[PATCH] target/riscv: Force to set mstatus_hs.[SD|FS] bits in mark_fs_dirty()

frank.chang@sifive.com posted 1 patch 2 years, 6 months ago
Test checkpatch passed
Patches applied successfully (tree, apply log)
git fetch https://github.com/patchew-project/qemu tags/patchew/20210914013732.881754-1-frank.chang@sifive.com
Maintainers: Bin Meng <bin.meng@windriver.com>, Palmer Dabbelt <palmer@dabbelt.com>, Alistair Francis <alistair.francis@wdc.com>
target/riscv/translate.c | 19 ++++++++++---------
1 file changed, 10 insertions(+), 9 deletions(-)
[PATCH] target/riscv: Force to set mstatus_hs.[SD|FS] bits in mark_fs_dirty()
Posted by frank.chang@sifive.com 2 years, 6 months ago
From: Frank Chang <frank.chang@sifive.com>

When V=1, both vsstauts.FS and HS-level sstatus.FS are in effect.
Modifying the floating-point state when V=1 causes both fields to
be set to 3 (Dirty).

However, it's possible that HS-level sstatus.FS is Clean and VS-level
vsstatus.FS is Dirty at the time mark_fs_dirty() is called when V=1.
We can't early return for this case because we still need to set
sstatus.FS to Dirty according to spec.

Signed-off-by: Frank Chang <frank.chang@sifive.com>
Reviewed-by: Vincent Chen <vincent.chen@sifive.com>
Tested-by: Vincent Chen <vincent.chen@sifive.com>
---
 target/riscv/translate.c | 19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

diff --git a/target/riscv/translate.c b/target/riscv/translate.c
index e356fc6c46c..0096b098738 100644
--- a/target/riscv/translate.c
+++ b/target/riscv/translate.c
@@ -280,26 +280,27 @@ static void gen_jal(DisasContext *ctx, int rd, target_ulong imm)
 static void mark_fs_dirty(DisasContext *ctx)
 {
     TCGv tmp;
-    target_ulong sd;
+    target_ulong sd = is_32bit(ctx) ? MSTATUS32_SD : MSTATUS64_SD;
+
+    if (ctx->virt_enabled) {
+        tmp = tcg_temp_new();
+        tcg_gen_ld_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus_hs));
+        tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | sd);
+        tcg_gen_st_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus_hs));
+        tcg_temp_free(tmp);
+    }
 
     if (ctx->mstatus_fs == MSTATUS_FS) {
         return;
     }
+
     /* Remember the state change for the rest of the TB.  */
     ctx->mstatus_fs = MSTATUS_FS;
 
     tmp = tcg_temp_new();
-    sd = is_32bit(ctx) ? MSTATUS32_SD : MSTATUS64_SD;
-
     tcg_gen_ld_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus));
     tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | sd);
     tcg_gen_st_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus));
-
-    if (ctx->virt_enabled) {
-        tcg_gen_ld_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus_hs));
-        tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | sd);
-        tcg_gen_st_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus_hs));
-    }
     tcg_temp_free(tmp);
 }
 #else
-- 
2.25.1


Re: [PATCH] target/riscv: Force to set mstatus_hs.[SD|FS] bits in mark_fs_dirty()
Posted by Richard Henderson 2 years, 6 months ago
On 9/13/21 6:37 PM, frank.chang@sifive.com wrote:
> From: Frank Chang <frank.chang@sifive.com>
> 
> When V=1, both vsstauts.FS and HS-level sstatus.FS are in effect.
> Modifying the floating-point state when V=1 causes both fields to
> be set to 3 (Dirty).
> 
> However, it's possible that HS-level sstatus.FS is Clean and VS-level
> vsstatus.FS is Dirty at the time mark_fs_dirty() is called when V=1.
> We can't early return for this case because we still need to set
> sstatus.FS to Dirty according to spec.
> 
> Signed-off-by: Frank Chang <frank.chang@sifive.com>
> Reviewed-by: Vincent Chen <vincent.chen@sifive.com>
> Tested-by: Vincent Chen <vincent.chen@sifive.com>
> ---
>   target/riscv/translate.c | 19 ++++++++++---------
>   1 file changed, 10 insertions(+), 9 deletions(-)
> 
> diff --git a/target/riscv/translate.c b/target/riscv/translate.c
> index e356fc6c46c..0096b098738 100644
> --- a/target/riscv/translate.c
> +++ b/target/riscv/translate.c
> @@ -280,26 +280,27 @@ static void gen_jal(DisasContext *ctx, int rd, target_ulong imm)
>   static void mark_fs_dirty(DisasContext *ctx)
>   {
>       TCGv tmp;
> -    target_ulong sd;
> +    target_ulong sd = is_32bit(ctx) ? MSTATUS32_SD : MSTATUS64_SD;
> +
> +    if (ctx->virt_enabled) {
> +        tmp = tcg_temp_new();
> +        tcg_gen_ld_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus_hs));
> +        tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | sd);
> +        tcg_gen_st_tl(tmp, cpu_env, offsetof(CPURISCVState, mstatus_hs));
> +        tcg_temp_free(tmp);
> +    }
>   
>       if (ctx->mstatus_fs == MSTATUS_FS) {
>           return;
>       }

You should introduce a ctx->mstatus_hs field to track the code that you moved.  Otherwise 
you'll be setting this dirty bit for every fp insn.


r~

Re: [PATCH] target/riscv: Force to set mstatus_hs.[SD|FS] bits in mark_fs_dirty()
Posted by Frank Chang 2 years, 6 months ago
On Tue, Sep 14, 2021 at 10:10 AM Richard Henderson <
richard.henderson@linaro.org> wrote:

> On 9/13/21 6:37 PM, frank.chang@sifive.com wrote:
> > From: Frank Chang <frank.chang@sifive.com>
> >
> > When V=1, both vsstauts.FS and HS-level sstatus.FS are in effect.
> > Modifying the floating-point state when V=1 causes both fields to
> > be set to 3 (Dirty).
> >
> > However, it's possible that HS-level sstatus.FS is Clean and VS-level
> > vsstatus.FS is Dirty at the time mark_fs_dirty() is called when V=1.
> > We can't early return for this case because we still need to set
> > sstatus.FS to Dirty according to spec.
> >
> > Signed-off-by: Frank Chang <frank.chang@sifive.com>
> > Reviewed-by: Vincent Chen <vincent.chen@sifive.com>
> > Tested-by: Vincent Chen <vincent.chen@sifive.com>
> > ---
> >   target/riscv/translate.c | 19 ++++++++++---------
> >   1 file changed, 10 insertions(+), 9 deletions(-)
> >
> > diff --git a/target/riscv/translate.c b/target/riscv/translate.c
> > index e356fc6c46c..0096b098738 100644
> > --- a/target/riscv/translate.c
> > +++ b/target/riscv/translate.c
> > @@ -280,26 +280,27 @@ static void gen_jal(DisasContext *ctx, int rd,
> target_ulong imm)
> >   static void mark_fs_dirty(DisasContext *ctx)
> >   {
> >       TCGv tmp;
> > -    target_ulong sd;
> > +    target_ulong sd = is_32bit(ctx) ? MSTATUS32_SD : MSTATUS64_SD;
> > +
> > +    if (ctx->virt_enabled) {
> > +        tmp = tcg_temp_new();
> > +        tcg_gen_ld_tl(tmp, cpu_env, offsetof(CPURISCVState,
> mstatus_hs));
> > +        tcg_gen_ori_tl(tmp, tmp, MSTATUS_FS | sd);
> > +        tcg_gen_st_tl(tmp, cpu_env, offsetof(CPURISCVState,
> mstatus_hs));
> > +        tcg_temp_free(tmp);
> > +    }
> >
> >       if (ctx->mstatus_fs == MSTATUS_FS) {
> >           return;
> >       }
>
> You should introduce a ctx->mstatus_hs field to track the code that you
> moved.  Otherwise
> you'll be setting this dirty bit for every fp insn.
>
>
Thanks, Richard, I was struggling with whether to introduce a new field
in DisasContext.
I will update my patch.

Regards,
Frank Chang


>
> r~
>