iplb) {
-            cpu_physical_memory_write(addr, iplb, be32_to_cpu(iplb->len));
-            env->regs[r1 + 1] = DIAG_308_RC_OK;
-        } else {
+        if (!iplb) {
             env->regs[r1 + 1] = DIAG_308_RC_NO_CONF;
+            return;
         }
+
+        if (!env->pv) {
+            cpu_physical_memory_write(addr, iplb, be32_to_cpu(iplb->len));
+        } else {
+            s390_cpu_pv_mem_write(cpu, 0, iplb, be32_to_cpu(iplb->len));
+        }
+
+        env->regs[r1 + 1] = DIAG_308_RC_OK;
         break;
     case DIAG308_PV_START:
         iplb = s390_ipl_get_iplb_secure();
Signature · 1 line
2.20.1
#51 lore raw

PATCH v5 14/18 s390x: protvirt: Disable address checks for PV guest IO emulation

In reply to Janosch Frank

IO instruction data is routed through SIDAD for protected guests, so adresses do not need to be checked, as this is kernel memory.

---
 target/s390x/ioinst.c | 26 +++++++++++++++++++-------
 1 file changed, 19 insertions(+), 7 deletions(-)
target/s390x/ioinst.c +19 -7
index c437a1d8c6..e4102430aa 100644
@@ -17,6 +17,16 @@
 #include "trace.h"
 #include "hw/s390x/s390-pci-bus.h"
 
+static uint64_t get_address_from_regs(CPUS390XState *env, uint32_t ipb,
+                                      uint8_t *ar)
+{
+    if (env->pv) {
+        *ar = 0;
+        return 0;
+    }
+    return decode_basedisp_s(env, ipb, ar);
+}
+
 int ioinst_disassemble_sch_ident(uint32_t value, int *m, int *cssid, int *ssid,
                                  int *schid)
 {
@@ -114,7 +124,7 @@ void ioinst_handle_msch(S390CPU *cpu, uint64_t reg1, uint32_t ipb, uintptr_t ra)
     CPUS390XState *env = &cpu->env;
     uint8_t ar;
 
-    addr = decode_basedisp_s(env, ipb, &ar);
+    addr = get_address_from_regs(env, ipb, &ar);
     if (addr & 3) {
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         return;
@@ -171,7 +181,7 @@ void ioinst_handle_ssch(S390CPU *cpu, uint64_t reg1, uint32_t ipb, uintptr_t ra)
     CPUS390XState *env = &cpu->env;
     uint8_t ar;
 
-    addr = decode_basedisp_s(env, ipb, &ar);
+    addr = get_address_from_regs(env, ipb, &ar);
     if (addr & 3) {
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         return;
@@ -203,7 +213,7 @@ void ioinst_handle_stcrw(S390CPU *cpu, uint32_t ipb, uintptr_t ra)
     CPUS390XState *env = &cpu->env;
     uint8_t ar;
 
-    addr = decode_basedisp_s(env, ipb, &ar);
+    addr = get_address_from_regs(env, ipb, &ar);
     if (addr & 3) {
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         return;
@@ -234,7 +244,7 @@ void ioinst_handle_stsch(S390CPU *cpu, uint64_t reg1, uint32_t ipb,
     CPUS390XState *env = &cpu->env;
     uint8_t ar;
 
-    addr = decode_basedisp_s(env, ipb, &ar);
+    addr = get_address_from_regs(env, ipb, &ar);
     if (addr & 3) {
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         return;
@@ -303,7 +313,7 @@ int ioinst_handle_tsch(S390CPU *cpu, uint64_t reg1, uint32_t ipb, uintptr_t ra)
         return -EIO;
     }
     trace_ioinst_sch_id("tsch", cssid, ssid, schid);
-    addr = decode_basedisp_s(env, ipb, &ar);
+    addr = get_address_from_regs(env, ipb, &ar);
     if (addr & 3) {
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         return -EIO;
@@ -601,7 +611,7 @@ void ioinst_handle_chsc(S390CPU *cpu, uint32_t ipb, uintptr_t ra)
 {
     ChscReq *req;
     ChscResp *res;
-    uint64_t addr;
+    uint64_t addr = 0;
     int reg;
     uint16_t len;
     uint16_t command;
@@ -610,7 +620,9 @@ void ioinst_handle_chsc(S390CPU *cpu, uint32_t ipb, uintptr_t ra)
 
     trace_ioinst("chsc");
     reg = (ipb >> 20) & 0x00f;
-    addr = env->regs[reg];
+    if (!env->pv) {
+        addr = env->regs[reg];
+    }
     /* Page boundary? */
     if (addr & 0xfff) {
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
Signature · 1 line
2.20.1
#52 lore raw

PATCH v5 15/18 s390x: protvirt: Move IO control structures over SIDA

In reply to Janosch Frank

For protected guests, we need to put the IO emulation results into the SIDA, so SIE will write them into the guest at the next entry.

---
 target/s390x/ioinst.c | 87 ++++++++++++++++++++++++++++++-------------
 1 file changed, 61 insertions(+), 26 deletions(-)
target/s390x/ioinst.c +61 -26
index e4102430aa..330b04d79a 100644
@@ -129,9 +129,13 @@ void ioinst_handle_msch(S390CPU *cpu, uint64_t reg1, uint32_t ipb, uintptr_t ra)
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         return;
     }
-    if (s390_cpu_virt_mem_read(cpu, addr, ar, &schib, sizeof(schib))) {
-        s390_cpu_virt_mem_handle_exc(cpu, ra);
-        return;
+    if (env->pv) {
+        s390_cpu_pv_mem_read(cpu, addr, &schib, sizeof(schib));
+    } else {
+        if (s390_cpu_virt_mem_read(cpu, addr, ar, &schib, sizeof(schib))) {
+            s390_cpu_virt_mem_handle_exc(cpu, ra);
+            return;
+        }
     }
     if (ioinst_disassemble_sch_ident(reg1, &m, &cssid, &ssid, &schid) ||
         !ioinst_schib_valid(&schib)) {
@@ -186,9 +190,13 @@ void ioinst_handle_ssch(S390CPU *cpu, uint64_t reg1, uint32_t ipb, uintptr_t ra)
         s390_program_interrupt(env, PGM_SPECIFICATION, ra);
         return;
     }
-    if (s390_cpu_virt_mem_read(cpu, addr, ar, &orig_orb, sizeof(orb))) {
-        s390_cpu_virt_mem_handle_exc(cpu, ra);
-        return;
+    if (env->pv) {
+        s390_cpu_pv_mem_read(cpu, addr, &orig_orb, sizeof(orb));
+    } else {
+        if (s390_cpu_virt_mem_read(cpu, addr, ar, &orig_orb, sizeof(orb))) {
+            s390_cpu_virt_mem_handle_exc(cpu, ra);
+            return;
+        }
     }
     copy_orb_from_guest(&orb, &orig_orb);
     if (ioinst_disassemble_sch_ident(reg1, &m, &cssid, &ssid, &schid) ||
@@ -222,14 +230,19 @@ void ioinst_handle_stcrw(S390CPU *cpu, uint32_t ipb, uintptr_t ra)
     cc = css_do_stcrw(&crw);
     /* 0 - crw stored, 1 - zeroes stored */
 
-    if (s390_cpu_virt_mem_write(cpu, addr, ar, &crw, sizeof(crw)) == 0) {
+    if (env->pv) {
+        s390_cpu_pv_mem_write(cpu, addr, &crw, sizeof(crw));
         setcc(cpu, cc);
     } else {
-        if (cc == 0) {
-            /* Write failed: requeue CRW since STCRW is suppressing */
-            css_undo_stcrw(&crw);
+        if (s390_cpu_virt_mem_write(cpu, addr, ar, &crw, sizeof(crw)) == 0) {
+            setcc(cpu, cc);
+        } else {
+            if (cc == 0) {
+                /* Write failed: requeue CRW since STCRW is suppressing */
+                css_undo_stcrw(&crw);
+            }
+            s390_cpu_virt_mem_handle_exc(cpu, ra);
         }
-        s390_cpu_virt_mem_handle_exc(cpu, ra);
     }
 }
 
@@ -251,6 +264,9 @@ void ioinst_handle_stsch(S390CPU *cpu, uint64_t reg1, uint32_t ipb,
     }
 
     if (ioinst_disassemble_sch_ident(reg1, &m, &cssid, &ssid, &schid)) {
+        if (env->pv) {
+            return;
+        }
         /*
          * As operand exceptions have a lower priority than access exceptions,
          * we check whether the memory area is writeable (injecting the
@@ -283,14 +299,19 @@ void ioinst_handle_stsch(S390CPU *cpu, uint64_t reg1, uint32_t ipb,
         }
     }
     if (cc != 3) {
-        if (s390_cpu_virt_mem_write(cpu, addr, ar, &schib,
-                                    sizeof(schib)) != 0) {
-            s390_cpu_virt_mem_handle_exc(cpu, ra);
-            return;
+        if (env->pv) {
+            s390_cpu_pv_mem_write(cpu, addr, &schib, sizeof(schib));
+        } else {
+            if (s390_cpu_virt_mem_write(cpu, addr, ar, &schib,
+                                        sizeof(schib)) != 0) {
+                s390_cpu_virt_mem_handle_exc(cpu, ra);
+                return;
+            }
         }
     } else {
         /* Access exceptions have a higher priority than cc3 */
-        if (s390_cpu_virt_mem_check_write(cpu, addr, ar, sizeof(schib)) != 0) {
+        if (!env->pv &&
+            s390_cpu_virt_mem_check_write(cpu, addr, ar, sizeof(schib)) != 0) {
             s390_cpu_virt_mem_handle_exc(cpu, ra);
             return;
         }
@@ -327,15 +348,20 @@ int ioinst_handle_tsch(S390CPU *cpu, uint64_t reg1, uint32_t ipb, uintptr_t ra)
     }
     /* 0 - status pending, 1 - not status pending, 3 - not operational */
     if (cc != 3) {
-        if (s390_cpu_virt_mem_write(cpu, addr, ar, &irb, irb_len) != 0) {
-            s390_cpu_virt_mem_handle_exc(cpu, ra);
-            return -EFAULT;
+        if (env->pv) {
+            s390_cpu_pv_mem_write(cpu, addr, &irb, irb_len);
+        } else {
+            if (s390_cpu_virt_mem_write(cpu, addr, ar, &irb, irb_len) != 0) {
+                s390_cpu_virt_mem_handle_exc(cpu, ra);
+                return -EFAULT;
+            }
         }
         css_do_tsch_update_subch(sch);
     } else {
         irb_len = sizeof(irb) - sizeof(irb.emw);
         /* Access exceptions have a higher priority than cc3 */
-        if (s390_cpu_virt_mem_check_write(cpu, addr, ar, irb_len) != 0) {
+        if (!env->pv &&
+            s390_cpu_virt_mem_check_write(cpu, addr, ar, irb_len) != 0) {
             s390_cpu_virt_mem_handle_exc(cpu, ra);
             return -EFAULT;
         }
@@ -633,9 +659,13 @@ void ioinst_handle_chsc(S390CPU *cpu, uint32_t ipb, uintptr_t ra)
      * present CHSC sub-handlers ... if we ever need more, we should take
      * care of req->len here first.
      */
-    if (s390_cpu_virt_mem_read(cpu, addr, reg, buf, sizeof(ChscReq))) {
-        s390_cpu_virt_mem_handle_exc(cpu, ra);
-        return;
+    if (env->pv) {
+        s390_cpu_pv_mem_read(cpu, addr, buf, sizeof(ChscReq));
+    } else {
+        if (s390_cpu_virt_mem_read(cpu, addr, reg, buf, sizeof(ChscReq))) {
+            s390_cpu_virt_mem_handle_exc(cpu, ra);
+            return;
+        }
     }
     req = (ChscReq *)buf;
     len = be16_to_cpu(req->len);
@@ -666,11 +696,16 @@ void ioinst_handle_chsc(S390CPU *cpu, uint32_t ipb, uintptr_t ra)
         break;
     }
 
-    if (!s390_cpu_virt_mem_write(cpu, addr + len, reg, res,
-                                 be16_to_cpu(res->len))) {
+    if (env->pv) {
+        s390_cpu_pv_mem_write(cpu, addr + len, res, be16_to_cpu(res->len));
         setcc(cpu, 0);    /* Command execution complete */
     } else {
-        s390_cpu_virt_mem_handle_exc(cpu, ra);
+        if (!s390_cpu_virt_mem_write(cpu, addr + len, reg, res,
+                                     be16_to_cpu(res->len))) {
+            setcc(cpu, 0);    /* Command execution complete */
+        } else {
+            s390_cpu_virt_mem_handle_exc(cpu, ra);
+        }
     }
 }
 
Signature · 1 line
2.20.1
#53 lore raw

PATCH v5 16/18 s390x: protvirt: Handle SIGP store status correctly

In reply to Janosch Frank

Status storing is not done by QEMU anymore, but is handled by SIE.

---
 target/s390x/helper.c | 4 ++++
 target/s390x/sigp.c   | 1 +
 2 files changed, 5 insertions(+)
target/s390x/helper.c +4 -0
index 6808dfda01..ec3185151f 100644
@@ -246,6 +246,10 @@ int s390_store_status(S390CPU *cpu, hwaddr addr, bool store_arch)
     hwaddr len = sizeof(*sa);
     int i;
 
+    if (cpu->env.pv) {
+        return 0;
+    }
+
     sa = cpu_physical_memory_map(addr, &len, 1);
     if (!sa) {
         return -EFAULT;
target/s390x/sigp.c +1 -0
index c604f17710..e1c8071464 100644
@@ -497,6 +497,7 @@ void do_stop_interrupt(CPUS390XState *env)
     if (s390_cpu_set_state(S390_CPU_STATE_STOPPED, cpu) == 0) {
         qemu_system_shutdown_request(SHUTDOWN_CAUSE_GUEST_SHUTDOWN);
     }
+    /* Storing will occur on next SIE entry for protected VMs */
     if (cpu->env.sigp_order == SIGP_STOP_STORE_STATUS) {
         s390_store_status(cpu, S390_STORE_STATUS_DEF_ADDR, true);
     }
Signature · 1 line
2.20.1
#54 lore raw

PATCH v5 17/18 s390x: Add unpack facility feature to GA1

In reply to Janosch Frank

From: Christian Borntraeger <[email protected]>

The unpack facility is an indication that diagnose 308 subcodes 8-10 are available to the guest. That means, that the guest can put itself into protected mode.

Once it is in protected mode, the hardware stops any attempt of VM introspection by the hypervisor.

Some features are currently not supported in protected mode:
* Passthrough devices
* Migration
* Huge page backings

---
 target/s390x/gen-features.c | 1 +
 target/s390x/kvm.c          | 5 +++++
 2 files changed, 6 insertions(+)
target/s390x/gen-features.c +1 -0
index 6278845b12..8ddeebc544 100644
@@ -562,6 +562,7 @@ static uint16_t full_GEN15_GA1[] = {
     S390_FEAT_GROUP_MSA_EXT_9,
     S390_FEAT_GROUP_MSA_EXT_9_PCKMO,
     S390_FEAT_ETOKEN,
+    S390_FEAT_UNPACK,
 };
 
 /* Default features (in order of release)
target/s390x/kvm.c +5 -0
index a4cbdc5fc6..bf807793bc 100644
@@ -2396,6 +2396,11 @@ void kvm_s390_get_host_cpu_model(S390CPUModel *model, Error **errp)
         clear_bit(S390_FEAT_BPB, model->features);
     }
 
+    /* we do have the IPL enhancements */
+    if (cap_protected) {
+        set_bit(S390_FEAT_UNPACK, model->features);
+    }
+
     /* We emulate a zPCI bus and AEN, therefore we don't need HW support */
     set_bit(S390_FEAT_ZPCI, model->features);
     set_bit(S390_FEAT_ADAPTER_EVENT_NOTIFICATION, model->features);
Signature · 1 line
2.20.1
#55 lore raw

PATCH v5 18/18 docs: Add protvirt docs

In reply to Janosch Frank

Lets add some documentation for the Protected VM functionality.

---
 docs/system/index.rst    |  1 +
 docs/system/protvirt.rst | 57 ++++++++++++++++++++++++++++++++++++++++
 2 files changed, 58 insertions(+)
 create mode 100644 docs/system/protvirt.rst
docs/system/index.rst +1 -0
index 1a4b2c82ac..d2dc63b973 100644
@@ -16,3 +16,4 @@ Contents:
 
    qemu-block-drivers
    vfio-ap
+   protvirt
docs/system/protvirt.rst new file +57 -0

57 lines not on this page · raw message

Signature · 1 line
2.20.1
#56 lore raw
In reply to Janosch Frank
Quoted 72 lines · On Wed, 26 Feb 2020 07:20:20 -0500 Janosch Frank <[email protected]> wrote:

Most of the QEMU changes for PV are related to the new IPL type with subcodes 8 - 10 and the execution of the necessary Ultravisor calls to IPL secure guests. Note that we can only boot into secure mode from normal mode, i.e. stfle 161 is not active in secure mode.

The other changes related to data gathering for emulation and disabling addressing checks in secure mode, as well as CPU resets.

v5:
* Added balloon inhibition
* Moved docs into docs/system
* Some more enable/disable changes
* Moved enablement/disablement of pv in separate functions
* Some review fixes

v4:
* Sync with KVM changes
* Review changes

V3:
* Use dedicated functions to access SIDA
* Smaller cleanups and segfault fixes
* Error reporting for Ultravisor calls
* Inject of RC of diag308 subcode 10 fails

V2:
* Split out cleanups
* Internal PV state tracking
* Review feedback

Christian Borntraeger (1):
s390x: Add unpack facility feature to GA1

Janosch Frank (17):
s390x: Use constant for ESA PSW address
Sync pv
s390x: protvirt: Add diag308 subcodes 8 - 10
s390x: protvirt: Support unpack facility
s390x: protvirt: Add migration blocker
s390x: protvirt: Handle diag 308 subcodes 0,1,3,4
s390x: protvirt: Inhibit balloon when switching to protected mode
s390x: protvirt: KVM intercept changes
s390x: Add SIDA memory ops
s390x: protvirt: Move STSI data over SIDAD
s390x: protvirt: SCLP interpretation
s390x: protvirt: Set guest IPL PSW
s390x: protvirt: Move diag 308 data over SIDAD
s390x: protvirt: Disable address checks for PV guest IO emulation
s390x: protvirt: Move IO control structures over SIDA
s390x: protvirt: Handle SIGP store status correctly
docs: Add protvirt docs

 docs/system/index.rst               |   1 +
 docs/system/protvirt.rst            |  57 +++++++++++
 hw/s390x/Makefile.objs              |   1 +
 hw/s390x/ipl.c                      |  82 +++++++++++++++-
 hw/s390x/ipl.h                      |  33 +++++++
 hw/s390x/pv.c                       | 106 +++++++++++++++++++++
 hw/s390x/pv.h                       |  34 +++++++
 hw/s390x/s390-virtio-ccw.c          | 143 +++++++++++++++++++++++++++-
 hw/s390x/sclp.c                     |  17 ++++
 include/hw/s390x/s390-virtio-ccw.h  |   1 +
 include/hw/s390x/sclp.h             |   2 +
 linux-headers/linux/kvm.h           |  45 ++++++++-
 target/s390x/cpu.c                  |  27 ++++--
 target/s390x/cpu.h                  |   9 +-
 target/s390x/cpu_features_def.inc.h |   1 +
 target/s390x/diag.c                 |  61 ++++++++++--
 target/s390x/gen-features.c         |   1 +
 target/s390x/helper.c               |   4 +
 target/s390x/ioinst.c               | 113 +++++++++++++++-------
Quoted 9 lines
 target/s390x/kvm.c                  |  54 ++++++++++-
 target/s390x/kvm_s390x.h            |   2 +
 target/s390x/mmu_helper.c           |  14 +++
 target/s390x/sigp.c                 |   1 +
 23 files changed, 744 insertions(+), 65 deletions(-)
 create mode 100644 docs/system/protvirt.rst
 create mode 100644 hw/s390x/pv.c
 create mode 100644 hw/s390x/pv.h

I guess this is on top of my published s390-next branch? (At least I can apply it there...)

When I try to build the result on x86 with clang, I get

In file included from /home/cohuck/git/qemu/hw/s390x/ipl.c:36:
/home/cohuck/git/qemu/hw/s390x/pv.h:25:5: error: no previous prototype for
function 's390_pv_vm_enable' [-Werror,-Wmissing-prototypes]
int s390_pv_vm_enable(void) { return 0; }

    ^
/home/cohuck/git/qemu/hw/s390x/pv.h:26:6: error: no previous prototype for
      function 's390_pv_vm_disable' [-Werror,-Wmissing-prototypes]
void s390_pv_vm_disable(void) {}
     ^
/home/cohuck/git/qemu/hw/s390x/pv.h:27:5: error: no previous prototype for
      function 's390_pv_set_sec_parms' [-Werror,-Wmissing-prototypes]
int s390_pv_set_sec_parms(uint64_t origin, uint64_t length) { return 0; }
    ^
/home/cohuck/git/qemu/hw/s390x/pv.h:28:5: error: no previous prototype for
      function 's390_pv_unpack' [-Werror,-Wmissing-prototypes]
int s390_pv_unpack(uint64_t addr, uint64_t size, uint64_t tweak) { return 0; }
    ^
/home/cohuck/git/qemu/hw/s390x/pv.h:29:6: error: no previous prototype for
      function 's390_pv_perf_clear_reset' [-Werror,-Wmissing-prototypes]
void s390_pv_perf_clear_reset(void) {}
     ^
/home/cohuck/git/qemu/hw/s390x/pv.h:30:5: error: no previous prototype for
      function 's390_pv_verify' [-Werror,-Wmissing-prototypes]
int s390_pv_verify(void) { return 0; }
    ^
/home/cohuck/git/qemu/hw/s390x/pv.h:31:6: error: no previous prototype for
      function 's390_pv_unshare' [-Werror,-Wmissing-prototypes]
void s390_pv_unshare(void) {}
     ^

and

/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:355:9: error: variable 'rc' is

      used uninitialized whenever 'if' condition is true
      [-Werror,-Wsometimes-uninitialized]
    if (local_err) {
        ^~~~~~~~~
/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:391:12: note: uninitialized use
      occurs here
    return rc;
           ^~
/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:355:5: note: remove the 'if' if
      its condition is always false
    if (local_err) {
    ^~~~~~~~~~~~~~~~
/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:347:11: note: initialize the
      variable 'rc' to silence this warning
    int rc;
          ^
           = 0
/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:396:26: error: incomplete
      definition of type 'struct kvm_run'
    int r1 = (cs->kvm_run->s390_sieic.ipa & 0x00f0) >> 4;
              ~~~~~~~~~~~^
/home/cohuck/git/qemu/include/hw/core/cpu.h:265:8: note: forward declaration of
      'struct kvm_run'
struct kvm_run;
       ^
#57 lore raw
Quoted 85 lines · Cornelia Huck
Quoted 72 lines · On Wed, 26 Feb 2020 07:20:20 -0500 Janosch Frank <[email protected]> wrote:

Most of the QEMU changes for PV are related to the new IPL type with subcodes 8 - 10 and the execution of the necessary Ultravisor calls to IPL secure guests. Note that we can only boot into secure mode from normal mode, i.e. stfle 161 is not active in secure mode.

The other changes related to data gathering for emulation and disabling addressing checks in secure mode, as well as CPU resets.

v5:
* Added balloon inhibition
* Moved docs into docs/system
* Some more enable/disable changes
* Moved enablement/disablement of pv in separate functions
* Some review fixes

v4:
* Sync with KVM changes
* Review changes

V3:
* Use dedicated functions to access SIDA
* Smaller cleanups and segfault fixes
* Error reporting for Ultravisor calls
* Inject of RC of diag308 subcode 10 fails

V2:
* Split out cleanups
* Internal PV state tracking
* Review feedback

Christian Borntraeger (1):
s390x: Add unpack facility feature to GA1

Janosch Frank (17):
s390x: Use constant for ESA PSW address
Sync pv
s390x: protvirt: Add diag308 subcodes 8 - 10
s390x: protvirt: Support unpack facility
s390x: protvirt: Add migration blocker
s390x: protvirt: Handle diag 308 subcodes 0,1,3,4
s390x: protvirt: Inhibit balloon when switching to protected mode
s390x: protvirt: KVM intercept changes
s390x: Add SIDA memory ops
s390x: protvirt: Move STSI data over SIDAD
s390x: protvirt: SCLP interpretation
s390x: protvirt: Set guest IPL PSW
s390x: protvirt: Move diag 308 data over SIDAD
s390x: protvirt: Disable address checks for PV guest IO emulation
s390x: protvirt: Move IO control structures over SIDA
s390x: protvirt: Handle SIGP store status correctly
docs: Add protvirt docs

 docs/system/index.rst               |   1 +
 docs/system/protvirt.rst            |  57 +++++++++++
 hw/s390x/Makefile.objs              |   1 +
 hw/s390x/ipl.c                      |  82 +++++++++++++++-
 hw/s390x/ipl.h                      |  33 +++++++
 hw/s390x/pv.c                       | 106 +++++++++++++++++++++
 hw/s390x/pv.h                       |  34 +++++++
 hw/s390x/s390-virtio-ccw.c          | 143 +++++++++++++++++++++++++++-
 hw/s390x/sclp.c                     |  17 ++++
 include/hw/s390x/s390-virtio-ccw.h  |   1 +
 include/hw/s390x/sclp.h             |   2 +
 linux-headers/linux/kvm.h           |  45 ++++++++-
 target/s390x/cpu.c                  |  27 ++++--
 target/s390x/cpu.h                  |   9 +-
 target/s390x/cpu_features_def.inc.h |   1 +
 target/s390x/diag.c                 |  61 ++++++++++--
 target/s390x/gen-features.c         |   1 +
 target/s390x/helper.c               |   4 +
 target/s390x/ioinst.c               | 113 +++++++++++++++-------
Quoted 9 lines
 target/s390x/kvm.c                  |  54 ++++++++++-
 target/s390x/kvm_s390x.h            |   2 +
 target/s390x/mmu_helper.c           |  14 +++
 target/s390x/sigp.c                 |   1 +
 23 files changed, 744 insertions(+), 65 deletions(-)
 create mode 100644 docs/system/protvirt.rst
 create mode 100644 hw/s390x/pv.c
 create mode 100644 hw/s390x/pv.h
Quoted 2 lines

I guess this is on top of my published s390-next branch? (At least I can apply it there...)

Yes, you can also pick from here:
https://github.com/frankjaa/qemu/commits/protvirt

Quoted 2 lines

When I try to build the result on x86 with clang, I get

I forgot to add static inline to the non kvm functions... The latest github branch state has that fixed

Quoted 31 lines

In file included from /home/cohuck/git/qemu/hw/s390x/ipl.c:36:
/home/cohuck/git/qemu/hw/s390x/pv.h:25:5: error: no previous prototype for
function 's390_pv_vm_enable' [-Werror,-Wmissing-prototypes]
int s390_pv_vm_enable(void) { return 0; }

    ^
/home/cohuck/git/qemu/hw/s390x/pv.h:26:6: error: no previous prototype for
      function 's390_pv_vm_disable' [-Werror,-Wmissing-prototypes]
void s390_pv_vm_disable(void) {}
     ^
/home/cohuck/git/qemu/hw/s390x/pv.h:27:5: error: no previous prototype for
      function 's390_pv_set_sec_parms' [-Werror,-Wmissing-prototypes]
int s390_pv_set_sec_parms(uint64_t origin, uint64_t length) { return 0; }
    ^
/home/cohuck/git/qemu/hw/s390x/pv.h:28:5: error: no previous prototype for
      function 's390_pv_unpack' [-Werror,-Wmissing-prototypes]
int s390_pv_unpack(uint64_t addr, uint64_t size, uint64_t tweak) { return 0; }
    ^
/home/cohuck/git/qemu/hw/s390x/pv.h:29:6: error: no previous prototype for
      function 's390_pv_perf_clear_reset' [-Werror,-Wmissing-prototypes]
void s390_pv_perf_clear_reset(void) {}
     ^
/home/cohuck/git/qemu/hw/s390x/pv.h:30:5: error: no previous prototype for
      function 's390_pv_verify' [-Werror,-Wmissing-prototypes]
int s390_pv_verify(void) { return 0; }
    ^
/home/cohuck/git/qemu/hw/s390x/pv.h:31:6: error: no previous prototype for
      function 's390_pv_unshare' [-Werror,-Wmissing-prototypes]
void s390_pv_unshare(void) {}
     ^
Quoted 1 line

and

I'll look into it

Quoted 28 lines

/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:355:9: error: variable 'rc' is

      used uninitialized whenever 'if' condition is true
      [-Werror,-Wsometimes-uninitialized]
    if (local_err) {
        ^~~~~~~~~
/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:391:12: note: uninitialized use
      occurs here
    return rc;
           ^~
/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:355:5: note: remove the 'if' if
      its condition is always false
    if (local_err) {
    ^~~~~~~~~~~~~~~~
/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:347:11: note: initialize the
      variable 'rc' to silence this warning
    int rc;
          ^
           = 0
/home/cohuck/git/qemu/hw/s390x/s390-virtio-ccw.c:396:26: error: incomplete
      definition of type 'struct kvm_run'
    int r1 = (cs->kvm_run->s390_sieic.ipa & 0x00f0) >> 4;
              ~~~~~~~~~~~^
/home/cohuck/git/qemu/include/hw/core/cpu.h:265:8: note: forward declaration of
      'struct kvm_run'
struct kvm_run;
       ^
#58 lore raw

PATCH 1/1 pc-bios: s390x: Save iplb location in lowcore

In reply to Janosch Frank

The POP states that for a list directed IPL the IPLB is stored into memory by the machine loader and its address is stored at offset 0x14 of the lowcore.

ZIPL currently uses the address in offset 0x14 to access the IPLB and acquire flags about secure boot. If the IPLB address points into memory which has an unsupported mix of flags set, ZIPL will panic instead of booting the OS.

As the lowcore can have quite a high entropy for a guest that did drop out of protected mode (i.e. rebooted) we encountered the ZIPL panic quite often.

---
 pc-bios/s390-ccw/jump2ipl.c  |  1 +
 pc-bios/s390-ccw/main.c      |  8 +++++++-
 pc-bios/s390-ccw/netmain.c   |  1 +
 pc-bios/s390-ccw/s390-arch.h | 10 ++++++++--
 pc-bios/s390-ccw/s390-ccw.h  |  1 +
 5 files changed, 18 insertions(+), 3 deletions(-)
pc-bios/s390-ccw/jump2ipl.c +1 -0
index da13c43cc0..4eba2510b0 100644
@@ -35,6 +35,7 @@ void jump_to_IPL_code(uint64_t address)
 {
     /* store the subsystem information _after_ the bootmap was loaded */
     write_subsystem_identification();
+    write_iplb_location();
 
     /* prevent unknown IPL types in the guest */
     if (iplb.pbt == S390_IPL_TYPE_QEMU_SCSI) {
pc-bios/s390-ccw/main.c +7 -1
index a21b386280..4e65b411e1 100644
@@ -9,6 +9,7 @@
  */
 
 #include "libc.h"
+#include "helper.h"
 #include "s390-arch.h"
 #include "s390-ccw.h"
 #include "cio.h"
@@ -22,7 +23,7 @@ QemuIplParameters qipl;
 IplParameterBlock iplb __attribute__((__aligned__(PAGE_SIZE)));
 static bool have_iplb;
 static uint16_t cutype;
-LowCore const *lowcore; /* Yes, this *is* a pointer to address 0 */
+LowCore *lowcore; /* Yes, this *is* a pointer to address 0 */
 
 #define LOADPARM_PROMPT "PROMPT  "
 #define LOADPARM_EMPTY  "        "
@@ -42,6 +43,11 @@ void write_subsystem_identification(void)
     *zeroes = 0;
 }
 
+void write_iplb_location(void)
+{
+    lowcore->ptr_iplb = ptr2u32(&iplb);
+}
+
 void panic(const char *string)
 {
     sclp_print(string);
pc-bios/s390-ccw/netmain.c +1 -0
index f2dcc01e27..309ffa30d9 100644
@@ -40,6 +40,7 @@
 #define DEFAULT_TFTP_RETRIES 20
 
 extern char _start[];
+void write_iplb_location(void) {}
 
 #define KERNEL_ADDR             ((void *)0L)
 #define KERNEL_MAX_SIZE         ((long)_start)
pc-bios/s390-ccw/s390-arch.h +8 -2
index 504fc7c2f0..5f36361c02 100644
@@ -36,7 +36,13 @@ typedef struct LowCore {
     /* prefix area: defined by architecture */
     PSWLegacy       ipl_psw;                  /* 0x000 */
     uint32_t        ccw1[2];                  /* 0x008 */
-    uint32_t        ccw2[2];                  /* 0x010 */
+    union {
+        uint32_t        ccw2[2];                  /* 0x010 */
+        struct {
+            uint32_t reserved10;
+            uint32_t ptr_iplb;
+        };
+    };
     uint8_t         pad1[0x80 - 0x18];        /* 0x018 */
     uint32_t        ext_params;               /* 0x080 */
     uint16_t        cpu_addr;                 /* 0x084 */
@@ -85,7 +91,7 @@ typedef struct LowCore {
     PSW             io_new_psw;               /* 0x1f0 */
 } __attribute__((packed, aligned(8192))) LowCore;
 
-extern LowCore const *lowcore;
+extern LowCore *lowcore;
 
 static inline void set_prefix(uint32_t address)
 {
pc-bios/s390-ccw/s390-ccw.h +1 -0
index 11bce7d73c..21f27e7990 100644
@@ -57,6 +57,7 @@ void consume_io_int(void);
 /* main.c */
 void panic(const char *string);
 void write_subsystem_identification(void);
+void write_iplb_location(void);
 extern char stack[PAGE_SIZE * 8] __attribute__((__aligned__(PAGE_SIZE)));
 unsigned int get_loadparm_index(void);
 
Signature · 1 line
2.20.1
#59 lore raw

Re: [PATCH] pc-bios: s390x: Save iplb location in lowcore

Quoted 4 lines · Janosch Frank

The POP states that for a list directed IPL the IPLB is stored into memory by the machine loader and its address is stored at offset 0x14 of the lowcore.

Quoted 8 lines

ZIPL currently uses the address in offset 0x14 to access the IPLB and acquire flags about secure boot. If the IPLB address points into memory which has an unsupported mix of flags set, ZIPL will panic instead of booting the OS.

As the lowcore can have quite a high entropy for a guest that did drop out of protected mode (i.e. rebooted) we encountered the ZIPL panic quite often.

How did this ever work? Or does this only become relevant with secure boot?

Fixes ? Or has this always been broken?
Quoted 101 lines
---
 pc-bios/s390-ccw/jump2ipl.c  |  1 +
 pc-bios/s390-ccw/main.c      |  8 +++++++-
 pc-bios/s390-ccw/netmain.c   |  1 +
 pc-bios/s390-ccw/s390-arch.h | 10 ++++++++--
 pc-bios/s390-ccw/s390-ccw.h  |  1 +
 5 files changed, 18 insertions(+), 3 deletions(-)
pc-bios/s390-ccw/jump2ipl.c +1 -0
index da13c43cc0..4eba2510b0 100644
@@ -35,6 +35,7 @@ void jump_to_IPL_code(uint64_t address)
 {
     /* store the subsystem information _after_ the bootmap was loaded */
     write_subsystem_identification();
+    write_iplb_location();
 
     /* prevent unknown IPL types in the guest */
     if (iplb.pbt == S390_IPL_TYPE_QEMU_SCSI) {
pc-bios/s390-ccw/main.c +7 -1
index a21b386280..4e65b411e1 100644
@@ -9,6 +9,7 @@
  */
 
 #include "libc.h"
+#include "helper.h"
 #include "s390-arch.h"
 #include "s390-ccw.h"
 #include "cio.h"
@@ -22,7 +23,7 @@ QemuIplParameters qipl;
 IplParameterBlock iplb __attribute__((__aligned__(PAGE_SIZE)));
 static bool have_iplb;
 static uint16_t cutype;
-LowCore const *lowcore; /* Yes, this *is* a pointer to address 0 */
+LowCore *lowcore; /* Yes, this *is* a pointer to address 0 */
 
 #define LOADPARM_PROMPT "PROMPT  "
 #define LOADPARM_EMPTY  "        "
@@ -42,6 +43,11 @@ void write_subsystem_identification(void)
     *zeroes = 0;
 }
 
+void write_iplb_location(void)
+{
+    lowcore->ptr_iplb = ptr2u32(&iplb);
+}
+
 void panic(const char *string)
 {
     sclp_print(string);
pc-bios/s390-ccw/netmain.c +1 -0
index f2dcc01e27..309ffa30d9 100644
@@ -40,6 +40,7 @@
 #define DEFAULT_TFTP_RETRIES 20
 
 extern char _start[];
+void write_iplb_location(void) {}
 
 #define KERNEL_ADDR             ((void *)0L)
 #define KERNEL_MAX_SIZE         ((long)_start)
pc-bios/s390-ccw/s390-arch.h +8 -2
index 504fc7c2f0..5f36361c02 100644
@@ -36,7 +36,13 @@ typedef struct LowCore {
     /* prefix area: defined by architecture */
     PSWLegacy       ipl_psw;                  /* 0x000 */
     uint32_t        ccw1[2];                  /* 0x008 */
-    uint32_t        ccw2[2];                  /* 0x010 */
+    union {
+        uint32_t        ccw2[2];                  /* 0x010 */
+        struct {
+            uint32_t reserved10;
+            uint32_t ptr_iplb;
+        };
+    };
     uint8_t         pad1[0x80 - 0x18];        /* 0x018 */
     uint32_t        ext_params;               /* 0x080 */
     uint16_t        cpu_addr;                 /* 0x084 */
@@ -85,7 +91,7 @@ typedef struct LowCore {
     PSW             io_new_psw;               /* 0x1f0 */
 } __attribute__((packed, aligned(8192))) LowCore;
 
-extern LowCore const *lowcore;
+extern LowCore *lowcore;
 
 static inline void set_prefix(uint32_t address)
 {
pc-bios/s390-ccw/s390-ccw.h +0 -0
index 11bce7d73c..21f27e7990 100644

pc-bios/s390-ccw/s390-ccw.h void consume_io_int(void); · line 62

Quoted 6 lines
pc-bios/s390-ccw/s390-ccw.h +1 -0
@@ -57,6 +57,7 @@ void consume_io_int(void);
 /* main.c */
 void panic(const char *string);
 void write_subsystem_identification(void);
+void write_iplb_location(void);
 extern char stack[PAGE_SIZE * 8] __attribute__((__aligned__(PAGE_SIZE)));
 unsigned int get_loadparm_index(void);

In general LGTM.

Side note: I wonder if the assert in ptr2u32() should actually be

IPL_assert((uint64_t)ptr <= 0xfffffffful, "ptr2u32: ptr too large");
				      ^

or even better

IPL_assert((uint64_t)ptr != (uint32_t)ptr, "ptr2u32: ptr too large");

Otherwise, sign extension will simply always make this pass.

Signature · 3 lines
Thanks,

David / dhildenb
#60 lore raw

Re: [PATCH] pc-bios: s390x: Save iplb location in lowcore

Quoted 10 lines · David Hildenbrand
Quoted 9 lines · Janosch Frank

The POP states that for a list directed IPL the IPLB is stored into memory by the machine loader and its address is stored at offset 0x14 of the lowcore.

ZIPL currently uses the address in offset 0x14 to access the IPLB and acquire flags about secure boot. If the IPLB address points into memory which has an unsupported mix of flags set, ZIPL will panic instead of booting the OS.

Quoted 5 lines
Quoted 3 lines

As the lowcore can have quite a high entropy for a guest that did drop out of protected mode (i.e. rebooted) we encountered the ZIPL panic quite often.

How did this ever work? Or does this only become relevant with secure boot?

I'd guess that until secure boot ZIPL never touched this and with it we never hit the right combination of flags to trigger a ZIPL panic.

This way of getting to the IPLB was used before diag308 was available, i.e. way before KVM got to IBM Z. It looks like ZIPL only uses it for secure boot for some reason and hence we never implemented it before.

I'm also in discussion with the ZIPL developers to make this more robust.

Quoted 2 lines
Fixes ? Or has this always been broken?

See above

Quoted 114 lines
Quoted 102 lines
---
 pc-bios/s390-ccw/jump2ipl.c  |  1 +
 pc-bios/s390-ccw/main.c      |  8 +++++++-
 pc-bios/s390-ccw/netmain.c   |  1 +
 pc-bios/s390-ccw/s390-arch.h | 10 ++++++++--
 pc-bios/s390-ccw/s390-ccw.h  |  1 +
 5 files changed, 18 insertions(+), 3 deletions(-)
pc-bios/s390-ccw/jump2ipl.c +1 -0
index da13c43cc0..4eba2510b0 100644
@@ -35,6 +35,7 @@ void jump_to_IPL_code(uint64_t address)
 {
     /* store the subsystem information _after_ the bootmap was loaded */
     write_subsystem_identification();
+    write_iplb_location();
 
     /* prevent unknown IPL types in the guest */
     if (iplb.pbt == S390_IPL_TYPE_QEMU_SCSI) {
pc-bios/s390-ccw/main.c +7 -1
index a21b386280..4e65b411e1 100644
@@ -9,6 +9,7 @@
  */
 
 #include "libc.h"
+#include "helper.h"
 #include "s390-arch.h"
 #include "s390-ccw.h"
 #include "cio.h"
@@ -22,7 +23,7 @@ QemuIplParameters qipl;
 IplParameterBlock iplb __attribute__((__aligned__(PAGE_SIZE)));
 static bool have_iplb;
 static uint16_t cutype;
-LowCore const *lowcore; /* Yes, this *is* a pointer to address 0 */
+LowCore *lowcore; /* Yes, this *is* a pointer to address 0 */
 
 #define LOADPARM_PROMPT "PROMPT  "
 #define LOADPARM_EMPTY  "        "
@@ -42,6 +43,11 @@ void write_subsystem_identification(void)
     *zeroes = 0;
 }
 
+void write_iplb_location(void)
+{
+    lowcore->ptr_iplb = ptr2u32(&iplb);
+}
+
 void panic(const char *string)
 {
     sclp_print(string);
pc-bios/s390-ccw/netmain.c +1 -0
index f2dcc01e27..309ffa30d9 100644
@@ -40,6 +40,7 @@
 #define DEFAULT_TFTP_RETRIES 20
 
 extern char _start[];
+void write_iplb_location(void) {}
 
 #define KERNEL_ADDR             ((void *)0L)
 #define KERNEL_MAX_SIZE         ((long)_start)
pc-bios/s390-ccw/s390-arch.h +8 -2
index 504fc7c2f0..5f36361c02 100644
@@ -36,7 +36,13 @@ typedef struct LowCore {
     /* prefix area: defined by architecture */
     PSWLegacy       ipl_psw;                  /* 0x000 */
     uint32_t        ccw1[2];                  /* 0x008 */
-    uint32_t        ccw2[2];                  /* 0x010 */
+    union {
+        uint32_t        ccw2[2];                  /* 0x010 */
+        struct {
+            uint32_t reserved10;
+            uint32_t ptr_iplb;
+        };
+    };
     uint8_t         pad1[0x80 - 0x18];        /* 0x018 */
     uint32_t        ext_params;               /* 0x080 */
     uint16_t        cpu_addr;                 /* 0x084 */
@@ -85,7 +91,7 @@ typedef struct LowCore {
     PSW             io_new_psw;               /* 0x1f0 */
 } __attribute__((packed, aligned(8192))) LowCore;
 
-extern LowCore const *lowcore;
+extern LowCore *lowcore;
 
 static inline void set_prefix(uint32_t address)
 {
pc-bios/s390-ccw/s390-ccw.h +0 -0
index 11bce7d73c..21f27e7990 100644

pc-bios/s390-ccw/s390-ccw.h void consume_io_int(void); · line 62

Quoted 6 lines
pc-bios/s390-ccw/s390-ccw.h +1 -0
@@ -57,6 +57,7 @@ void consume_io_int(void);
 /* main.c */
 void panic(const char *string);
 void write_subsystem_identification(void);
+void write_iplb_location(void);
 extern char stack[PAGE_SIZE * 8] __attribute__((__aligned__(PAGE_SIZE)));
 unsigned int get_loadparm_index(void);

In general LGTM.

Side note: I wonder if the assert in ptr2u32() should actually be

Quoted 8 lines
IPL_assert((uint64_t)ptr <= 0xfffffffful, "ptr2u32: ptr too large");
				      ^

or even better

IPL_assert((uint64_t)ptr != (uint32_t)ptr, "ptr2u32: ptr too large");

Otherwise, sign extension will simply always make this pass.

Do you want to add a patch or shall I add it to my cleanup series?

#61 lore raw

Re: [PATCH] pc-bios: s390x: Save iplb location in lowcore

Quoted 17 lines · Janosch Frank
Quoted 10 lines · David Hildenbrand
Quoted 9 lines · Janosch Frank

The POP states that for a list directed IPL the IPLB is stored into memory by the machine loader and its address is stored at offset 0x14 of the lowcore.

ZIPL currently uses the address in offset 0x14 to access the IPLB and acquire flags about secure boot. If the IPLB address points into memory which has an unsupported mix of flags set, ZIPL will panic instead of booting the OS.

Quoted 5 lines
Quoted 3 lines

As the lowcore can have quite a high entropy for a guest that did drop out of protected mode (i.e. rebooted) we encountered the ZIPL panic quite often.

How did this ever work? Or does this only become relevant with secure boot?

Quoted 9 lines

I'd guess that until secure boot ZIPL never touched this and with it we never hit the right combination of flags to trigger a ZIPL panic.

This way of getting to the IPLB was used before diag308 was available, i.e. way before KVM got to IBM Z. It looks like ZIPL only uses it for secure boot for some reason and hence we never implemented it before.

I'm also in discussion with the ZIPL developers to make this more robust.

Thanks for the clarification!

[...]

Quoted 2 lines

Do you want to add a patch or shall I add it to my cleanup series?

Would be great if you could add a cleanup.

Signature · 3 lines
Thanks,

David / dhildenb