From 4b917390acce021ffd7f015661a447bf61ceb381 Mon Sep 17 00:00:00 2001
From: Bananymous <oskari.alaranta@bananymous.com>
Date: Tue, 16 Jul 2024 23:15:11 +0300
Subject: [PATCH] Kernel: Fix sse state saving

This was broken when I added SMP support. This patch makes sse kind of
dumb as it is saved and restored on every interrupt, but now it at least
works properly... I'll have to look into how sse can get optimized
nicely with SMP. Simple way would be pinning each thread to a specific
processor and doing pretty much what I had before, but sse thread saved
in processor rather than static global.
---
 .vscode/c_cpp_properties.json  | 14 +++++++-----
 kernel/include/kernel/Thread.h |  1 -
 kernel/kernel/IDT.cpp          | 41 +++++++++++++---------------------
 kernel/kernel/Thread.cpp       | 35 +----------------------------
 4 files changed, 25 insertions(+), 66 deletions(-)

diff --git a/.vscode/c_cpp_properties.json b/.vscode/c_cpp_properties.json
index fd7a95183e..f6bf985c0b 100644
--- a/.vscode/c_cpp_properties.json
+++ b/.vscode/c_cpp_properties.json
@@ -8,8 +8,9 @@
                 "${workspaceFolder}/userspace/libraries/*/include"
             ],
             "defines": [
-				"__arch=x86_64"
-			],
+                "__arch=x86_64",
+                "__enable_sse=1"
+            ],
             "compilerPath": "${workspaceFolder}/toolchain/local/bin/x86_64-banan_os-gcc",
             "cStandard": "c17",
             "cppStandard": "gnu++20",
@@ -23,9 +24,10 @@
                 "${workspaceFolder}/userspace/libraries/*/include"
             ],
             "defines": [
-				"__arch=x86_64",
-				"__is_kernel"
-			],
+                "__arch=x86_64",
+                "__is_kernel",
+                "__enable_sse=1"
+            ],
             "compilerPath": "${workspaceFolder}/toolchain/local/bin/x86_64-banan_os-gcc",
             "cStandard": "c17",
             "cppStandard": "gnu++20",
@@ -33,4 +35,4 @@
         }
     ],
     "version": 4
-}
\ No newline at end of file
+}
diff --git a/kernel/include/kernel/Thread.h b/kernel/include/kernel/Thread.h
index ff11e345bc..701fab7a64 100644
--- a/kernel/include/kernel/Thread.h
+++ b/kernel/include/kernel/Thread.h
@@ -82,7 +82,6 @@ namespace Kernel
 #if __enable_sse
 		void save_sse();
 		void load_sse();
-		static Thread* sse_thread();
 #endif
 
 		void add_mutex() { m_mutex_count++; }
diff --git a/kernel/kernel/IDT.cpp b/kernel/kernel/IDT.cpp
index b3fa74c36d..455f35b8a2 100644
--- a/kernel/kernel/IDT.cpp
+++ b/kernel/kernel/IDT.cpp
@@ -173,6 +173,10 @@ namespace Kernel
 
 		if (tid)
 		{
+#if __enable_sse
+			Thread::current().save_sse();
+#endif
+
 			if (isr == ISR::PageFault)
 			{
 				// Check if stack is OOB
@@ -218,31 +222,6 @@ namespace Kernel
 					}
 				}
 			}
-#if __enable_sse
-			else if (isr == ISR::DeviceNotAvailable)
-			{
-#if ARCH(x86_64)
-				asm volatile(
-					"movq %cr0, %rax;"
-					"andq $~(1 << 3), %rax;"
-					"movq %rax, %cr0;"
-				);
-#elif ARCH(i686)
-				asm volatile(
-					"movl %cr0, %eax;"
-					"andl $~(1 << 3), %eax;"
-					"movl %eax, %cr0;"
-				);
-#endif
-				if (auto* current = &Thread::current(); current != Thread::sse_thread())
-				{
-					if (auto* sse = Thread::sse_thread())
-						sse->save_sse();
-					current->load_sse();
-				}
-				goto done;
-			}
-#endif
 		}
 
 		Debug::s_debug_lock.lock();
@@ -334,7 +313,11 @@ namespace Kernel
 		ASSERT(Thread::current().state() != Thread::State::Terminated);
 
 done:
+#if __enable_sse
+		Thread::current().load_sse();
+#else
 		return;
+#endif
 	}
 
 	extern "C" void cpp_yield_handler(InterruptStack* interrupt_stack, InterruptRegisters* interrupt_registers)
@@ -357,6 +340,10 @@ done:
 			asm volatile("cli; 1: hlt; jmp 1b");
 		}
 
+#if __enable_sse
+		Thread::current().save_sse();
+#endif
+
 		if (!InterruptController::get().is_in_service(irq))
 			dprintln("spurious irq 0x{2H}", irq);
 		else
@@ -377,6 +364,10 @@ done:
 		Scheduler::get().reschedule_if_idling();
 
 		ASSERT(Thread::current().state() != Thread::State::Terminated);
+
+#if __enable_sse
+		Thread::current().load_sse();
+#endif
 	}
 
 	void IDT::register_interrupt_handler(uint8_t index, void (*handler)())
diff --git a/kernel/kernel/Thread.cpp b/kernel/kernel/Thread.cpp
index 75c6057037..c7d255b0dd 100644
--- a/kernel/kernel/Thread.cpp
+++ b/kernel/kernel/Thread.cpp
@@ -113,33 +113,8 @@ namespace Kernel
 		: m_tid(tid), m_process(process)
 	{
 #if __enable_sse
-	#if ARCH(x86_64)
-		uintptr_t cr0;
-		asm volatile(
-			"movq %%cr0, %%rax;"
-			"movq %%rax, %[cr0];"
-			"andq $~(1 << 3), %%rax;"
-			"movq %%rax, %%cr0;"
-			: [cr0]"=r"(cr0)
-			:: "rax"
-		);
+		// initializes sse storage to valid state
 		save_sse();
-		asm volatile("movq %0, %%cr0" :: "r"(cr0));
-	#elif ARCH(i686)
-		uintptr_t cr0;
-		asm volatile(
-			"movl %%cr0, %%eax;"
-			"movl %%eax, %[cr0];"
-			"andl $~(1 << 3), %%eax;"
-			"movl %%eax, %%cr0;"
-			: [cr0]"=r"(cr0)
-			:: "eax"
-		);
-		save_sse();
-		asm volatile("movl %0, %%cr0" :: "r"(cr0));
-	#else
-		#error
-	#endif
 #endif
 	}
 
@@ -482,8 +457,6 @@ namespace Kernel
 	}
 
 #if __enable_sse
-	static Thread* s_sse_thread = nullptr;
-
 	void Thread::save_sse()
 	{
 		asm volatile("fxsave %0" :: "m"(m_sse_storage));
@@ -492,12 +465,6 @@ namespace Kernel
 	void Thread::load_sse()
 	{
 		asm volatile("fxrstor %0" :: "m"(m_sse_storage));
-		s_sse_thread = this;
-	}
-
-	Thread* Thread::sse_thread()
-	{
-		return s_sse_thread;
 	}
 #endif