Skip to contentSkip to conversation

Error handling fixes, may contain 4.2 material

PATCH In review Markus Armbruster 16 participants 74 messages Nov 30, 2019 – Jan 13, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover Error handling fixes, may contain 4.2 material

PATCH 2-4 fix crash bugs. Including them would be a no-brainer at -rc0. But we're post -rc3, and even for crash bugs we require a certain likelihood of users getting bitten.

Jens, please assess impact of PATCH 2's crash bug.

Kevin, please do the same for PATCH 3.

Daniel, please do the same for PATCH 4.

The remainder is definitely not 4.2 material.

Cc "Daniel P. Berrangé" <[email protected]>
Cc "Michael S. Tsirkin" <[email protected]>
Cc Christian Borntraeger <[email protected]>
Cc Corey Minyard <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Igor Mammedov <[email protected]>
Cc Jens Freimann <[email protected]>
Cc Kevin Wolf <[email protected]>
Cc Michael Roth <[email protected]>
Cc Michael S. Tsirkin <[email protected]>
Cc Nishanth Aravamudan <[email protected]>
Cc Stefan Hajnoczi <[email protected]>

Markus Armbruster (21):
net/virtio: Drop useless n->primary_dev not null checks net/virtio: Fix failover error handling crash bugs block/file-posix: Fix laio_init() error handling crash bug
crypto: Fix certificate file error handling crash bug
crypto: Fix typo in QCryptoTLSSession's <example> comment
io: Fix Error usage in a comment <example>
tests: Clean up initialization of Error *err variables
exec: Fix latent file_ram_alloc() error handling bug hw/acpi: Fix latent legacy CPU plug error handling bug hw/core: Fix latent fit_load_fdt() error handling bug hw/ipmi: Fix latent realize() error handling bugs
qga: Fix latent guest-get-fsinfo error handling bug
memory-device: Fix latent memory pre-plug error handling bugs s390x/event-facility: Fix latent realize() error handling bug s390x/cpu_models: Fix latent feature property error handling bugs s390/cpu_modules: Fix latent realize() error handling bugs
s390x: Fix latent query-cpu-model-FOO error handling bugs
s390x: Fix latent query-cpu-definitions error handling bug
error: Clean up unusual names of Error * variables hw/intc/s390: Simplify error handling in kvm_s390_flic_realize()
tests-blockjob: Use error_free_or_abort()

 include/crypto/tlssession.h         |  2 +-
 include/io/task.h                   |  2 +-
 block/file-posix.c                  |  2 +-
 crypto/tlscredsx509.c               |  2 +-
 exec.c                              |  6 +-
 hw/acpi/cpu_hotplug.c               | 10 +--
 hw/core/loader-fit.c                | 15 ++---
 hw/intc/s390_flic_kvm.c             | 16 +++--
 hw/ipmi/isa_ipmi_bt.c               |  7 ++-
 hw/ipmi/isa_ipmi_kcs.c              |  7 ++-
 hw/ipmi/pci_ipmi_bt.c               |  6 +-
 hw/ipmi/pci_ipmi_kcs.c              |  6 +-
 hw/mem/memory-device.c              |  6 +-
 hw/net/virtio-net.c                 | 27 ++++----
 hw/ppc/spapr_pci.c                  | 16 ++---
 hw/ppc/spapr_pci_nvlink2.c          | 10 +--
 hw/s390x/event-facility.c           |  6 +-
 qga/commands-posix.c                |  6 +-
 target/s390x/cpu_models.c           | 98 +++++++++++++++++------------
 tests/test-blockjob.c               | 15 +++--
 tests/test-qobject-output-visitor.c |  8 +--
 tests/test-string-output-visitor.c  |  4 +-
 22 files changed, 154 insertions(+), 123 deletions(-)
Signature · 1 line
2.21.0
#1 lore raw

PATCH 1/21 net/virtio: Drop useless n->primary_dev not null checks

virtio_net_handle_migration_primary() returns early when it can't ensure n->primary_dev is non-null. Checking it again right after that early return is redundant. Drop.

If n->primary_dev is null on entering failover_replug_primary(), @pdev will become null, and pdev->partially_hotplugged will crash. Checking n->primary_dev later is useless. It can't actually be null, because its caller virtio_net_handle_migration_primary() ensures it isn't. Drop the useless check.

Cc Jens Freimann <[email protected]>
Cc Michael S. Tsirkin <[email protected]>
---
 hw/net/virtio-net.c | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)
hw/net/virtio-net.c +1 -7
index 3c31471026..87088ba374 100644
@@ -2810,11 +2810,6 @@ static bool failover_replug_primary(VirtIONet *n, Error **errp)
             goto out;
         }
     }
-    if (!n->primary_dev) {
-            error_setg(errp, "virtio_net: couldn't find primary device");
-            goto out;
-    }
-
     n->primary_bus = n->primary_dev->parent_bus;
     if (!n->primary_bus) {
         error_setg(errp, "virtio_net: couldn't find primary bus");
@@ -2849,8 +2844,7 @@ static void virtio_net_handle_migration_primary(VirtIONet *n,
         }
     }
 
-    if (migration_in_setup(s) && !should_be_hidden &&
-        n->primary_dev) {
+    if (migration_in_setup(s) && !should_be_hidden) {
         if (failover_unplug_primary(n)) {
             vmstate_unregister(n->primary_dev, qdev_get_vmsd(n->primary_dev),
                     n->primary_dev);
Signature · 1 line
2.21.0
Reviewed-by on v1 1/21
#2 lore raw

Re: [PATCH 01/21] net/virtio: Drop useless n->primary_dev not null checks

Quoted 10 lines · Markus Armbruster

virtio_net_handle_migration_primary() returns early when it can't ensure n->primary_dev is non-null. Checking it again right after that early return is redundant. Drop.

If n->primary_dev is null on entering failover_replug_primary(), @pdev will become null, and pdev->partially_hotplugged will crash. Checking n->primary_dev later is useless. It can't actually be null, because its caller virtio_net_handle_migration_primary() ensures it isn't. Drop the useless check.

Quoted 7 lines
Cc Jens Freimann <[email protected]>
Cc Michael S. Tsirkin <[email protected]>

---
hw/net/virtio-net.c | 8 +-------

1 file changed, 1 insertion(+), 7 deletions(-)

Thanks Markus!

regards
Jens

#3 lore raw

PATCH 2/21 net/virtio: Fix failover error handling crash bugs

In reply to Markus Armbruster

Functions that take an Error ** parameter to pass an error to the caller expect the parameter to point to null. failover_replug_primary() violates this precondition in several places:

* After qemu_opts_from_qdict() failed, *errp is no longer null. Passing it to error_setg() is wrong, and will trip the assertion in error_setv(). Messed up in commit 150ab54aa6 "net/virtio: fix re-plugging of primary device". Simply drop the error_setg().

* Passing @errp to qemu_opt_set_bool(), hotplug_handler_pre_plug(), and hotplug_handler_plug() is wrong. If one of the first two fails, *errp is no longer null. Risks tripping the same assertion. Moreover, continuing after such errors is unsafe. Messed up in commit 9711cd0dfc "net/virtio: add failover support". Fix by handling each error properly.

failover_replug_primary() crashes when passed a null @errp. Also messed up in commit 9711cd0dfc. This bug can't bite as no caller actually passes null. Fix it anyway.

Fixes 9711cd0dfc3fa414f7f64935713c07134ae67971
Fixes 150ab54aa6934583180f88a2bd540bc6fc4fbff3
Cc Jens Freimann <[email protected]>
Cc Michael S. Tsirkin <[email protected]>
---
 hw/net/virtio-net.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)
hw/net/virtio-net.c +13 -6
index 87088ba374..db3d7c38e6 100644
@@ -2795,6 +2795,7 @@ static bool failover_unplug_primary(VirtIONet *n)
 
 static bool failover_replug_primary(VirtIONet *n, Error **errp)
 {
+    Error *err = NULL;
     HotplugHandler *hotplug_ctrl;
     PCIDevice *pdev = PCI_DEVICE(n->primary_dev);
 
@@ -2806,27 +2807,33 @@ static bool failover_replug_primary(VirtIONet *n, Error **errp)
                 qemu_find_opts("device"),
                 n->primary_device_dict, errp);
         if (!n->primary_device_opts) {
-            error_setg(errp, "virtio_net: couldn't find primary device opts");
-            goto out;
+            return false;
         }
     }
     n->primary_bus = n->primary_dev->parent_bus;
     if (!n->primary_bus) {
         error_setg(errp, "virtio_net: couldn't find primary bus");
-        goto out;
+        return false;
     }
     qdev_set_parent_bus(n->primary_dev, n->primary_bus);
     n->primary_should_be_hidden = false;
     qemu_opt_set_bool(n->primary_device_opts,
-                      "partially_hotplugged", true, errp);
+                      "partially_hotplugged", true, &err);
+    if (err) {
+        goto out;
+    }
     hotplug_ctrl = qdev_get_hotplug_handler(n->primary_dev);
     if (hotplug_ctrl) {
-        hotplug_handler_pre_plug(hotplug_ctrl, n->primary_dev, errp);
+        hotplug_handler_pre_plug(hotplug_ctrl, n->primary_dev, &err);
+        if (err) {
+            goto out;
+        }
         hotplug_handler_plug(hotplug_ctrl, n->primary_dev, errp);
     }
 
 out:
-    return *errp == NULL;
+    error_propagate(errp, err);
+    return !err;
 }
 
 static void virtio_net_handle_migration_primary(VirtIONet *n,
Signature · 1 line
2.21.0
Reviewed-by on v1 2/21
#4 lore raw

Re: [PATCH 02/21] net/virtio: Fix failover error handling crash bugs

Quoted 21 lines · Markus Armbruster

Functions that take an Error ** parameter to pass an error to the caller expect the parameter to point to null. failover_replug_primary() violates this precondition in several places:

* After qemu_opts_from_qdict() failed, *errp is no longer null. Passing it to error_setg() is wrong, and will trip the assertion in error_setv(). Messed up in commit 150ab54aa6 "net/virtio: fix re-plugging of primary device". Simply drop the error_setg().

* Passing @errp to qemu_opt_set_bool(), hotplug_handler_pre_plug(), and hotplug_handler_plug() is wrong. If one of the first two fails, *errp is no longer null. Risks tripping the same assertion. Moreover, continuing after such errors is unsafe. Messed up in commit 9711cd0dfc "net/virtio: add failover support". Fix by handling each error properly.

failover_replug_primary() crashes when passed a null @errp. Also messed up in commit 9711cd0dfc. This bug can't bite as no caller actually passes null. Fix it anyway.

Quoted 9 lines
Fixes 9711cd0dfc3fa414f7f64935713c07134ae67971
Fixes 150ab54aa6934583180f88a2bd540bc6fc4fbff3
Cc Jens Freimann <[email protected]>
Cc Michael S. Tsirkin <[email protected]>

---
hw/net/virtio-net.c | 19 +++++++++++++------

1 file changed, 13 insertions(+), 6 deletions(-)

Thanks Markus!

#5 lore raw

PATCH 3/21 block/file-posix: Fix laio_init() error handling crash bug

In reply to Markus Armbruster

raw_aio_attach_aio_context() passes uninitialized Error *local_err by reference to laio_init() via aio_setup_linux_aio(). When laio_init() fails, it passes it on to error_setg_errno(), tripping error_setv()'s assertion unless @local_err is null by dumb luck.

Fix by initializing @local_err properly.

Fixes ed6e2161715c527330f936d44af4c547f25f687e
Cc Nishanth Aravamudan <[email protected]>
Cc Stefan Hajnoczi <[email protected]>
Cc Kevin Wolf <[email protected]>
---
 block/file-posix.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/file-posix.c +1 -1
index 1f0f61a02b..1b805bd938 100644
@@ -1973,7 +1973,7 @@ static void raw_aio_attach_aio_context(BlockDriverState *bs,
 #ifdef CONFIG_LINUX_AIO
     BDRVRawState *s = bs->opaque;
     if (s->use_linux_aio) {
-        Error *local_err;
+        Error *local_err = NULL;
         if (!aio_setup_linux_aio(new_context, &local_err)) {
             error_reportf_err(local_err, "Unable to use native AIO, "
                                          "falling back to thread pool: ");
Signature · 1 line
2.21.0
Reviewed-by on v1 3/21
#6 lore raw

Re: [PATCH 03/21] block/file-posix: Fix laio_init() error handling crash bug

Quoted 7 lines · Markus Armbruster

raw_aio_attach_aio_context() passes uninitialized Error *local_err by reference to laio_init() via aio_setup_linux_aio(). When laio_init() fails, it passes it on to error_setg_errno(), tripping error_setv()'s assertion unless @local_err is null by dumb luck.

Fix by initializing @local_err properly.

Quoted 8 lines
Fixes ed6e2161715c527330f936d44af4c547f25f687e
Cc Nishanth Aravamudan <[email protected]>
Cc Stefan Hajnoczi <[email protected]>
Cc Kevin Wolf <[email protected]>
---
 block/file-posix.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Worth including in QEMU 4.2.

Reviewed-by on v1 3/21
#7 lore raw

Re: [PATCH 03/21] block/file-posix: Fix laio_init() error handling crash bug

In reply to Markus Armbruster

Am 30.11.2019 um 20:42 hat Markus Armbruster geschrieben:

Quoted 5 lines

raw_aio_attach_aio_context() passes uninitialized Error *local_err by reference to laio_init() via aio_setup_linux_aio(). When laio_init() fails, it passes it on to error_setg_errno(), tripping error_setv()'s assertion unless @local_err is null by dumb luck.

Quoted 7 lines

Fix by initializing @local_err properly.

Fixes ed6e2161715c527330f936d44af4c547f25f687e
Cc Nishanth Aravamudan <[email protected]>
Cc Stefan Hajnoczi <[email protected]>
Cc Kevin Wolf <[email protected]>
#8 lore raw

PATCH 4/21 crypto: Fix certificate file error handling crash bug

In reply to Markus Armbruster

qcrypto_tls_creds_load_cert() passes uninitialized GError *gerr by reference to g_file_get_contents(). When g_file_get_contents() fails, it'll try to set a GError. Unless @gerr is null by dumb luck, this logs a ERROR_OVERWRITTEN_WARNING warning message and leaves @gerr unchanged. qcrypto_tls_creds_load_cert() then dereferences the uninitialized @gerr.

Fix by initializing @gerr properly.

Fixes 9a2fd4347c40321f5cbb4ab4220e759fcbf87d03
Cc "Daniel P. Berrangé" <[email protected]>
---
 crypto/tlscredsx509.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
crypto/tlscredsx509.c +1 -1
index 01fc304e5d..53a4368f49 100644
@@ -380,7 +380,7 @@ qcrypto_tls_creds_load_cert(QCryptoTLSCredsX509 *creds,
     gnutls_x509_crt_t cert = NULL;
     g_autofree char *buf = NULL;
     gsize buflen;
-    GError *gerr;
+    GError *gerr = NULL;
     int ret = -1;
     int err;
 
Signature · 1 line
2.21.0
Reviewed-by on v1 4/21
#9 lore raw

Re: [PATCH 04/21] crypto: Fix certificate file error handling crash bug

Quoted 22 lines · Markus Armbruster

qcrypto_tls_creds_load_cert() passes uninitialized GError *gerr by reference to g_file_get_contents(). When g_file_get_contents() fails, it'll try to set a GError. Unless @gerr is null by dumb luck, this logs a ERROR_OVERWRITTEN_WARNING warning message and leaves @gerr unchanged. qcrypto_tls_creds_load_cert() then dereferences the uninitialized @gerr.

Fix by initializing @gerr properly.

Fixes 9a2fd4347c40321f5cbb4ab4220e759fcbf87d03
Cc "Daniel P. Berrangé" <[email protected]>
---
  crypto/tlscredsx509.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
crypto/tlscredsx509.c +0 -0
index 01fc304e5d..53a4368f49 100644

crypto/tlscredsx509.c qcrypto_tls_creds_load_cert(QCryptoTLSCredsX509 *creds, · line 386

Quoted 8 lines
crypto/tlscredsx509.c +1 -1
@@ -380,7 +380,7 @@ qcrypto_tls_creds_load_cert(QCryptoTLSCredsX509 *creds,
      gnutls_x509_crt_t cert = NULL;
      g_autofree char *buf = NULL;
      gsize buflen;
-    GError *gerr;
+    GError *gerr = NULL;
      int ret = -1;
      int err;
 

I didn't check the logic in commit message (and I don't know how GError works), but initializing local pointer to NULL never hurts:

Signature · 2 lines
Best regards,
Vladimir
#10 lore raw

PATCH 5/21 crypto: Fix typo in QCryptoTLSSession's <example> comment

In reply to Markus Armbruster
Cc "Daniel P. Berrangé" <[email protected]>
---
 include/crypto/tlssession.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/crypto/tlssession.h +1 -1
index e01e1a9dc2..15b9cef086 100644
@@ -56,7 +56,7 @@
  *
  * static int mysock_run_tls(int sockfd,
  *                           QCryptoTLSCreds *creds,
- *                           Error *errp)
+ *                           Error **errp)
  * {
  *    QCryptoTLSSession *sess;
  *
Signature · 1 line
2.21.0
Reviewed-by on v1 5/21
#11 lore raw

Re: [PATCH 05/21] crypto: Fix typo in QCryptoTLSSession's <example> comment

Quoted 12 lines · Markus Armbruster
Cc "Daniel P. Berrangé" <[email protected]>
---
  include/crypto/tlssession.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
include/crypto/tlssession.h +0 -0
index e01e1a9dc2..15b9cef086 100644

include/crypto/tlssession.h line 62

Quoted 8 lines
include/crypto/tlssession.h +1 -1
@@ -56,7 +56,7 @@
   *
   * static int mysock_run_tls(int sockfd,
   *                           QCryptoTLSCreds *creds,
- *                           Error *errp)
+ *                           Error **errp)
   * {
   *    QCryptoTLSSession *sess;
   *
Signature · 2 lines
Best regards,
Vladimir
#12 lore raw

PATCH 6/21 io: Fix Error usage in a comment <example>

In reply to Markus Armbruster
Cc "Daniel P. Berrangé" <[email protected]>
---
 include/io/task.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/io/task.h +1 -1
index 5cb9faf9f2..1abbfb8b65 100644
@@ -119,7 +119,7 @@ typedef void (*QIOTaskWorker)(QIOTask *task,
  *   gboolean myobject_operation_timer(gpointer opaque)
  *   {
  *      QIOTask *task = QIO_TASK(opaque);
- *      Error *err;*
+ *      Error *err = NULL;
  *
  *      ...check something important...
  *       if (err) {
Signature · 1 line
2.21.0
Reviewed-by on v1 6/21
#13 lore raw

Re: [PATCH 06/21] io: Fix Error usage in a comment <example>

Quoted 12 lines · Markus Armbruster
Cc "Daniel P. Berrangé" <[email protected]>
---
  include/io/task.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
include/io/task.h +0 -0
index 5cb9faf9f2..1abbfb8b65 100644

include/io/task.h typedef void (*QIOTaskWorker)(QIOTask *task, · line 125

Quoted 8 lines
include/io/task.h +1 -1
@@ -119,7 +119,7 @@ typedef void (*QIOTaskWorker)(QIOTask *task,
   *   gboolean myobject_operation_timer(gpointer opaque)
   *   {
   *      QIOTask *task = QIO_TASK(opaque);
- *      Error *err;*
+ *      Error *err = NULL;
   *
   *      ...check something important...
   *       if (err) {

(I also think you can safely s/Fix Error usage/Fix typo/ in subject)

Signature · 2 lines
Best regards,
Vladimir
#14 lore raw

Re: [PATCH 06/21] io: Fix Error usage in a comment <example>

Quoted 24 lines · Vladimir Sementsov-Ogievskiy
Quoted 12 lines · Markus Armbruster
Cc "Daniel P. Berrangé" <[email protected]>
---
  include/io/task.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
include/io/task.h +0 -0
index 5cb9faf9f2..1abbfb8b65 100644

include/io/task.h typedef void (*QIOTaskWorker)(QIOTask *task, · line 125

Quoted 8 lines
include/io/task.h +1 -1
@@ -119,7 +119,7 @@ typedef void (*QIOTaskWorker)(QIOTask *task,
   *   gboolean myobject_operation_timer(gpointer opaque)
   *   {
   *      QIOTask *task = QIO_TASK(opaque);
- *      Error *err;*
+ *      Error *err = NULL;
   *
   *      ...check something important...
   *       if (err) {
Quoted 1 line

(I also think you can safely s/Fix Error usage/Fix typo/ in subject)

I'd say it's both: the trailing * is clearly a typo, but the missing initializer is asking Murphy for random crashes at inopportune times. Examples better set *good* examples :)

Thanks!

#15 lore raw

PATCH 7/21 tests: Clean up initialization of Error *err variables

In reply to Markus Armbruster

Declaring a local Error *err without initializer looks suspicious. Fuse the declaration with the initialization to avoid that.

---
 tests/test-qobject-output-visitor.c | 8 ++++----
 tests/test-string-output-visitor.c  | 4 ++--
 2 files changed, 6 insertions(+), 6 deletions(-)
tests/test-qobject-output-visitor.c +4 -4
index 3e993e5ba8..d7761ebf84 100644
@@ -145,10 +145,10 @@ static void test_visitor_out_enum_errors(TestOutputVisitorData *data,
                                          const void *unused)
 {
     EnumOne i, bad_values[] = { ENUM_ONE__MAX, -1 };
-    Error *err;
 
     for (i = 0; i < ARRAY_SIZE(bad_values) ; i++) {
-        err = NULL;
+        Error *err = NULL;
+
         visit_type_EnumOne(data->ov, "unused", &bad_values[i], &err);
         error_free_or_abort(&err);
         visitor_reset(data);
@@ -240,11 +240,11 @@ static void test_visitor_out_struct_errors(TestOutputVisitorData *data,
     EnumOne bad_values[] = { ENUM_ONE__MAX, -1 };
     UserDefOne u = {0};
     UserDefOne *pu = &u;
-    Error *err;
     int i;
 
     for (i = 0; i < ARRAY_SIZE(bad_values) ; i++) {
-        err = NULL;
+        Error *err = NULL;
+
         u.has_enum1 = true;
         u.enum1 = bad_values[i];
         visit_type_UserDefOne(data->ov, "unused", &pu, &err);
tests/test-string-output-visitor.c +2 -2
index 02766c0f65..1be1540767 100644
@@ -207,10 +207,10 @@ static void test_visitor_out_enum_errors(TestOutputVisitorData *data,
                                          const void *unused)
 {
     EnumOne i, bad_values[] = { ENUM_ONE__MAX, -1 };
-    Error *err;
 
     for (i = 0; i < ARRAY_SIZE(bad_values) ; i++) {
-        err = NULL;
+        Error *err = NULL;
+
         visit_type_EnumOne(data->ov, "unused", &bad_values[i], &err);
         error_free_or_abort(&err);
     }
Signature · 1 line
2.21.0
#17 lore raw

PATCH 8/21 exec: Fix latent file_ram_alloc() error handling bug

In reply to Markus Armbruster

When os_mem_prealloc() fails, file_ram_alloc() calls qemu_ram_munmap() and returns null. Except it doesn't when its @errp argument is null, because it checks for failure with (errp && *errp). Messed up in commit 056b68af77 "fix qemu exit on memory hotplug when allocation fails at prealloc time".

The bug can't bite as no caller actually passes null. Fix it anyway.

---
 exec.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
exec.c +4 -2
index ffdb518535..45695a5f2d 100644
@@ -1841,6 +1841,7 @@ static void *file_ram_alloc(RAMBlock *block,
                             bool truncate,
                             Error **errp)
 {
+    Error *err = NULL;
     MachineState *ms = MACHINE(qdev_get_machine());
     void *area;
 
@@ -1898,8 +1899,9 @@ static void *file_ram_alloc(RAMBlock *block,
     }
 
     if (mem_prealloc) {
-        os_mem_prealloc(fd, area, memory, ms->smp.cpus, errp);
-        if (errp && *errp) {
+        os_mem_prealloc(fd, area, memory, ms->smp.cpus, &err);
+        if (err) {
+            error_propagate(errp, err);
             qemu_ram_munmap(fd, area, memory);
             return NULL;
         }
Signature · 1 line
2.21.0
Reviewed-by on v1 8/21
#18 lore raw

Re: [PATCH 08/21] exec: Fix latent file_ram_alloc() error handling bug

Quoted 10 lines · On Sat, 30 Nov 2019 20:42:27 +0100 Markus Armbruster <[email protected]> wrote:

When os_mem_prealloc() fails, file_ram_alloc() calls qemu_ram_munmap() and returns null. Except it doesn't when its @errp argument is null, because it checks for failure with (errp && *errp). Messed up in commit 056b68af77 "fix qemu exit on memory hotplug when allocation fails at prealloc time".

The bug can't bite as no caller actually passes null. Fix it anyway.

Quoted 28 lines
---
 exec.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
exec.c +4 -2
index ffdb518535..45695a5f2d 100644
@@ -1841,6 +1841,7 @@ static void *file_ram_alloc(RAMBlock *block,
                             bool truncate,
                             Error **errp)
 {
+    Error *err = NULL;
     MachineState *ms = MACHINE(qdev_get_machine());
     void *area;
 
@@ -1898,8 +1899,9 @@ static void *file_ram_alloc(RAMBlock *block,
     }
 
     if (mem_prealloc) {
-        os_mem_prealloc(fd, area, memory, ms->smp.cpus, errp);
-        if (errp && *errp) {
+        os_mem_prealloc(fd, area, memory, ms->smp.cpus, &err);
+        if (err) {
+            error_propagate(errp, err);
             qemu_ram_munmap(fd, area, memory);
             return NULL;
         }
#19 lore raw

PATCH 9/21 hw/acpi: Fix latent legacy CPU plug error handling bug

In reply to Markus Armbruster

legacy_acpi_cpu_plug_cb() crashes when acpi_set_cpu_present_bit() fails and its @errp argument is null. Messed up in commit cc43364de7 "acpi/cpu-hotplug: introduce helper function to keep bit setting in one place".

The bug can't bite as no caller actually passes null, and acpi_set_cpu_present_bit() can't actually fail. Fix it anyway.

Cc "Michael S. Tsirkin" <[email protected]>
Cc Igor Mammedov <[email protected]>
---
 hw/acpi/cpu_hotplug.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)
hw/acpi/cpu_hotplug.c +3 -7
index 3ac2045a95..9c3bcc84de 100644
@@ -55,8 +55,7 @@ static const MemoryRegionOps AcpiCpuHotplug_ops = {
     },
 };
 
-static void acpi_set_cpu_present_bit(AcpiCpuHotplug *g, CPUState *cpu,
-                                     Error **errp)
+static void acpi_set_cpu_present_bit(AcpiCpuHotplug *g, CPUState *cpu)
 {
     CPUClass *k = CPU_GET_CLASS(cpu);
     int64_t cpu_id;
@@ -74,10 +73,7 @@ static void acpi_set_cpu_present_bit(AcpiCpuHotplug *g, CPUState *cpu,
 void legacy_acpi_cpu_plug_cb(HotplugHandler *hotplug_dev,
                              AcpiCpuHotplug *g, DeviceState *dev, Error **errp)
 {
-    acpi_set_cpu_present_bit(g, CPU(dev), errp);
-    if (*errp != NULL) {
-        return;
-    }
+    acpi_set_cpu_present_bit(g, CPU(dev));
     acpi_send_event(DEVICE(hotplug_dev), ACPI_CPU_HOTPLUG_STATUS);
 }
 
@@ -92,7 +88,7 @@ void legacy_acpi_cpu_hotplug_init(MemoryRegion *parent, Object *owner,
     gpe_cpu->device = owner;
 
     CPU_FOREACH(cpu) {
-        acpi_set_cpu_present_bit(gpe_cpu, cpu, &error_abort);
+        acpi_set_cpu_present_bit(gpe_cpu, cpu);
     }
 }
 
Signature · 1 line
2.21.0
Reviewed-by on v1 9/21
#20 lore raw

Re: [PATCH 09/21] hw/acpi: Fix latent legacy CPU plug error handling bug

Quoted 11 lines · On Sat, 30 Nov 2019 20:42:28 +0100 Markus Armbruster <[email protected]> wrote:

legacy_acpi_cpu_plug_cb() crashes when acpi_set_cpu_present_bit() fails and its @errp argument is null. Messed up in commit cc43364de7 "acpi/cpu-hotplug: introduce helper function to keep bit setting in one place".

The bug can't bite as no caller actually passes null, and acpi_set_cpu_present_bit() can't actually fail. Fix it anyway.

Cc "Michael S. Tsirkin" <[email protected]>
Cc Igor Mammedov <[email protected]>
Quoted 39 lines
---
 hw/acpi/cpu_hotplug.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)
hw/acpi/cpu_hotplug.c +3 -7
index 3ac2045a95..9c3bcc84de 100644
@@ -55,8 +55,7 @@ static const MemoryRegionOps AcpiCpuHotplug_ops = {
     },
 };
 
-static void acpi_set_cpu_present_bit(AcpiCpuHotplug *g, CPUState *cpu,
-                                     Error **errp)
+static void acpi_set_cpu_present_bit(AcpiCpuHotplug *g, CPUState *cpu)
 {
     CPUClass *k = CPU_GET_CLASS(cpu);
     int64_t cpu_id;
@@ -74,10 +73,7 @@ static void acpi_set_cpu_present_bit(AcpiCpuHotplug *g, CPUState *cpu,
 void legacy_acpi_cpu_plug_cb(HotplugHandler *hotplug_dev,
                              AcpiCpuHotplug *g, DeviceState *dev, Error **errp)
 {
-    acpi_set_cpu_present_bit(g, CPU(dev), errp);
-    if (*errp != NULL) {
-        return;
-    }
+    acpi_set_cpu_present_bit(g, CPU(dev));
     acpi_send_event(DEVICE(hotplug_dev), ACPI_CPU_HOTPLUG_STATUS);
 }
 
@@ -92,7 +88,7 @@ void legacy_acpi_cpu_hotplug_init(MemoryRegion *parent, Object *owner,
     gpe_cpu->device = owner;
 
     CPU_FOREACH(cpu) {
-        acpi_set_cpu_present_bit(gpe_cpu, cpu, &error_abort);
+        acpi_set_cpu_present_bit(gpe_cpu, cpu);
     }
 }
 
#21 lore raw

PATCH 10/21 hw/core: Fix latent fit_load_fdt() error handling bug

In reply to Markus Armbruster

fit_load_fdt() recovers from fit_image_addr() failing with ENOENT. Except it doesn't when its @errp argument is &error_fatal or &error_abort, because it blindly passes @errp to fit_image_addr(). Messed up in commit 3eb99edb48 "loader-fit: Wean off error_printf()".

The bug can't bite as no caller actually passes &error_fatal or &error_abort. Fix it anyway.

---
 hw/core/loader-fit.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
hw/core/loader-fit.c +8 -7
index 953b16bc82..c465921b8f 100644
@@ -178,11 +178,12 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
                         int cfg, void *opaque, const void *match_data,
                         hwaddr kernel_end, Error **errp)
 {
+    Error *err = NULL;
     const char *name;
     const void *data;
     const void *load_data;
     hwaddr load_addr;
-    int img_off, err;
+    int img_off;
     size_t sz;
     int ret;
 
@@ -197,13 +198,13 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
         return -EINVAL;
     }
 
-    err = fit_image_addr(itb, img_off, "load", &load_addr, errp);
-    if (err == -ENOENT) {
+    ret = fit_image_addr(itb, img_off, "load", &load_addr, &err);
+    if (ret == -ENOENT) {
         load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
-        error_free(*errp);
-    } else if (err) {
-        error_prepend(errp, "unable to read FDT load address from FIT: ");
-        ret = err;
+        error_free(err);
+    } else if (ret) {
+        error_propagate_prepend(errp, err,
+                                "unable to read FDT load address from FIT: ");
         goto out;
     }
 
Signature · 1 line
2.21.0
Reviewed-by on v1 10/21
#22 lore raw

Re: [PATCH 10/21] hw/core: Fix latent fit_load_fdt() error handling bug

Quoted 9 lines · Markus Armbruster

fit_load_fdt() recovers from fit_image_addr() failing with ENOENT. Except it doesn't when its @errp argument is &error_fatal or &error_abort, because it blindly passes @errp to fit_image_addr(). Messed up in commit 3eb99edb48 "loader-fit: Wean off error_printf()".

The bug can't bite as no caller actually passes &error_fatal or &error_abort. Fix it anyway.

Hmm, actually it makes my
"[PATCH v7 01/21] hw/core/loader-fit: fix freeing errp in fit_load_fdt" unnecessary. If you want you may drop my 01 (as it covers less problems), and in this case you may note in your cover letter, that
errp = NULL is broken here too (may be nobady pass it?),
and normal errp is handled wrong, as *errp doesn't set to NULL after error_free(*errp) (still, the only caller rely on return value more than on err, so the problem can't be triggered with current code)

Quoted 33 lines
---
  hw/core/loader-fit.c | 15 ++++++++-------
  1 file changed, 8 insertions(+), 7 deletions(-)
hw/core/loader-fit.c +4 -4
index 953b16bc82..c465921b8f 100644
@@ -178,11 +178,12 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
                          int cfg, void *opaque, const void *match_data,
                          hwaddr kernel_end, Error **errp)
  {
+    Error *err = NULL;
      const char *name;
      const void *data;
      const void *load_data;
      hwaddr load_addr;
-    int img_off, err;
+    int img_off;
      size_t sz;
      int ret;
 
@@ -197,13 +198,13 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
          return -EINVAL;
      }
 
-    err = fit_image_addr(itb, img_off, "load", &load_addr, errp);
-    if (err == -ENOENT) {
+    ret = fit_image_addr(itb, img_off, "load", &load_addr, &err);
+    if (ret == -ENOENT) {
          load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
-        error_free(*errp);

hw/core/loader-fit.c static int fit_load_fdt(const struct fit_loader *ldr, const void *itb, · line 210

Quoted 10 lines
hw/core/loader-fit.c +4 -3
-    } else if (err) {
-        error_prepend(errp, "unable to read FDT load address from FIT: ");
-        ret = err;
+        error_free(err);
+    } else if (ret) {
+        error_propagate_prepend(errp, err,
+                                "unable to read FDT load address from FIT: ");
          goto out;
      }
 

So much attention to that function :)

I'd also propose the following:

hw/core/loader-fit.c +4 -4
index c465921b8f..2c9efeef7a 100644
@@ -180,8 +180,8 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
  {
      Error *err = NULL;
      const char *name;
-    const void *data;
-    const void *load_data;
+    void *data;
+    void *load_data;
      hwaddr load_addr;
      int img_off;
      size_t sz;
@@ -218,9 +218,9 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
 
      ret = 0;
  out:
-    g_free((void *) data);
+    g_free(data);
      if (data != load_data) {
-        g_free((void *) load_data);
+        g_free(load_data);
      }
      return ret;
  }

Or, even better:

hw/core/loader-fit.c +7 -10
@@ -180,7 +180,8 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
  {
      Error *err = NULL;
      const char *name;
-    const void *data;
+    g_autofree void *data = NULL;
+    g_autofree void *fdt_filter_data = NULL;
      const void *load_data;
      hwaddr load_addr;
      int img_off;
@@ -202,27 +203,23 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
      if (ret == -ENOENT) {
          load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
          error_free(err);
+        return 0;
      } else if (ret) {
          error_propagate_prepend(errp, err,
                                  "unable to read FDT load address from FIT: ");
-        goto out;
+        return ret;
      }
 
      if (ldr->fdt_filter) {
-        load_data = ldr->fdt_filter(opaque, data, match_data, &load_addr);
+        load_data = fdt_filter_data =
+                ldr->fdt_filter(opaque, data, match_data, &load_addr);
      }
 
      load_addr = ldr->addr_to_phys(opaque, load_addr);
      sz = fdt_totalsize(load_data);
      rom_add_blob_fixed(name, load_data, sz, load_addr);
 
-    ret = 0;
-out:
-    g_free((void *) data);
-    if (data != load_data) {
-        g_free((void *) load_data);
-    }
-    return ret;
+    return 0;
  }
 
 
Signature · 2 lines
Best regards,
Vladimir
#23 lore raw

Re: [PATCH 10/21] hw/core: Fix latent fit_load_fdt() error handling bug

Quoted 9 lines · Vladimir Sementsov-Ogievskiy
Quoted 8 lines · Markus Armbruster

fit_load_fdt() recovers from fit_image_addr() failing with ENOENT. Except it doesn't when its @errp argument is &error_fatal or &error_abort, because it blindly passes @errp to fit_image_addr(). Messed up in commit 3eb99edb48 "loader-fit: Wean off error_printf()".

The bug can't bite as no caller actually passes &error_fatal or &error_abort. Fix it anyway.

Quoted 7 lines
Quoted 1 line

Hmm, actually it makes my
"[PATCH v7 01/21] hw/core/loader-fit: fix freeing errp in fit_load_fdt" unnecessary. If you want you may drop my 01 (as it covers less problems),

Yes.

Quoted 2 lines

and in this case you may note in your cover letter, that errp = NULL is broken here too (may be nobady pass it?),

You're right, null @errp is wrong, too.

Quoted 2 lines

and normal errp is handled wrong, as *errp doesn't set to NULL after error_free(*errp)

Yes, that's wrong, too. fit_load_fdt() itself doesn't use *errp after freeing it, but it sets a trap for its callers.

Quoted 2 lines

(still, the only caller rely on return value more than on
err, so the problem can't be triggered with current code)

True.

New commit message (based on v2's):

hw/core: Fix fit_load_fdt() error API violations

fit_load_fdt() passes @errp to fit_image_addr(), then recovers from ENOENT failures. Passing @errp is wrong, because it works only as long as @errp is neither @error_fatal nor @error_abort. Error recovery dereferences @errp. That's also wrong; see the big comment in error.h. Error recovery can leave *errp pointing to a freed Error object. Wrong, it must be null on success. Messed up in commit 3eb99edb48 "loader-fit: Wean off error_printf()".

No caller actually passes such values, or uses *errp on success.

Fix anyway: splice in a local Error *err, and error_propagate().

Okay?

Quoted 113 lines
Quoted 34 lines
---
  hw/core/loader-fit.c | 15 ++++++++-------
  1 file changed, 8 insertions(+), 7 deletions(-)
hw/core/loader-fit.c +4 -4
index 953b16bc82..c465921b8f 100644
@@ -178,11 +178,12 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
                          int cfg, void *opaque, const void *match_data,
                          hwaddr kernel_end, Error **errp)
  {
+    Error *err = NULL;
      const char *name;
      const void *data;
      const void *load_data;
      hwaddr load_addr;
-    int img_off, err;
+    int img_off;
      size_t sz;
      int ret;
 
@@ -197,13 +198,13 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
          return -EINVAL;
      }
 
-    err = fit_image_addr(itb, img_off, "load", &load_addr, errp);
-    if (err == -ENOENT) {
+    ret = fit_image_addr(itb, img_off, "load", &load_addr, &err);
+    if (ret == -ENOENT) {
          load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
-        error_free(*errp);

hw/core/loader-fit.c static int fit_load_fdt(const struct fit_loader *ldr, const void *itb, · line 210

Quoted 10 lines
hw/core/loader-fit.c +4 -3
-    } else if (err) {
-        error_prepend(errp, "unable to read FDT load address from FIT: ");
-        ret = err;
+        error_free(err);
+    } else if (ret) {
+        error_propagate_prepend(errp, err,
+                                "unable to read FDT load address from FIT: ");
          goto out;
      }
 

So much attention to that function :)

I'd also propose the following:

hw/core/loader-fit.c +4 -4
index c465921b8f..2c9efeef7a 100644
@@ -180,8 +180,8 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
  {
      Error *err = NULL;
      const char *name;
-    const void *data;
-    const void *load_data;
+    void *data;
+    void *load_data;
      hwaddr load_addr;
      int img_off;
      size_t sz;
@@ -218,9 +218,9 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
 
      ret = 0;
  out:
-    g_free((void *) data);
+    g_free(data);
      if (data != load_data) {
-        g_free((void *) load_data);
+        g_free(load_data);
      }
      return ret;
  }

Or, even better:

hw/core/loader-fit.c +6 -3
@@ -180,7 +180,8 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
  {
      Error *err = NULL;
      const char *name;
-    const void *data;
+    g_autofree void *data = NULL;
+    g_autofree void *fdt_filter_data = NULL;
      const void *load_data;
      hwaddr load_addr;
      int img_off;
@@ -202,27 +203,23 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
      if (ret == -ENOENT) {
          load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
          error_free(err);
+        return 0;
      } else if (ret) {
          error_propagate_prepend(errp, err,
                                  "unable to read FDT load address from FIT: ");
-        goto out;
+        return ret;
      }
 
      if (ldr->fdt_filter) {
-        load_data = ldr->fdt_filter(opaque, data, match_data, &load_addr);
+        load_data = fdt_filter_data =
+                ldr->fdt_filter(opaque, data, match_data, &load_addr);
      }
 
      load_addr = ldr->addr_to_phys(opaque, load_addr);
      sz = fdt_totalsize(load_data);
      rom_add_blob_fixed(name, load_data, sz, load_addr);

hw/core/loader-fit.c static int fit_load_fdt(const struct fit_loader *ldr, const void *itb, · line 223

Quoted 10 lines
hw/core/loader-fit.c +1 -7
 
-    ret = 0;
-out:
-    g_free((void *) data);
-    if (data != load_data) {
-        g_free((void *) load_data);
-    }
-    return ret;
+    return 0;
  }

Looks like a sensible separate cleanup patch to me. Care to post it?

Thanks!

#24 lore raw

Re: [PATCH 10/21] hw/core: Fix latent fit_load_fdt() error handling bug

Quoted 49 lines · Markus Armbruster
Quoted 9 lines · Vladimir Sementsov-Ogievskiy
Quoted 8 lines · Markus Armbruster

fit_load_fdt() recovers from fit_image_addr() failing with ENOENT. Except it doesn't when its @errp argument is &error_fatal or &error_abort, because it blindly passes @errp to fit_image_addr(). Messed up in commit 3eb99edb48 "loader-fit: Wean off error_printf()".

The bug can't bite as no caller actually passes &error_fatal or &error_abort. Fix it anyway.

Quoted 7 lines
Quoted 1 line

Hmm, actually it makes my
"[PATCH v7 01/21] hw/core/loader-fit: fix freeing errp in fit_load_fdt" unnecessary. If you want you may drop my 01 (as it covers less problems),

Yes.

Quoted 2 lines

and in this case you may note in your cover letter, that errp = NULL is broken here too (may be nobady pass it?),

You're right, null @errp is wrong, too.

Quoted 2 lines

and normal errp is handled wrong, as *errp doesn't set to NULL after error_free(*errp)

Yes, that's wrong, too. fit_load_fdt() itself doesn't use *errp after freeing it, but it sets a trap for its callers.

Quoted 2 lines

(still, the only caller rely on return value more than on
err, so the problem can't be triggered with current code)

True.

New commit message (based on v2's):

hw/core: Fix fit_load_fdt() error API violations

fit_load_fdt() passes @errp to fit_image_addr(), then recovers from ENOENT failures. Passing @errp is wrong, because it works only as long as @errp is neither @error_fatal nor @error_abort. Error recovery dereferences @errp. That's also wrong; see the big comment in error.h. Error recovery can leave *errp pointing to a freed Error object. Wrong, it must be null on success. Messed up in commit 3eb99edb48 "loader-fit: Wean off error_printf()".

Quoted 7 lines

No caller actually passes such values, or uses *errp on success.

Fix anyway: splice in a local Error *err, and error_propagate().

Okay?

Yes) also add

Quoted 115 lines
Quoted 115 lines
Quoted 35 lines

---
   hw/core/loader-fit.c | 15 ++++++++-------
   1 file changed, 8 insertions(+), 7 deletions(-)
hw/core/loader-fit.c +4 -4
index 953b16bc82..c465921b8f 100644
@@ -178,11 +178,12 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
                           int cfg, void *opaque, const void *match_data,
                           hwaddr kernel_end, Error **errp)
   {
+    Error *err = NULL;
       const char *name;
       const void *data;
       const void *load_data;
       hwaddr load_addr;
-    int img_off, err;
+    int img_off;
       size_t sz;
       int ret;
 
@@ -197,13 +198,13 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
           return -EINVAL;
       }
 
-    err = fit_image_addr(itb, img_off, "load", &load_addr, errp);
-    if (err == -ENOENT) {
+    ret = fit_image_addr(itb, img_off, "load", &load_addr, &err);
+    if (ret == -ENOENT) {
           load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
-        error_free(*errp);

hw/core/loader-fit.c static int fit_load_fdt(const struct fit_loader *ldr, const void *itb, · line 210

Quoted 10 lines
hw/core/loader-fit.c +4 -3
-    } else if (err) {
-        error_prepend(errp, "unable to read FDT load address from FIT: ");
-        ret = err;
+        error_free(err);
+    } else if (ret) {
+        error_propagate_prepend(errp, err,
+                                "unable to read FDT load address from FIT: ");
           goto out;
       }
 

So much attention to that function :)

I'd also propose the following:

hw/core/loader-fit.c +4 -4
index c465921b8f..2c9efeef7a 100644
@@ -180,8 +180,8 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
   {
       Error *err = NULL;
       const char *name;
-    const void *data;
-    const void *load_data;
+    void *data;
+    void *load_data;
       hwaddr load_addr;
       int img_off;
       size_t sz;
@@ -218,9 +218,9 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
 
       ret = 0;
   out:
-    g_free((void *) data);
+    g_free(data);
       if (data != load_data) {
-        g_free((void *) load_data);
+        g_free(load_data);
       }
       return ret;
   }

Or, even better:

hw/core/loader-fit.c +6 -3
@@ -180,7 +180,8 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
   {
       Error *err = NULL;
       const char *name;
-    const void *data;
+    g_autofree void *data = NULL;
+    g_autofree void *fdt_filter_data = NULL;
       const void *load_data;
       hwaddr load_addr;
       int img_off;
@@ -202,27 +203,23 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
       if (ret == -ENOENT) {
           load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
           error_free(err);
+        return 0;
       } else if (ret) {
           error_propagate_prepend(errp, err,
                                   "unable to read FDT load address from FIT: ");
-        goto out;
+        return ret;
       }
 
       if (ldr->fdt_filter) {
-        load_data = ldr->fdt_filter(opaque, data, match_data, &load_addr);
+        load_data = fdt_filter_data =
+                ldr->fdt_filter(opaque, data, match_data, &load_addr);
       }
 
       load_addr = ldr->addr_to_phys(opaque, load_addr);
       sz = fdt_totalsize(load_data);
       rom_add_blob_fixed(name, load_data, sz, load_addr);
 
Quoted 12 lines

hw/core/loader-fit.c static int fit_load_fdt(const struct fit_loader *ldr, const void *itb, · line 223

Quoted 9 lines
hw/core/loader-fit.c +1 -7
-    ret = 0;
-out:
-    g_free((void *) data);
-    if (data != load_data) {
-        g_free((void *) load_data);
-    }
-    return ret;
+    return 0;
   }

Looks like a sensible separate cleanup patch to me. Care to post it?

Yes, I'll send

Signature · 2 lines
Best regards,
Vladimir
#25 lore raw

Re: [PATCH 10/21] hw/core: Fix latent fit_load_fdt() error handling bug

Quoted 187 lines · Vladimir Sementsov-Ogievskiy
Quoted 49 lines · Markus Armbruster
Quoted 9 lines · Vladimir Sementsov-Ogievskiy
Quoted 8 lines · Markus Armbruster

fit_load_fdt() recovers from fit_image_addr() failing with ENOENT. Except it doesn't when its @errp argument is &error_fatal or &error_abort, because it blindly passes @errp to fit_image_addr(). Messed up in commit 3eb99edb48 "loader-fit: Wean off error_printf()".

The bug can't bite as no caller actually passes &error_fatal or &error_abort.  Fix it anyway.

Quoted 7 lines
Quoted 1 line

Hmm, actually it makes my
"[PATCH v7 01/21] hw/core/loader-fit: fix freeing errp in fit_load_fdt" unnecessary. If you want you may drop my 01 (as it covers less problems),

Yes.

Quoted 2 lines

and in this case you may note in your cover letter, that errp = NULL is broken here too (may be nobady pass it?),

You're right, null @errp is wrong, too.

Quoted 2 lines

and normal errp is handled wrong, as *errp doesn't set to NULL after error_free(*errp)

Yes, that's wrong, too.  fit_load_fdt() itself doesn't use *errp after freeing it, but it sets a trap for its callers.

Quoted 2 lines

                   (still, the only caller rely on return value more than on err, so the problem can't be triggered with current code)

True.

New commit message (based on v2's):

     hw/core: Fix fit_load_fdt() error API violations

     fit_load_fdt() passes @errp to fit_image_addr(), then recovers from ENOENT failures.  Passing @errp is wrong, because it works only as long as @errp is neither @error_fatal nor @error_abort.  Error recovery dereferences @errp.  That's also wrong; see the big comment in error.h.  Error recovery can leave *errp pointing to a freed Error object.  Wrong, it must be null on success.  Messed up in commit 3eb99edb48 "loader-fit: Wean off error_printf()".

Quoted 7 lines

     No caller actually passes such values, or uses *errp on success.

     Fix anyway: splice in a local Error *err, and error_propagate().

Okay?

Yes) also add

Quoted 112 lines
Quoted 112 lines
Quoted 32 lines

---
   hw/core/loader-fit.c | 15 ++++++++-------

   1 file changed, 8 insertions(+), 7 deletions(-)
hw/core/loader-fit.c +4 -4
index 953b16bc82..c465921b8f 100644
@@ -178,11 +178,12 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
                            int cfg, void *opaque, const void *match_data,
                            hwaddr kernel_end, Error **errp)
    {
+    Error *err = NULL;
        const char *name;
        const void *data;
        const void *load_data;
        hwaddr load_addr;
-    int img_off, err;
+    int img_off;
        size_t sz;
        int ret;
@@ -197,13 +198,13 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
            return -EINVAL;
        }
-    err = fit_image_addr(itb, img_off, "load", &load_addr, errp);
-    if (err == -ENOENT) {
+    ret = fit_image_addr(itb, img_off, "load", &load_addr, &err);
+    if (ret == -ENOENT) {
            load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
-        error_free(*errp);

hw/core/loader-fit.c static int fit_load_fdt(const struct fit_loader *ldr, const void *itb, · line 209

Quoted 10 lines
hw/core/loader-fit.c +4 -3
-    } else if (err) {
-        error_prepend(errp, "unable to read FDT load address from FIT: ");
-        ret = err;
+        error_free(err);
+    } else if (ret) {
+        error_propagate_prepend(errp, err,
+                                "unable to read FDT load address from FIT: ");
            goto out;
        }
 

So much attention to that function :)

I'd also propose the following:

hw/core/loader-fit.c +2 -2
index c465921b8f..2c9efeef7a 100644
@@ -180,8 +180,8 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
    {
        Error *err = NULL;
        const char *name;
-    const void *data;
-    const void *load_data;
+    void *data;
+    void *load_data;
        hwaddr load_addr;
        int img_off;
        size_t sz;
@@ -218,9 +218,9 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
 
        ret = 0;

   out:
-    g_free((void *) data);
+    g_free(data);
       if (data != load_data) {
-        g_free((void *) load_data);
+        g_free(load_data);
       }
       return ret;
   }

Or, even better:

hw/core/loader-fit.c +6 -3
@@ -180,7 +180,8 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
    {
        Error *err = NULL;
        const char *name;
-    const void *data;
+    g_autofree void *data = NULL;
+    g_autofree void *fdt_filter_data = NULL;
        const void *load_data;
        hwaddr load_addr;
        int img_off;
@@ -202,27 +203,23 @@ static int fit_load_fdt(const struct fit_loader *ldr, const void *itb,
        if (ret == -ENOENT) {
            load_addr = ROUND_UP(kernel_end, 64 * KiB) + (10 * MiB);
            error_free(err);
+        return 0;
        } else if (ret) {
            error_propagate_prepend(errp, err,
                                    "unable to read FDT load address from FIT: ");
-        goto out;
+        return ret;
        }
 
        if (ldr->fdt_filter) {
-        load_data = ldr->fdt_filter(opaque, data, match_data, &load_addr);
+        load_data = fdt_filter_data =
+                ldr->fdt_filter(opaque, data, match_data, &load_addr);
        }
 
        load_addr = ldr->addr_to_phys(opaque, load_addr);
        sz = fdt_totalsize(load_data);
        rom_add_blob_fixed(name, load_data, sz, load_addr);
 
Quoted 12 lines

hw/core/loader-fit.c static int fit_load_fdt(const struct fit_loader *ldr, const void *itb, · line 223

Quoted 9 lines
hw/core/loader-fit.c +1 -7
-    ret = 0;
-out:
-    g_free((void *) data);
-    if (data != load_data) {
-        g_free((void *) load_data);
-    }
-    return ret;
+    return 0;
    }

Looks like a sensible separate cleanup patch to me.  Care to post it?

Quoted 2 lines

Yes, I'll send

Hm, it doesn't compile, as fit_load_image_alloc return type is const pointer.. So, I just don't understand the logic of the code (for me it's strange to free pointer, returned by function, returning const pointer)

And what are the reasons, to make return type of fit_load_image_alloc constant, I don't see.. IMHO, there are no reasons.
Same thing with .fdt_filter ...

I don't want to dig in, so, sorry, I'll not send :)

Signature · 2 lines
Best regards,
Vladimir
#26 lore raw

Re: [PATCH 10/21] hw/core: Fix latent fit_load_fdt() error handling bug

Quoted 3 lines · Vladimir Sementsov-Ogievskiy
Quoted 2 lines · Vladimir Sementsov-Ogievskiy
Quoted 1 line · Markus Armbruster

Vladimir Sementsov-Ogievskiy <[email protected]> writes:

[...]

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

So much attention to that function :)

I'd also propose the following:

[...]

Quoted 12 lines
Quoted 6 lines
Quoted 3 lines

Looks like a sensible separate cleanup patch to me.  Care to post it?

Yes, I'll send

Hm, it doesn't compile, as fit_load_image_alloc return type is const pointer.. So, I just don't understand the logic of the code (for me it's strange to free pointer, returned by function, returning const pointer)

Quoted 5 lines

And what are the reasons, to make return type of fit_load_image_alloc constant, I don't see.. IMHO, there are no reasons.
Same thing with .fdt_filter ...

I don't want to dig in, so, sorry, I'll not send :)

That's okay.

#27 lore raw

PATCH 11/21 hw/ipmi: Fix latent realize() error handling bugs

In reply to Markus Armbruster

isa_ipmi_bt_realize(), ipmi_isa_realize(), pci_ipmi_bt_realize(), and pci_ipmi_kcs_realize() crash when IPMIInterfaceClass method init() fails and their @errp argument is null. First messed up in commit 0719029c47 "ipmi: Add an ISA KCS low-level interface", then imitated in commit a9b74079cb "ipmi: Add a BT low-level interface" and commit 12f983c6aa "ipmi: Add PCI IPMI interfaces".

The bug can't bite as no caller actually passes null, and none of the init() methods can actually fail. Fix it anyway.

---
 hw/ipmi/isa_ipmi_bt.c  | 7 +++++--
 hw/ipmi/isa_ipmi_kcs.c | 7 +++++--
 hw/ipmi/pci_ipmi_bt.c  | 6 ++++--
 hw/ipmi/pci_ipmi_kcs.c | 6 ++++--
 4 files changed, 18 insertions(+), 8 deletions(-)
hw/ipmi/isa_ipmi_bt.c +5 -2
index 9a87ffd3f0..9fba5ed383 100644
@@ -70,6 +70,7 @@ static void isa_ipmi_bt_lower_irq(IPMIBT *ib)
 
 static void isa_ipmi_bt_realize(DeviceState *dev, Error **errp)
 {
+    Error *err = NULL;
     ISADevice *isadev = ISA_DEVICE(dev);
     ISAIPMIBTDevice *iib = ISA_IPMI_BT(dev);
     IPMIInterface *ii = IPMI_INTERFACE(dev);
@@ -85,9 +86,11 @@ static void isa_ipmi_bt_realize(DeviceState *dev, Error **errp)
     iib->bt.bmc->intf = ii;
     iib->bt.opaque = iib;
 
-    iic->init(ii, 0, errp);
-    if (*errp)
+    iic->init(ii, 0, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
+    }
 
     if (iib->isairq > 0) {
         isa_init_irq(isadev, &iib->irq, iib->isairq);
hw/ipmi/isa_ipmi_kcs.c +5 -2
index ca3ea36a3f..cc6bd817f2 100644
@@ -69,6 +69,7 @@ static void isa_ipmi_kcs_lower_irq(IPMIKCS *ik)
 
 static void ipmi_isa_realize(DeviceState *dev, Error **errp)
 {
+    Error *err = NULL;
     ISADevice *isadev = ISA_DEVICE(dev);
     ISAIPMIKCSDevice *iik = ISA_IPMI_KCS(dev);
     IPMIInterface *ii = IPMI_INTERFACE(dev);
@@ -84,9 +85,11 @@ static void ipmi_isa_realize(DeviceState *dev, Error **errp)
     iik->kcs.bmc->intf = ii;
     iik->kcs.opaque = iik;
 
-    iic->init(ii, 0, errp);
-    if (*errp)
+    iic->init(ii, 0, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
+    }
 
     if (iik->isairq > 0) {
         isa_init_irq(isadev, &iik->irq, iik->isairq);
hw/ipmi/pci_ipmi_bt.c +4 -2
index 6ed925a665..ba9cf016b5 100644
@@ -54,6 +54,7 @@ static void pci_ipmi_lower_irq(IPMIBT *ik)
 
 static void pci_ipmi_bt_realize(PCIDevice *pd, Error **errp)
 {
+    Error *err = NULL;
     PCIIPMIBTDevice *pik = PCI_IPMI_BT(pd);
     IPMIInterface *ii = IPMI_INTERFACE(pd);
     IPMIInterfaceClass *iic = IPMI_INTERFACE_GET_CLASS(ii);
@@ -74,8 +75,9 @@ static void pci_ipmi_bt_realize(PCIDevice *pd, Error **errp)
     pik->bt.raise_irq = pci_ipmi_raise_irq;
     pik->bt.lower_irq = pci_ipmi_lower_irq;
 
-    iic->init(ii, 8, errp);
-    if (*errp) {
+    iic->init(ii, 8, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
     pci_register_bar(pd, 0, PCI_BASE_ADDRESS_SPACE_IO, &pik->bt.io);
hw/ipmi/pci_ipmi_kcs.c +4 -2
index eeba63baa4..99f46152f4 100644
@@ -54,6 +54,7 @@ static void pci_ipmi_lower_irq(IPMIKCS *ik)
 
 static void pci_ipmi_kcs_realize(PCIDevice *pd, Error **errp)
 {
+    Error *err = NULL;
     PCIIPMIKCSDevice *pik = PCI_IPMI_KCS(pd);
     IPMIInterface *ii = IPMI_INTERFACE(pd);
     IPMIInterfaceClass *iic = IPMI_INTERFACE_GET_CLASS(ii);
@@ -74,8 +75,9 @@ static void pci_ipmi_kcs_realize(PCIDevice *pd, Error **errp)
     pik->kcs.raise_irq = pci_ipmi_raise_irq;
     pik->kcs.lower_irq = pci_ipmi_lower_irq;
 
-    iic->init(ii, 8, errp);
-    if (*errp) {
+    iic->init(ii, 8, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
     pci_register_bar(pd, 0, PCI_BASE_ADDRESS_SPACE_IO, &pik->kcs.io);
Signature · 1 line
2.21.0
#28 lore raw

Re: [PATCH 11/21] hw/ipmi: Fix latent realize() error handling bugs

Quoted 9 lines · Markus Armbruster

isa_ipmi_bt_realize(), ipmi_isa_realize(), pci_ipmi_bt_realize(), and pci_ipmi_kcs_realize() crash when IPMIInterfaceClass method init() fails and their @errp argument is null. First messed up in commit 0719029c47 "ipmi: Add an ISA KCS low-level interface", then imitated in commit a9b74079cb "ipmi: Add a BT low-level interface" and commit 12f983c6aa "ipmi: Add PCI IPMI interfaces".

The bug can't bite as no caller actually passes null, and none of the init() methods can actually fail. Fix it anyway.

Well, whatever. It looks correct and is better style. I've added this to my tree.

-corey

Quoted 113 lines
---
 hw/ipmi/isa_ipmi_bt.c  | 7 +++++--
 hw/ipmi/isa_ipmi_kcs.c | 7 +++++--
 hw/ipmi/pci_ipmi_bt.c  | 6 ++++--
 hw/ipmi/pci_ipmi_kcs.c | 6 ++++--
 4 files changed, 18 insertions(+), 8 deletions(-)
hw/ipmi/isa_ipmi_bt.c +5 -2
index 9a87ffd3f0..9fba5ed383 100644
@@ -70,6 +70,7 @@ static void isa_ipmi_bt_lower_irq(IPMIBT *ib)
 
 static void isa_ipmi_bt_realize(DeviceState *dev, Error **errp)
 {
+    Error *err = NULL;
     ISADevice *isadev = ISA_DEVICE(dev);
     ISAIPMIBTDevice *iib = ISA_IPMI_BT(dev);
     IPMIInterface *ii = IPMI_INTERFACE(dev);
@@ -85,9 +86,11 @@ static void isa_ipmi_bt_realize(DeviceState *dev, Error **errp)
     iib->bt.bmc->intf = ii;
     iib->bt.opaque = iib;
 
-    iic->init(ii, 0, errp);
-    if (*errp)
+    iic->init(ii, 0, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
+    }
 
     if (iib->isairq > 0) {
         isa_init_irq(isadev, &iib->irq, iib->isairq);
hw/ipmi/isa_ipmi_kcs.c +5 -2
index ca3ea36a3f..cc6bd817f2 100644
@@ -69,6 +69,7 @@ static void isa_ipmi_kcs_lower_irq(IPMIKCS *ik)
 
 static void ipmi_isa_realize(DeviceState *dev, Error **errp)
 {
+    Error *err = NULL;
     ISADevice *isadev = ISA_DEVICE(dev);
     ISAIPMIKCSDevice *iik = ISA_IPMI_KCS(dev);
     IPMIInterface *ii = IPMI_INTERFACE(dev);
@@ -84,9 +85,11 @@ static void ipmi_isa_realize(DeviceState *dev, Error **errp)
     iik->kcs.bmc->intf = ii;
     iik->kcs.opaque = iik;
 
-    iic->init(ii, 0, errp);
-    if (*errp)
+    iic->init(ii, 0, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
+    }
 
     if (iik->isairq > 0) {
         isa_init_irq(isadev, &iik->irq, iik->isairq);
hw/ipmi/pci_ipmi_bt.c +4 -2
index 6ed925a665..ba9cf016b5 100644
@@ -54,6 +54,7 @@ static void pci_ipmi_lower_irq(IPMIBT *ik)
 
 static void pci_ipmi_bt_realize(PCIDevice *pd, Error **errp)
 {
+    Error *err = NULL;
     PCIIPMIBTDevice *pik = PCI_IPMI_BT(pd);
     IPMIInterface *ii = IPMI_INTERFACE(pd);
     IPMIInterfaceClass *iic = IPMI_INTERFACE_GET_CLASS(ii);
@@ -74,8 +75,9 @@ static void pci_ipmi_bt_realize(PCIDevice *pd, Error **errp)
     pik->bt.raise_irq = pci_ipmi_raise_irq;
     pik->bt.lower_irq = pci_ipmi_lower_irq;
 
-    iic->init(ii, 8, errp);
-    if (*errp) {
+    iic->init(ii, 8, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
     pci_register_bar(pd, 0, PCI_BASE_ADDRESS_SPACE_IO, &pik->bt.io);
hw/ipmi/pci_ipmi_kcs.c +4 -2
index eeba63baa4..99f46152f4 100644
@@ -54,6 +54,7 @@ static void pci_ipmi_lower_irq(IPMIKCS *ik)
 
 static void pci_ipmi_kcs_realize(PCIDevice *pd, Error **errp)
 {
+    Error *err = NULL;
     PCIIPMIKCSDevice *pik = PCI_IPMI_KCS(pd);
     IPMIInterface *ii = IPMI_INTERFACE(pd);
     IPMIInterfaceClass *iic = IPMI_INTERFACE_GET_CLASS(ii);
@@ -74,8 +75,9 @@ static void pci_ipmi_kcs_realize(PCIDevice *pd, Error **errp)
     pik->kcs.raise_irq = pci_ipmi_raise_irq;
     pik->kcs.lower_irq = pci_ipmi_lower_irq;
 
-    iic->init(ii, 8, errp);
-    if (*errp) {
+    iic->init(ii, 8, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
     pci_register_bar(pd, 0, PCI_BASE_ADDRESS_SPACE_IO, &pik->kcs.io);
Signature · 1 line
2.21.0
#29 lore raw

Re: [PATCH 11/21] hw/ipmi: Fix latent realize() error handling bugs

Quoted 8 lines · Corey Minyard
Quoted 7 lines · Markus Armbruster

isa_ipmi_bt_realize(), ipmi_isa_realize(), pci_ipmi_bt_realize(), and pci_ipmi_kcs_realize() crash when IPMIInterfaceClass method init() fails and their @errp argument is null. First messed up in commit 0719029c47 "ipmi: Add an ISA KCS low-level interface", then imitated in commit a9b74079cb "ipmi: Add a BT low-level interface" and commit 12f983c6aa "ipmi: Add PCI IPMI interfaces".

Quoted 5 lines
Quoted 2 lines

The bug can't bite as no caller actually passes null, and none of the init() methods can actually fail. Fix it anyway.

Well, whatever. It looks correct and is better style. I've added this to my tree.

I've since posted v2 with a revamped commit message, and I'm ready to post a pull request. I really want the whole thing committed before the Christmas break, so Vladimir can base on it more easily. Options:

* You post a pull request before me.

* Ask me to drop this patch from my pull request, so you can take it through your tree at your leisure.

* Post your Reviewed-by or Acked-by for me to include in my pull request.

* Do nothing; I'll post my pull request later this week.

#30 lore raw

Re: [PATCH 11/21] hw/ipmi: Fix latent realize() error handling bugs

Quoted 8 lines · Markus Armbruster

Corey Minyard <[email protected]> writes:

I've since posted v2 with a revamped commit message, and I'm ready to post a pull request. I really want the whole thing committed before the Christmas break, so Vladimir can base on it more easily. Options:

* You post a pull request before me.

Quoted 8 lines

* Ask me to drop this patch from my pull request, so you can take it through your tree at your leisure.

* Post your Reviewed-by or Acked-by for me to include in my pull request.

* Do nothing; I'll post my pull request later this week.

Since you already have it ready, I'll choose the "do nothing" patch and let you post the pull request. Thanks.

-corey

#31 lore raw

Re: [PATCH 11/21] hw/ipmi: Fix latent realize() error handling bugs

Quoted 12 lines · Corey Minyard
Quoted 11 lines · Markus Armbruster

Corey Minyard <[email protected]> writes:

I've since posted v2 with a revamped commit message, and I'm ready to post a pull request. I really want the whole thing committed before the Christmas break, so Vladimir can base on it more easily. Options:

* You post a pull request before me.

* Ask me to drop this patch from my pull request, so you can take it through your tree at your leisure.

Quoted 8 lines
Quoted 5 lines

* Post your Reviewed-by or Acked-by for me to include in my pull request.

* Do nothing; I'll post my pull request later this week.

Since you already have it ready, I'll choose the "do nothing" patch and let you post the pull request. Thanks.

Okay, sent. Thanks for your quick reply!

#32 lore raw

PATCH 12/21 qga: Fix latent guest-get-fsinfo error handling bug

In reply to Markus Armbruster

build_guest_fsinfo_for_virtual_device() crashes when build_guest_fsinfo_for_device() fails and its @errp argument is null. Messed up in commit 46d4c5723e "qga: Add guest-get-fsinfo command".

The bug can't bite as no caller actually passes null. Fix it anyway.

---
 qga/commands-posix.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
qga/commands-posix.c +4 -2
index 1c1a165dae..0be527ccb8 100644
@@ -1049,6 +1049,7 @@ static void build_guest_fsinfo_for_virtual_device(char const *syspath,
                                                   GuestFilesystemInfo *fs,
                                                   Error **errp)
 {
+    Error *err = NULL;
     DIR *dir;
     char *dirpath;
     struct dirent *entry;
@@ -1078,10 +1079,11 @@ static void build_guest_fsinfo_for_virtual_device(char const *syspath,
 
             g_debug(" slave device '%s'", entry->d_name);
             path = g_strdup_printf("%s/slaves/%s", syspath, entry->d_name);
-            build_guest_fsinfo_for_device(path, fs, errp);
+            build_guest_fsinfo_for_device(path, fs, &err);
             g_free(path);
 
-            if (*errp) {
+            if (err) {
+                error_propagate(errp, err);
                 break;
             }
         }
Signature · 1 line
2.21.0
Reviewed-by on v1 12/21
#33 lore raw

Re: [PATCH 12/21] qga: Fix latent guest-get-fsinfo error handling bug

Quoted 8 lines · Markus Armbruster

build_guest_fsinfo_for_virtual_device() crashes when build_guest_fsinfo_for_device() fails and its @errp argument is null. Messed up in commit 46d4c5723e "qga: Add guest-get-fsinfo command".

The bug can't bite as no caller actually passes null. Fix it anyway.

Actually, all such bugs should be fixed by my auto-generated series.. And, if fixing by hand, it may be better to teach this function to return int, than propagation is not needed.

Quoted 31 lines
---
  qga/commands-posix.c | 6 ++++--
  1 file changed, 4 insertions(+), 2 deletions(-)
qga/commands-posix.c +4 -2
index 1c1a165dae..0be527ccb8 100644
@@ -1049,6 +1049,7 @@ static void build_guest_fsinfo_for_virtual_device(char const *syspath,
                                                    GuestFilesystemInfo *fs,
                                                    Error **errp)
  {
+    Error *err = NULL;
      DIR *dir;
      char *dirpath;
      struct dirent *entry;
@@ -1078,10 +1079,11 @@ static void build_guest_fsinfo_for_virtual_device(char const *syspath,
 
              g_debug(" slave device '%s'", entry->d_name);
              path = g_strdup_printf("%s/slaves/%s", syspath, entry->d_name);
-            build_guest_fsinfo_for_device(path, fs, errp);
+            build_guest_fsinfo_for_device(path, fs, &err);
              g_free(path);
 
-            if (*errp) {
+            if (err) {
+                error_propagate(errp, err);
                  break;
              }
          }
Signature · 2 lines
Best regards,
Vladimir
#34 lore raw

Re: [PATCH 12/21] qga: Fix latent guest-get-fsinfo error handling bug

Quoted 7 lines · Vladimir Sementsov-Ogievskiy
Quoted 6 lines · Markus Armbruster

build_guest_fsinfo_for_virtual_device() crashes when build_guest_fsinfo_for_device() fails and its @errp argument is null. Messed up in commit 46d4c5723e "qga: Add guest-get-fsinfo command".

The bug can't bite as no caller actually passes null. Fix it anyway.

Quoted 7 lines
Quoted 2 lines

Actually, all such bugs should be fixed by my auto-generated series..

I see. I didn't consider that.

One advantage of my manual fixing is a clearer record of the flaws in git. It should also keep your work simpler, which is always a good idea for massive, mechanical patches.

Quoted 2 lines

And, if fixing by hand, it may be better to teach this function to return int, than propagation is not needed.

I went for the minimal fix.

I believe returning something useful is better. It also matches the GError conventions. Deviating from them in this point was a mistake.

Touching up functions to return something useful by hand is a lot of work, though: functions have many returns, and we have many functions in need of this touch-up. Some clever Coccinellery might be able to pull it off. I haven't tried.

However, your clever "auto propagation" Coccinellery makes such a change less compelling, because it hides the propagation.

Thanks!

#35 lore raw

PATCH 13/21 memory-device: Fix latent memory pre-plug error handling bugs

In reply to Markus Armbruster

memory_device_get_free_addr() crashes when memory_device_check_addable() fails and its @errp argument is null. Messed up in commit 1b6d6af21b "pc-dimm: factor out capacity and slot checks into MemoryDevice".

The bug can't bite as no caller actually passes null. Fix it anyway.

---
 hw/mem/memory-device.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
hw/mem/memory-device.c +4 -2
index aef148c1d7..4bc9cf0917 100644
@@ -99,6 +99,7 @@ static uint64_t memory_device_get_free_addr(MachineState *ms,
                                             uint64_t align, uint64_t size,
                                             Error **errp)
 {
+    Error *err = NULL;
     GSList *list = NULL, *item;
     Range as, new = range_empty;
 
@@ -123,8 +124,9 @@ static uint64_t memory_device_get_free_addr(MachineState *ms,
         return 0;
     }
 
-    memory_device_check_addable(ms, size, errp);
-    if (*errp) {
+    memory_device_check_addable(ms, size, &err);
+    if (err) {
+        error_propagate(errp, err);
         return 0;
     }
 
Signature · 1 line
2.21.0
Reviewed-by on v1 13/21
#36 lore raw

Re: [PATCH 13/21] memory-device: Fix latent memory pre-plug error handling bugs

Quoted 26 lines · Markus Armbruster

memory_device_get_free_addr() crashes when memory_device_check_addable() fails and its @errp argument is null. Messed up in commit 1b6d6af21b "pc-dimm: factor out capacity and slot checks into MemoryDevice".

The bug can't bite as no caller actually passes null. Fix it anyway.

---
 hw/mem/memory-device.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
hw/mem/memory-device.c +1 -0
index aef148c1d7..4bc9cf0917 100644
@@ -99,6 +99,7 @@ static uint64_t memory_device_get_free_addr(MachineState *ms,
                                             uint64_t align, uint64_t size,
                                             Error **errp)
 {
+    Error *err = NULL;
     GSList *list = NULL, *item;
     Range as, new = range_empty;
 

hw/mem/memory-device.c static uint64_t memory_device_get_free_addr(MachineState *ms, · line 131

Quoted 10 lines
hw/mem/memory-device.c +3 -2
@@ -123,8 +124,9 @@ static uint64_t memory_device_get_free_addr(MachineState *ms,
         return 0;
     }
 
-    memory_device_check_addable(ms, size, errp);
-    if (*errp) {
+    memory_device_check_addable(ms, size, &err);
+    if (err) {
+        error_propagate(errp, err);
         return 0;
     }

I remember that some time ago, the best practice was to use "local_err", what is the latest state of that?

I still disagree that these are BUGs or even latent BUGs. If somebody things these are BUGs and not cleanups, then we should probably have proper "Fixes: " tags instead.

Signature · 3 lines
Thanks,

David / dhildenb
#37 lore raw

Re: [PATCH 13/21] memory-device: Fix latent memory pre-plug error handling bugs

Quoted 30 lines · David Hildenbrand
Quoted 29 lines · Markus Armbruster

memory_device_get_free_addr() crashes when memory_device_check_addable() fails and its @errp argument is null. Messed up in commit 1b6d6af21b "pc-dimm: factor out capacity and slot checks into MemoryDevice".

The bug can't bite as no caller actually passes null. Fix it anyway.

---
 hw/mem/memory-device.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
hw/mem/memory-device.c +1 -0
index aef148c1d7..4bc9cf0917 100644
@@ -99,6 +99,7 @@ static uint64_t memory_device_get_free_addr(MachineState *ms,
                                             uint64_t align, uint64_t size,
                                             Error **errp)
 {
+    Error *err = NULL;
     GSList *list = NULL, *item;
     Range as, new = range_empty;
 
@@ -123,8 +124,9 @@ static uint64_t memory_device_get_free_addr(MachineState *ms,
         return 0;
     }
 
Quoted 10 lines

hw/mem/memory-device.c static uint64_t memory_device_get_free_addr(MachineState *ms, · line 131

Quoted 7 lines
hw/mem/memory-device.c +3 -2
-    memory_device_check_addable(ms, size, errp);
-    if (*errp) {
+    memory_device_check_addable(ms, size, &err);
+    if (err) {
+        error_propagate(errp, err);
         return 0;
     }

I remember that some time ago, the best practice was to use "local_err", what is the latest state of that?

Hundreds of local Error * variables are named @local_err, and hundreds more are named @err.

For what it's worth, the big comment in error.h uses @err, except in one place where it needs two of them.

Quoted 3 lines

I still disagree that these are BUGs or even latent BUGs. If somebody things these are BUGs and not cleanups, then we should probably have proper "Fixes: " tags instead.

Let's continue that discussion in the sub-thread where you first raised this objection.

Quoted 1 line

Thanks!

#38 lore raw

Re: [PATCH 13/21] memory-device: Fix latent memory pre-plug error handling bugs

Quoted 11 lines · Markus Armbruster
Quoted 5 lines
Quoted 2 lines

{
+ Error *err = NULL;

I remember that some time ago, the best practice was to use "local_err", what is the latest state of that?

Hundreds of local Error * variables are named @local_err, and hundreds more are named @err.

For what it's worth, the big comment in error.h uses @err, except in one place where it needs two of them.

What's more, if we go through with Vladimir's Coccinelle cleanup to use ERRP_AUTO_PROPAGATE, then we don't have either name to worry about (both '&local_err' and '&err' are replaced by 'errp').

Quoted 7 lines
Quoted 4 lines

I still disagree that these are BUGs or even latent BUGs. If somebody things these are BUGs and not cleanups, then we should probably have proper "Fixes: " tags instead.

Let's continue that discussion in the sub-thread where you first raised this objection.

One benefit of fixing the style (whether or not you count it as a bug fix) is that the Coccinelle script for updating to a new style is more likely to apply correctly.

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
#39 lore raw

PATCH 14/21 s390x/event-facility: Fix latent realize() error handling bug

In reply to Markus Armbruster

sclp_events_bus_realize() crashes when object_property_set_bool() fails and its @errp argument is null. Messed up in commit f6102c329c "s390/sclp: rework sclp event facility initialization + device realization".

The bug can't bite as no caller actually passes null. Fix it anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 hw/s390x/event-facility.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
hw/s390x/event-facility.c +4 -2
index 66205697ae..cdcf9154c4 100644
@@ -339,14 +339,16 @@ out:
 
 static void sclp_events_bus_realize(BusState *bus, Error **errp)
 {
+    Error *err = NULL;
     BusChild *kid;
 
     /* TODO: recursive realization has to be done in common code */
     QTAILQ_FOREACH(kid, &bus->children, sibling) {
         DeviceState *dev = kid->child;
 
-        object_property_set_bool(OBJECT(dev), true, "realized", errp);
-        if (*errp) {
+        object_property_set_bool(OBJECT(dev), true, "realized", &err);
+        if (errp) {
+            error_propagate(errp, err);
             return;
         }
     }
Signature · 1 line
2.21.0
Reviewed-by on v1 14/21
#40 lore raw

Re: [PATCH 14/21] s390x/event-facility: Fix latent realize() error handling bug

Quoted 5 lines · Markus Armbruster

sclp_events_bus_realize() crashes when object_property_set_bool() fails and its @errp argument is null. Messed up in commit f6102c329c "s390/sclp: rework sclp event facility initialization + device realization".

s/crashes .../would crash .../ ...

Quoted 33 lines

The bug can't bite as no caller actually passes null. Fix it anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 hw/s390x/event-facility.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
hw/s390x/event-facility.c +4 -2
index 66205697ae..cdcf9154c4 100644
@@ -339,14 +339,16 @@ out:
 
 static void sclp_events_bus_realize(BusState *bus, Error **errp)
 {
+    Error *err = NULL;
     BusChild *kid;
 
     /* TODO: recursive realization has to be done in common code */
     QTAILQ_FOREACH(kid, &bus->children, sibling) {
         DeviceState *dev = kid->child;
 
-        object_property_set_bool(OBJECT(dev), true, "realized", errp);
-        if (*errp) {
+        object_property_set_bool(OBJECT(dev), true, "realized", &err);
+        if (errp) {
+            error_propagate(errp, err);
             return;
         }
     }
Signature · 3 lines
Thanks,

David / dhildenb
#41 lore raw

PATCH 15/21 s390x/cpu_models: Fix latent feature property error handling bugs

In reply to Markus Armbruster

s390x-cpu property setters set_feature() and set_feature_group() crash when the visitor fails and its @errp argument is null. Messed up in commit 0754f60429 "s390x/cpumodel: expose features and feature groups as properties".

The bug can't bite as no caller actually passes null. Fix it anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
target/s390x/cpu_models.c +8 -4
index 7e92fb2e15..6a29fd3ab1 100644
@@ -987,6 +987,7 @@ static void get_feature(Object *obj, Visitor *v, const char *name,
 static void set_feature(Object *obj, Visitor *v, const char *name,
                         void *opaque, Error **errp)
 {
+    Error *err = NULL;
     S390Feat feat = (S390Feat) opaque;
     DeviceState *dev = DEVICE(obj);
     S390CPU *cpu = S390_CPU(obj);
@@ -1002,8 +1003,9 @@ static void set_feature(Object *obj, Visitor *v, const char *name,
         return;
     }
 
-    visit_type_bool(v, name, &value, errp);
-    if (*errp) {
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
     if (value) {
@@ -1043,6 +1045,7 @@ static void get_feature_group(Object *obj, Visitor *v, const char *name,
 static void set_feature_group(Object *obj, Visitor *v, const char *name,
                               void *opaque, Error **errp)
 {
+    Error *err = NULL;
     S390FeatGroup group = (S390FeatGroup) opaque;
     const S390FeatGroupDef *def = s390_feat_group_def(group);
     DeviceState *dev = DEVICE(obj);
@@ -1059,8 +1062,9 @@ static void set_feature_group(Object *obj, Visitor *v, const char *name,
         return;
     }
 
-    visit_type_bool(v, name, &value, errp);
-    if (*errp) {
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
     if (value) {
Signature · 1 line
2.21.0
Reviewed-by on v1 15/21
#42 lore raw

Re: [PATCH 15/21] s390x/cpu_models: Fix latent feature property error handling bugs

Quoted 4 lines · Markus Armbruster

s390x-cpu property setters set_feature() and set_feature_group() crash when the visitor fails and its @errp argument is null. Messed up in commit 0754f60429 "s390x/cpumodel: expose features and feature groups as properties".

Same comment as to the other patches :)

I think we usually use "s390x/cpumodels", but that's just nitpicking.

Quoted 55 lines

The bug can't bite as no caller actually passes null. Fix it anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
target/s390x/cpu_models.c +8 -4
index 7e92fb2e15..6a29fd3ab1 100644
@@ -987,6 +987,7 @@ static void get_feature(Object *obj, Visitor *v, const char *name,
 static void set_feature(Object *obj, Visitor *v, const char *name,
                         void *opaque, Error **errp)
 {
+    Error *err = NULL;
     S390Feat feat = (S390Feat) opaque;
     DeviceState *dev = DEVICE(obj);
     S390CPU *cpu = S390_CPU(obj);
@@ -1002,8 +1003,9 @@ static void set_feature(Object *obj, Visitor *v, const char *name,
         return;
     }
 
-    visit_type_bool(v, name, &value, errp);
-    if (*errp) {
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
     if (value) {
@@ -1043,6 +1045,7 @@ static void get_feature_group(Object *obj, Visitor *v, const char *name,
 static void set_feature_group(Object *obj, Visitor *v, const char *name,
                               void *opaque, Error **errp)
 {
+    Error *err = NULL;
     S390FeatGroup group = (S390FeatGroup) opaque;
     const S390FeatGroupDef *def = s390_feat_group_def(group);
     DeviceState *dev = DEVICE(obj);
@@ -1059,8 +1062,9 @@ static void set_feature_group(Object *obj, Visitor *v, const char *name,
         return;
     }
 
-    visit_type_bool(v, name, &value, errp);
-    if (*errp) {
+    visit_type_bool(v, name, &value, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
     if (value) {
Signature · 3 lines
Thanks,

David / dhildenb
#43 lore raw

Re: [PATCH 15/21] s390x/cpu_models: Fix latent feature property error handling bugs

Quoted 9 lines · David Hildenbrand
Quoted 4 lines · Markus Armbruster

s390x-cpu property setters set_feature() and set_feature_group() crash when the visitor fails and its @errp argument is null. Messed up in commit 0754f60429 "s390x/cpumodel: expose features and feature groups as properties".

Same comment as to the other patches :)

I think we usually use "s390x/cpumodels", but that's just nitpicking.

$ git-log --oneline target/s390x/cpu_models.c | awk '$2 ~ /:$/ { print $2 }' | sort | uniq -c
      1 S390:
      6 qapi:
      1 qemu-common:
      1 qmp:
      2 qobject:
      1 qom:
      1 s390/cpumodel:
      1 s390x/ccw:
     21 s390x/cpumodel:
      1 s390x/cpumodels:
      1 s390x/kvm:
      4 s390x/tcg:
      7 s390x:
      1 target/s390x/cpu_models:
     17 target/s390x:
      1 target:

You're right, except for the plural vs. singular. I should've browsed git-log.

Quoted 1 line

Thanks!

#44 lore raw

PATCH 16/21 s390/cpu_modules: Fix latent realize() error handling bugs

In reply to Markus Armbruster

get_max_cpu_model() crashes when kvm_s390_get_host_cpu_model() fails and its @errp argument is null.

apply_cpu_model() crashes when kvm_s390_apply_cpu_model() fails and its @errp argument is null.

s390_realize_cpu_model() crashes when get_max_cpu_model() or check_compatibility() fail, and its @errp argument is null.

All three messed up in commit 80560137cf "s390x/cpumodel: check and apply the CPU model".

The bugs can't bite as no caller actually passes null. Fix them anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 37 ++++++++++++++++++++++---------------
 1 file changed, 22 insertions(+), 15 deletions(-)
target/s390x/cpu_models.c +22 -15
index 6a29fd3ab1..c702e34a26 100644
@@ -870,6 +870,7 @@ static void check_compatibility(const S390CPUModel *max_model,
 
 static S390CPUModel *get_max_cpu_model(Error **errp)
 {
+    Error *err = NULL;
     static S390CPUModel max_model;
     static bool cached;
 
@@ -878,22 +879,24 @@ static S390CPUModel *get_max_cpu_model(Error **errp)
     }
 
     if (kvm_enabled()) {
-        kvm_s390_get_host_cpu_model(&max_model, errp);
+        kvm_s390_get_host_cpu_model(&max_model, &err);
     } else {
         max_model.def = s390_find_cpu_def(QEMU_MAX_CPU_TYPE, QEMU_MAX_CPU_GEN,
                                           QEMU_MAX_CPU_EC_GA, NULL);
         bitmap_copy(max_model.features, qemu_max_cpu_feat, S390_FEAT_MAX);
-   }
-    if (!*errp) {
-        cached = true;
-        return &max_model;
     }
-    return NULL;
+    if (err) {
+        error_propagate(errp, err);
+        return NULL;
+    }
+    cached = true;
+    return &max_model;
 }
 
 static inline void apply_cpu_model(const S390CPUModel *model, Error **errp)
 {
 #ifndef CONFIG_USER_ONLY
+    Error *err = NULL;
     static S390CPUModel applied_model;
     static bool applied;
 
@@ -909,20 +912,23 @@ static inline void apply_cpu_model(const S390CPUModel *model, Error **errp)
     }
 
     if (kvm_enabled()) {
-        kvm_s390_apply_cpu_model(model, errp);
+        kvm_s390_apply_cpu_model(model, &err);
+        if (err) {
+            error_propagate(errp, err);
+            return;
+        }
     }
 
-    if (!*errp) {
-        applied = true;
-        if (model) {
-            applied_model = *model;
-        }
+    applied = true;
+    if (model) {
+        applied_model = *model;
     }
 #endif
 }
 
 void s390_realize_cpu_model(CPUState *cs, Error **errp)
 {
+    Error *err = NULL;
     S390CPUClass *xcc = S390_CPU_GET_CLASS(cs);
     S390CPU *cpu = S390_CPU(cs);
     const S390CPUModel *max_model;
@@ -939,7 +945,7 @@ void s390_realize_cpu_model(CPUState *cs, Error **errp)
     }
 
     max_model = get_max_cpu_model(errp);
-    if (*errp) {
+    if (!max_model) {
         error_prepend(errp, "CPU models are not available: ");
         return;
     }
@@ -951,8 +957,9 @@ void s390_realize_cpu_model(CPUState *cs, Error **errp)
     cpu->model->cpu_ver = max_model->cpu_ver;
 
     check_consistency(cpu->model);
-    check_compatibility(max_model, cpu->model, errp);
-    if (*errp) {
+    check_compatibility(max_model, cpu->model, &err);
+    if (err) {
+        error_propagate(errp, err);
         return;
     }
 
Signature · 1 line
2.21.0
Reviewed-by on v1 16/21
#45 lore raw

Re: [PATCH 16/21] s390/cpu_modules: Fix latent realize() error handling bugs

Quoted 6 lines · Markus Armbruster

get_max_cpu_model() crashes when kvm_s390_get_host_cpu_model() fails and its @errp argument is null.

apply_cpu_model() crashes when kvm_s390_apply_cpu_model() fails and its @errp argument is null.

Quoted 9 lines

s390_realize_cpu_model() crashes when get_max_cpu_model() or check_compatibility() fail, and its @errp argument is null.

All three messed up in commit 80560137cf "s390x/cpumodel: check and apply the CPU model".

The bugs can't bite as no caller actually passes null. Fix them anyway.

Subject is wrong, should e.g., start with "s390x/cpumodels". (I am not aware of CPU modules :) )

[...]

Same comment regarding "local_err" and "BUG".

Signature · 3 lines
Thanks,

David / dhildenb
#46 lore raw

Re: [PATCH 16/21] s390/cpu_modules: Fix latent realize() error handling bugs

Quoted 13 lines · David Hildenbrand
Quoted 12 lines · Markus Armbruster

get_max_cpu_model() crashes when kvm_s390_get_host_cpu_model() fails and its @errp argument is null.

apply_cpu_model() crashes when kvm_s390_apply_cpu_model() fails and its @errp argument is null.

s390_realize_cpu_model() crashes when get_max_cpu_model() or check_compatibility() fail, and its @errp argument is null.

All three messed up in commit 80560137cf "s390x/cpumodel: check and apply the CPU model".

Quoted 6 lines
Quoted 3 lines

The bugs can't bite as no caller actually passes null. Fix them anyway.

Subject is wrong, should e.g., start with "s390x/cpumodels". (I am not aware of CPU modules :) )

Of course.

Quoted 5 lines

[...]

Same comment regarding "local_err" and "BUG".

Thanks!

#47 lore raw

PATCH 17/21 s390x: Fix latent query-cpu-model-FOO error handling bugs

In reply to Markus Armbruster

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It crashes when the visitor or the QOM setter fails, and its @errp argument is null. Messed up in commit 137974cea3 's390x/cpumodel:
implement QMP interface "query-cpu-model-expansion"'.

Its three callers have the same bug. Messed up in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel: implement QMP interface "query-cpu-model-baseline"'.

The bugs can't bite as no caller actually passes null. Fix them anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 43 ++++++++++++++++++++++++---------------
 1 file changed, 27 insertions(+), 16 deletions(-)
target/s390x/cpu_models.c +27 -16
index c702e34a26..3ed301b5e5 100644
@@ -477,6 +477,7 @@ CpuDefinitionInfoList *qmp_query_cpu_definitions(Error **errp)
 static void cpu_model_from_info(S390CPUModel *model, const CpuModelInfo *info,
                                 Error **errp)
 {
+    Error *err = NULL;
     const QDict *qdict = NULL;
     const QDictEntry *e;
     Visitor *visitor;
@@ -513,24 +514,26 @@ static void cpu_model_from_info(S390CPUModel *model, const CpuModelInfo *info,
 
     if (qdict) {
         visitor = qobject_input_visitor_new(info->props);
-        visit_start_struct(visitor, NULL, NULL, 0, errp);
-        if (*errp) {
+        visit_start_struct(visitor, NULL, NULL, 0, &err);
+        if (err) {
+            error_propagate(errp, err);
             visit_free(visitor);
             object_unref(obj);
             return;
         }
         for (e = qdict_first(qdict); e; e = qdict_next(qdict, e)) {
-            object_property_set(obj, visitor, e->key, errp);
-            if (*errp) {
+            object_property_set(obj, visitor, e->key, &err);
+            if (err) {
                 break;
             }
         }
-        if (!*errp) {
+        if (!err) {
             visit_check_struct(visitor, errp);
         }
         visit_end_struct(visitor, NULL);
         visit_free(visitor);
-        if (*errp) {
+        if (err) {
+            error_propagate(errp, err);
             object_unref(obj);
             return;
         }
@@ -595,13 +598,15 @@ CpuModelExpansionInfo *qmp_query_cpu_model_expansion(CpuModelExpansionType type,
                                                       CpuModelInfo *model,
                                                       Error **errp)
 {
+    Error *err = NULL;
     CpuModelExpansionInfo *expansion_info = NULL;
     S390CPUModel s390_model;
     bool delta_changes = false;
 
     /* convert it to our internal representation */
-    cpu_model_from_info(&s390_model, model, errp);
-    if (*errp) {
+    cpu_model_from_info(&s390_model, model, &err);
+    if (err) {
+        error_propagate(errp, err);
         return NULL;
     }
 
@@ -634,18 +639,21 @@ CpuModelCompareInfo *qmp_query_cpu_model_comparison(CpuModelInfo *infoa,
                                                      CpuModelInfo *infob,
                                                      Error **errp)
 {
+    Error *err = NULL;
     CpuModelCompareResult feat_result, gen_result;
     CpuModelCompareInfo *compare_info;
     S390FeatBitmap missing, added;
     S390CPUModel modela, modelb;
 
     /* convert both models to our internal representation */
-    cpu_model_from_info(&modela, infoa, errp);
-    if (*errp) {
+    cpu_model_from_info(&modela, infoa, &err);
+    if (err) {
+        error_propagate(errp, err);
         return NULL;
     }
-    cpu_model_from_info(&modelb, infob, errp);
-    if (*errp) {
+    cpu_model_from_info(&modelb, infob, &err);
+    if (err) {
+        error_propagate(errp, err);
         return NULL;
     }
     compare_info = g_new0(CpuModelCompareInfo, 1);
@@ -707,6 +715,7 @@ CpuModelBaselineInfo *qmp_query_cpu_model_baseline(CpuModelInfo *infoa,
                                                     CpuModelInfo *infob,
                                                     Error **errp)
 {
+    Error *err = NULL;
     CpuModelBaselineInfo *baseline_info;
     S390CPUModel modela, modelb, model;
     uint16_t cpu_type;
@@ -714,13 +723,15 @@ CpuModelBaselineInfo *qmp_query_cpu_model_baseline(CpuModelInfo *infoa,
     uint8_t max_gen;
 
     /* convert both models to our internal representation */
-    cpu_model_from_info(&modela, infoa, errp);
-    if (*errp) {
+    cpu_model_from_info(&modela, infoa, &err);
+    if (err) {
+        error_propagate(errp, err);
         return NULL;
     }
 
-    cpu_model_from_info(&modelb, infob, errp);
-    if (*errp) {
+    cpu_model_from_info(&modelb, infob, &err);
+    if (err) {
+        error_propagate(errp, err);
         return NULL;
     }
 
Signature · 1 line
2.21.0
#48 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 10 lines

Am 30.11.2019 um 20:42 schrieb Markus Armbruster <[email protected]>:

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It crashes when the visitor or the QOM setter fails, and its @errp argument is null. Messed up in commit 137974cea3 's390x/cpumodel:
implement QMP interface "query-cpu-model-expansion"'.

Quoted 7 lines

Its three callers have the same bug. Messed up in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel: implement QMP interface "query-cpu-model-baseline"'.

The bugs can't bite as no caller actually passes null. Fix them anyway.

https://en.m.wikipedia.org/wiki/Software_bug

„ A software bug is an error, flaw or fault in a computer program or system that causes it to produce an incorrect or unexpected result, or to behave in unintended ways. „

Please make it clear in the descriptions that these are cleanups and not bugfixes. It might be very confusing for people looking out for real bugs. Also, please change the terminology „messed up“ to „introduced in“ or similar.

(applies to all s390x patches)

Thanks.

#49 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 18 lines · David Hildenbrand
Quoted 10 lines

Am 30.11.2019 um 20:42 schrieb Markus Armbruster <[email protected]>:

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It crashes when the visitor or the QOM setter fails, and its @errp argument is null. Messed up in commit 137974cea3 's390x/cpumodel:
implement QMP interface "query-cpu-model-expansion"'.

Quoted 7 lines

Its three callers have the same bug. Messed up in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel: implement QMP interface "query-cpu-model-baseline"'.

The bugs can't bite as no caller actually passes null. Fix them anyway.

Quoted 11 lines

https://en.m.wikipedia.org/wiki/Software_bug

„ A software bug is an error, flaw or fault in a computer program or
system that causes it to produce an incorrect or unexpected result, or to behave in unintended ways. „

Please make it clear in the descriptions that these are cleanups and not bugfixes. It might be very confusing for people looking out for real bugs.

Disclaimer: I am not entirely familiar with the code in question, so take my opinion with reasonablereservation.

It looks that we here deal with latent bugs. As you probably know from experience, a latent bugs, when they are activated with some ostensibly unrelated code change, can be much more difficult to diagnose and fix than regular bugs.

In that light, this change is not a clean up. It is a fix of a latent bugs, and Markus' aproach to treat it as a bug fix looks right to me. I would just add a word "latent" or similar, which would even more distance the patch from "cleanup" meaning.

David, if I understand well, this patch fixes the commit done by you. I definitely understand this is not a pleasant position, but we all (definitelly including myself too) should learn to handle such situations as gracefully as we can.

Yours,
Aleksandar

Quoted 9 lines

Also, please change the terminology „messed up“ to „introduced in“ or similar.

(applies to all s390x patches)

Thanks.

#50 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 34 lines · Aleksandar Markovic
Quoted 18 lines · David Hildenbrand
Quoted 10 lines

Am 30.11.2019 um 20:42 schrieb Markus Armbruster <[email protected]>:

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It crashes when the visitor or the QOM setter fails, and its @errp argument is null. Messed up in commit 137974cea3 's390x/cpumodel:
implement QMP interface "query-cpu-model-expansion"'.

Quoted 7 lines

Its three callers have the same bug. Messed up in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel: implement QMP interface "query-cpu-model-baseline"'.

The bugs can't bite as no caller actually passes null. Fix them anyway.

Quoted 11 lines

https://en.m.wikipedia.org/wiki/Software_bug

„ A software bug is an error, flaw or fault in a computer program or
system that causes it to produce an incorrect or unexpected result, or to behave in unintended ways. „

Please make it clear in the descriptions that these are cleanups and not bugfixes. It might be very confusing for people looking out for real bugs.

Quoted 9 lines

Disclaimer: I am not entirely familiar with the code in question, so take my opinion with reasonablereservation.

It looks that we here deal with latent bugs. As you probably know from experience, a latent bugs, when they are activated with some ostensibly unrelated code change, can be much more difficult to diagnose and fix than regular bugs.

Oops, I didn't even realize that the patch title contains the word "latent". (I wrote the previous message without that knowledge. For some strange reason, my email client doesn't display email subject while replying.)

In this case, I would suggest usage of phrase "latent bug" instead of "latent error" or so in the message title, to strenghten the point that this is not a cleanup.

Yours, Aleksandar

Quoted 26 lines

In that light, this change is not a clean up. It is a fix of a latent bugs, and Markus' aproach to treat it as a bug fix looks right to me. I would just add a word "latent" or similar, which would even more distance the patch from "cleanup" meaning.

David, if I understand well, this patch fixes the commit done by you. I definitely understand this is not a pleasant position, but we all (definitelly including myself too) should learn to handle such situations as gracefully as we can.

Yours,
Aleksandar

Quoted 9 lines

Also, please change the terminology „messed up“ to „introduced in“ or similar.

(applies to all s390x patches)

Thanks.

#51 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 49 lines · Aleksandar Markovic
Quoted 35 lines · Aleksandar Markovic
Quoted 19 lines · David Hildenbrand
Quoted 5 lines

Am 30.11.2019 um 20:42 schrieb Markus Armbruster <[email protected]>:

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(

),

Quoted 5 lines

qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It crashes when the visitor or the QOM setter fails, and its @errp argument is null. Messed up in commit 137974cea3 's390x/cpumodel:
implement QMP interface "query-cpu-model-expansion"'.

Quoted 7 lines

Its three callers have the same bug. Messed up in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel: implement QMP interface "query-cpu-model-baseline"'.

The bugs can't bite as no caller actually passes null. Fix them anyway.

Quoted 11 lines

https://en.m.wikipedia.org/wiki/Software_bug

„ A software bug is an error, flaw or fault in a computer program or
system that causes it to produce an incorrect or unexpected result, or to behave in unintended ways. „

Please make it clear in the descriptions that these are cleanups and not bugfixes. It might be very confusing for people looking out for real bugs.

Quoted 9 lines

Disclaimer: I am not entirely familiar with the code in question, so take my opinion with reasonablereservation.

It looks that we here deal with latent bugs. As you probably know from experience, a latent bugs, when they are activated with some ostensibly unrelated code change, can be much more difficult to diagnose and fix than regular bugs.

Quoted 10 lines

Oops, I didn't even realize that the patch title contains the word "latent". (I wrote the previous message without that knowledge. For some strange reason, my email client doesn't display email subject while replying.)

In this case, I would suggest usage of phrase "latent bug" instead of "latent error" or so in the message title, to strenghten the point that this is not a cleanup.

Actually, the message title already does use "latent .... bugs". So it is fine - in my opinion.

Quoted 30 lines

Yours, Aleksandar

Quoted 26 lines

In that light, this change is not a clean up. It is a fix of a latent bugs, and Markus' aproach to treat it as a bug fix looks right to me. I would just add a word "latent" or similar, which would even more distance the patch from "cleanup" meaning.

David, if I understand well, this patch fixes the commit done by you. I definitely understand this is not a pleasant position, but we all (definitelly including myself too) should learn to handle such situations as gracefully as we can.

Yours,
Aleksandar

Quoted 9 lines

Also, please change the terminology „messed up“ to „introduced in“ or similar.

(applies to all s390x patches)

Thanks.

#52 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

In reply to Aleksandar Markovic
Quoted 38 lines · Aleksandar Markovic
Quoted 1 line · On Saturday, November 30, 2019, David Hildenbrand <[email protected]…

Am 30.11.2019 um 20:42 schrieb Markus Armbruster

<[email protected] <mailto:[email protected]>>:
>
> cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(),
> qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline().  It > crashes when the visitor or the QOM setter fails, and its @errp > argument is null.  Messed up in commit 137974cea3 's390x/cpumodel:
> implement QMP interface "query-cpu-model-expansion"'.
>
> Its three callers have the same bug.  Messed up in commit 4e82ef0502 > 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' > and commit f1a47d08ef 's390x/cpumodel: implement QMP interface > "query-cpu-model-baseline"'.
>
> The bugs can't bite as no caller actually passes null.  Fix them > anyway.

https://en.m.wikipedia.org/wiki/Software_bug
<https://en.m.wikipedia.org/wiki/Software_bug>

  „ A software bug is an error, flaw or fault in a computer program or system that causes it to produce an incorrect or unexpected result, or to behave in unintended ways. „

Please make it clear in the descriptions that these are cleanups and not bugfixes. It might be very confusing for people looking out for real bugs.

Quoted 7 lines

Disclaimer: I am not entirely familiar with the code in question, so take my opinion with reasonablereservation.

It looks that we here deal with latent bugs. As you probably know from experience, a latent bugs, when they are activated with some ostensibly unrelated code change, can be much more difficult to diagnose and fix than regular bugs.

"https://economictimes.indiatimes.com/definition/latent-bug

"Definition: An uncovered or unidentified bug which exists in the system over a period of time is known as the Latent Bug. The bug may persist in the system in one or more versions of the software."

AFAIK, a latent BUG can be triggered, it simply was never triggered.

Do you think the following code is buggy?

static int get_val(int *ptr)

{
	return *ptr;
}

int main()

{
	int a = 0;

	return get_val(&a);
}

I claim, no, although we could access a NULL pointer if ever reworked. There is no invalid system state possible.

Quoted 5 lines

In that light, this change is not a clean up. It is a fix of a latent bugs, and Markus' aproach to treat it as a bug fix looks right to me. I would just add a word "latent" or similar, which would even more distance the patch from "cleanup" meaning.

I agree iff there is some way to trigger it. Otherwise, to me it is a cleanup.If it's a BUG, it deserves proper Fixes tags and some description how it can be triggered.

Signature · 3 lines
Thanks,

David / dhildenb
#53 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 47 lines · David Hildenbrand
Quoted 38 lines · Aleksandar Markovic
Quoted 1 line · On Saturday, November 30, 2019, David Hildenbrand <[email protected]…

Am 30.11.2019 um 20:42 schrieb Markus Armbruster

<[email protected] <mailto:[email protected]>>:
>
> cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(),
> qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline().  It > crashes when the visitor or the QOM setter fails, and its @errp > argument is null.  Messed up in commit 137974cea3 's390x/cpumodel:
> implement QMP interface "query-cpu-model-expansion"'.
>
> Its three callers have the same bug.  Messed up in commit 4e82ef0502 > 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' > and commit f1a47d08ef 's390x/cpumodel: implement QMP interface > "query-cpu-model-baseline"'.
>
> The bugs can't bite as no caller actually passes null.  Fix them > anyway.

https://en.m.wikipedia.org/wiki/Software_bug
<https://en.m.wikipedia.org/wiki/Software_bug>

  „ A software bug is an error, flaw or fault in a computer program or system that causes it to produce an incorrect or unexpected result, or to behave in unintended ways. „

Please make it clear in the descriptions that these are cleanups and not bugfixes. It might be very confusing for people looking out for real bugs.

Quoted 7 lines

Disclaimer: I am not entirely familiar with the code in question, so take my opinion with reasonablereservation.

It looks that we here deal with latent bugs. As you probably know from experience, a latent bugs, when they are activated with some ostensibly unrelated code change, can be much more difficult to diagnose and fix than regular bugs.

Quoted 7 lines

"https://economictimes.indiatimes.com/definition/latent-bug

"Definition: An uncovered or unidentified bug which exists in the system over a period of time is known as the Latent Bug. The bug may persist in the system in one or more versions of the software."

AFAIK, a latent BUG can be triggered, it simply was never triggered.

First search hit. Here's my second one:

Q: What are latent bugs?

A: These bugs do not cause problems today. However, they are lurking just waiting to reveal themselves later. The Ariane 5 rocket failure was caused by a float->int conversion error that lay dormant when previous rockets were slower; but the faster Ariane 5 triggered the problem. The Millennium bug is another example of a latent bug that came to light when circumstances changed. Latent bugs are much harder to test using conventional testing techniques, and finding them requires someone with foresight to ask.

http://www.geekinterview.com/question_details/36689

My point is: common usage of the term is not as clear-cut as your quote makes it seem.

Quoted 8 lines

Do you think the following code is buggy?

static int get_val(int *ptr)

{
	return *ptr;
}

int main()

Quoted 8 lines
{
	int a = 0;

	return get_val(&a);
}

I claim, no, although we could access a NULL pointer if ever reworked. There is no invalid system state possible.

get_val() is silent on how it wants to be used. error.h is anything
but: it spells out how it wantes to be used in quite some detail. In particular:

* 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!

My patch fixes this exact misuse of the interface.

Quoted 8 lines
Quoted 4 lines

In that light, this change is not a clean up. It is a fix of a latent bugs, and Markus' aproach to treat it as a bug fix looks right to me. I would just add a word "latent" or similar, which would even more distance the patch from "cleanup" meaning.

I agree iff there is some way to trigger it. Otherwise, to me it is a cleanup.If it's a BUG, it deserves proper Fixes tags and some description how it can be triggered.

Yes, a bug that can bite deserves a reproducer and a formal Fixes: tag.

The thing we're discussing (however we may want to call it) does not have a reproducer, and I think we're in agreement that it doesn't need a
Fixes: tag.

However, my patch is not cleaning up something that's dirty, it's fixing something that's unequivocally wrong: a violation of a stated interface contract.

The violation happens to have no ill effects at this time due to the way the violating code is being used.

I call that a "latent bug". git-log has quite a few occurences of "latent bug", by Richard Henderson, Daniel Berrangé, Paolo, ...

Your point that the commit message should not confuse people looking for real bugs is well taken. I think "latent bug" is clear enough, and also concise. I'm of course open to better phrasings.

s390x: Fix currently harmless query-cpu-model-FOO error API violations

feels no clearer to me than

s390x: Fix latent query-cpu-model-FOO error handling bugs

It's also too long.

I tried. Your turn :)

#54 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

[...]

Quoted 13 lines

First search hit. Here's my second one:

Q: What are latent bugs?

A: These bugs do not cause problems today. However, they are lurking just waiting to reveal themselves later. The Ariane 5 rocket failure was caused by a float->int conversion error that lay dormant when previous rockets were slower; but the faster Ariane 5 triggered the problem. The Millennium bug is another example of a latent bug that came to light when circumstances changed. Latent bugs are much harder to test using conventional testing techniques, and finding them requires someone with foresight to ask.

Quoted 1 line

http://www.geekinterview.com/question_details/36689

Google search "latent software BUG"

Hit 1: What I posted

Hit 2:
https://www.quora.com/What-are-some-examples-for-a-latent-defect-in-software-testing

"The problems will not cause the damage currently, but wait to reveal themselves at a later time. ... E.g. February has 28 days. ... These defects do not cause damage to the system immediately but wait for a particular event sometime to cause damage and show their presence."

"Mostly, these types of bugs are unexpected outcome of any corner/edge case scenarios which was executed with some specific set of test data."

Hit 3: https://sqa.stackexchange.com/questions/9170/what-is-a-latent-bug

"Latent bugs are bugs which exist, but have not yet been discovered.

They are bugs waiting to be found."

"In Software Quality Assurance:

Latent defects are the those which arises in the field, and unknown until they reported by the field staff."

Hit 4:
https://sqa.stackexchange.com/questions/13980/what-is-the-difference-between-gold-bug-and-latent-bug

"A latent bug is a bug which is present in the system from previous iterations or release (in your scenario Sprint 1). They are either low priority bugs, which either went undetected or were not reported."

Same at least for Hit 5, 6, 7 (then I got tired ;) )

https://blog.qatestlab.com/2011/10/21/latent-and-masked-software-bugs-what-is-the-difference/
https://www.testing-whiz.com/blog/3-types-of-unusual-software-defects-you-should-not-miss
https://www.360logica.com/blog/latent-defect-hide-n-seek-defect-software-testing/

Which contain perfect examples and descriptions for latent bugs :)

e.g.,

"Let’s imagine that an application is able to print a document either by laser printer or by dot matrix printer. To reach this, the application first searches for the laser printer. In this case if it finds a laser printer (used by default) it uses this one and prints. In case if it does not find a laser printer, the application searches for dot matrix printer. And if the application finds a dot matrix printer, it gives an error message. This unleashes a latent defect. Therefore this application will never search for the dot matrix printer. And the application never got tested for the dot matrix printer. That means the accurate conditions were never met for the dot matrix printer. This is what we call a latent software bug."

[...]

Quoted 11 lines
Quoted 9 lines
Quoted 5 lines

In that light, this change is not a clean up. It is a fix of a latent bugs, and Markus' aproach to treat it as a bug fix looks right to me. I would just add a word "latent" or similar, which would even more distance the patch from "cleanup" meaning.

I agree iff there is some way to trigger it. Otherwise, to me it is a cleanup.If it's a BUG, it deserves proper Fixes tags and some description how it can be triggered.

Yes, a bug that can bite deserves a reproducer and a formal Fixes: tag.

A "BUG that cannot be triggered" is an oxymoron. :)

The code might be "error prone", it might "violate API guidelines", it might "not obey coding practices". If it can't be triggered, it's - by definition - not a (latent) BUG.

Quoted 8 lines

The thing we're discussing (however we may want to call it) does not have a reproducer, and I think we're in agreement that it doesn't need a
Fixes: tag.

However, my patch is not cleaning up something that's dirty, it's fixing something that's unequivocally wrong: a violation of a stated interface contract.

In general, I don't care about these minor details, but if somebody literally tells the world for all eternity in a handful of patch descriptions "David messed up the code by introducing a (latent) BUG (that cannot be triggered)" I get suspicious. Something nicer (IMHO) would be "Commit X introduced error prone code, let's rework that to make it more stable in the future and obey the "error API" guidelines. I hope you see the difference ;)

Quoted 7 lines

Your point that the commit message should not confuse people looking for real bugs is well taken. I think "latent bug" is clear enough, and also concise. I'm of course open to better phrasings.

s390x: Fix currently harmless query-cpu-model-FOO error API violations

Quoted 7 lines

feels no clearer to me than

s390x: Fix latent query-cpu-model-FOO error handling bugs

It's also too long.

I tried. Your turn :)

s390x: Fix query-cpu-model-FOO error API violations

s390x: Make query-cpu-model-FOO error handling less error prone

s390x: Cleanup error handling in query-cpu-model-FOO

s390x: Rework error handling in query-cpu-model-FOO

...

but, enough time spent on this, feel free to continue with this however you want.

Signature · 3 lines
Thanks,

David / dhildenb
#55 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 15 lines · David Hildenbrand

[...]

Quoted 13 lines

First search hit. Here's my second one:

Q: What are latent bugs?

A: These bugs do not cause problems today. However, they are lurking just waiting to reveal themselves later. The Ariane 5 rocket failure was caused by a float->int conversion error that lay dormant when previous rockets were slower; but the faster Ariane 5 triggered the problem. The Millennium bug is another example of a latent bug that came to light when circumstances changed. Latent bugs are much harder to test using conventional testing techniques, and finding them requires someone with foresight to ask.

Quoted 3 lines
Quoted 1 line

http://www.geekinterview.com/question_details/36689

Google search "latent software BUG"

I think this argument has run its course. Let's agree to differ on the finer meaning and possible uses of "latent", and craft a mutually agreeable commit message instead. I'll reply to Cornelia's message.

[...]

#56 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

In reply to Markus Armbruster

On Sat, 30 Nov 2019 20:42:36 +0100
Markus Armbruster <[email protected]> wrote:

I don't really want to restart the discussion :), but what about:

Quoted 4 lines

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It crashes when the visitor or the QOM setter fails, and its @errp argument is null.

"It would crash when the visitor or the QOM setter fails if its @errp argument were NULL." ?

(Hope I got my conditionals right...)

Quoted 1 line

Messed up in commit 137974cea3 's390x/cpumodel:

I agree that "Introduced" is a bit nicer than "Messed up".

Quoted 6 lines

implement QMP interface "query-cpu-model-expansion"'.

Its three callers have the same bug. Messed up in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel: implement QMP interface "query-cpu-model-baseline"'.

If we agree, I can tweak the various commit messages for the s390x patches and apply them.

Quoted 9 lines

The bugs can't bite as no caller actually passes null. Fix them anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 43 ++++++++++++++++++++++++---------------
 1 file changed, 27 insertions(+), 16 deletions(-)

David, I don't think you gave a R-b for that one yet?

#57 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 10 lines · Cornelia Huck

On Sat, 30 Nov 2019 20:42:36 +0100
Markus Armbruster <[email protected]> wrote:

I don't really want to restart the discussion :), but what about:

Quoted 4 lines

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It crashes when the visitor or the QOM setter fails, and its @errp argument is null.

Quoted 4 lines

"It would crash when the visitor or the QOM setter fails if its @errp argument were NULL." ?

(Hope I got my conditionals right...)

I don't think this is an improvement.

The commit message matches a pattern "what's wrong, since when, impact, how is it fixed". The pattern has become habit for me. Its "what's wrong" part is strictly local. The non-local argument comes in only when we assess impact.

Use of "would" in the what part's conditional signals the condition is unlikely. True (it's actually impossible), but distracting (because it involves the non-local argument I'm not yet ready to make).

Let me try a different phrasing below.

Quoted 3 lines
Quoted 1 line

Messed up in commit 137974cea3 's390x/cpumodel:

I agree that "Introduced" is a bit nicer than "Messed up".

Works fine for me. I didn't mean any disrespect --- I'd have to disrespect myself: the mess corrected by PATCH 10 is mine.

Quoted 3 lines
Quoted 3 lines

implement QMP interface "query-cpu-model-expansion"'.

Its three callers have the same bug. Messed up in commit 4e82ef0502

Feel free to call it "issue" rather than "bug". I don't care, but David might.

Quoted 6 lines
Quoted 3 lines

's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel: implement QMP interface "query-cpu-model-baseline"'.

If we agree, I can tweak the various commit messages for the s390x patches and apply them.

Tweaking the non-s390x commit messages as well would be nicer, but requires a respin.

Let's try to craft a mutually agreeable commit message for this patch. Here's my attempt:

s390x: Fix query-cpu-model-FOO error API violations

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It dereferences @errp when the visitor or the QOM setter fails. That's wrong; see the big comment in error.h. Introduced in commit 137974cea3 's390x/cpumodel: implement QMP interface "query-cpu-model-expansion"'.

Its three callers have the same issue. Introduced in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel:
implement QMP interface "query-cpu-model-baseline"'.

No caller actually passes null. To fix, splice in a local Error *err, and error_propagate().

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>

Adapting it to other patches should be straightforward.

Quoted 11 lines
Quoted 9 lines

The bugs can't bite as no caller actually passes null. Fix them anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 43 ++++++++++++++++++++++++---------------
 1 file changed, 27 insertions(+), 16 deletions(-)

David, I don't think you gave a R-b for that one yet?

#58 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 19 lines · On Tue, 03 Dec 2019 08:49:58 +0100 Markus Armbruster <[email protected]> wrote:
Quoted 10 lines · Cornelia Huck

On Sat, 30 Nov 2019 20:42:36 +0100
Markus Armbruster <[email protected]> wrote:

I don't really want to restart the discussion :), but what about:

Quoted 4 lines

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It crashes when the visitor or the QOM setter fails, and its @errp argument is null.

Quoted 4 lines

"It would crash when the visitor or the QOM setter fails if its @errp argument were NULL." ?

(Hope I got my conditionals right...)

I don't think this is an improvement.

Quoted 8 lines

The commit message matches a pattern "what's wrong, since when, impact, how is it fixed". The pattern has become habit for me. Its "what's wrong" part is strictly local. The non-local argument comes in only when we assess impact.

Use of "would" in the what part's conditional signals the condition is unlikely. True (it's actually impossible), but distracting (because it involves the non-local argument I'm not yet ready to make).

I think we'll have to agree to disagree here...

Quoted 2 lines

Let me try a different phrasing below.

...but also see below :)

Quoted 42 lines
Quoted 4 lines
Quoted 2 lines

Messed up in commit 137974cea3 's390x/cpumodel:

I agree that "Introduced" is a bit nicer than "Messed up".

Works fine for me. I didn't mean any disrespect --- I'd have to disrespect myself: the mess corrected by PATCH 10 is mine.

Quoted 3 lines
Quoted 3 lines

implement QMP interface "query-cpu-model-expansion"'.

Its three callers have the same bug. Messed up in commit 4e82ef0502

Feel free to call it "issue" rather than "bug". I don't care, but David might.

Quoted 6 lines
Quoted 3 lines

's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel: implement QMP interface "query-cpu-model-baseline"'.

If we agree, I can tweak the various commit messages for the s390x patches and apply them.

Tweaking the non-s390x commit messages as well would be nicer, but requires a respin.

Let's try to craft a mutually agreeable commit message for this patch. Here's my attempt:

s390x: Fix query-cpu-model-FOO error API violations

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It dereferences @errp when the visitor or the QOM setter fails. That's wrong; see the big comment in error.h. Introduced in commit 137974cea3 's390x/cpumodel: implement QMP interface "query-cpu-model-expansion"'.

Its three callers have the same issue. Introduced in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel:
implement QMP interface "query-cpu-model-baseline"'.

Quoted 6 lines

No caller actually passes null. To fix, splice in a local Error *err, and error_propagate().

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>

That sounds good to me.

Quoted 2 lines

Adapting it to other patches should be straightforward.

Ok, so how to proceed? I'm happy to tweak the commit messages for s390x, but that is bound to get messy.

Quoted 12 lines
Quoted 12 lines
Quoted 10 lines

The bugs can't bite as no caller actually passes null. Fix them anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 43 ++++++++++++++++++++++++---------------
 1 file changed, 27 insertions(+), 16 deletions(-)  

David, I don't think you gave a R-b for that one yet?

Reviewed-by
#59 lore raw

Re: [PATCH 17/21] s390x: Fix latent query-cpu-model-FOO error handling bugs

Quoted 14 lines
Quoted 14 lines

s390x: Fix query-cpu-model-FOO error API violations

cpu_model_from_info() is a helper for qmp_query_cpu_model_expansion(), qmp_query_cpu_model_comparison(), qmp_query_cpu_model_baseline(). It dereferences @errp when the visitor or the QOM setter fails. That's wrong; see the big comment in error.h. Introduced in commit 137974cea3 's390x/cpumodel: implement QMP interface "query-cpu-model-expansion"'.

Its three callers have the same issue. Introduced in commit 4e82ef0502 's390x/cpumodel: implement QMP interface "query-cpu-model-comparison"' and commit f1a47d08ef 's390x/cpumodel:
implement QMP interface "query-cpu-model-baseline"'.

Quoted 8 lines
Quoted 6 lines

No caller actually passes null. To fix, splice in a local Error *err, and error_propagate().

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>

That sounds good to me.

Yes, something I would have enjoyed reading from the first sentence.

Signature · 3 lines
Thanks,

David / dhildenb
#60 lore raw

PATCH 18/21 s390x: Fix latent query-cpu-definitions error handling bug

In reply to Markus Armbruster

qmp_query_cpu_definitions() tries to ignore get_max_cpu_model()'s errors. However, it crashes when its @errp argument is null or &error_abort, and exit(1)s when it's &error_fatal. Messed up in commit 38cba1f4d8 "s390x: return unavailable features via query-cpu-definitions".

The bug can't bite as no caller actually passes such @errp values. Fix it anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
target/s390x/cpu_models.c +1 -5
index 3ed301b5e5..547bab8ac3 100644
@@ -462,11 +462,7 @@ CpuDefinitionInfoList *qmp_query_cpu_definitions(Error **errp)
         .list = NULL,
     };
 
-    list_data.model = get_max_cpu_model(errp);
-    if (*errp) {
-        error_free(*errp);
-        *errp = NULL;
-    }
+    list_data.model = get_max_cpu_model(NULL);
 
     object_class_foreach(create_cpu_model_list, TYPE_S390_CPU, false,
                          &list_data);
Signature · 1 line
2.21.0
Reviewed-by on v1 18/21
#61 lore raw

Re: [PATCH 18/21] s390x: Fix latent query-cpu-definitions error handling bug

Quoted 5 lines · Markus Armbruster

qmp_query_cpu_definitions() tries to ignore get_max_cpu_model()'s errors. However, it crashes when its @errp argument is null or &error_abort, and exit(1)s when it's &error_fatal. Messed up in commit 38cba1f4d8 "s390x: return unavailable features via query-cpu-definitions".

s/crashes .../would crash .../ etc.

Quoted 29 lines

The bug can't bite as no caller actually passes such @errp values. Fix it anyway.

Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
---
 target/s390x/cpu_models.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
target/s390x/cpu_models.c +1 -5
index 3ed301b5e5..547bab8ac3 100644
@@ -462,11 +462,7 @@ CpuDefinitionInfoList *qmp_query_cpu_definitions(Error **errp)
         .list = NULL,
     };
 
-    list_data.model = get_max_cpu_model(errp);
-    if (*errp) {
-        error_free(*errp);
-        *errp = NULL;
-    }
+    list_data.model = get_max_cpu_model(NULL);
 
     object_class_foreach(create_cpu_model_list, TYPE_S390_CPU, false,
                          &list_data);
Signature · 3 lines
Thanks,

David / dhildenb
#62 lore raw

PATCH 19/21 error: Clean up unusual names of Error * variables

In reply to Markus Armbruster

Local Error * variables are conventionally named @err or @local_err, and Error ** parameters @errp. Naming local variables like parameters is confusing. Clean that up.

Naming parameters like local variables is also confusing. Left for another day.

---
 hw/intc/s390_flic_kvm.c    | 10 +++++-----
 hw/ppc/spapr_pci.c         | 16 ++++++++--------
 hw/ppc/spapr_pci_nvlink2.c | 10 +++++-----
 tests/test-blockjob.c      | 16 ++++++++--------
 4 files changed, 26 insertions(+), 26 deletions(-)
hw/intc/s390_flic_kvm.c +5 -5
index c9ee80eaae..30d50c2369 100644
@@ -582,10 +582,10 @@ static void kvm_s390_flic_realize(DeviceState *dev, Error **errp)
     struct kvm_create_device cd = {0};
     struct kvm_device_attr test_attr = {0};
     int ret;
-    Error *errp_local = NULL;
+    Error *err = NULL;
 
-    KVM_S390_FLIC_GET_CLASS(dev)->parent_realize(dev, &errp_local);
-    if (errp_local) {
+    KVM_S390_FLIC_GET_CLASS(dev)->parent_realize(dev, &err);
+    if (err) {
         goto fail;
     }
     flic_state->fd = -1;
@@ -593,7 +593,7 @@ static void kvm_s390_flic_realize(DeviceState *dev, Error **errp)
     cd.type = KVM_DEV_TYPE_FLIC;
     ret = kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &cd);
     if (ret < 0) {
-        error_setg_errno(&errp_local, errno, "Creating the KVM device failed");
+        error_setg_errno(&err, errno, "Creating the KVM device failed");
         trace_flic_create_device(errno);
         goto fail;
     }
@@ -605,7 +605,7 @@ static void kvm_s390_flic_realize(DeviceState *dev, Error **errp)
                                             KVM_HAS_DEVICE_ATTR, test_attr);
     return;
 fail:
-    error_propagate(errp, errp_local);
+    error_propagate(errp, err);
 }
 
 static void kvm_s390_flic_reset(DeviceState *dev)
hw/ppc/spapr_pci.c +8 -8
index f6fbcf99ed..723373de73 100644
@@ -2042,13 +2042,13 @@ void spapr_phb_dma_reset(SpaprPhbState *sphb)
 static void spapr_phb_reset(DeviceState *qdev)
 {
     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev);
-    Error *errp = NULL;
+    Error *err = NULL;
 
     spapr_phb_dma_reset(sphb);
     spapr_phb_nvgpu_free(sphb);
-    spapr_phb_nvgpu_setup(sphb, &errp);
-    if (errp) {
-        error_report_err(errp);
+    spapr_phb_nvgpu_setup(sphb, &err);
+    if (err) {
+        error_report_err(err);
     }
 
     /* Reset the IOMMU state */
@@ -2326,7 +2326,7 @@ int spapr_dt_phb(SpaprMachineState *spapr, SpaprPhbState *phb,
                                 cpu_to_be32(phb->numa_node)};
     SpaprTceTable *tcet;
     SpaprDrc *drc;
-    Error *errp = NULL;
+    Error *err = NULL;
 
     /* Start populating the FDT */
     _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname));
@@ -2408,9 +2408,9 @@ int spapr_dt_phb(SpaprMachineState *spapr, SpaprPhbState *phb,
         return ret;
     }
 
-    spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &errp);
-    if (errp) {
-        error_report_err(errp);
+    spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &err);
+    if (err) {
+        error_report_err(err);
     }
     spapr_phb_nvgpu_ram_populate_dt(phb, fdt);
 
hw/ppc/spapr_pci_nvlink2.c +5 -5
index 4aa89ede23..8332d5694e 100644
@@ -57,7 +57,7 @@ struct SpaprPhbPciNvGpuConfig {
     uint64_t nv2_atsd_current;
     int num; /* number of non empty (i.e. tgt!=0) entries in slots[] */
     SpaprPhbPciNvGpuSlot slots[NVGPU_MAX_NUM];
-    Error *errp;
+    Error *err;
 };
 
 static SpaprPhbPciNvGpuSlot *
@@ -153,7 +153,7 @@ static void spapr_phb_pci_collect_nvgpu(PCIBus *bus, PCIDevice *pdev,
             spapr_pci_collect_nvnpu(nvgpus, pdev, tgt, MEMORY_REGION(mr_npu),
                                     &local_err);
         }
-        error_propagate(&nvgpus->errp, local_err);
+        error_propagate(&nvgpus->err, local_err);
     }
     if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) !=
          PCI_HEADER_TYPE_BRIDGE)) {
@@ -187,9 +187,9 @@ void spapr_phb_nvgpu_setup(SpaprPhbState *sphb, Error **errp)
     pci_for_each_device(bus, pci_bus_num(bus),
                         spapr_phb_pci_collect_nvgpu, sphb->nvgpus);
 
-    if (sphb->nvgpus->errp) {
-        error_propagate(errp, sphb->nvgpus->errp);
-        sphb->nvgpus->errp = NULL;
+    if (sphb->nvgpus->err) {
+        error_propagate(errp, sphb->nvgpus->err);
+        sphb->nvgpus->err = NULL;
         goto cleanup_exit;
     }
 
tests/test-blockjob.c +8 -8
index 7844c9ffcb..e670a20617 100644
@@ -34,13 +34,13 @@ static BlockJob *mk_job(BlockBackend *blk, const char *id,
                         int flags)
 {
     BlockJob *job;
-    Error *errp = NULL;
+    Error *err = NULL;
 
     job = block_job_create(id, drv, NULL, blk_bs(blk),
                            0, BLK_PERM_ALL, 0, flags, block_job_cb,
-                           NULL, &errp);
+                           NULL, &err);
     if (should_succeed) {
-        g_assert_null(errp);
+        g_assert_null(err);
         g_assert_nonnull(job);
         if (id) {
             g_assert_cmpstr(job->job.id, ==, id);
@@ -48,9 +48,9 @@ static BlockJob *mk_job(BlockBackend *blk, const char *id,
             g_assert_cmpstr(job->job.id, ==, blk_name(blk));
         }
     } else {
-        g_assert_nonnull(errp);
+        g_assert_nonnull(err);
         g_assert_null(job);
-        error_free(errp);
+        error_free(err);
     }
 
     return job;
@@ -80,9 +80,9 @@ static BlockBackend *create_blk(const char *name)
     bdrv_unref(bs);
 
     if (name) {
-        Error *errp = NULL;
-        monitor_add_blk(blk, name, &errp);
-        g_assert_null(errp);
+        Error *err = NULL;
+        monitor_add_blk(blk, name, &err);
+        g_assert_null(err);
     }
 
     return blk;
Signature · 1 line
2.21.0
Reviewed-by on v1 19/21
#63 lore raw

Re: [PATCH 19/21] error: Clean up unusual names of Error * variables

Quoted 7 lines · Markus Armbruster

Local Error * variables are conventionally named @err or @local_err, and Error ** parameters @errp. Naming local variables like parameters is confusing. Clean that up.

Naming parameters like local variables is also confusing. Left for another day.

Quoted 7 lines
---
  hw/intc/s390_flic_kvm.c    | 10 +++++-----
  hw/ppc/spapr_pci.c         | 16 ++++++++--------
  hw/ppc/spapr_pci_nvlink2.c | 10 +++++-----
  tests/test-blockjob.c      | 16 ++++++++--------
  4 files changed, 26 insertions(+), 26 deletions(-)
#64 lore raw

PATCH 20/21 hw/intc/s390: Simplify error handling in kvm_s390_flic_realize()

In reply to Markus Armbruster
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Christian Borntraeger <[email protected]>
---
 hw/intc/s390_flic_kvm.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
hw/intc/s390_flic_kvm.c +4 -6
index 30d50c2369..dddd33ea61 100644
@@ -586,16 +586,17 @@ static void kvm_s390_flic_realize(DeviceState *dev, Error **errp)
 
     KVM_S390_FLIC_GET_CLASS(dev)->parent_realize(dev, &err);
     if (err) {
-        goto fail;
+        error_propagate(errp, err);
+        return;
     }
     flic_state->fd = -1;
 
     cd.type = KVM_DEV_TYPE_FLIC;
     ret = kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &cd);
     if (ret < 0) {
-        error_setg_errno(&err, errno, "Creating the KVM device failed");
+        error_setg_errno(errp, errno, "Creating the KVM device failed");
         trace_flic_create_device(errno);
-        goto fail;
+        return;
     }
     flic_state->fd = cd.fd;
 
@@ -603,9 +604,6 @@ static void kvm_s390_flic_realize(DeviceState *dev, Error **errp)
     test_attr.group = KVM_DEV_FLIC_CLEAR_IO_IRQ;
     flic_state->clear_io_supported = !ioctl(flic_state->fd,
                                             KVM_HAS_DEVICE_ATTR, test_attr);
-    return;
-fail:
-    error_propagate(errp, err);
 }
 
 static void kvm_s390_flic_reset(DeviceState *dev)
Signature · 1 line
2.21.0
Reviewed-by on v1 20/21
#65 lore raw

Re: [PATCH 20/21] hw/intc/s390: Simplify error handling in kvm_s390_flic_realize()

Quoted 7 lines · On Sat, 30 Nov 2019 20:42:39 +0100 Markus Armbruster <[email protected]> wrote:
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Christian Borntraeger <[email protected]>
---
 hw/intc/s390_flic_kvm.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

...someone else wants to take a look before I queue this?

Acked-by
#66 lore raw

Re: [PATCH 20/21] hw/intc/s390: Simplify error handling in kvm_s390_flic_realize()

Quoted 11 lines · On Mon, 2 Dec 2019 17:40:07 +0100 Cornelia Huck <[email protected]> wrote:
Quoted 7 lines · On Sat, 30 Nov 2019 20:42:39 +0100 Markus Armbruster <[email protected]> wrote:
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Christian Borntraeger <[email protected]>
---
 hw/intc/s390_flic_kvm.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
Quoted 4 lines

...someone else wants to take a look before I queue this?

I guess it is a matter of taste.

The only difference I can see is if the **errp argument where to already contain an error (*errp). I guess the old code would just keep the first error, and discard the next one, while error_setv() does an assert(*errp == NULL).

I assume calling with *errp != NULL is not a valid scenario. But then, I can't say I understand the use-case behind this discard the newer error behavior of error_propagate().

Regards,
Halil

#67 lore raw

Re: [PATCH 20/21] hw/intc/s390: Simplify error handling in kvm_s390_flic_realize()

Quoted 14 lines · Halil Pasic
Quoted 11 lines · On Mon, 2 Dec 2019 17:40:07 +0100 Cornelia Huck <[email protected]> wrote:
Quoted 7 lines · On Sat, 30 Nov 2019 20:42:39 +0100 Markus Armbruster <[email protected]> wrote:
Cc Halil Pasic <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Christian Borntraeger <[email protected]>
---
 hw/intc/s390_flic_kvm.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
Quoted 8 lines
Quoted 4 lines

...someone else wants to take a look before I queue this?

I guess it is a matter of taste.

Thanks!

Quoted 4 lines

The only difference I can see is if the **errp argument where to already contain an error (*errp). I guess the old code would just keep the first error, and discard the next one, while error_setv() does an assert(*errp == NULL).

Correct.

Quoted 1 line

I assume calling with *errp != NULL is not a valid scenario.

Correct again. On function entry, @errp must either be null, @error_fatal, @error_abort, or point to null.

Quoted 3 lines

But then, I
can't say I understand the use-case behind this discard the newer error behavior of error_propagate().

The documented[1] use case is "receive and accumulate multiple errors (first one wins)". See the big comment in include/qapi/error.h.

For what it's worth, GError explicitly disallows such accumulation: "The error variable dest points to must be NULL"[2]. If you do it anyway, it logs a warning nobody reads[3], then throws away the newer error.

[1] It's "ex post facto" documentation, like much of the Error API documentation.

[2] https://developer.gnome.org/glib/stable/glib-Error-Reporting.html#g-propagate-error

[3] First order approximation based on the amount of crap supposedly stable applications log.

#68 lore raw

PATCH 21/21 tests-blockjob: Use error_free_or_abort()

In reply to Markus Armbruster
---
 tests/test-blockjob.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
tests/test-blockjob.c +1 -2
index e670a20617..4eeb184caf 100644
@@ -48,9 +48,8 @@ static BlockJob *mk_job(BlockBackend *blk, const char *id,
             g_assert_cmpstr(job->job.id, ==, blk_name(blk));
         }
     } else {
-        g_assert_nonnull(err);
+        error_free_or_abort(&err);
         g_assert_null(job);
-        error_free(err);
     }
 
     return job;
Signature · 1 line
2.21.0
Reviewed-by on v1 21/21
#69 lore raw

Re: [PATCH 21/21] tests-blockjob: Use error_free_or_abort()

Quoted 5 lines · Markus Armbruster
---
  tests/test-blockjob.c | 3 +--
  1 file changed, 1 insertion(+), 2 deletions(-)
Quoted 16 lines
tests/test-blockjob.c +1 -2
index e670a20617..4eeb184caf 100644
@@ -48,9 +48,8 @@ static BlockJob *mk_job(BlockBackend *blk, const char *id,
              g_assert_cmpstr(job->job.id, ==, blk_name(blk));
          }
      } else {
-        g_assert_nonnull(err);
+        error_free_or_abort(&err);
          g_assert_null(job);
-        error_free(err);
      }
 
      return job;
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
Reviewed-by on the v1 series
#70 lore raw
In reply to Markus Armbruster
Quoted 9 lines · Markus Armbruster

PATCH 2-4 fix crash bugs. Including them would be a no-brainer at -rc0. But we're post -rc3, and even for crash bugs we require a certain likelihood of users getting bitten.

Jens, please assess impact of PATCH 2's crash bug.

Kevin, please do the same for PATCH 3.

Daniel, please do the same for PATCH 4.

virtio things:

Jason do you want to pick these?

Quoted 66 lines

The remainder is definitely not 4.2 material.

Cc "Daniel P. Berrangé" <[email protected]>
Cc "Michael S. Tsirkin" <[email protected]>
Cc Christian Borntraeger <[email protected]>
Cc Corey Minyard <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Halil Pasic <[email protected]>
Cc Igor Mammedov <[email protected]>
Cc Jens Freimann <[email protected]>
Cc Kevin Wolf <[email protected]>
Cc Michael Roth <[email protected]>
Cc Michael S. Tsirkin <[email protected]>
Cc Nishanth Aravamudan <[email protected]>
Cc Stefan Hajnoczi <[email protected]>

Markus Armbruster (21):
net/virtio: Drop useless n->primary_dev not null checks net/virtio: Fix failover error handling crash bugs block/file-posix: Fix laio_init() error handling crash bug
crypto: Fix certificate file error handling crash bug
crypto: Fix typo in QCryptoTLSSession's <example> comment
io: Fix Error usage in a comment <example>
tests: Clean up initialization of Error *err variables
exec: Fix latent file_ram_alloc() error handling bug hw/acpi: Fix latent legacy CPU plug error handling bug hw/core: Fix latent fit_load_fdt() error handling bug hw/ipmi: Fix latent realize() error handling bugs
qga: Fix latent guest-get-fsinfo error handling bug
memory-device: Fix latent memory pre-plug error handling bugs s390x/event-facility: Fix latent realize() error handling bug s390x/cpu_models: Fix latent feature property error handling bugs s390/cpu_modules: Fix latent realize() error handling bugs
s390x: Fix latent query-cpu-model-FOO error handling bugs
s390x: Fix latent query-cpu-definitions error handling bug
error: Clean up unusual names of Error * variables hw/intc/s390: Simplify error handling in kvm_s390_flic_realize()
tests-blockjob: Use error_free_or_abort()

 include/crypto/tlssession.h         |  2 +-
 include/io/task.h                   |  2 +-
 block/file-posix.c                  |  2 +-
 crypto/tlscredsx509.c               |  2 +-
 exec.c                              |  6 +-
 hw/acpi/cpu_hotplug.c               | 10 +--
 hw/core/loader-fit.c                | 15 ++---
 hw/intc/s390_flic_kvm.c             | 16 +++--
 hw/ipmi/isa_ipmi_bt.c               |  7 ++-
 hw/ipmi/isa_ipmi_kcs.c              |  7 ++-
 hw/ipmi/pci_ipmi_bt.c               |  6 +-
 hw/ipmi/pci_ipmi_kcs.c              |  6 +-
 hw/mem/memory-device.c              |  6 +-
 hw/net/virtio-net.c                 | 27 ++++----
 hw/ppc/spapr_pci.c                  | 16 ++---
 hw/ppc/spapr_pci_nvlink2.c          | 10 +--
 hw/s390x/event-facility.c           |  6 +-
 qga/commands-posix.c                |  6 +-
 target/s390x/cpu_models.c           | 98 +++++++++++++++++------------
 tests/test-blockjob.c               | 15 +++--
 tests/test-qobject-output-visitor.c |  8 +--
 tests/test-string-output-visitor.c  |  4 +-
 22 files changed, 154 insertions(+), 123 deletions(-)
Signature · 1 line
2.21.0
#71 lore raw
Quoted 7 lines · Michael S. Tsirkin
Quoted 6 lines · Markus Armbruster

PATCH 2-4 fix crash bugs. Including them would be a no-brainer at -rc0. But we're post -rc3, and even for crash bugs we require a certain likelihood of users getting bitten.

Jens, please assess impact of PATCH 2's crash bug.

Quoted 7 lines
Quoted 3 lines

Kevin, please do the same for PATCH 3.

Daniel, please do the same for PATCH 4.

virtio things:

In my haste to get this into -rc4, I lost your r-bys. Sorry about that!

Quoted 1 line

Jason do you want to pick these?

Merged in commit 39032981fa851d25fb27527f25f046fed800e585.

#72 lore raw
In reply to Markus Armbruster
Quoted 9 lines · Markus Armbruster

PATCH 2-4 fix crash bugs. Including them would be a no-brainer at -rc0. But we're post -rc3, and even for crash bugs we require a certain likelihood of users getting bitten.

Jens, please assess impact of PATCH 2's crash bug.

Kevin, please do the same for PATCH 3.

Daniel, please do the same for PATCH 4.

The code has existed like this since 2.5.0, and its not a security issue, so I can't see any justification for putting it into 4.2 at the last minute.

Regards,
Daniel

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#73 lore raw
In reply to Markus Armbruster
Quoted 5 lines · Markus Armbruster

PATCH 2-4 fix crash bugs. Including them would be a no-brainer at -rc0. But we're post -rc3, and even for crash bugs we require a certain likelihood of users getting bitten.

Jens, please assess impact of PATCH 2's crash bug.

Guest can't use it to trigger a qemu crash because the hotplug_handler called after the pre-plug doesn't do anything with errp.

qemu_opt_set_bool() is also unlikely to trigger the bug because both reasons for why it would set errp are not true in this path.

I would prefer for this fix to go into 4.2 but if it is the only one and triggers a large amount of work then don't do it.

regards
Jens