Re: [PATCH v3 05/10] riscv: Add support for the tagged address ABI

From: Alexandre Ghiti
Date: Wed Aug 14 2024 - 15:29:22 EST



On 14/08/2024 18:14, Samuel Holland wrote:
Hi Alex,

On 2024-08-14 10:10 AM, Alexandre Ghiti wrote:
On 14/08/2024 10:13, Samuel Holland wrote:
When pointer masking is enabled for userspace, the kernel can accept
tagged pointers as arguments to some system calls. Allow this by
untagging the pointers in access_ok() and the uaccess routines. The
uaccess routines must peform untagging in software because U-mode and
S-mode have entirely separate pointer masking configurations. In fact,
hardware may not even implement pointer masking for S-mode.

Since the number of tag bits is variable, untagged_addr_remote() needs
to know what PMLEN to use for the remote mm. Therefore, the pointer
masking mode must be the same for all threads sharing an mm. Enforce
this with a lock flag in the mm context, as x86 does for LAM. The flag
gets reset in init_new_context() during fork(), as the new mm is no
longer multithreaded.

Signed-off-by: Samuel Holland <samuel.holland@xxxxxxxxxx>
---

Changes in v3:
  - Use IS_ENABLED instead of #ifdef when possible
  - Implement mm_untag_mask()
  - Remove pmlen from struct thread_info (now only in mm_context_t)

Changes in v2:
  - Implement untagged_addr_remote()
  - Restrict PMLEN changes once a process is multithreaded

  arch/riscv/include/asm/mmu.h         |  7 +++
  arch/riscv/include/asm/mmu_context.h | 13 +++++
  arch/riscv/include/asm/uaccess.h     | 58 ++++++++++++++++++++--
  arch/riscv/kernel/process.c          | 73 ++++++++++++++++++++++++++--
  4 files changed, 141 insertions(+), 10 deletions(-)

diff --git a/arch/riscv/include/asm/mmu.h b/arch/riscv/include/asm/mmu.h
index c9e03e9da3dc..1cc90465d75b 100644
--- a/arch/riscv/include/asm/mmu.h
+++ b/arch/riscv/include/asm/mmu.h
@@ -25,9 +25,16 @@ typedef struct {
  #ifdef CONFIG_BINFMT_ELF_FDPIC
      unsigned long exec_fdpic_loadmap;
      unsigned long interp_fdpic_loadmap;
+#endif
+    unsigned long flags;
+#ifdef CONFIG_RISCV_ISA_SUPM
+    u8 pmlen;
  #endif
  } mm_context_t;
  +/* Lock the pointer masking mode because this mm is multithreaded */
+#define MM_CONTEXT_LOCK_PMLEN    0
+
  #define cntx2asid(cntx)        ((cntx) & SATP_ASID_MASK)
  #define cntx2version(cntx)    ((cntx) & ~SATP_ASID_MASK)
  diff --git a/arch/riscv/include/asm/mmu_context.h
b/arch/riscv/include/asm/mmu_context.h
index 7030837adc1a..8c4bc49a3a0f 100644
--- a/arch/riscv/include/asm/mmu_context.h
+++ b/arch/riscv/include/asm/mmu_context.h
@@ -20,6 +20,9 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,
  static inline void activate_mm(struct mm_struct *prev,
                     struct mm_struct *next)
  {
+#ifdef CONFIG_RISCV_ISA_SUPM
+    next->context.pmlen = 0;
+#endif
      switch_mm(prev, next, NULL);
  }
  @@ -30,11 +33,21 @@ static inline int init_new_context(struct task_struct *tsk,
  #ifdef CONFIG_MMU
      atomic_long_set(&mm->context.id, 0);
  #endif
+    if (IS_ENABLED(CONFIG_RISCV_ISA_SUPM))
+        clear_bit(MM_CONTEXT_LOCK_PMLEN, &mm->context.flags);
      return 0;
  }
    DECLARE_STATIC_KEY_FALSE(use_asid_allocator);
  +#ifdef CONFIG_RISCV_ISA_SUPM
+#define mm_untag_mask mm_untag_mask
+static inline unsigned long mm_untag_mask(struct mm_struct *mm)
+{
+    return -1UL >> mm->context.pmlen;
+}
+#endif
+
  #include <asm-generic/mmu_context.h>
    #endif /* _ASM_RISCV_MMU_CONTEXT_H */
diff --git a/arch/riscv/include/asm/uaccess.h b/arch/riscv/include/asm/uaccess.h
index 72ec1d9bd3f3..6416559232a2 100644
--- a/arch/riscv/include/asm/uaccess.h
+++ b/arch/riscv/include/asm/uaccess.h
@@ -9,8 +9,56 @@
  #define _ASM_RISCV_UACCESS_H
    #include <asm/asm-extable.h>
+#include <asm/cpufeature.h>
  #include <asm/pgtable.h>        /* for TASK_SIZE */
  +#ifdef CONFIG_RISCV_ISA_SUPM
+static inline unsigned long __untagged_addr(unsigned long addr)
+{
+    if (riscv_has_extension_unlikely(RISCV_ISA_EXT_SUPM)) {
+        u8 pmlen = current->mm->context.pmlen;
+
+        /* Virtual addresses are sign-extended; physical addresses are
zero-extended. */
+        if (IS_ENABLED(CONFIG_MMU))
+            return (long)(addr << pmlen) >> pmlen;
+        else
+            return (addr << pmlen) >> pmlen;
+    }
+
+    return addr;
+}
+
+#define untagged_addr(addr) ({                        \
+    unsigned long __addr = (__force unsigned long)(addr);        \
+    (__force __typeof__(addr))__untagged_addr(__addr);        \
+})
+
+static inline unsigned long __untagged_addr_remote(struct mm_struct *mm,
unsigned long addr)
+{
+    if (riscv_has_extension_unlikely(RISCV_ISA_EXT_SUPM)) {
+        u8 pmlen = mm->context.pmlen;
+
+        /* Virtual addresses are sign-extended; physical addresses are
zero-extended. */
+        if (IS_ENABLED(CONFIG_MMU))
+            return (long)(addr << pmlen) >> pmlen;
+        else
+            return (addr << pmlen) >> pmlen;
+    }
+
+    return addr;
+}

I should have mentioned that in v2: now that you removed the thread_info pmlen
field, __untagged_addr_remote() and __untagged_addr() are almost the same, can
you merge them?
I can merge them, but this places the load of current->mm outside the static
branch. If you think that is okay, then I'll merge them for v4.


I think it's ok, it's not a big overhead :)

Thanks,

Alex


Regards,
Samuel

+
+#define untagged_addr_remote(mm, addr) ({                \
+    unsigned long __addr = (__force unsigned long)(addr);        \
+    mmap_assert_locked(mm);                        \
+    (__force __typeof__(addr))__untagged_addr_remote(mm, __addr);    \
+})
+
+#define access_ok(addr, size) likely(__access_ok(untagged_addr(addr), size))
+#else
+#define untagged_addr(addr) (addr)
+#endif
+
  /*
   * User space memory access functions
   */
@@ -130,7 +178,7 @@ do {                                \
   */
  #define __get_user(x, ptr)                    \
  ({                                \
-    const __typeof__(*(ptr)) __user *__gu_ptr = (ptr);    \
+    const __typeof__(*(ptr)) __user *__gu_ptr = untagged_addr(ptr); \
      long __gu_err = 0;                    \
                                  \
      __chk_user_ptr(__gu_ptr);                \
@@ -246,7 +294,7 @@ do {                                \
   */
  #define __put_user(x, ptr)                    \
  ({                                \
-    __typeof__(*(ptr)) __user *__gu_ptr = (ptr);        \
+    __typeof__(*(ptr)) __user *__gu_ptr = untagged_addr(ptr); \
      __typeof__(*__gu_ptr) __val = (x);            \
      long __pu_err = 0;                    \
                                  \
@@ -293,13 +341,13 @@ unsigned long __must_check __asm_copy_from_user(void *to,
  static inline unsigned long
  raw_copy_from_user(void *to, const void __user *from, unsigned long n)
  {
-    return __asm_copy_from_user(to, from, n);
+    return __asm_copy_from_user(to, untagged_addr(from), n);
  }
    static inline unsigned long
  raw_copy_to_user(void __user *to, const void *from, unsigned long n)
  {
-    return __asm_copy_to_user(to, from, n);
+    return __asm_copy_to_user(untagged_addr(to), from, n);
  }
    extern long strncpy_from_user(char *dest, const char __user *src, long
count);
@@ -314,7 +362,7 @@ unsigned long __must_check clear_user(void __user *to,
unsigned long n)
  {
      might_fault();
      return access_ok(to, n) ?
-        __clear_user(to, n) : n;
+        __clear_user(untagged_addr(to), n) : n;
  }
    #define __get_kernel_nofault(dst, src, type, err_label)            \
diff --git a/arch/riscv/kernel/process.c b/arch/riscv/kernel/process.c
index 1280a7c4a412..f4d8e5c3bb84 100644
--- a/arch/riscv/kernel/process.c
+++ b/arch/riscv/kernel/process.c
@@ -203,6 +203,10 @@ int copy_thread(struct task_struct *p, const struct
kernel_clone_args *args)
      unsigned long tls = args->tls;
      struct pt_regs *childregs = task_pt_regs(p);
  +    /* Ensure all threads in this mm have the same pointer masking mode. */
+    if (IS_ENABLED(CONFIG_RISCV_ISA_SUPM) && p->mm && (clone_flags & CLONE_VM))
+        set_bit(MM_CONTEXT_LOCK_PMLEN, &p->mm->context.flags);
+
      memset(&p->thread.s, 0, sizeof(p->thread.s));
        /* p->thread holds context to be restored by __switch_to() */
@@ -248,10 +252,16 @@ enum {
  static bool have_user_pmlen_7;
  static bool have_user_pmlen_16;
  +/*
+ * Control the relaxed ABI allowing tagged user addresses into the kernel.
+ */
+static unsigned int tagged_addr_disabled;
+
  long set_tagged_addr_ctrl(struct task_struct *task, unsigned long arg)
  {
-    unsigned long valid_mask = PR_PMLEN_MASK;
+    unsigned long valid_mask = PR_PMLEN_MASK | PR_TAGGED_ADDR_ENABLE;
      struct thread_info *ti = task_thread_info(task);
+    struct mm_struct *mm = task->mm;
      unsigned long pmm;
      u8 pmlen;
  @@ -266,16 +276,41 @@ long set_tagged_addr_ctrl(struct task_struct *task,
unsigned long arg)
       * in case choosing a larger PMLEN has a performance impact.
       */
      pmlen = FIELD_GET(PR_PMLEN_MASK, arg);
-    if (pmlen == PMLEN_0)
+    if (pmlen == PMLEN_0) {
          pmm = ENVCFG_PMM_PMLEN_0;
-    else if (pmlen <= PMLEN_7 && have_user_pmlen_7)
+    } else if (pmlen <= PMLEN_7 && have_user_pmlen_7) {
+        pmlen = PMLEN_7;
          pmm = ENVCFG_PMM_PMLEN_7;
-    else if (pmlen <= PMLEN_16 && have_user_pmlen_16)
+    } else if (pmlen <= PMLEN_16 && have_user_pmlen_16) {
+        pmlen = PMLEN_16;
          pmm = ENVCFG_PMM_PMLEN_16;
-    else
+    } else {
          return -EINVAL;
+    }
+
+    /*
+     * Do not allow the enabling of the tagged address ABI if globally
+     * disabled via sysctl abi.tagged_addr_disabled, if pointer masking
+     * is disabled for userspace.
+     */
+    if (arg & PR_TAGGED_ADDR_ENABLE && (tagged_addr_disabled || !pmlen))
+        return -EINVAL;
+
+    if (!(arg & PR_TAGGED_ADDR_ENABLE))
+        pmlen = PMLEN_0;
+
+    if (mmap_write_lock_killable(mm))
+        return -EINTR;
+
+    if (test_bit(MM_CONTEXT_LOCK_PMLEN, &mm->context.flags) &&
mm->context.pmlen != pmlen) {
+        mmap_write_unlock(mm);
+        return -EBUSY;
+    }
        envcfg_update_bits(task, ENVCFG_PMM, pmm);
+    mm->context.pmlen = pmlen;
+
+    mmap_write_unlock(mm);
        return 0;
  }
@@ -288,6 +323,10 @@ long get_tagged_addr_ctrl(struct task_struct *task)
      if (is_compat_thread(ti))
          return -EINVAL;
  +    /*
+     * The mm context's pmlen is set only when the tagged address ABI is
+     * enabled, so the effective PMLEN must be extracted from envcfg.PMM.
+     */
      switch (task->thread.envcfg & ENVCFG_PMM) {
      case ENVCFG_PMM_PMLEN_7:
          ret = FIELD_PREP(PR_PMLEN_MASK, PMLEN_7);
@@ -297,6 +336,9 @@ long get_tagged_addr_ctrl(struct task_struct *task)
          break;
      }
  +    if (task->mm->context.pmlen)
+        ret |= PR_TAGGED_ADDR_ENABLE;
+
      return ret;
  }
  @@ -306,6 +348,24 @@ static bool try_to_set_pmm(unsigned long value)
      return (csr_read_clear(CSR_ENVCFG, ENVCFG_PMM) & ENVCFG_PMM) == value;
  }
  +/*
+ * Global sysctl to disable the tagged user addresses support. This control
+ * only prevents the tagged address ABI enabling via prctl() and does not
+ * disable it for tasks that already opted in to the relaxed ABI.
+ */
+
+static struct ctl_table tagged_addr_sysctl_table[] = {
+    {
+        .procname    = "tagged_addr_disabled",
+        .mode        = 0644,
+        .data        = &tagged_addr_disabled,
+        .maxlen        = sizeof(int),
+        .proc_handler    = proc_dointvec_minmax,
+        .extra1        = SYSCTL_ZERO,
+        .extra2        = SYSCTL_ONE,
+    },
+};
+
  static int __init tagged_addr_init(void)
  {
      if (!riscv_has_extension_unlikely(RISCV_ISA_EXT_SUPM))
@@ -319,6 +379,9 @@ static int __init tagged_addr_init(void)
      have_user_pmlen_7 = try_to_set_pmm(ENVCFG_PMM_PMLEN_7);
      have_user_pmlen_16 = try_to_set_pmm(ENVCFG_PMM_PMLEN_16);
  +    if (!register_sysctl("abi", tagged_addr_sysctl_table))
+        return -EINVAL;
+
      return 0;
  }
  core_initcall(tagged_addr_init);

_______________________________________________
linux-riscv mailing list
linux-riscv@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/linux-riscv