fix(vkernel): 补 ISR/调度 call 的 rsi/rdi clobber、rep stosb 内存 clobber，修 mm 别名笔误

N8 paging.vp：AllocPage 的 rep stosb 写 4096B 补 CLOBBER_MEMORY，避免优化器重排/消除其前后的内存读写 N9 coroutine.vp / sched.vp：_coro_entry/_thread_entry 的 call rax 补 CLOBBER_RSI/CLOBBER_RDI（RSP 由 sub rsp,8 / add rsp,8 配对抵消，无需声明） N15 process.vp：mm_mm.malloc/free → mm.malloc/free（该文件只有 import mm.mm as mm） N16 elf.vp：读 RAX 的 c.Asm 补 CLOBBER_RAX（[DIAG] 探针） N20 elf.vp / idt.vp（12 处）：外部 ELF 入口 / ISR handler 的 call 补 CLOBBER_RSI/CLOBBER_RDI（被调方破坏全部 caller-saved 寄存器）
2026-09-11 17:40:21 +08:00
1 parent c1dd55d606
commit 6a54373c26
@@ -133,7 +133,11 @@ def load_elf(path: t.CConst | str, do_map: int = 1) -> t.CPtr:
    # [DIAG] read_len 实际值 + call 后 RAX 快照：区分"call 返回错"（rl=0 rax=0）
    # vs "store 后 RAX 被覆盖"（rl=0 rax=64）
    _rax_v2: uint64 = 0
-    c.Asm(f"""mov {c.AsmOut(_rax_v2, t.ASM_DESCR.OUTPUT_REG)}, rax""", op=[t.ASM_DESCR.CLOBBER_MEMORY])
+    # 读 RAX 未声明输入：至少须声明 CLOBBER_RAX，让编译器知道本块不保留
+    # RAX（此前只声明 ~{memory}，编译器可假设 RAX 存活 → 读到的值不保证是
+    # 紧邻 call 的返回值）。N16。（本块为 [DIAG] 探针，问题定位后建议删除。）
+    c.Asm(f"""mov {c.AsmOut(_rax_v2, t.ASM_DESCR.OUTPUT_REG)}, rax""",
+          op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX])
    _dbrl: t.CArray[t.CChar, 96]
    viperlib.snprintf(c.Addr(_dbrl), 96, "[elf] rl=%lu rax=0x%lx\n", read_len, _rax_v2)
    serial.puts(c.Addr(_dbrl))
@@ -302,9 +306,12 @@ def run_elf(path: t.CConst | str) -> int:
        serial.puts("[elf] load failed\n")
        return -1
    serial.puts("[elf] jumping to entry...\n")
+    # 被调方是任意 ELF 入口（外部代码）：`call` 破坏全部 caller-saved 寄存器，
+    # 除 rax/rcx/rdx/r8~r11 外必须声明 rsi/rdi（N20）。
    c.Asm(f"""call {c.AsmInp(entry, t.ASM_DESCR.REG_ANY)}""",
          op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
              t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+              t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
              t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
              t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    serial.puts("[elf] returned from entry\n")
@@ -325,6 +325,7 @@ def isr33_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(1)
@@ -336,6 +337,7 @@ def isr34_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler2, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(2)
@@ -347,6 +349,7 @@ def isr35_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler3, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(3)
@@ -358,6 +361,7 @@ def isr36_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler4, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(4)
@@ -369,6 +373,7 @@ def isr37_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler5, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(5)
@@ -380,6 +385,7 @@ def isr38_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler6, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(6)
@@ -393,6 +399,7 @@ def isr39_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler7, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(7)
@@ -404,6 +411,7 @@ def isr40_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler8, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(8)
@@ -415,6 +423,7 @@ def isr41_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler9, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(9)
@@ -426,6 +435,7 @@ def isr42_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler10, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(10)
@@ -437,6 +447,7 @@ def isr43_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler11, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(11)
@@ -447,6 +458,7 @@ def isr44_handler() -> t.CExport | t.CVoid:
        c.Asm(f"""call {c.AsmInp(handler, t.ASM_DESCR.REG_ANY)}""",
              op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
                  t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+                  t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
                  t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
                  t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    pic.eoi(12)
@@ -65,11 +65,14 @@ def AllocPage() -> t.CPtr:
    if pt_pool_used >= uint64(PT_POOL_PAGES): return None
    page: t.CPtr = t.CVoid(pt_pool_start + pt_pool_used * uint64(4096), t.CPtr)
    pt_pool_used = pt_pool_used + uint64(1)
+    # rep stosb 向 [rdi] 写 4096 字节：必须声明 CLOBBER_MEMORY，否则优化器
+    # 无法得知本块改写了内存，可能把其前后的内存读写重排/消除（N8）。
    c.Asm(f"""mov al, 0
            mov ecx, 4096
            mov rdi, {c.AsmInp(page, t.ASM_DESCR.REG_ANY)}
            rep stosb""",
-        op = [t.ASM_DESCR.CLOBBER_AL, t.ASM_DESCR.CLOBBER_ECX, t.ASM_DESCR.CLOBBER_RDI])
+        op = [t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_AL,
+              t.ASM_DESCR.CLOBBER_ECX, t.ASM_DESCR.CLOBBER_RDI])
    AllocatedPages = AllocatedPages + uint64(1)
    return page

@@ -197,8 +197,12 @@ def _coro_entry():
            mov dword ptr [rax + rcx + 40], 3
            mov dword ptr [rax + rcx + 56], eax
            add rsp, 8""",
+          # `call rax` 破坏全部 caller-saved 寄存器（x64 SysV）：除 RAX/RCX/RDX/
+          # R8-R11 外，RSI 与 RDI 也必须声明，否则编译器可能假设其值存活（N9）。
+          # RSP 由 `sub rsp,8` / `add rsp,8` 配对抵消，净变化为 0，无需声明。
          op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
              t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+              t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
              t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
              t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    Coroutine._yield()
@@ -190,7 +190,7 @@ class ProcessManager:
        for j in range(PROC_MAX_DIRS):
            p.dir_used[j] = 0
            p.dir_table[j] = t.CVoid(0, t.CPtr)
-        ks: t.CPtr = mm_mm.malloc(KERN_STACK_SIZE)
+        ks: t.CPtr = mm.malloc(KERN_STACK_SIZE)
        # [DIAG] 定位跨模块 malloc 返回值（mm 成功但调用方收到 None）
        _dbg_ks: t.CArray[t.CChar, 64]
        viperlib.snprintf(c.Addr(_dbg_ks), 64, "[proc] ks=0x%lx\n", uint64(ks))
@@ -201,14 +201,14 @@ class ProcessManager:
            return None
        ks_size: uint64 = KERN_STACK_SIZE
        p.kern_stack = uint64(ks) + ks_size
-        us: t.CPtr = mm_mm.malloc(USER_STACK_SIZE)
+        us: t.CPtr = mm.malloc(USER_STACK_SIZE)
        # [DIAG] 定位 user_stack 收到 None（mm 内部打印成功但此处收 0x0）
        _dbg_us: t.CArray[t.CChar, 64]
        viperlib.snprintf(c.Addr(_dbg_us), 64, "[proc] us=0x%lx\n", uint64(us))
        serial.puts(c.Addr(_dbg_us))
        if us is None:
            serial.puts("[process] user stack alloc failed\n")
-            mm_mm.free(ks)
+            mm.free(ks)
            m.count = m.count - 1
            return None
        us_size: uint64 = USER_STACK_SIZE
@@ -278,8 +278,12 @@ def _thread_entry():
            imul rcx, rcx, 56
            mov dword ptr [rax + rcx + 40], 3
            add rsp, 8""",
+          # `call rax` 破坏全部 caller-saved 寄存器（x64 SysV）：除 RAX/RCX/RDX/
+          # R8-R11 外，RSI 与 RDI 也必须声明，否则编译器可能假设其值存活（N9）。
+          # RSP 由 `sub rsp,8` / `add rsp,8` 配对抵消，净变化为 0，无需声明。
          op=[t.ASM_DESCR.CLOBBER_MEMORY, t.ASM_DESCR.CLOBBER_RAX,
              t.ASM_DESCR.CLOBBER_RCX, t.ASM_DESCR.CLOBBER_RDX,
+              t.ASM_DESCR.CLOBBER_RSI, t.ASM_DESCR.CLOBBER_RDI,
              t.ASM_DESCR.CLOBBER_R8, t.ASM_DESCR.CLOBBER_R9,
              t.ASM_DESCR.CLOBBER_R10, t.ASM_DESCR.CLOBBER_R11])
    Scheduler._yield()