Skip to contentSkip to conversation

s390x/tcg: fix and optimize SPX (SET PREFIX)

PATCH Applied David Hildenbrand 4 participants 6 messages Aug 5, 2021 – Aug 10, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 s390x/tcg: fix and optimize SPX (SET PREFIX)

We not only invalidate the translation of the range 0x0-0x2000, we also invalidate the translation of the new prefix range and the translation of the old prefix range -- because real2abs would return different results for all of these ranges when changing the prefix location.

This fixes the kvm-unit-tests "edat" test that just hangs before this patch because we end up clearing the new prefix area instead of the old prefix area.

While at it, let's not do anything in case the prefix doesn't change.

Cc Richard Henderson <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Thomas Huth <[email protected]>
Cc Claudio Imbrenda <[email protected]>
---
 target/s390x/tcg/misc_helper.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
target/s390x/tcg/misc_helper.c +14 -1
index 33e6999e15..aab9c47747 100644
@@ -151,13 +151,26 @@ void HELPER(diag)(CPUS390XState *env, uint32_t r1, uint32_t r3, uint32_t num)
 /* Set Prefix */
 void HELPER(spx)(CPUS390XState *env, uint64_t a1)
 {
+    const uint32_t prefix = a1 & 0x7fffe000;
+    const uint32_t old_prefix = env->psa;
     CPUState *cs = env_cpu(env);
-    uint32_t prefix = a1 & 0x7fffe000;
+
+    if (prefix == old_prefix) {
+        return;
+    }
 
     env->psa = prefix;
     HELPER_LOG("prefix: %#x\n", prefix);
     tlb_flush_page(cs, 0);
     tlb_flush_page(cs, TARGET_PAGE_SIZE);
+    if (prefix != 0) {
+        tlb_flush_page(cs, prefix);
+        tlb_flush_page(cs, prefix + TARGET_PAGE_SIZE);
+    }
+    if (old_prefix != 0) {
+        tlb_flush_page(cs, old_prefix);
+        tlb_flush_page(cs, old_prefix + TARGET_PAGE_SIZE);
+    }
 }
 
 static void update_ckc_timer(CPUS390XState *env)
Signature · 1 line
2.31.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 5 lines · On Thu, 5 Aug 2021 14:59:38 +0200 David Hildenbrand <[email protected]> wrote:

We not only invalidate the translation of the range 0x0-0x2000, we also invalidate the translation of the new prefix range and the translation of the old prefix range -- because real2abs would return different results for all of these ranges when changing the prefix location.

looks like a good idea

Quoted 5 lines

This fixes the kvm-unit-tests "edat" test that just hangs before this patch because we end up clearing the new prefix area instead of the old prefix area.

While at it, let's not do anything in case the prefix doesn't change.

also looks like a good idea

Quoted 43 lines
Cc Richard Henderson <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Thomas Huth <[email protected]>
Cc Claudio Imbrenda <[email protected]>
---
 target/s390x/tcg/misc_helper.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)
(quoted diff) +0 -0

b/target/s390x/tcg/misc_helper.c index 33e6999e15..aab9c47747 100644

target/s390x/tcg/misc_helper.c +0 -0
@@ -151,13 +151,26 @@ void HELPER(diag)(CPUS390XState *env, uint32_t

r1, uint32_t r3, uint32_t num) /* Set Prefix */
void HELPER(spx)(CPUS390XState *env, uint64_t a1) {

target/s390x/tcg/misc_helper.c +14 -1
+    const uint32_t prefix = a1 & 0x7fffe000;
+    const uint32_t old_prefix = env->psa;
     CPUState *cs = env_cpu(env);
-    uint32_t prefix = a1 & 0x7fffe000;
+
+    if (prefix == old_prefix) {
+        return;
+    }
 
     env->psa = prefix;
     HELPER_LOG("prefix: %#x\n", prefix);
     tlb_flush_page(cs, 0);
     tlb_flush_page(cs, TARGET_PAGE_SIZE);
+    if (prefix != 0) {
+        tlb_flush_page(cs, prefix);
+        tlb_flush_page(cs, prefix + TARGET_PAGE_SIZE);
+    }
+    if (old_prefix != 0) {
+        tlb_flush_page(cs, old_prefix);
+        tlb_flush_page(cs, old_prefix + TARGET_PAGE_SIZE);
+    }
 }
 
 static void update_ckc_timer(CPUS390XState *env)
Reviewed-by on v1 1/1
#2 lore raw
In reply to David Hildenbrand
Quoted 42 lines · David Hildenbrand

We not only invalidate the translation of the range 0x0-0x2000, we also invalidate the translation of the new prefix range and the translation of the old prefix range -- because real2abs would return different results for all of these ranges when changing the prefix location.

This fixes the kvm-unit-tests "edat" test that just hangs before this patch because we end up clearing the new prefix area instead of the old prefix area.

While at it, let's not do anything in case the prefix doesn't change.

Cc Richard Henderson <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Thomas Huth <[email protected]>
Cc Claudio Imbrenda <[email protected]>
---
  target/s390x/tcg/misc_helper.c | 15 ++++++++++++++-
  1 file changed, 14 insertions(+), 1 deletion(-)
target/s390x/tcg/misc_helper.c +6 -1
index 33e6999e15..aab9c47747 100644
@@ -151,13 +151,26 @@ void HELPER(diag)(CPUS390XState *env, uint32_t r1, uint32_t r3, uint32_t num)
  /* Set Prefix */
  void HELPER(spx)(CPUS390XState *env, uint64_t a1)
  {
+    const uint32_t prefix = a1 & 0x7fffe000;
+    const uint32_t old_prefix = env->psa;
      CPUState *cs = env_cpu(env);
-    uint32_t prefix = a1 & 0x7fffe000;
+
+    if (prefix == old_prefix) {
+        return;
+    }
 
      env->psa = prefix;
      HELPER_LOG("prefix: %#x\n", prefix);
      tlb_flush_page(cs, 0);

target/s390x/tcg/misc_helper.c void HELPER(diag)(CPUS390XState *env, uint32_t r1, uint32_t r3, uint32_t num) · line 174

Quoted 10 lines
target/s390x/tcg/misc_helper.c +8 -0
      tlb_flush_page(cs, TARGET_PAGE_SIZE);
+    if (prefix != 0) {
+        tlb_flush_page(cs, prefix);
+        tlb_flush_page(cs, prefix + TARGET_PAGE_SIZE);
+    }
+    if (old_prefix != 0) {
+        tlb_flush_page(cs, old_prefix);
+        tlb_flush_page(cs, old_prefix + TARGET_PAGE_SIZE);
+    }
  }

Sounds reasonable.

#3 lore raw
In reply to David Hildenbrand
Quoted 18 lines · David Hildenbrand

We not only invalidate the translation of the range 0x0-0x2000, we also invalidate the translation of the new prefix range and the translation of the old prefix range -- because real2abs would return different results for all of these ranges when changing the prefix location.

This fixes the kvm-unit-tests "edat" test that just hangs before this patch because we end up clearing the new prefix area instead of the old prefix area.

While at it, let's not do anything in case the prefix doesn't change.

Cc Richard Henderson <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Thomas Huth <[email protected]>
Cc Claudio Imbrenda <[email protected]>
Quoted 3 lines
---
 target/s390x/tcg/misc_helper.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

Is this still -rc material? Especially as we are late in the -rc cycle already.

#4 lore raw
Quoted 24 lines · Cornelia Huck
Quoted 18 lines · David Hildenbrand

We not only invalidate the translation of the range 0x0-0x2000, we also invalidate the translation of the new prefix range and the translation of the old prefix range -- because real2abs would return different results for all of these ranges when changing the prefix location.

This fixes the kvm-unit-tests "edat" test that just hangs before this patch because we end up clearing the new prefix area instead of the old prefix area.

While at it, let's not do anything in case the prefix doesn't change.

Cc Richard Henderson <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Thomas Huth <[email protected]>
Cc Claudio Imbrenda <[email protected]>
Quoted 3 lines
---
  target/s390x/tcg/misc_helper.c | 15 ++++++++++++++-
  1 file changed, 14 insertions(+), 1 deletion(-)
Quoted 3 lines

Is this still -rc material? Especially as we are late in the -rc cycle already.

It's been broken forever, this can wait I guess.

Signature · 3 lines
Thanks,

David / dhildenb
#5 lore raw
In reply to David Hildenbrand
Quoted 18 lines · David Hildenbrand

We not only invalidate the translation of the range 0x0-0x2000, we also invalidate the translation of the new prefix range and the translation of the old prefix range -- because real2abs would return different results for all of these ranges when changing the prefix location.

This fixes the kvm-unit-tests "edat" test that just hangs before this patch because we end up clearing the new prefix area instead of the old prefix area.

While at it, let's not do anything in case the prefix doesn't change.

Cc Richard Henderson <[email protected]>
Cc David Hildenbrand <[email protected]>
Cc Cornelia Huck <[email protected]>
Cc Thomas Huth <[email protected]>
Cc Claudio Imbrenda <[email protected]>
Quoted 3 lines
---
 target/s390x/tcg/misc_helper.c | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

Thanks, applied.