Protecting Memory Pages for Underflow Detection — Handmade Hero — Episode Guide

0:01Recap and set the stage for the day

1:16Blackboard: Buffer Overruns

2:47Blackboard: VirtualAlloc() and virtual memory mapping

6:23Blackboard: Playing with how this memory works

8:23Blackboard: Guard Page with VirtualProtect()

9:38Blackboard: Faulting at the smallest error

13:29handmade_platform.h: Add PlatformMemory_OverflowCheck and PlatformMemory_UnderflowCheck flags

14:51handmade_memory.h: Enable PushSize_() to perform bounds checking by passing every allocation down to the system every time

17:26Run the game and see nothing in particular happen

17:37handmade_memory.h: Make PushSize_() set the OverflowCheck flag on everything, run the game and grind to a halt

19:24"You're an operating system. Have some self-respect"^α

19:51"It's 2016, and all you have to do to kill Windows is just allocate some memory"^β

20:35Return with the determination to provide a way to free our memory

21:48handmade_memory.h: Change PushSize_() to ensure arenas only keep the exact size they need

23:44Run the game to give that a shot

24:32handmade_memory.h: Rename memory_block_footer to memory_block_chain

27:02Blackboard: Positioning that memprotect in order to catch underflows

27:26win32_handmade.cpp: Make PLATFORM_ALLOCATE_MEMORY() allocate one blank 4K page at the beginning

30:13handmade_memory.h: Consider unifying the arena and allocator's data structures

31:40win32_handmade.cpp: Make PLATFORM_ALLOCATE_MEMORY() allocate

32:34Run the game to see it run exactly the same

32:45win32_handmade.cpp: Make PLATFORM_ALLOCATE_MEMORY() call VirtualProtect()¹^,2

37:03Blackboard: What actually happens with the guard page

39:00win32_handmade.cpp: Pass PAGE_NOACCESS to that VirtualProtect() call

39:38win32_handmade.cpp: Verify that the guard pages are happening properly

40:11Run the game with the expectation to crash and investigate why we don't

40:38handmade_memory.h: Make PushSize_() set the UnderflowCheck flag, run the game and successfully crash

41:28win32_handmade.cpp: Enable PLATFORM_ALLOCATE_MEMORY() to perform underflow checking

47:02win32_handmade.cpp: Enable PLATFORM_DEALLOCATE_MEMORY() to handle the checking

51:06handmade_world.cpp: Make GetWorldChunk() write off the beginning of the buffer, run the game and see nothing being detected

51:57handmade_memory.cpp: Toggle on the underflow checking in PushSize_(), run the game and immediately catch that offending write

52:49handmade_world.cpp: Make GetWorldChunk() write off the end of the buffer, run the game and see nothing being detected

53:34handmade_memory.cpp: Determine to enable PushSize_() to perform overflow checking

55:16todo.txt: Update the TODO list

55:39Q&A

🗩

55:39Q&A

🗩

55:39Q&A

🗩

55:59insobot !qa

🗪

55:59insobot !qa

🗪

55:59insobot !qa

🗪

56:46insofaras Yes

🗪

56:46insofaras Yes

🗪

56:46insofaras Yes

🗪

58:16k2t0f12d I'm following an Linux with SDL because I'm not sure if it's worth building on X11 or if Wayland is ready yet. Have you thought about building on X11, Wayland or both?

🗪

58:16k2t0f12d I'm following an Linux with SDL because I'm not sure if it's worth building on X11 or if Wayland is ready yet. Have you thought about building on X11, Wayland or both?

🗪

58:16k2t0f12d I'm following an Linux with SDL because I'm not sure if it's worth building on X11 or if Wayland is ready yet. Have you thought about building on X11, Wayland or both?

🗪

59:28drjeats Before when you had fixed allocation, you could do a between-runs save state kind of thing (if I recall correctly). How would you make that happen with the dynamic blocks?

🗪

59:28drjeats Before when you had fixed allocation, you could do a between-runs save state kind of thing (if I recall correctly). How would you make that happen with the dynamic blocks?

🗪

59:28drjeats Before when you had fixed allocation, you could do a between-runs save state kind of thing (if I recall correctly). How would you make that happen with the dynamic blocks?

🗪

59:48flirtychair Do you have any insight into if you are developing this game in a similar timeflow to major devs, e.g. you spent some time today working on under / overflow before such a thing came into play. Is this just because of the format of the show?

🗪

59:48flirtychair Do you have any insight into if you are developing this game in a similar timeflow to major devs, e.g. you spent some time today working on under / overflow before such a thing came into play. Is this just because of the format of the show?

🗪

59:48flirtychair Do you have any insight into if you are developing this game in a similar timeflow to major devs, e.g. you spent some time today working on under / overflow before such a thing came into play. Is this just because of the format of the show?

🗪

1:01:06ab1093996300 Why do you look so different from Handmade Hero Day 001?

🗪

1:01:06ab1093996300 Why do you look so different from Handmade Hero Day 001?

🗪

1:01:06ab1093996300 Why do you look so different from Handmade Hero Day 001?

🗪

1:01:45drjeats Does it really work between runs of the program? I saw it work with looped editing, but not different runs. Wouldn't you need to muck with base pointers?

🗪

1:01:45drjeats Does it really work between runs of the program? I saw it work with looped editing, but not different runs. Wouldn't you need to muck with base pointers?

🗪

1:01:45drjeats Does it really work between runs of the program? I saw it work with looped editing, but not different runs. Wouldn't you need to muck with base pointers?

🗪

1:03:46zamar037 I know this isn't relevant to the code, but does it seem strange streaming in a different location?

🗪

1:03:46zamar037 I know this isn't relevant to the code, but does it seem strange streaming in a different location?

🗪

1:03:46zamar037 I know this isn't relevant to the code, but does it seem strange streaming in a different location?

🗪

1:04:05enemymouse Seems that Arena started out as a fast, stack-style allocator and is now moving towards a dynamic heap. You've demonstrated the benefits of VirtualAlloc etc. but can you comment on the performance hit and why you choose it over malloc?

🗪

1:04:05enemymouse Seems that Arena started out as a fast, stack-style allocator and is now moving towards a dynamic heap. You've demonstrated the benefits of VirtualAlloc etc. but can you comment on the performance hit and why you choose it over malloc?

🗪

1:04:05enemymouse Seems that Arena started out as a fast, stack-style allocator and is now moving towards a dynamic heap. You've demonstrated the benefits of VirtualAlloc etc. but can you comment on the performance hit and why you choose it over malloc?

🗪

1:05:22williambundy Don't recognize the tune

🗪

1:05:22williambundy Don't recognize the tune

🗪

1:05:22williambundy Don't recognize the tune

🗪

1:05:45windowl Are any days (streams) missing from the YouTube archive?

🗪

1:05:45windowl Are any days (streams) missing from the YouTube archive?

🗪

1:05:45windowl Are any days (streams) missing from the YouTube archive?

🗪

1:07:05drjeats Was memprotect the primary motivation for moving to dynamic memory, or were there other reasons?

🗪

1:07:05drjeats Was memprotect the primary motivation for moving to dynamic memory, or were there other reasons?

🗪

1:07:05drjeats Was memprotect the primary motivation for moving to dynamic memory, or were there other reasons?

🗪

1:09:22Wrap it up

🗩

1:09:22Wrap it up

🗩

1:09:22Wrap it up

🗩

Keyboard Navigation

Global Keys

Menu toggling

In-Menu and Index Controls

Quotes and References Menus and Index

Quotes, References and Credits Menus

Filter Menu

Filter and Link Menus

Credits Menu