[PATCH v2 0/3] tcg: Fixes for temp_allocate_frame

[PATCH v2 0/3] tcg: Fixes for temp_allocate_frame

[Richard Henderson]

Yay! Is works for purpose.  Couple of tweaks:

  * Use fixed 16 as alignment for TCG_TYPE_V32.  Just an itsy bit
    of future-proofing against any host that has a stack frame
    aligned larger than 32.

  * We now have the ability to restart translation when exhausting
    resources.  Do that when exhausting the stack frame.


r~


Richard Henderson (3):
  tcg/sparc: Fix temp_allocate_frame vs sparc stack bias
  tcg: Allocate sufficient storage in temp_allocate_frame
  tcg: Restart when exhausting the stack frame

 tcg/tcg.c                  | 45 ++++++++++++++++++++++++++++----------
 tcg/sparc/tcg-target.c.inc | 16 +++++++++-----
 2 files changed, 44 insertions(+), 17 deletions(-)

-- 
2.25.1

[PATCH v2 1/3] tcg/sparc: Fix temp_allocate_frame vs sparc stack bias

[Richard Henderson]

We should not be aligning the offset in temp_allocate_frame,
because the odd offset produces an aligned address in the end.
Instead, pass the logical offset into tcg_set_frame and add
the stack bias last.

Cc: qemu-stable@nongnu.org
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg.c                  |  9 +++------
 tcg/sparc/tcg-target.c.inc | 16 ++++++++++------
 2 files changed, 13 insertions(+), 12 deletions(-)

diff --git a/tcg/tcg.c b/tcg/tcg.c
index dd584f3bba..52e858523c 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -3015,17 +3015,14 @@ static void check_regs(TCGContext *s)
 
 static void temp_allocate_frame(TCGContext *s, TCGTemp *ts)
 {
-#if !(defined(__sparc__) && TCG_TARGET_REG_BITS == 64)
-    /* Sparc64 stack is accessed with offset of 2047 */
-    s->current_frame_offset = (s->current_frame_offset +
-                               (tcg_target_long)sizeof(tcg_target_long) - 1) &
-        ~(sizeof(tcg_target_long) - 1);
-#endif
     if (s->current_frame_offset + (tcg_target_long)sizeof(tcg_target_long) >
         s->frame_end) {
         tcg_abort();
     }
     ts->mem_offset = s->current_frame_offset;
+#if defined(__sparc__)
+    ts->mem_offset += TCG_TARGET_STACK_BIAS;
+#endif
     ts->mem_base = s->frame_temp;
     ts->mem_allocated = 1;
     s->current_frame_offset += sizeof(tcg_target_long);
diff --git a/tcg/sparc/tcg-target.c.inc b/tcg/sparc/tcg-target.c.inc
index ce39ac2d86..a6ec94a094 100644
--- a/tcg/sparc/tcg-target.c.inc
+++ b/tcg/sparc/tcg-target.c.inc
@@ -984,14 +984,18 @@ static void tcg_target_qemu_prologue(TCGContext *s)
 {
     int tmp_buf_size, frame_size;
 
-    /* The TCG temp buffer is at the top of the frame, immediately
-       below the frame pointer.  */
+    /*
+     * The TCG temp buffer is at the top of the frame, immediately
+     * below the frame pointer.  Use the logical (aligned) offset here;
+     * the stack bias is applied in temp_allocate_frame().
+     */
     tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long);
-    tcg_set_frame(s, TCG_REG_I6, TCG_TARGET_STACK_BIAS - tmp_buf_size,
-                  tmp_buf_size);
+    tcg_set_frame(s, TCG_REG_I6, -tmp_buf_size, tmp_buf_size);
 
-    /* TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
-       otherwise the minimal frame usable by callees.  */
+    /*
+     * TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
+     * otherwise the minimal frame usable by callees.
+     */
     frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS;
     frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size;
     frame_size += TCG_TARGET_STACK_ALIGN - 1;
-- 
2.25.1

[PATCH v2 2/3] tcg: Allocate sufficient storage in temp_allocate_frame

[Richard Henderson]

This function should have been updated for vector types
when they were introduced.

Fixes: d2fd745fe8b
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/367
Cc: qemu-stable@nongnu.org
Tested-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg.c | 32 +++++++++++++++++++++++++++-----
 1 file changed, 27 insertions(+), 5 deletions(-)

diff --git a/tcg/tcg.c b/tcg/tcg.c
index 52e858523c..47cc66f159 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -3015,17 +3015,39 @@ static void check_regs(TCGContext *s)
 
 static void temp_allocate_frame(TCGContext *s, TCGTemp *ts)
 {
-    if (s->current_frame_offset + (tcg_target_long)sizeof(tcg_target_long) >
-        s->frame_end) {
-        tcg_abort();
+    size_t size, align;
+    intptr_t off;
+
+    switch (ts->type) {
+    case TCG_TYPE_I32:
+        size = align = 4;
+        break;
+    case TCG_TYPE_I64:
+    case TCG_TYPE_V64:
+        size = align = 8;
+        break;
+    case TCG_TYPE_V128:
+        size = align = 16;
+        break;
+    case TCG_TYPE_V256:
+        /* Note that we do not require aligned storage for V256. */
+        size = 32, align = 16;
+        break;
+    default:
+        g_assert_not_reached();
     }
-    ts->mem_offset = s->current_frame_offset;
+
+    assert(align <= TCG_TARGET_STACK_ALIGN);
+    off = ROUND_UP(s->current_frame_offset, align);
+    assert(off + size <= s->frame_end);
+    s->current_frame_offset = off + size;
+
+    ts->mem_offset = off;
 #if defined(__sparc__)
     ts->mem_offset += TCG_TARGET_STACK_BIAS;
 #endif
     ts->mem_base = s->frame_temp;
     ts->mem_allocated = 1;
-    s->current_frame_offset += sizeof(tcg_target_long);
 }
 
 static void temp_load(TCGContext *, TCGTemp *, TCGRegSet, TCGRegSet, TCGRegSet);
-- 
2.25.1

[PATCH v2 3/3] tcg: Restart when exhausting the stack frame

[Richard Henderson]

Assume that we'll have fewer temps allocated after
restarting with a fewer number of instructions.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tcg.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/tcg/tcg.c b/tcg/tcg.c
index 47cc66f159..fec922fa9e 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -3039,7 +3039,11 @@ static void temp_allocate_frame(TCGContext *s, TCGTemp *ts)
 
     assert(align <= TCG_TARGET_STACK_ALIGN);
     off = ROUND_UP(s->current_frame_offset, align);
-    assert(off + size <= s->frame_end);
+
+    /* If we've exhausted the stack frame, restart with a smaller TB. */
+    if (off + size > s->frame_end) {
+        tcg_raise_tb_overflow(s);
+    }
     s->current_frame_offset = off + size;
 
     ts->mem_offset = off;
-- 
2.25.1

Re: [PATCH v2 3/3] tcg: Restart when exhausting the stack frame

[Philippe Mathieu-Daudé]

On 6/19/21 3:42 PM, Richard Henderson wrote:
> Assume that we'll have fewer temps allocated after
> restarting with a fewer number of instructions.
> 
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/tcg.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

Re: [PATCH v2 1/3] tcg/sparc: Fix temp_allocate_frame vs sparc stack bias

[Philippe Mathieu-Daudé]

On 6/19/21 3:42 PM, Richard Henderson wrote:
> We should not be aligning the offset in temp_allocate_frame,
> because the odd offset produces an aligned address in the end.
> Instead, pass the logical offset into tcg_set_frame and add
> the stack bias last.
> 
> Cc: qemu-stable@nongnu.org
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/tcg.c                  |  9 +++------
>  tcg/sparc/tcg-target.c.inc | 16 ++++++++++------
>  2 files changed, 13 insertions(+), 12 deletions(-)

LGTM but I'd rather have another review...

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>

> diff --git a/tcg/tcg.c b/tcg/tcg.c
> index dd584f3bba..52e858523c 100644
> --- a/tcg/tcg.c
> +++ b/tcg/tcg.c
> @@ -3015,17 +3015,14 @@ static void check_regs(TCGContext *s)
>  
>  static void temp_allocate_frame(TCGContext *s, TCGTemp *ts)
>  {
> -#if !(defined(__sparc__) && TCG_TARGET_REG_BITS == 64)
> -    /* Sparc64 stack is accessed with offset of 2047 */
> -    s->current_frame_offset = (s->current_frame_offset +
> -                               (tcg_target_long)sizeof(tcg_target_long) - 1) &
> -        ~(sizeof(tcg_target_long) - 1);
> -#endif
>      if (s->current_frame_offset + (tcg_target_long)sizeof(tcg_target_long) >
>          s->frame_end) {
>          tcg_abort();
>      }
>      ts->mem_offset = s->current_frame_offset;
> +#if defined(__sparc__)
> +    ts->mem_offset += TCG_TARGET_STACK_BIAS;
> +#endif
>      ts->mem_base = s->frame_temp;
>      ts->mem_allocated = 1;
>      s->current_frame_offset += sizeof(tcg_target_long);
> diff --git a/tcg/sparc/tcg-target.c.inc b/tcg/sparc/tcg-target.c.inc
> index ce39ac2d86..a6ec94a094 100644
> --- a/tcg/sparc/tcg-target.c.inc
> +++ b/tcg/sparc/tcg-target.c.inc
> @@ -984,14 +984,18 @@ static void tcg_target_qemu_prologue(TCGContext *s)
>  {
>      int tmp_buf_size, frame_size;
>  
> -    /* The TCG temp buffer is at the top of the frame, immediately
> -       below the frame pointer.  */
> +    /*
> +     * The TCG temp buffer is at the top of the frame, immediately
> +     * below the frame pointer.  Use the logical (aligned) offset here;
> +     * the stack bias is applied in temp_allocate_frame().
> +     */
>      tmp_buf_size = CPU_TEMP_BUF_NLONGS * (int)sizeof(long);
> -    tcg_set_frame(s, TCG_REG_I6, TCG_TARGET_STACK_BIAS - tmp_buf_size,
> -                  tmp_buf_size);
> +    tcg_set_frame(s, TCG_REG_I6, -tmp_buf_size, tmp_buf_size);
>  
> -    /* TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
> -       otherwise the minimal frame usable by callees.  */
> +    /*
> +     * TCG_TARGET_CALL_STACK_OFFSET includes the stack bias, but is
> +     * otherwise the minimal frame usable by callees.
> +     */
>      frame_size = TCG_TARGET_CALL_STACK_OFFSET - TCG_TARGET_STACK_BIAS;
>      frame_size += TCG_STATIC_CALL_ARGS_SIZE + tmp_buf_size;
>      frame_size += TCG_TARGET_STACK_ALIGN - 1;
>