POE 2 Complete PC Freeze while loading screen
" Before we start: Please run 0:015> r. My observations: Thread 15 (502c): Raising an Exception but more importantly it is not waiting, instead it is actively crashing. If this thread was supposed to signal one of the waiting threads, the wake-up signal is lost, causing the other threads to remain stuck. In our case it is. Threads 16, 17, 18, 19: These are stuck in SleepConditionVariableSRW(). Since Thread 15 crashed it will never come and they will remain stuck forever. Them sharing the same address means thread 15 was performing a critical task but failed, leaving the others in limbo. The next block is an exception handling method as evidenced by the existence of CxxFrameHandler4. After returning from sub_1401CAF60, the function: 1) Executes a nop (likely for alignment or debugging). 2) Clears the eax register (xor eax, eax), meaning it is returning 0. Returning 0 indicates success if they are following the standard coding practices. 3) Restores the stack (add rsp, 58h). 4)Returns (retn). Return is 0. Since sub_140C560B0 only calls sub_1401CAF60 and returns, it is unlikely that sub_140C560B0 itself caused the exception. Instead, sub_1401CAF60 is the function that either crashed or explicitly threw an exception. If sub_1401CAF60 performed an invalid operation (for example dereferencing a null pointer, accessing an invalid memory location, or causing a divide-by-zero error), it could trigger an access violation (0xC0000005). If sub_1401CAF60 encountered an assertion failure or logic error, it could have explicitly called RaiseException(), causing Thread 15 to crash. If sub_1401CAF60 was supposed to signal a condition variable before returning, but instead crashed, then: Thread 15 stopped before waking up waiting threads (Threads 16-19). This left those threads stuck in SleepConditionVariableSRW(). The result: a deadlock, leading to a system freeze. Continuing on for the block "this is the code that creates those threads": The thread stack size is being set (dwStackSize = 0x80000 or 512KB). This means the thread is given a dedicated stack size, but not an unusually large one. Now this in itself can give us actual details about what sub_1401CAF60 which is called by sub_140C560B0 supposed to do because Explicit Stack Size is often used only on certain things and we know from your previous post that there is a recurring pattern of function calls for DX12 so this is most likely: Shader compilation. Other possibilities are: Command buffer generation, scene rendering traversal. For a Vulkan/DX12 worker thread handling deep recursion, large GPU structures, or multi-threaded rendering 512KB is quite low. Now... I both want this to be true and false at the same time but... 0x80000 is 512KB 0x800000 is 8MB. 8MB is also the suggested stack size for heavy workloads such as... Deep recursion. It is also small enough to be missed. If this is not it then 512KB is just a red herring. But still it is quite low in my opinion since even the default is 1MB. A well optimized code could be able to run this in a way that Toyota Hilux was running at the end of Top Gear. Why I'm focused on this: If the thread overflows its stack and corrupts kernel memory, it can cause a system-wide freeze instead of just crashing the application. Some Windows drivers use the same kernel stack memory region, meaning an out-of-bounds write could cause a hard lockup. Anyway... Continuing on from "those other threads that don't generate an exception are waiting and here is the code": This is a System API. And it points out to the following scenario: 1) If Thread 15 is on deep recursion and exhausted its 512 KB stack, it would crash before completing execution. 2) It would crash before calling WakeConditionVariable() 3) Without WakeConditionVariable() no threads are ever woken up, leaving the others stuck. 4) If these threads were rendering threads, the entire graphics pipeline would stall, leading to a system freeze. Continuing from "this code is called from here ( from sub_140C55D10 )": This function appears to be preparing and waiting on a condition variable, ensuring that a certain event or resource becomes available before proceeding. The key is that it eventually calls sub_1422D69F4, which in turn calls SleepConditionVariableSRW(). This means that if the expected condition is never met, the thread will wait indefinitely. rdi is the first function parameter, meaning this function is likely operating on an object (for example a synchronization object, a queue, or a resource handler). Continuing from "The Array": I believe this is a thread pool managing GPU work. A GPU task scheduler could assign worker threads to functions in this list. A failure (such as stack overflow in Thread 15) could disrupt the entire system, leaving worker threads waiting. Continuing from "Then I've searched for that array in the code": I'm now 99% sure this is the thread pool managing GPU work. The function iterates through a list of work items or jobs (rsi linked structure). It loads a function pointer table (off_1427D3C30), meaning it dynamically selects which function to execute. There’s exception handling (pExceptionObject), but if a failure is not properly handled, it could leave threads stuck. We know it is not properly handling. The final box: YOU MIGHT HAVE SOLVED IT. OR COME VERY CLOSE TO IT. Please run 0:015> r. It would print all register values, including Rip, Rsp, and Rbp, which could confirm if: If Rip is in RaiseException(), the crash was likely triggered deliberately (assertion failure, runtime abort, etc.). If Rip is an invalid address, the stack may have overflown, corrupting return addresses. If Rip points to a function in PathOfExileSteam, the game’s threading code is failing internally. If Rsp is abnormally low, it strengthens the case that 512 KB was too small, leading to a stack overflow. TLDR: There might be a typo at the point where they set the stack sizes. 0x80000 is 512KB 0x800000 is 8MB. The size is set as 0x80000. Obligatory: If only GGG were honest we would not even have to have this conversation... We could simply look at GGGs explanation of the situation with proper technical information but we can never know why exactly this is happening as GGG keeps being dodgy and dishonest. By the way, Microsoft is still not acknowledging this is a Windows 11 24H2 issue as claimed by GGG. As long as GGG is not interested in open and honest communication our only other option is to increase the pressure and warn others by giving a negative review for Path of Exile 2 on Steam in a detailed and informative manner. I cannot send/reply to direct messages because my in-game character has not finished Act 1. What to do: 1)Write a short review about the hard crashes in notepad. 2)Copy and paste it to steam reviews, put up a negative review. 3)Copy and paste it to steam discussions, put it up there. Zuletzt bearbeitet von Cainrith#2807 um 18.02.2025, 20:22:47
|
|
" Not only will it not work, it's completely non-applicable. This is an early access game. Early access is defined as: Early access, also known as alpha access, alpha founding, paid alpha, or game preview, is a funding model in the video game industry by which consumers can purchase and play a game in the various pre-release development cycles, such as pre-alpha, alpha, and/or beta, while the developer is able to use those funds to continue further development on the game. Those that pay to participate typically help to debug the game, provide feedback and suggestions, may have access to special materials in the game. The early-access approach is a common way to obtain funding for indie games, and may also be used along with other funding mechanisms, including crowdfunding. Many crowdfunding projects promise to offer access to alpha and/or beta versions of the game as development progresses; however, unlike some of these projects which solicit funds but do not yet have a playable game, all early access games offer an immediately playable version of the unfinished game to players. So, you didn't "buy" anything. You contributed to the development of a game slated to be free to play later this year, and in exchange for that, you got a work in progress to play and bug test. I get the frustration of spending money and having issues. I do. But this misconception of "I should get a finished game" needs to end. You bought access to an alpha/beta. If that's too much for some people and uninstalling is what's right for their peace of mind, then aces. But let's stop trying to incite legal this and legal that when there's no leg to stand on when people went in with at least the correct understanding made available to them. If they were in a rush to play and didn't read through what they signed up for, then that's on them, not GGG. |
|
" Not GPU related : it is a thread stack that is handling some sort of watchdog for internet connection and responsible for server hop in case it didn't find any availabie instance. Except the thread crap itself with jumbo frame and modern way of handling network. POE1 is plagged with exactly the same code and it's been ages it hasn't been changed. You can somehow make the code run even worse tinkering with your MTU size and/or playing with your DNS such as using google DNS. And it is not a deadlock, as if the thread goes non responding, the main loop create another one while the old is zombified in the CPU command queue : it is litteral fork bomb (as it only clear all of its instances if a successful connection happens) |
|
" So, basically, you know exactly where this freeze happens and originates from and in what circumstances it may happen. So in summary, according to you, some piece of code that is suppose to handle the case where there are no available instances to host the map (could be server issues, or personal internet problems, or just random lag etc) is not working properly, which in turn makes the main loop start to mass-produce threads until a successful connection is made. So if a connection is made before the computer crashes we are good, and if not then total freeze. This explains why the freeze only happens when loading a new map. It explains why it seems to happen at random times. It kind of explains why there seems to be no correlation between hardware and software and the likelihood of getting a crash. nvida, amd, intel, drivers, bios, windows version etc, makes no difference. It all comes down to how stable your connection is, how well your DNS is performing that day, your ISP, and how well the servers are feeling etc. Does this mean that freezes are more likely to happen during times where many ppl play, since it may take longer to get a free instance to host the map on? It it worth trying to play with a cable instead of wifi (I think not but every millisecond matters I guess)? If this is true its honestly incredible. GGG is disabling multi-threading on load screens as a temporary solution to this and seems to think its Microsofts fault somehow. To me, with my very limited knowledge, this seems like it should be relatively easy to find and fix right? Or do you think its a complicated bug? Im so confused. |
|
" First, thank you guys for spending time to look into that. Now, Freeze may not only happen when loading maps, but mainly. Outside of loading screen seems to be very rare, but they do seem to occur. I had it after 0.0.1e, when the monsters got spawned on the map by trialmaster and also this week when my friend ported to me at the beginning of the vanguard dreadnought (when his char got spawned maybe?). I can't be 100% that it is the same bug, but, those were also full freeze, with slow mouse cursor and nothing else responding after something got loaded. Pretty hard to reproduce outside of loading screen though. Big patch fixed the issue for me, so I was very surprised when it happened, but yeah, outside of loading screen, as far as I know the workaround is not used. Hope this help even a little bit. |
|
" Well. I also had complete freeze when opening the caravan map for example. I don't know why, but I have a feeling the map is hosted on a different server or instance somehow compared to the main game. It takes a really long time to load sometimes and it even comes with the cogweels of death. Bottom line is, it probably need to contact something from outside to fetch the info it needs to render the map. So its possible that the same code runs there. Again, this is pure speculation. When your friend ported to you, maybe the game needed both of you "moved" to a new instance for some reason? Or maybe its a completely unrelated crash as you mentioned. The amount of random freezes outside of actual map loads are extremely limited. It happed to me once. Freezes during map loads however... Im saying this as someone who know absolutely nothing about how the game works. So take everything I say and have said regarding this with healthy scoop of salt. |
|
|
Oh my bad I got things mixed up with all the quotes and replies, I thought you were the person analysing the code/connection. Sorry.
I am not sure how instances work either, but it just felt weird to have the infamous freeze with only my friend spawning. Also yes the map is funny sometimes, especially the caravan one, it opens fast for me now, but get the checker texture for half a second most of the time before it changes to what it needs to be. |
|
" bad take, thats just outright moronic. If i buy a product in early access i expect bugs and crashes, not for that product to actively HARM my pc by playing it because it CRASHES the SYSTEM not the game, under full cpu load, requiring a hard reset of the pc itself. that's not what i signed up for and i have NEVER in my 2 decades of pc gaming EVER encountered an issue this malicious or dangerous to pc's in a video game. i didn't sign up to test a game early so they could fuck my computer in the process. no other game devs have forced their community to send in useless dxdiags that wont pinpoint the issue any further than what we already know. its not windows, its not nvidia or amd or any other fuckin random company. it's an issue with the code in GGG's engine. and it is actively destroying computers which is outright unacceptable regardless of what state the game is in. game crashes should be confined to THE GAME. not my entire fucking system thats just ridiculous. |
|
|
LMFAO only ones left in this thread are the ones demanding downvotes for GGG experiencing a bug that they couldnt fix over night.
[Removed by Support] "Well people voted with their wallets and reviews. And that twitch watch numbers..." HAHAHAHAHAHAHAHAHAHAHA, yea every EA and beta since the beginning of time has retained its player base from day 1, nobody ever stops playing something when it stops being new, we followed your voice of reasoning and hit the devs where it hurts most! Just lol Zuletzt bearbeitet von Vash_GGG#0000 um 19.02.2025, 18:09:32
|
|
" Poe1 just got patched, where is the fix |
|
