thermosphere: major refactor of memory map

- use recursive stage 1 page table (thanks @fincs for this idea)
- NULL now unmapped
- no identity mapping
- image + GICv2 now mapped at the same address for every platform
- tempbss mapped just after "real" bss, can now steal unused mem from
the latter
- no hardcoded VAs for other MMIO devices
- tegra: remove timers, use the generic timer instead

thermosphere/src/start.s

@@ -32,16 +32,13 @@ g_initialKernelEntrypoint:
 start:
     mov     x19, #1
     b       _startCommon
-.global start2
-.type   start2, %function
 start2:
     mov     x19, xzr
 _startCommon:
-    // Disable interrupts, select sp_el2
-    msr     daifset, 0b1111
-    msr     spsel, #1
-
+    // Disable interrupts, select sp_el0 before mmu is enabled
     mrs     x20, cntpct_el0
+    msr     daifset, 0b1111
+    msr     spsel, #0
 
     // Set sctlr_el2 ASAP to disable mmu/caching if not already done.
     mov     x1, #0x0838
@@ -52,25 +49,39 @@ _startCommon:
 
     bl      cacheClearLocalDataCacheOnBoot
     cbz     x19, 1f
+
+    // "Boot core only" stuff:
     bl      cacheClearSharedDataCachesOnBoot
+    // Temporarily use temp end region as stack, then create the translation table
+    // The stack top is also equal to the mmu table address...
+    adr     x0, g_loadImageLayout
+    ldp     x2, x3, [x0, #0x18]
+    add     x1, x2, x3
+    mov     sp, x1
+    bl      memoryMapSetupMmu
 
 1:
+    // Enable MMU, note that the function is not allowed to use any stack
+    adr     x0, g_loadImageLayout
+    ldr     x18, =_postMmuEnableReturnAddr
+    bl      memoryMapEnableMmu
+
+    // This is where we will land on exception return after enabling the MMU:
+_postMmuEnableReturnAddr:
+    // Select sp_el2
+    msr     spsel, #1
+
     // Get core ID
-    // Ensure Aff0 is 4-1 at most (4 cores), and that Aff1, 2 and 3 are 0 (1 cluster only)
-    mrs     x0, mpidr_el1
-    tst     x0, #(0xFF << 32)
-    bne     .
-    and     x0, x0, #0x00FFFFFF        // Aff0 to 2
-    cmp     x0, #4
-    bhs     .
+    mrs     x8, mpidr_el1
+    and     x8, x8, #0xFF
 
-    // Set stack pointer
-    adrp    x8, __stacks_top__
-    lsl     x9, x0, #10
-    sub     sp, x8, x9
+    mov     w0, w8
+    bl      memoryMapGetStackTop
+    mov     sp, x0
 
-    // Set up x18, other sysregs, BSS, MMU, etc.
+    // Set up x18, other sysregs, BSS, etc.
     // Don't call init array to save space?
+    mov     w0, w8
     mov     w1, w19
     bl      initSystem
 
@@ -92,3 +103,26 @@ _startCommon:
     b       _restoreAllRegisters
 
 .pool
+
+/*
+    typedef struct LoadImageLayout {
+        uintptr_t startPa;
+        size_t imageSize; // "image" includes "real" BSS but not tempbss
+        size_t maxImageSize;
+
+        uintptr_t tempPa;
+        size_t maxTempSize;
+        size_t tempSize;
+
+        uintptr_t vbar;
+    } LoadImageLayout;
+*/
+.global     g_loadImageLayout
+g_loadImageLayout:
+    .quad       __start_pa__
+    .quad       __max_image_size__
+    .quad       __image_size__
+    .quad       __temp_pa__
+    .quad       __max_temp_size__
+    .quad       __temp_size__
+    .quad       __vectors_start__