Merge branch 'x86/cleanups' into x86/mm
authorIngo Molnar <mingo@kernel.org>
Fri, 12 Feb 2021 12:39:32 +0000 (13:39 +0100)
committerIngo Molnar <mingo@kernel.org>
Fri, 12 Feb 2021 12:40:02 +0000 (13:40 +0100)
Merge recent cleanups to the x86 MM code to resolve a conflict.

Conflicts:
arch/x86/mm/fault.c

Signed-off-by: Ingo Molnar <mingo@kernel.org>
15 files changed:
arch/x86/Kconfig
arch/x86/include/asm/irq.h
arch/x86/include/asm/page_64_types.h
arch/x86/include/asm/pgtable_types.h
arch/x86/include/asm/vm86.h
arch/x86/include/uapi/asm/vm86.h
arch/x86/kernel/cpu/mtrr/cleanup.c
arch/x86/kernel/cpu/mtrr/generic.c
arch/x86/kernel/cpu/mtrr/mtrr.c
arch/x86/kernel/pci-iommu_table.c
arch/x86/kernel/ptrace.c
arch/x86/kernel/sys_x86_64.c
arch/x86/kernel/vm86_32.c
arch/x86/mm/fault.c
arch/x86/mm/mmio-mod.c

index 7b6dd10..865c1e7 100644 (file)
@@ -889,7 +889,7 @@ config HPET_TIMER
 
 config HPET_EMULATE_RTC
        def_bool y
-       depends on HPET_TIMER && (RTC=y || RTC=m || RTC_DRV_CMOS=m || RTC_DRV_CMOS=y)
+       depends on HPET_TIMER && (RTC_DRV_CMOS=m || RTC_DRV_CMOS=y)
 
 config APB_TIMER
        def_bool y if X86_INTEL_MID
index 528c8a7..76d3896 100644 (file)
@@ -40,8 +40,6 @@ extern void native_init_IRQ(void);
 
 extern void __handle_irq(struct irq_desc *desc, struct pt_regs *regs);
 
-extern __visible void do_IRQ(struct pt_regs *regs, unsigned long vector);
-
 extern void init_ISA_irqs(void);
 
 extern void __init init_IRQ(void);
index 645bd1d..64297ea 100644 (file)
@@ -66,7 +66,7 @@
  * On Intel CPUs, if a SYSCALL instruction is at the highest canonical
  * address, then that syscall will enter the kernel with a
  * non-canonical return address, and SYSRET will explode dangerously.
- * We avoid this particular problem by preventing anything executable
+ * We avoid this particular problem by preventing anything
  * from being mapped at the maximum canonical address.
  *
  * On AMD CPUs in the Ryzen family, there's a nasty bug in which the
index 394757e..f24d7ef 100644 (file)
@@ -177,8 +177,6 @@ enum page_cache_mode {
 #define __pgprot(x)            ((pgprot_t) { (x) } )
 #define __pg(x)                        __pgprot(x)
 
-#define _PAGE_PAT_LARGE                (_AT(pteval_t, 1) << _PAGE_BIT_PAT_LARGE)
-
 #define PAGE_NONE           __pg(   0|   0|   0|___A|   0|   0|   0|___G)
 #define PAGE_SHARED         __pg(__PP|__RW|_USR|___A|__NX|   0|   0|   0)
 #define PAGE_SHARED_EXEC     __pg(__PP|__RW|_USR|___A|   0|   0|   0|   0)
index 26efbec..9e8ac50 100644 (file)
@@ -36,7 +36,6 @@ struct vm86 {
        unsigned long saved_sp0;
 
        unsigned long flags;
-       unsigned long screen_bitmap;
        unsigned long cpu_type;
        struct revectored_struct int_revectored;
        struct revectored_struct int21_revectored;
index d2ee4e3..18909b8 100644 (file)
@@ -97,7 +97,7 @@ struct revectored_struct {
 struct vm86_struct {
        struct vm86_regs regs;
        unsigned long flags;
-       unsigned long screen_bitmap;
+       unsigned long screen_bitmap;            /* unused, preserved by vm86() */
        unsigned long cpu_type;
        struct revectored_struct int_revectored;
        struct revectored_struct int21_revectored;
@@ -106,7 +106,7 @@ struct vm86_struct {
 /*
  * flags masks
  */
-#define VM86_SCREEN_BITMAP     0x0001
+#define VM86_SCREEN_BITMAP     0x0001        /* no longer supported */
 
 struct vm86plus_info_struct {
        unsigned long force_return_for_pic:1;
index 5bd0117..9231640 100644 (file)
@@ -537,9 +537,9 @@ static void __init print_out_mtrr_range_state(void)
                if (!size_base)
                        continue;
 
-               size_base = to_size_factor(size_base, &size_factor),
+               size_base = to_size_factor(size_base, &size_factor);
                start_base = range_state[i].base_pfn << (PAGE_SHIFT - 10);
-               start_base = to_size_factor(start_base, &start_factor),
+               start_base = to_size_factor(start_base, &start_factor);
                type = range_state[i].type;
 
                pr_debug("reg %d, base: %ld%cB, range: %ld%cB, type %s\n",
index 23ad8e9..2cf4465 100644 (file)
@@ -3,7 +3,6 @@
  * This only handles 32bit MTRR on 32bit hosts. This is strictly wrong
  * because MTRRs can span up to 40 bits (36bits on most modern x86)
  */
-#define DEBUG
 
 #include <linux/export.h>
 #include <linux/init.h>
index 61eb26e..28c8a23 100644 (file)
@@ -31,8 +31,6 @@
     System Programming Guide; Section 9.11. (1997 edition - PPro).
 */
 
-#define DEBUG
-
 #include <linux/types.h> /* FIXME: kvm_para.h needs this */
 
 #include <linux/stop_machine.h>
index 2e9006c..42e92ec 100644 (file)
@@ -4,9 +4,6 @@
 #include <linux/string.h>
 #include <linux/kallsyms.h>
 
-
-#define DEBUG 1
-
 static struct iommu_table_entry * __init
 find_dependents_of(struct iommu_table_entry *start,
                   struct iommu_table_entry *finish,
index bedca01..87a4143 100644 (file)
@@ -704,6 +704,9 @@ void ptrace_disable(struct task_struct *child)
 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 #endif
+#ifdef CONFIG_X86_64
+static const struct user_regset_view user_x86_64_view; /* Initialized below. */
+#endif
 
 long arch_ptrace(struct task_struct *child, long request,
                 unsigned long addr, unsigned long data)
@@ -711,6 +714,14 @@ long arch_ptrace(struct task_struct *child, long request,
        int ret;
        unsigned long __user *datap = (unsigned long __user *)data;
 
+#ifdef CONFIG_X86_64
+       /* This is native 64-bit ptrace() */
+       const struct user_regset_view *regset_view = &user_x86_64_view;
+#else
+       /* This is native 32-bit ptrace() */
+       const struct user_regset_view *regset_view = &user_x86_32_view;
+#endif
+
        switch (request) {
        /* read the word at location addr in the USER area. */
        case PTRACE_PEEKUSR: {
@@ -749,28 +760,28 @@ long arch_ptrace(struct task_struct *child, long request,
 
        case PTRACE_GETREGS:    /* Get all gp regs from the child. */
                return copy_regset_to_user(child,
-                                          task_user_regset_view(current),
+                                          regset_view,
                                           REGSET_GENERAL,
                                           0, sizeof(struct user_regs_struct),
                                           datap);
 
        case PTRACE_SETREGS:    /* Set all gp regs in the child. */
                return copy_regset_from_user(child,
-                                            task_user_regset_view(current),
+                                            regset_view,
                                             REGSET_GENERAL,
                                             0, sizeof(struct user_regs_struct),
                                             datap);
 
        case PTRACE_GETFPREGS:  /* Get the child FPU state. */
                return copy_regset_to_user(child,
-                                          task_user_regset_view(current),
+                                          regset_view,
                                           REGSET_FP,
                                           0, sizeof(struct user_i387_struct),
                                           datap);
 
        case PTRACE_SETFPREGS:  /* Set the child FPU state. */
                return copy_regset_from_user(child,
-                                            task_user_regset_view(current),
+                                            regset_view,
                                             REGSET_FP,
                                             0, sizeof(struct user_i387_struct),
                                             datap);
@@ -1152,28 +1163,28 @@ static long x32_arch_ptrace(struct task_struct *child,
 
        case PTRACE_GETREGS:    /* Get all gp regs from the child. */
                return copy_regset_to_user(child,
-                                          task_user_regset_view(current),
+                                          &user_x86_64_view,
                                           REGSET_GENERAL,
                                           0, sizeof(struct user_regs_struct),
                                           datap);
 
        case PTRACE_SETREGS:    /* Set all gp regs in the child. */
                return copy_regset_from_user(child,
-                                            task_user_regset_view(current),
+                                            &user_x86_64_view,
                                             REGSET_GENERAL,
                                             0, sizeof(struct user_regs_struct),
                                             datap);
 
        case PTRACE_GETFPREGS:  /* Get the child FPU state. */
                return copy_regset_to_user(child,
-                                          task_user_regset_view(current),
+                                          &user_x86_64_view,
                                           REGSET_FP,
                                           0, sizeof(struct user_i387_struct),
                                           datap);
 
        case PTRACE_SETFPREGS:  /* Set the child FPU state. */
                return copy_regset_from_user(child,
-                                            task_user_regset_view(current),
+                                            &user_x86_64_view,
                                             REGSET_FP,
                                             0, sizeof(struct user_i387_struct),
                                             datap);
@@ -1309,6 +1320,25 @@ void __init update_regset_xstate_info(unsigned int size, u64 xstate_mask)
        xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
 }
 
+/*
+ * This is used by the core dump code to decide which regset to dump.  The
+ * core dump code writes out the resulting .e_machine and the corresponding
+ * regsets.  This is suboptimal if the task is messing around with its CS.L
+ * field, but at worst the core dump will end up missing some information.
+ *
+ * Unfortunately, it is also used by the broken PTRACE_GETREGSET and
+ * PTRACE_SETREGSET APIs.  These APIs look at the .regsets field but have
+ * no way to make sure that the e_machine they use matches the caller's
+ * expectations.  The result is that the data format returned by
+ * PTRACE_GETREGSET depends on the returned CS field (and even the offset
+ * of the returned CS field depends on its value!) and the data format
+ * accepted by PTRACE_SETREGSET is determined by the old CS value.  The
+ * upshot is that it is basically impossible to use these APIs correctly.
+ *
+ * The best way to fix it in the long run would probably be to add new
+ * improved ptrace() APIs to read and write registers reliably, possibly by
+ * allowing userspace to select the ELF e_machine variant that they expect.
+ */
 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
 {
 #ifdef CONFIG_IA32_EMULATION
index 504fa54..660b788 100644 (file)
@@ -90,14 +90,10 @@ SYSCALL_DEFINE6(mmap, unsigned long, addr, unsigned long, len,
                unsigned long, prot, unsigned long, flags,
                unsigned long, fd, unsigned long, off)
 {
-       long error;
-       error = -EINVAL;
        if (off & ~PAGE_MASK)
-               goto out;
+               return -EINVAL;
 
-       error = ksys_mmap_pgoff(addr, len, prot, flags, fd, off >> PAGE_SHIFT);
-out:
-       return error;
+       return ksys_mmap_pgoff(addr, len, prot, flags, fd, off >> PAGE_SHIFT);
 }
 
 static void find_start_end(unsigned long addr, unsigned long flags,
index 764573d..e5a7a10 100644 (file)
@@ -134,7 +134,11 @@ void save_v86_state(struct kernel_vm86_regs *regs, int retval)
        unsafe_put_user(regs->ds, &user->regs.ds, Efault_end);
        unsafe_put_user(regs->fs, &user->regs.fs, Efault_end);
        unsafe_put_user(regs->gs, &user->regs.gs, Efault_end);
-       unsafe_put_user(vm86->screen_bitmap, &user->screen_bitmap, Efault_end);
+
+       /*
+        * Don't write screen_bitmap in case some user had a value there
+        * and expected it to remain unchanged.
+        */
 
        user_access_end();
 
@@ -160,49 +164,6 @@ Efault:
        do_exit(SIGSEGV);
 }
 
-static void mark_screen_rdonly(struct mm_struct *mm)
-{
-       struct vm_area_struct *vma;
-       spinlock_t *ptl;
-       pgd_t *pgd;
-       p4d_t *p4d;
-       pud_t *pud;
-       pmd_t *pmd;
-       pte_t *pte;
-       int i;
-
-       mmap_write_lock(mm);
-       pgd = pgd_offset(mm, 0xA0000);
-       if (pgd_none_or_clear_bad(pgd))
-               goto out;
-       p4d = p4d_offset(pgd, 0xA0000);
-       if (p4d_none_or_clear_bad(p4d))
-               goto out;
-       pud = pud_offset(p4d, 0xA0000);
-       if (pud_none_or_clear_bad(pud))
-               goto out;
-       pmd = pmd_offset(pud, 0xA0000);
-
-       if (pmd_trans_huge(*pmd)) {
-               vma = find_vma(mm, 0xA0000);
-               split_huge_pmd(vma, pmd, 0xA0000);
-       }
-       if (pmd_none_or_clear_bad(pmd))
-               goto out;
-       pte = pte_offset_map_lock(mm, pmd, 0xA0000, &ptl);
-       for (i = 0; i < 32; i++) {
-               if (pte_present(*pte))
-                       set_pte(pte, pte_wrprotect(*pte));
-               pte++;
-       }
-       pte_unmap_unlock(pte, ptl);
-out:
-       mmap_write_unlock(mm);
-       flush_tlb_mm_range(mm, 0xA0000, 0xA0000 + 32*PAGE_SIZE, PAGE_SHIFT, false);
-}
-
-
-
 static int do_vm86_irq_handling(int subfunction, int irqnumber);
 static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus);
 
@@ -282,6 +243,15 @@ static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
                        offsetof(struct vm86_struct, int_revectored)))
                return -EFAULT;
 
+
+       /* VM86_SCREEN_BITMAP had numerous bugs and appears to have no users. */
+       if (v.flags & VM86_SCREEN_BITMAP) {
+               char comm[TASK_COMM_LEN];
+
+               pr_info_once("vm86: '%s' uses VM86_SCREEN_BITMAP, which is no longer supported\n", get_task_comm(comm, current));
+               return -EINVAL;
+       }
+
        memset(&vm86regs, 0, sizeof(vm86regs));
 
        vm86regs.pt.bx = v.regs.ebx;
@@ -302,7 +272,6 @@ static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
        vm86regs.gs = v.regs.gs;
 
        vm86->flags = v.flags;
-       vm86->screen_bitmap = v.screen_bitmap;
        vm86->cpu_type = v.cpu_type;
 
        if (copy_from_user(&vm86->int_revectored,
@@ -370,9 +339,6 @@ static long do_sys_vm86(struct vm86plus_struct __user *user_vm86, bool plus)
        update_task_stack(tsk);
        preempt_enable();
 
-       if (vm86->flags & VM86_SCREEN_BITMAP)
-               mark_screen_rdonly(tsk->mm);
-
        memcpy((struct kernel_vm86_regs *)regs, &vm86regs, sizeof(vm86regs));
        return regs->ax;
 }
index 7b3a125..5251973 100644 (file)
@@ -282,25 +282,6 @@ void arch_sync_kernel_mappings(unsigned long start, unsigned long end)
        }
 }
 
-/*
- * Did it hit the DOS screen memory VA from vm86 mode?
- */
-static inline void
-check_v8086_mode(struct pt_regs *regs, unsigned long address,
-                struct task_struct *tsk)
-{
-#ifdef CONFIG_VM86
-       unsigned long bit;
-
-       if (!v8086_mode(regs) || !tsk->thread.vm86)
-               return;
-
-       bit = (address - 0xA0000) >> PAGE_SHIFT;
-       if (bit < 32)
-               tsk->thread.vm86->screen_bitmap |= 1 << bit;
-#endif
-}
-
 static bool low_pfn(unsigned long pfn)
 {
        return pfn < max_low_pfn;
@@ -355,15 +336,6 @@ KERN_ERR
 "******* Disabling USB legacy in the BIOS may also help.\n";
 #endif
 
-/*
- * No vm86 mode in 64-bit mode:
- */
-static inline void
-check_v8086_mode(struct pt_regs *regs, unsigned long address,
-                struct task_struct *tsk)
-{
-}
-
 static int bad_address(void *p)
 {
        unsigned long dummy;
@@ -1465,8 +1437,6 @@ good_area:
                else
                        BUG();
        }
-
-       check_v8086_mode(regs, address, tsk);
 }
 NOKPROBE_SYMBOL(do_user_addr_fault);
 
index bd7aff5..cd768da 100644 (file)
@@ -10,8 +10,6 @@
 
 #define pr_fmt(fmt) "mmiotrace: " fmt
 
-#define DEBUG 1
-
 #include <linux/moduleparam.h>
 #include <linux/debugfs.h>
 #include <linux/slab.h>