[PATCH 0/2] tcg: Fixes for temp_allocate_frame

[PATCH 0/2] tcg: Fixes for temp_allocate_frame

[Richard Henderson]

Stefan, I think this will resolve #367 for you -- please test.


r~


Richard Henderson (2):
  tcg/sparc: Fix temp_allocate_frame vs sparc stack bias
  tcg: Allocate sufficient storage in temp_allocate_frame

 tcg/tcg.c                  | 41 ++++++++++++++++++++++++++++----------
 tcg/sparc/tcg-target.c.inc | 16 +++++++++------
 2 files changed, 40 insertions(+), 17 deletions(-)

-- 
2.25.1

[PATCH 1/2] 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
Cc: mark.cave-ayland@ilande.co.uk
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 2/2] tcg: Allocate sufficient storage in temp_allocate_frame

[Richard Henderson]

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

Cc: qemu-stable@nongnu.org
Fixes: d2fd745fe8b
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/367
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..1bad423bde 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 = TCG_TARGET_STACK_ALIGN;
+        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

Re: [PATCH 0/2] tcg: Fixes for temp_allocate_frame

[Stefan Weil]

Am 19.06.21 um 06:56 schrieb Richard Henderson:

> Stefan, I think this will resolve #367 for you -- please test.


Tested-by: Stefan Weil <sw@weilnetz.de>


Thank you, yes, it seems to resolve that issue.

I had to make two additional changes for building on Windows:

- Downgrade to support glib 2.54 (that's the version provided by Cygwin 
for mingw-w64). Daniel, could we go back from 2.56 to 2.54?

- Fix assertion in oslib-win32.c (I already have sent a patch for that, 
see 
http://patchwork.ozlabs.org/project/qemu-devel/patch/20210611105846.347954-1-sw@weilnetz.de/)

Stefan

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

[Philippe Mathieu-Daudé]

On 6/19/21 6:57 AM, Richard Henderson wrote:
> This function should have been updated for vector types
> when they were introduced.
> 
> Cc: qemu-stable@nongnu.org
> Fixes: d2fd745fe8b
> Resolves: https://gitlab.com/qemu-project/qemu/-/issues/367
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/tcg.c | 32 +++++++++++++++++++++++++++-----
>  1 file changed, 27 insertions(+), 5 deletions(-)

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

Re: [PATCH 0/2] tcg: Fixes for temp_allocate_frame

[Daniel P. Berrangé]

On Sat, Jun 19, 2021 at 10:16:55AM +0200, Stefan Weil wrote:
> Am 19.06.21 um 06:56 schrieb Richard Henderson:
> 
> > Stefan, I think this will resolve #367 for you -- please test.
> 
> 
> Tested-by: Stefan Weil <sw@weilnetz.de>
> 
> 
> Thank you, yes, it seems to resolve that issue.
> 
> I had to make two additional changes for building on Windows:
> 
> - Downgrade to support glib 2.54 (that's the version provided by Cygwin for
> mingw-w64). Daniel, could we go back from 2.56 to 2.54?

2.55.1 introduced support for autoptr with GSList/GList and we have already
started using this functionality, so it isn't a clean downgrade.

I'm pretty surprised cygwin would be on 2.54, as that is almost a 3 year
old release at this point in time. What's their timeframe for actually
pulling in a new release ? 

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|