[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [RFC PATCH 01/13] target/riscv: Add UXL to tb flags

From: LIU Zhiwei
Subject: Re: [RFC PATCH 01/13] target/riscv: Add UXL to tb flags
Date: Fri, 6 Aug 2021 10:49:21 +0800
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101 Thunderbird/78.11.0

On 2021/8/6 上午3:01, Richard Henderson wrote:
On 8/4/21 4:53 PM, LIU Zhiwei wrote:
For 32-bit applications run on 64-bit cpu, it may share some code
with other 64-bit applictions. Thus we should distinguish the translated
cache of the share code with a tb flag.

Signed-off-by: LIU Zhiwei <zhiwei_liu@c-sky.com>
  target/riscv/cpu.h       | 15 +++++++++++++++
  target/riscv/translate.c |  3 +++
  2 files changed, 18 insertions(+)

diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
index bf1c899c00..2b3ba21a78 100644
--- a/target/riscv/cpu.h
+++ b/target/riscv/cpu.h
@@ -394,9 +394,20 @@ FIELD(TB_FLAGS, SEW, 5, 3)
  /* Is a Hypervisor instruction load/store allowed? */

Are you intending to reserve space for RV128 here?
Otherwise this could be a single bit.

Also, you probably don't want to name it "UXL", since it should indicate the current operating XLEN, taking MXL, SXL and UXL into account.

Hi Richard,

I don't have the ambitious to define a XLEN32 at least for this patch set. I think it is much more difficult.

The only purpose of this patch set is that we can run 32-bit application on  a 64 bit Linux from qemu-system-riscv64.

So  I didn't change the default behavior of every instruction except when

  1. Current CPU is 64 bit CPU, i.s. TARGET_LONG_BITS is 64.
  2. Current privileged is U-mode.
  3. UXL is 1.

I know that Alistair has done a lot to support 32-bit CPU on qemu-system-riscv64. But We are doing different things,
and it maybe a little confusing.

I still do not find a good to combine them. In my opinion, some code in this patch set can be reused for SXL32.
If you have any advice, please let me know.

Best Regards,

Perhaps just name the field XLEN32, and make it a single bit?

+static inline bool riscv_cpu_is_uxl32(CPURISCVState *env)
+    return (get_field(env->mstatus, MSTATUS64_UXL) == 1) &&
+           !riscv_cpu_is_32bit(env) &&
+           (env->priv == PRV_U);
+    return false;

Again, naming could be better?
It seems trivial to handle all of the fields here.  Perhaps

static inline bool riscv_cpu_is_xlen32(env)
#if defined(TARGET_RISCV32)
    return true;
#elif defined(CONFIG_USER_ONLY)
    return false;
    /* When emulating a 32-bit-only cpu, use RV32. */
    if (riscv_cpu_is_32bit(env)) {
        return true;
     * If MXL has been reduced to RV32, MSTATUSH doesn't have UXL/SXL,
     * therefore, XLEN cannot be widened back to RV64 for lower privs.
    if (get_field(env->misa, MISA64_MXL) == 1) {
        return true;
    switch (env->priv) {
    case PRV_M:
        return false;
    case PRV_U:
        return get_field(env->mstatus, MSTATUS64_UXL) == 1;
    default: /* PRV_S & PRV_H */
        return get_field(env->mstatus, MSTATUS64_SXL) == 1;

@@ -451,6 +462,10 @@ static inline void cpu_get_tb_cpu_state(CPURISCVState *env, target_ulong *pc,
              flags = FIELD_DP32(flags, TB_FLAGS, HLSX, 1);
+    if (riscv_cpu_is_uxl32(env)) {
+        flags = FIELD_DP32(flags, TB_FLAGS, UXL,
+                           get_field(env->mstatus, MSTATUS64_UXL));

  flags = FIELD_DP32(flags, TB_FLAGS, XLEN32,


reply via email to

[Prev in Thread] Current Thread [Next in Thread]