Skip to contentSkip to conversation

hw/i386/pc: Fix crash when hot-plugging nvdimm on older machine types

PATCH Applied Thomas Huth 6 participants 9 messages Apr 7, 2019 – Apr 11, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hw/i386/pc: Fix crash when hot-plugging nvdimm on older machine types

QEMU currently crashes when you try to hot-plug an "nvdimm" device on older machine types:

$ qemu-system-x86_64 -monitor stdio -M pc-1.1
QEMU 3.1.92 monitor - type 'help' for more information
(qemu) device_add nvdimm,id=nvdimmn1
qemu-system-x86_64: /home/thuth/devel/qemu/util/error.c:57: error_setv:
 Assertion `*errp == ((void *)0)' failed.
Aborted (core dumped)

The call to hotplug_handler_pre_plug() in pc_memory_pre_plug() has been added recently before the check whether nvdimm is enabled. It should be done after the check. And while we're at it, also check the errp after the hotplug_handler_pre_plug(), otherwise errors are silently ignored here.

Fixes 9040e6dfa8c3fed87695a3de555d2c775727bb51
---
 hw/i386/pc.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
hw/i386/pc.c +7 -2
index 6077d27361..f2c15bf1f2 100644
@@ -2078,6 +2078,7 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     const MachineState *ms = MACHINE(hotplug_dev);
     const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
     const uint64_t legacy_align = TARGET_PAGE_SIZE;
+    Error *local_err = NULL;
 
     /*
      * When -no-acpi is used with Q35 machine type, no ACPI is built,
@@ -2090,13 +2091,17 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
         return;
     }
 
-    hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
-
     if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
         error_setg(errp, "nvdimm is not enabled: missing 'nvdimm' in '-M'");
         return;
     }
 
+    hotplug_handler_pre_plug(pcms->acpi_dev, dev, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
     pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev),
                      pcmc->enforce_aligned_dimm ? NULL : &legacy_align, errp);
 }
Signature · 1 line
2.21.0
#1 lore raw
Quoted 45 lines · Thomas Huth

QEMU currently crashes when you try to hot-plug an "nvdimm" device on older machine types:

$ qemu-system-x86_64 -monitor stdio -M pc-1.1
QEMU 3.1.92 monitor - type 'help' for more information
(qemu) device_add nvdimm,id=nvdimmn1
qemu-system-x86_64: /home/thuth/devel/qemu/util/error.c:57: error_setv:
 Assertion `*errp == ((void *)0)' failed.
Aborted (core dumped)

The call to hotplug_handler_pre_plug() in pc_memory_pre_plug() has been added recently before the check whether nvdimm is enabled. It should be done after the check. And while we're at it, also check the errp after the hotplug_handler_pre_plug(), otherwise errors are silently ignored here.

Fixes 9040e6dfa8c3fed87695a3de555d2c775727bb51
---
 hw/i386/pc.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
hw/i386/pc.c +1 -2
index 6077d27361..f2c15bf1f2 100644
@@ -2078,6 +2078,7 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     const MachineState *ms = MACHINE(hotplug_dev);
     const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
     const uint64_t legacy_align = TARGET_PAGE_SIZE;
+    Error *local_err = NULL;
 
     /*
      * When -no-acpi is used with Q35 machine type, no ACPI is built,
@@ -2090,13 +2091,17 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
         return;
     }
 
-    hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
-
     if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
         error_setg(errp, "nvdimm is not enabled: missing 'nvdimm' in '-M'");
         return;
     }

hw/i386/pc.c static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, · line 2107

Quoted 10 lines
hw/i386/pc.c +6 -0
 
+    hotplug_handler_pre_plug(pcms->acpi_dev, dev, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
     pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev),
                      pcmc->enforce_aligned_dimm ? NULL : &legacy_align, errp);
 }

Queued, thanks.

Paolo

#2 lore raw
In reply to Thomas Huth
Quoted 10 lines · Thomas Huth

QEMU currently crashes when you try to hot-plug an "nvdimm" device on older machine types:

$ qemu-system-x86_64 -monitor stdio -M pc-1.1
QEMU 3.1.92 monitor - type 'help' for more information
(qemu) device_add nvdimm,id=nvdimmn1
qemu-system-x86_64: /home/thuth/devel/qemu/util/error.c:57: error_setv:
Assertion `*errp == ((void *)0)' failed.
Aborted (core dumped)
Quoted 5 lines

The call to hotplug_handler_pre_plug() in pc_memory_pre_plug() has been added recently before the check whether nvdimm is enabled. It should be done after the check. And while we're at it, also check the errp after the hotplug_handler_pre_plug(), otherwise errors are silently ignored here.

Thomas,

Thanks for pointing this out, while I have some different idea on how to fix this.

The reason of the core dump is errp already been set in hotplug_handler_pre_plug(), and this function check acpi hotplug capability. The order of this check is correct, while we should return when errp is set in hotplug_handler_pre_plug().

I got a fix like this, which I have tested and looks good to me.

hw/i386/pc.c +3 -0
index 6077d27361..b11f3b15c1 100644
@@ -2091,6 +2091,9 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     }
 
     hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
+    if (*errp) {
+        return;
+    }
 
     if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
         error_setg(errp, "nvdimm is not enabled: missing 'nvdimm' in '-M'");
Quoted 41 lines
Fixes 9040e6dfa8c3fed87695a3de555d2c775727bb51

---
hw/i386/pc.c | 9 +++++++--

1 file changed, 7 insertions(+), 2 deletions(-)
hw/i386/pc.c +7 -2
index 6077d27361..f2c15bf1f2 100644
@@ -2078,6 +2078,7 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
    const MachineState *ms = MACHINE(hotplug_dev);
    const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
    const uint64_t legacy_align = TARGET_PAGE_SIZE;
+    Error *local_err = NULL;
 
    /*
     * When -no-acpi is used with Q35 machine type, no ACPI is built,
@@ -2090,13 +2091,17 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
        return;
    }
 
-    hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
-
    if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
        error_setg(errp, "nvdimm is not enabled: missing 'nvdimm' in '-M'");
        return;
    }
 
+    hotplug_handler_pre_plug(pcms->acpi_dev, dev, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
    pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev),
                     pcmc->enforce_aligned_dimm ? NULL : &legacy_align, errp);
 }
Signature · 1 line
2.21.0
Signature · 2 lines
Wei Yang
Help you, Help me
#3 lore raw
Quoted 35 lines · Wei Yang
Quoted 10 lines · Thomas Huth

QEMU currently crashes when you try to hot-plug an "nvdimm" device on older machine types:

$ qemu-system-x86_64 -monitor stdio -M pc-1.1
QEMU 3.1.92 monitor - type 'help' for more information
(qemu) device_add nvdimm,id=nvdimmn1
qemu-system-x86_64: /home/thuth/devel/qemu/util/error.c:57: error_setv:
Assertion `*errp == ((void *)0)' failed.
Aborted (core dumped)
Quoted 5 lines

The call to hotplug_handler_pre_plug() in pc_memory_pre_plug() has been added recently before the check whether nvdimm is enabled. It should be done after the check. And while we're at it, also check the errp after the hotplug_handler_pre_plug(), otherwise errors are silently ignored here.

Thomas,

Thanks for pointing this out, while I have some different idea on how to fix this.

The reason of the core dump is errp already been set in hotplug_handler_pre_plug(), and this function check acpi hotplug capability. The order of this check is correct, while we should return when errp is set in hotplug_handler_pre_plug().

I got a fix like this, which I have tested and looks good to me.

hw/i386/pc.c +0 -0
index 6077d27361..b11f3b15c1 100644

hw/i386/pc.c static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, · line 2096

Quoted 6 lines
hw/i386/pc.c +3 -0
@@ -2091,6 +2091,9 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     }
 
     hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
+    if (*errp) {
+        return;
+    }

Not sure, but I think you can not rely on the fact that the caller set *errp = NULL already... that's why it is more common to use a local_err variable and error_propagate() for such cases (which is what I did in my patch).

Also, why don't you want the "nvdimm is not enabled: missing 'nvdimm' in '-M'" check to be done first?

Thomas

Quoted 45 lines
hw/i386/pc.c +0 -0
     if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
         error_setg(errp, "nvdimm is not enabled: missing 'nvdimm' in '-M'");
 
Quoted 41 lines
Fixes 9040e6dfa8c3fed87695a3de555d2c775727bb51

---
hw/i386/pc.c | 9 +++++++--

1 file changed, 7 insertions(+), 2 deletions(-)
hw/i386/pc.c +7 -2
index 6077d27361..f2c15bf1f2 100644
@@ -2078,6 +2078,7 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
    const MachineState *ms = MACHINE(hotplug_dev);
    const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
    const uint64_t legacy_align = TARGET_PAGE_SIZE;
+    Error *local_err = NULL;
 
    /*
     * When -no-acpi is used with Q35 machine type, no ACPI is built,
@@ -2090,13 +2091,17 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
        return;
    }
 
-    hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
-
    if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
        error_setg(errp, "nvdimm is not enabled: missing 'nvdimm' in '-M'");
        return;
    }
 
+    hotplug_handler_pre_plug(pcms->acpi_dev, dev, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
    pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev),
                     pcmc->enforce_aligned_dimm ? NULL : &legacy_align, errp);
 }
Signature · 1 line
2.21.0
#4 lore raw
Quoted 38 lines · Thomas Huth
Quoted 37 lines · Wei Yang
Quoted 10 lines · Thomas Huth

QEMU currently crashes when you try to hot-plug an "nvdimm" device on older machine types:

$ qemu-system-x86_64 -monitor stdio -M pc-1.1
QEMU 3.1.92 monitor - type 'help' for more information
(qemu) device_add nvdimm,id=nvdimmn1
qemu-system-x86_64: /home/thuth/devel/qemu/util/error.c:57: error_setv:
Assertion `*errp == ((void *)0)' failed.
Aborted (core dumped)
Quoted 5 lines

The call to hotplug_handler_pre_plug() in pc_memory_pre_plug() has been added recently before the check whether nvdimm is enabled. It should be done after the check. And while we're at it, also check the errp after the hotplug_handler_pre_plug(), otherwise errors are silently ignored here.

Thomas,

Thanks for pointing this out, while I have some different idea on how to fix this.

The reason of the core dump is errp already been set in hotplug_handler_pre_plug(), and this function check acpi hotplug capability. The order of this check is correct, while we should return when errp is set in hotplug_handler_pre_plug().

I got a fix like this, which I have tested and looks good to me.

hw/i386/pc.c +0 -0
index 6077d27361..b11f3b15c1 100644
@@ -2091,6 +2091,9 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     }
 
Quoted 10 lines

hw/i386/pc.c static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, · line 2096

Quoted 4 lines
hw/i386/pc.c +3 -0
     hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
+    if (*errp) {
+        return;
+    }

Not sure, but I think you can not rely on the fact that the caller set *errp = NULL already... that's why it is more common to use a local_err variable and error_propagate() for such cases (which is what I did in my patch).

Ok, that's fine for me.

Quoted 3 lines

Also, why don't you want the "nvdimm is not enabled: missing 'nvdimm' in '-M'" check to be done first?

Because this function pc_memory_pre_plug() will be called not only when nvdimm is hot-plugged but also dimm is hot-plugged. And hotplug_handler_pre_plug() here is to check the acpi(if it has) hot-plug capability.

So the check in pc_memory_pre_plug() is from generic to specific:
1. Do we have capability to hot-plug?
2. If the device is nvdimm, do we enabled nvdimm?

Quoted 50 lines

Thomas

Quoted 45 lines
hw/i386/pc.c +0 -0
     if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
         error_setg(errp, "nvdimm is not enabled: missing 'nvdimm' in '-M'");
 
Quoted 41 lines
Fixes 9040e6dfa8c3fed87695a3de555d2c775727bb51

---
hw/i386/pc.c | 9 +++++++--

1 file changed, 7 insertions(+), 2 deletions(-)
hw/i386/pc.c +7 -2
index 6077d27361..f2c15bf1f2 100644
@@ -2078,6 +2078,7 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
    const MachineState *ms = MACHINE(hotplug_dev);
    const bool is_nvdimm = object_dynamic_cast(OBJECT(dev), TYPE_NVDIMM);
    const uint64_t legacy_align = TARGET_PAGE_SIZE;
+    Error *local_err = NULL;
 
    /*
     * When -no-acpi is used with Q35 machine type, no ACPI is built,
@@ -2090,13 +2091,17 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
        return;
    }
 
-    hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
-
    if (is_nvdimm && !ms->nvdimms_state->is_enabled) {
        error_setg(errp, "nvdimm is not enabled: missing 'nvdimm' in '-M'");
        return;
    }
 
+    hotplug_handler_pre_plug(pcms->acpi_dev, dev, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
    pc_dimm_pre_plug(PC_DIMM(dev), MACHINE(hotplug_dev),
                     pcmc->enforce_aligned_dimm ? NULL : &legacy_align, errp);
 }
Signature · 1 line
2.21.0
Signature · 2 lines
Wei Yang
Help you, Help me
#5 lore raw
Quoted 38 lines · Wei Yang
Quoted 21 lines
Quoted 21 lines

Thomas,

Thanks for pointing this out, while I have some different idea on how to fix this.

The reason of the core dump is errp already been set in hotplug_handler_pre_plug(), and this function check acpi hotplug capability. The order of this check is correct, while we should return when errp is set in hotplug_handler_pre_plug().

I got a fix like this, which I have tested and looks good to me.

hw/i386/pc.c +0 -0
index 6077d27361..b11f3b15c1 100644
@@ -2091,6 +2091,9 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     }
 
Quoted 10 lines

hw/i386/pc.c static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, · line 2096

Quoted 4 lines
hw/i386/pc.c +3 -0
     hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
+    if (*errp) {
+        return;
+    }

Not sure, but I think you can not rely on the fact that the caller set *errp = NULL already... that's why it is more common to use a local_err variable and error_propagate() for such cases (which is what I did in my patch).

Ok, that's fine for me.

Quoted 3 lines

Also, why don't you want the "nvdimm is not enabled: missing 'nvdimm' in '-M'" check to be done first?

Quoted 9 lines

Because this function pc_memory_pre_plug() will be called not only when nvdimm is hot-plugged but also dimm is hot-plugged. And hotplug_handler_pre_plug() here is to check the acpi(if it has) hot-plug capability.

So the check in pc_memory_pre_plug() is from generic to specific:
1. Do we have capability to hot-plug?
2. If the device is nvdimm, do we enabled nvdimm?

Thomas

Do you think this is a reasonable explanation?

Signature · 2 lines
Wei Yang
Help you, Help me
#6 lore raw

Hi,

Quoted 44 lines · Wei Yang
Quoted 43 lines · Wei Yang
Quoted 21 lines
Quoted 21 lines

Thomas,

Thanks for pointing this out, while I have some different idea on how to fix this.

The reason of the core dump is errp already been set in hotplug_handler_pre_plug(), and this function check acpi hotplug capability. The order of this check is correct, while we should return when errp is set in hotplug_handler_pre_plug().

I got a fix like this, which I have tested and looks good to me.

hw/i386/pc.c +0 -0
index 6077d27361..b11f3b15c1 100644
@@ -2091,6 +2091,9 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     }
 
Quoted 10 lines

hw/i386/pc.c static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, · line 2096

Quoted 4 lines
hw/i386/pc.c +3 -0
     hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
+    if (*errp) {
+        return;
+    }

Not sure, but I think you can not rely on the fact that the caller set *errp = NULL already... that's why it is more common to use a local_err variable and error_propagate() for such cases (which is what I did in my patch).

Ok, that's fine for me.

Quoted 3 lines

Also, why don't you want the "nvdimm is not enabled: missing 'nvdimm' in '-M'" check to be done first?

Because this function pc_memory_pre_plug() will be called not only when nvdimm is hot-plugged but also dimm is hot-plugged. And hotplug_handler_pre_plug() here is to check the acpi(if it has) hot-plug capability.

Quoted 8 lines
Quoted 4 lines

So the check in pc_memory_pre_plug() is from generic to specific:
1. Do we have capability to hot-plug?
2. If the device is nvdimm, do we enabled nvdimm?

Thomas

Do you think this is a reasonable explanation?

Fine for me, I don't mind either way as long as the crash is fixed. Feel free to send a patch that restores the desired order.

Thomas

#7 lore raw
In reply to Thomas Huth
Quoted 1 line · Thomas Huth

On 08/04/2019 15.45, Wei Yang wrote:

[...]

Quoted 7 lines
Quoted 7 lines
hw/i386/pc.c +0 -0
index 6077d27361..b11f3b15c1 100644
@@ -2091,6 +2091,9 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     }
 
Quoted 9 lines

hw/i386/pc.c static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, · line 2096

Quoted 4 lines
hw/i386/pc.c +3 -0
     hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
+    if (*errp) {
+        return;
+    }

Not sure, but I think you can not rely on the fact that the caller set *errp = NULL already... that's why it is more common to use a local_err variable and error_propagate() for such cases (which is what I did in my patch).

*errp can't be non-NULL (otherwise functions calling error_setg() would crash). errp can be NULL, though, and that's why you need a local_err variable.

I'd love to eliminate NULL errp from our codebase, but I couldn't find a way to do it that is safe and simple (i.e. not letting us pass NULL errp by mistake and not requiring a macro wrapping every &local_err expression).

Signature · 1 line
Eduardo
#8 lore raw
Quoted 20 lines · Eduardo Habkost
Quoted 1 line · Thomas Huth

On 08/04/2019 15.45, Wei Yang wrote:

[...]

Quoted 7 lines
Quoted 7 lines
hw/i386/pc.c +0 -0
index 6077d27361..b11f3b15c1 100644
@@ -2091,6 +2091,9 @@ static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     }
 
Quoted 9 lines

hw/i386/pc.c static void pc_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev, · line 2096

Quoted 4 lines
hw/i386/pc.c +3 -0
     hotplug_handler_pre_plug(pcms->acpi_dev, dev, errp);
+    if (*errp) {
+        return;
+    }

Not sure, but I think you can not rely on the fact that the caller set *errp = NULL already... that's why it is more common to use a local_err variable and error_propagate() for such cases (which is what I did in my patch).

Quoted 3 lines

*errp can't be non-NULL (otherwise functions calling error_setg() would crash). errp can be NULL, though, and that's why you need a local_err variable.

Correct. The big comment in error.h advises:

* Receive an error and pass it on to the caller:

 *     Error *err = NULL;
 *     foo(arg, &err);
 *     if (err) {

* handle the error...
* error_propagate(errp, err);
* }
* where Error **errp is a parameter, by convention the last one. *
* Do *not* "optimize" this to
* foo(arg, errp);
* if (*errp) { // WRONG!
* handle the error...
* }
* because errp may be NULL!
*
* But when all you do with the error is pass it on, please use
* foo(arg, errp);
* for readability.

Quoted 4 lines

I'd love to eliminate NULL errp from our codebase, but I couldn't find a way to do it that is safe and simple (i.e. not letting us pass NULL errp by mistake and not requiring a macro wrapping every &local_err expression).

Also, I'd prefer not not deviate even more from GError.

Apropos NULL, we often pass NULL where we really ought to pass &error_abort.