blob: 102883a10deb031d741dc9b659169b7f48b59e53 [file] [log] [blame]
/*
*
* Copyright (C) 2015 LunarG, Inc.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*
* Authors:
* Jon Ashburn <jon@lunarg.com>
* Courtney Goeltzenleuchter <courtney@lunarg.com>
*/
#include "vk_loader_platform.h"
#include "loader.h"
#include "vk_debug_report_lunarg.h"
/*
* CreateMsgCallback is global and needs to be
* applied to all layers and ICDs.
* What happens if a layer is enabled on both the instance chain
* as well as the device chain and a call to CreateMsgCallback is made?
* Do we need to make sure that each layer / driver only gets called once?
* Should a layer implementing support for CreateMsgCallback only be allowed (?)
* to live on one chain? Or maybe make it the application's responsibility.
* If the app enables DRAW_STATE on at both CreateInstance time and CreateDevice
* time, CreateMsgCallback will call the DRAW_STATE layer twice. Once via
* the instance chain and once via the device chain.
* The loader should only return the DEBUG_REPORT extension as supported
* for the GetGlobalExtensionSupport call. That should help eliminate one
* duplication.
* Since the instance chain requires us iterating over the available ICDs
* and each ICD will have it's own unique MsgCallback object we need to
* track those objects to give back the right one.
* This also implies that the loader has to intercept vkDestroyObject and
* if the extension is enabled and the object type is a MsgCallback then
* we must translate the object into the proper ICD specific ones.
* DestroyObject works on a device chain. Should not be what's destroying
* the MsgCallback object. That needs to be an instance thing. So, since
* we used an instance to create it, we need a custom Destroy that also
* takes an instance. That way we can iterate over the ICDs properly.
* Example use:
* CreateInstance: DEBUG_REPORT
* Loader will create instance chain with enabled extensions.
* TODO: Should validation layers be enabled here? If not, they will not be in the instance chain.
* fn = GetProcAddr(INSTANCE, "vkCreateMsgCallback") -> point to loader's vkCreateMsgCallback
* App creates a callback object: fn(..., &MsgCallbackObject1)
* Have only established the instance chain so far. Loader will call the instance chain.
* Each layer in the instance chain will call down to the next layer, terminating with
* the CreateMsgCallback loader terminator function that creates the actual MsgCallbackObject1 object.
* The loader CreateMsgCallback terminator will iterate over the ICDs.
* Calling each ICD that supports vkCreateMsgCallback and collect answers in icd_msg_callback_map here.
* As result is sent back up the chain each layer has opportunity to record the callback operation and
* appropriate MsgCallback object.
* ...
* Any reports matching the flags set in MsgCallbackObject1 will generate the defined callback behavior
* in the layer / ICD that initiated that report.
* ...
* CreateDevice: MemTracker:...
* App does not include DEBUG_REPORT as that is a global extension.
* TODO: GetExtensionSupport must not report DEBUG_REPORT when using instance.
* App MUST include any desired validation layers or they will not participate in the device call chain.
* App creates a callback object: fn(..., &MsgCallbackObject2)
* Loader's vkCreateMsgCallback is called.
* Loader sends call down instance chain - this is a global extension - any validation layer that was
* enabled at CreateInstance will be able to register the callback. Loader will iterate over the ICDs and
* will record the ICD's version of the MsgCallback2 object here.
* ...
* Any report will go to the layer's report function and it will check the flags for MsgCallbackObject1
* and MsgCallbackObject2 and take the appropriate action as indicated by the app.
* ...
* App calls vkDestroyMsgCallback( MsgCallbackObject1 )
* Loader's DestroyMsgCallback is where call starts. DestroyMsgCallback will be sent down instance chain
* ending in the loader's DestroyMsgCallback terminator which will iterate over the ICD's destroying each
* ICD version of that MsgCallback object and then destroy the loader's version of the object.
* Any reports generated after this will only have MsgCallbackObject2 available.
*/
void debug_report_add_instance_extensions(
const struct loader_instance *inst,
struct loader_extension_list *ext_list);
void debug_report_create_instance(
struct loader_instance *ptr_instance,
const VkInstanceCreateInfo *pCreateInfo);
bool debug_report_instance_gpa(
struct loader_instance *ptr_instance,
const char* name,
void **addr);
VkResult VKAPI loader_DbgCreateMsgCallback(
VkInstance instance,
VkFlags msgFlags,
const PFN_vkDbgMsgCallback pfnMsgCallback,
void* pUserData,
VkDbgMsgCallback* pMsgCallback);
VkResult VKAPI loader_DbgDestroyMsgCallback(
VkInstance instance,
VkDbgMsgCallback msgCallback);