From f189457b79989543f65b8a4e8729eff2cdf9a758 Mon Sep 17 00:00:00 2001
From: Khem Raj <raj.khem@gmail.com>
Date: Sat, 13 Aug 2022 19:24:55 -0700
Subject: [PATCH] core: Define section attributes for clang

Clang's attribute section is not same as gcc, here we need to add flags
to sections so they can be eventually collected by linker into final
output segments. Only way to do so with clang is to use

pragma clang section ...

The behavious is described here [1], this allows us to define names bss
sections. This was not an issue until clang-15 where LLD linker starts
to detect the section flags before merging them and throws the following
errors

| ld.lld: error: section type mismatch for .nozi.kdata_page
| >>> /mnt/b/yoe/master/build/tmp/work/qemuarm64-yoe-linux/optee-os-tadevkit/3.17.0-r0/build/core/arch/arm/kernel/thread.o:(.nozi.kdata_page): SHT_PROGBITS
| >>> output section .nozi: SHT_NOBITS
|
| ld.lld: error: section type mismatch for .nozi.mmu.l2
| >>> /mnt/b/yoe/master/build/tmp/work/qemuarm64-yoe-linux/optee-os-tadevkit/3.17.0-r0/build/core/arch/arm/mm/core_mmu_lpae.o:(.nozi.mmu.l2): SHT_PROGBITS
| >>> output section .nozi: SHT_NOBITS

These sections should be carrying SHT_NOBITS but so far it was not
possible to do so, this patch tries to use clangs pragma to get this
going and match the functionality with gcc.

[1] https://intel.github.io/llvm-docs/clang/LanguageExtensions.html#specifying-section-names-for-global-objects-pragma-clang-section

Upstream-Status: Pending
Signed-off-by: Khem Raj <raj.khem@gmail.com>
---
 core/arch/arm/kernel/thread.c    | 19 +++++++++++++++--
 core/arch/arm/mm/core_mmu_lpae.c | 35 ++++++++++++++++++++++++++++----
 core/arch/arm/mm/pgt_cache.c     | 12 ++++++++++-
 core/kernel/thread.c             | 13 +++++++++++-
 4 files changed, 71 insertions(+), 8 deletions(-)

--- a/core/arch/arm/kernel/thread.c
+++ b/core/arch/arm/kernel/thread.c
@@ -44,16 +44,31 @@ static size_t thread_user_kcode_size __n
 #if defined(CFG_CORE_UNMAP_CORE_AT_EL0) && \
 	defined(CFG_CORE_WORKAROUND_SPECTRE_BP_SEC) && defined(ARM64)
 long thread_user_kdata_sp_offset __nex_bss;
+#ifdef __clang__
+#ifndef CFG_VIRTUALIZATION
+#pragma clang section bss=".nozi.kdata_page"
+#else
+#pragma clang section bss=".nex_nozi.kdata_page"
+#endif
+#endif
 static uint8_t thread_user_kdata_page[
 	ROUNDUP(sizeof(struct thread_core_local) * CFG_TEE_CORE_NB_CORE,
 		SMALL_PAGE_SIZE)]
 	__aligned(SMALL_PAGE_SIZE)
+#ifndef __clang__
 #ifndef CFG_VIRTUALIZATION
-	__section(".nozi.kdata_page");
+	__section(".nozi.kdata_page")
 #else
-	__section(".nex_nozi.kdata_page");
+	__section(".nex_nozi.kdata_page")
 #endif
 #endif
+    ;
+#endif
+
+/* reset BSS section to default ( .bss ) */
+#ifdef __clang__
+#pragma clang section bss=""
+#endif
 
 #ifdef ARM32
 uint32_t __nostackcheck thread_get_exceptions(void)
--- a/core/arch/arm/mm/core_mmu_lpae.c
+++ b/core/arch/arm/mm/core_mmu_lpae.c
@@ -233,19 +233,46 @@ typedef uint16_t l1_idx_t;
 typedef uint64_t base_xlat_tbls_t[CFG_TEE_CORE_NB_CORE][NUM_BASE_LEVEL_ENTRIES];
 typedef uint64_t xlat_tbl_t[XLAT_TABLE_ENTRIES];
 
+#ifdef __clang__
+#pragma clang section bss=".nozi.mmu.base_table"
+#endif
 static base_xlat_tbls_t base_xlation_table[NUM_BASE_TABLES]
 	__aligned(NUM_BASE_LEVEL_ENTRIES * XLAT_ENTRY_SIZE)
-	__section(".nozi.mmu.base_table");
+#ifndef __clang__
+	__section(".nozi.mmu.base_table")
+#endif
+;
+#ifdef __clang__
+#pragma clang section bss=""
+#endif
 
+#ifdef __clang__
+#pragma clang section bss=".nozi.mmu.l2"
+#endif
 static xlat_tbl_t xlat_tables[MAX_XLAT_TABLES]
-	__aligned(XLAT_TABLE_SIZE) __section(".nozi.mmu.l2");
+	__aligned(XLAT_TABLE_SIZE)
+#ifndef __clang__
+	__section(".nozi.mmu.l2")
+#endif
+;
+#ifdef __clang__
+#pragma clang section bss=""
+#endif
 
 #define XLAT_TABLES_SIZE	(sizeof(xlat_tbl_t) * MAX_XLAT_TABLES)
 
+#ifdef __clang__
+#pragma clang section bss=".nozi.mmu.l2"
+#endif
 /* MMU L2 table for TAs, one for each thread */
 static xlat_tbl_t xlat_tables_ul1[CFG_NUM_THREADS]
-	__aligned(XLAT_TABLE_SIZE) __section(".nozi.mmu.l2");
-
+#ifndef __clang__
+	__aligned(XLAT_TABLE_SIZE) __section(".nozi.mmu.l2")
+#endif
+;
+#ifdef __clang__
+#pragma clang section bss=""
+#endif
 /*
  * TAs page table entry inside a level 1 page table.
  *
--- a/core/arch/arm/mm/pgt_cache.c
+++ b/core/arch/arm/mm/pgt_cache.c
@@ -104,8 +104,18 @@ void pgt_init(void)
 	 * has a large alignment, while .bss has a small alignment. The current
 	 * link script is optimized for small alignment in .bss
 	 */
+#ifdef __clang__
+#pragma clang section bss=".nozi.mmu.l2"
+#endif
 	static uint8_t pgt_tables[PGT_CACHE_SIZE][PGT_SIZE]
-			__aligned(PGT_SIZE) __section(".nozi.pgt_cache");
+			__aligned(PGT_SIZE)
+#ifndef __clang__
+			__section(".nozi.pgt_cache")
+#endif
+			;
+#ifdef __clang__
+#pragma clang section bss=""
+#endif
 	size_t n;
 
 	for (n = 0; n < ARRAY_SIZE(pgt_tables); n++) {
--- a/core/kernel/thread.c
+++ b/core/kernel/thread.c
@@ -37,13 +37,24 @@ struct thread_core_local thread_core_loc
 	name[stack_num][sizeof(name[stack_num]) / sizeof(uint32_t) - 1]
 #endif
 
+#define DO_PRAGMA(x) _Pragma (#x)
+
+#ifdef __clang__
+#define DECLARE_STACK(name, num_stacks, stack_size, linkage) \
+DO_PRAGMA (clang section bss=".nozi_stack." #name) \
+linkage uint32_t name[num_stacks] \
+		[ROUNDUP(stack_size + STACK_CANARY_SIZE + STACK_CHECK_EXTRA, \
+			 STACK_ALIGNMENT) / sizeof(uint32_t)] \
+		__attribute__((aligned(STACK_ALIGNMENT))); \
+DO_PRAGMA(clang section bss="")
+#else
 #define DECLARE_STACK(name, num_stacks, stack_size, linkage) \
 linkage uint32_t name[num_stacks] \
 		[ROUNDUP(stack_size + STACK_CANARY_SIZE + STACK_CHECK_EXTRA, \
 			 STACK_ALIGNMENT) / sizeof(uint32_t)] \
 		__attribute__((section(".nozi_stack." # name), \
 			       aligned(STACK_ALIGNMENT)))
-
+#endif
 #define GET_STACK(stack) ((vaddr_t)(stack) + STACK_SIZE(stack))
 
 DECLARE_STACK(stack_tmp, CFG_TEE_CORE_NB_CORE,
--- a/core/arch/arm/mm/core_mmu_v7.c
+++ b/core/arch/arm/mm/core_mmu_v7.c
@@ -204,16 +204,46 @@ typedef uint32_t l1_xlat_tbl_t[NUM_L1_EN
 typedef uint32_t l2_xlat_tbl_t[NUM_L2_ENTRIES];
 typedef uint32_t ul1_xlat_tbl_t[NUM_UL1_ENTRIES];
 
+#ifdef __clang__
+#pragma clang section bss=".nozi.mmu.l1"
+#endif
 static l1_xlat_tbl_t main_mmu_l1_ttb
-		__aligned(L1_ALIGNMENT) __section(".nozi.mmu.l1");
+		__aligned(L1_ALIGNMENT)
+#ifndef __clang__
+       __section(".nozi.mmu.l1")
+#endif
+;
+#ifdef __clang__
+#pragma clang section bss=""
+#endif
 
 /* L2 MMU tables */
+#ifdef __clang__
+#pragma clang section bss=".nozi.mmu.l2"
+#endif
 static l2_xlat_tbl_t main_mmu_l2_ttb[MAX_XLAT_TABLES]
-		__aligned(L2_ALIGNMENT) __section(".nozi.mmu.l2");
+		__aligned(L2_ALIGNMENT)
+#ifndef __clang__
+       __section(".nozi.mmu.l2")
+#endif
+;
+#ifdef __clang__
+#pragma clang section bss=""
+#endif
 
 /* MMU L1 table for TAs, one for each thread */
+#ifdef __clang__
+#pragma clang section bss=".nozi.mmu.ul1"
+#endif
 static ul1_xlat_tbl_t main_mmu_ul1_ttb[CFG_NUM_THREADS]
-		__aligned(UL1_ALIGNMENT) __section(".nozi.mmu.ul1");
+		__aligned(UL1_ALIGNMENT)
+#ifndef __clang__
+       __section(".nozi.mmu.ul1")
+#endif
+;
+#ifdef __clang__
+#pragma clang section bss=""
+#endif
 
 struct mmu_partition {
 	l1_xlat_tbl_t *l1_table;