efi_loader: add back optional efi_handler::open()

In some cases it is useful to defer creation of the protocol interface object. So add back an optional ->open() hook that is used if protcol_interface is NULL.

I've slightly simplified the fxn ptr signature to remove unneeded args, and so compiler will complain if patches that used the "old way" are, and which do not need this extra complexity, are rebased.

Signed-off-by Rob Clark <[email protected]>

---
v2: also handle this properly in locate_protocol as agraf pointed out

Note that I might be misunderstanding what the EFI application expects when looking up device-path on a file handle.. if this is only expected to be the device path up to the partition object (and not including the file-path part of the path), then maybe ->open() isn't *as* important as I thought it would be. OTOH I think it would be cleaner to use for loaded_image_info_obj and bootefi_device_obj.. currently I have to patch in the real boot device-path in efi_set_bootdev(), which is messy and fragile.

I don't fully understand Heinrich's concerns, especially since the EFI_OPEN_PROTOCOL_INFORMATION_ENTRY stuff looks like it only cares about already opened protocols, which should already have their open() called. (And either way, since open() is optional maybe it is enough to just not use it for any efiobj's where it would be problematic?)

So tl;dr, if this is really going to be a problem for some things that Heinrich has got in the pipeline, maybe (pending clarification of a file handle's device-path) we can get by without this. But if it is not a problem, then it certainly seems useful in a few places.

 include/efi_loader.h          |  9 ++++++++-
 lib/efi_loader/efi_boottime.c | 30 ++++++++++++++++++++++++------
 2 files changed, 32 insertions(+), 7 deletions(-)

include/efi_loader.h +8 -1

index 48eeb4cdd8..a9888afb04 100644
@@ -41,10 +41,17 @@ extern unsigned int __efi_runtime_rel_start, __efi_runtime_rel_stop;
 /*
  * When the UEFI payload wants to open a protocol on an object to get its
  * interface (usually a struct with callback functions), this struct maps the
- * protocol GUID to the respective protocol interface */
+ * protocol GUID to the respective protocol interface.
+ *
+ * The optional ->open() fxn can be used for cases where the protocol
+ * interface is constructed on-demand, and is called if protocol_interface
+ * is NULL.
+ */
 struct efi_handler {
 	const efi_guid_t *guid;
 	void *protocol_interface;
+	efi_status_t (EFIAPI *open)(void *handle, const efi_guid_t *protocol,
+			void **protocol_interface);
 };
 
 /*

lib/efi_loader/efi_boottime.c +17 -0

index 7d45c18ff7..80d9024a53 100644
@@ -903,6 +903,23 @@ out:
 	return EFI_EXIT(r);
 }
 
+static efi_status_t open_protocol(struct efi_object *efiobj,
+				  struct efi_handler *handler,
+				  void **protocol_interface)
+{
+	efi_status_t ret = EFI_SUCCESS;
+
+	if (!handler->protocol_interface) {
+		ret = handler->open(efiobj->handle,
+				handler->guid,
+				&handler->protocol_interface);
+	}
+	*protocol_interface =
+		handler->protocol_interface;
+
+	return ret;
+}
+
 static efi_status_t EFIAPI efi_locate_protocol(efi_guid_t *protocol,
 					       void *registration,
 					       void **protocol_interface)

lib/efi_loader/efi_boottime.c +4 -3

+				return EFI_EXIT(ret);
 			}
 		}
 	}

@@ -1061,12 +1079,12 @@ static efi_status_t EFIAPI efi_open_protocol(
 			if (!hprotocol)
 				continue;
 			if (!guidcmp(hprotocol, protocol)) {
+				r = EFI_SUCCESS;
 				if (attributes !=
 				    EFI_OPEN_PROTOCOL_TEST_PROTOCOL) {
-					*protocol_interface =
-						handler->protocol_interface;
+					r = open_protocol(efiobj, handler,
+							  protocol_interface);
 				}
-				r = EFI_SUCCESS;
 				goto out;
 			}
 		}

Quoted 88 lines · Rob Clark

In some cases it is useful to defer creation of the protocol interface object. So add back an optional ->open() hook that is used if protcol_interface is NULL.

I've slightly simplified the fxn ptr signature to remove unneeded args, and so compiler will complain if patches that used the "old way" are, and which do not need this extra complexity, are rebased.

Signed-off-by Rob Clark <[email protected]>

---
v2: also handle this properly in locate_protocol as agraf pointed out

Note that I might be misunderstanding what the EFI application expects when looking up device-path on a file handle.. if this is only expected to be the device path up to the partition object (and not including the file-path part of the path), then maybe ->open() isn't *as* important as I thought it would be. OTOH I think it would be cleaner to use for loaded_image_info_obj and bootefi_device_obj.. currently I have to patch in the real boot device-path in efi_set_bootdev(), which is messy and fragile.

I don't fully understand Heinrich's concerns, especially since the EFI_OPEN_PROTOCOL_INFORMATION_ENTRY stuff looks like it only cares about already opened protocols, which should already have their open() called. (And either way, since open() is optional maybe it is enough to just not use it for any efiobj's where it would be problematic?)

So tl;dr, if this is really going to be a problem for some things that Heinrich has got in the pipeline, maybe (pending clarification of a file handle's device-path) we can get by without this. But if it is not a problem, then it certainly seems useful in a few places.

 include/efi_loader.h          |  9 ++++++++-
 lib/efi_loader/efi_boottime.c | 30 ++++++++++++++++++++++++------
 2 files changed, 32 insertions(+), 7 deletions(-)

include/efi_loader.h +8 -1

index 48eeb4cdd8..a9888afb04 100644
@@ -41,10 +41,17 @@ extern unsigned int __efi_runtime_rel_start, __efi_runtime_rel_stop;
 /*
  * When the UEFI payload wants to open a protocol on an object to get its
  * interface (usually a struct with callback functions), this struct maps the
- * protocol GUID to the respective protocol interface */
+ * protocol GUID to the respective protocol interface.
+ *
+ * The optional ->open() fxn can be used for cases where the protocol
+ * interface is constructed on-demand, and is called if protocol_interface
+ * is NULL.
+ */
 struct efi_handler {
      const efi_guid_t *guid;
      void *protocol_interface;
+     efi_status_t (EFIAPI *open)(void *handle, const efi_guid_t *protocol,
+                     void **protocol_interface);
 };
 
 /*

lib/efi_loader/efi_boottime.c +17 -0

index 7d45c18ff7..80d9024a53 100644
@@ -903,6 +903,23 @@ out:
      return EFI_EXIT(r);
 }
 
+static efi_status_t open_protocol(struct efi_object *efiobj,
+                               struct efi_handler *handler,
+                               void **protocol_interface)
+{
+     efi_status_t ret = EFI_SUCCESS;
+
+     if (!handler->protocol_interface) {
+             ret = handler->open(efiobj->handle,
+                             handler->guid,
+                             &handler->protocol_interface);
+     }
+     *protocol_interface =
+             handler->protocol_interface;
+
+     return ret;
+}
+
 static efi_status_t EFIAPI efi_locate_protocol(efi_guid_t *protocol,
                                             void *registration,
                                             void **protocol_interface)

lib/efi_loader/efi_boottime.c static efi_status_t EFIAPI efi_locate_protocol(efi_guid_t *protocol, · line 947

Quoted 9 lines

lib/efi_loader/efi_boottime.c +3 -3

@@ -925,9 +942,10 @@ static efi_status_t EFIAPI efi_locate_protocol(efi_guid_t *protocol,
                      if (!handler->guid)
                              continue;
                      if (!guidcmp(handler->guid, protocol)) {
-                             *protocol_interface =
-                                     handler->protocol_interface;
-                             return EFI_EXIT(EFI_SUCCESS);
+                             efi_status_t ret;
+                             ret = open_protocol(efiobj, handler,
+                                                 protocol_interface);

NAK

With your design a driver could decide that on the first call of open_protocol it exposes a protocol and on the second it does not.

Or LocateHandleBuffer return an object meant to support a protocol and then opening is refused.

The UEFI standard assumes that the supported protocols can be reliably be determined once and for all.

If anything, please, use an init function of type:
void init(efi_guid guid);
Simply pass the protocol guid to the init function of the handler.

How many ms and kiB do you think you will save when starting grub with deferred initialization?

Best regards

Heinrich

Quoted 21 lines

lib/efi_loader/efi_boottime.c +4 -3

+                             return EFI_EXIT(ret);
                      }
              }
      }

@@ -1061,12 +1079,12 @@ static efi_status_t EFIAPI efi_open_protocol(
                      if (!hprotocol)
                              continue;
                      if (!guidcmp(hprotocol, protocol)) {
+                             r = EFI_SUCCESS;
                              if (attributes !=
                                  EFI_OPEN_PROTOCOL_TEST_PROTOCOL) {
-                                     *protocol_interface =
-                                             handler->protocol_interface;
+                                     r = open_protocol(efiobj, handler,
+                                                       protocol_interface);
                              }
-                             r = EFI_SUCCESS;
                              goto out;
                      }
              }