Skip to contentSkip to conversation

linux-user futex() and madvise() fixes

PATCH In review [email protected] 3 participants 9 messages Jan 23

lore
Share & tools
lore
#0 lore raw

cover linux-user futex() and madvise() fixes

From: Helge Deller <[email protected]>

Three small linux-user patches. Please review.

Helge Deller (3):
linux-user: strace: Fix 5th argument of futex syscall
linux-user: val3 parameter of futex() syscall needs endianess swapping
linux-user: Fix MADV_XXX constants on hppa target

 linux-user/strace.c  |  7 ++++++-
 linux-user/syscall.c | 22 +++++++++++++++++++++-
 2 files changed, 27 insertions(+), 2 deletions(-)
Signature · 1 line
2.52.0
#1 lore raw

PATCH 1/3 linux-user: strace: Fix 5th argument of futex syscall

From: Helge Deller <[email protected]>

The 5th argument isn't printed, but instead the 4th argument is shown twice. Fix this, and print the FUTEX_BITSET_MATCH_ANY string constant instead of -1 if the op is FUTEX_WAIT_BITSET or FUTEX_WAKE_BITSET.

---
 linux-user/strace.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
linux-user/strace.c +6 -1
index 18bc6c800c..ca67cfd09d 100644
@@ -4150,7 +4150,12 @@ print_futex(CPUArchState *cpu_env, const struct syscallname *name,
             break;
     }
     print_pointer(arg4, 0);
-    print_raw_param("%d", arg4, 1);
+    if ((op == FUTEX_WAIT_BITSET || (op == FUTEX_WAKE_BITSET)) &&
+        (arg5 == FUTEX_BITSET_MATCH_ANY)) {
+        qemu_log("FUTEX_BITSET_MATCH_ANY");
+    } else {
+        print_raw_param("%#x", arg5, 1);
+    }
     print_syscall_epilogue(name);
 }
 #endif
Signature · 1 line
2.52.0
Reviewed-by on v1 1/3
#2 lore raw

Re: [PATCH 1/3] linux-user: strace: Fix 5th argument of futex syscall

Quoted 20 lines · On Fri, 23 Jan 2026 at 15:16, <[email protected]> wrote:

From: Helge Deller <[email protected]>

The 5th argument isn't printed, but instead the 4th argument is shown twice. Fix this, and print the FUTEX_BITSET_MATCH_ANY string constant instead of -1 if the op is FUTEX_WAIT_BITSET or FUTEX_WAKE_BITSET.

---
 linux-user/strace.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
linux-user/strace.c +0 -0
index 18bc6c800c..ca67cfd09d 100644
@@ -4150,7 +4150,12 @@ print_futex(CPUArchState *cpu_env, const struct syscallname *name,
             break;
     }
     print_pointer(arg4, 0);

linux-user/strace.c print_futex(CPUArchState *cpu_env, const struct syscallname *name, · line 4161

Quoted 10 lines
linux-user/strace.c +6 -1
-    print_raw_param("%d", arg4, 1);
+    if ((op == FUTEX_WAIT_BITSET || (op == FUTEX_WAKE_BITSET)) &&
+        (arg5 == FUTEX_BITSET_MATCH_ANY)) {
+        qemu_log("FUTEX_BITSET_MATCH_ANY");
+    } else {
+        print_raw_param("%#x", arg5, 1);
+    }
     print_syscall_epilogue(name);
 }
 #endif

thanks
-- PMM

#3 lore raw

PATCH 2/3 linux-user: val3 parameter of futex() syscall needs endianess swapping

In reply to [email protected]

From: Helge Deller <[email protected]>

The FUTEX_WAIT_BITSET and FUTEX_WAKE_BITSET operations use a val3 parameter for comparism. Even if this 32-bit parameter is in most cases 0xffffffff, ensure that we do a required endianess swapping if host and guest endianess differ.

---
 linux-user/syscall.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
linux-user/syscall.c +7 -1
index c0a9a86529..7cb9de97e9 100644
@@ -8106,7 +8106,10 @@ static int do_futex(CPUState *cpu, bool time64, target_ulong uaddr,
 #endif
     switch (base_op) {
     case FUTEX_WAIT:
+        val = tswap32(val);
+        break;
     case FUTEX_WAIT_BITSET:
+        val3 = tswap32(val3);
         val = tswap32(val);
         break;
     case FUTEX_WAIT_REQUEUE_PI:
@@ -8116,8 +8119,11 @@ static int do_futex(CPUState *cpu, bool time64, target_ulong uaddr,
     case FUTEX_LOCK_PI:
     case FUTEX_LOCK_PI2:
         break;
-    case FUTEX_WAKE:
     case FUTEX_WAKE_BITSET:
+        val3 = tswap32(val3);
+        timeout = 0;
+        break;
+    case FUTEX_WAKE:
     case FUTEX_TRYLOCK_PI:
     case FUTEX_UNLOCK_PI:
         timeout = 0;
Signature · 1 line
2.52.0
#4 lore raw

Re: [PATCH 2/3] linux-user: val3 parameter of futex() syscall needs endianess swapping

Quoted 18 lines · On Fri, 23 Jan 2026 at 15:11, <[email protected]> wrote:

From: Helge Deller <[email protected]>

The FUTEX_WAIT_BITSET and FUTEX_WAKE_BITSET operations use a val3 parameter for comparism. Even if this 32-bit parameter is in most cases 0xffffffff, ensure that we do a required endianess swapping if host and guest endianess differ.

---
 linux-user/syscall.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
linux-user/syscall.c +0 -0
index c0a9a86529..7cb9de97e9 100644

linux-user/syscall.c static int do_futex(CPUState *cpu, bool time64, target_ulong uaddr, · line 8114

Quoted 9 lines
linux-user/syscall.c +3 -0
@@ -8106,7 +8106,10 @@ static int do_futex(CPUState *cpu, bool time64, target_ulong uaddr,
 #endif
     switch (base_op) {
     case FUTEX_WAIT:
+        val = tswap32(val);
+        break;
     case FUTEX_WAIT_BITSET:
+        val3 = tswap32(val3);
         val = tswap32(val);
         break;

I see why we need to swap the "val" for these operations:
it is the value that the host kernel will be comparing against the in-memory value. It's currently in host order but it will be in memory in guest order, so we must swap it.

But why do we need to swap the bitmask in val3 ? That is in host order now and I would have thought the host kernel also wants it in host order.

thanks
-- PMM

#5 lore raw

Re: [PATCH 2/3] linux-user: val3 parameter of futex() syscall needs endianess swapping

Quoted 35 lines · Peter Maydell
Quoted 18 lines · On Fri, 23 Jan 2026 at 15:11, <[email protected]> wrote:

From: Helge Deller <[email protected]>

The FUTEX_WAIT_BITSET and FUTEX_WAKE_BITSET operations use a val3 parameter for comparism. Even if this 32-bit parameter is in most cases 0xffffffff, ensure that we do a required endianess swapping if host and guest endianess differ.

---
  linux-user/syscall.c | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
linux-user/syscall.c +0 -0
index c0a9a86529..7cb9de97e9 100644

linux-user/syscall.c static int do_futex(CPUState *cpu, bool time64, target_ulong uaddr, · line 8114

Quoted 9 lines
linux-user/syscall.c +3 -0
@@ -8106,7 +8106,10 @@ static int do_futex(CPUState *cpu, bool time64, target_ulong uaddr,
  #endif
      switch (base_op) {
      case FUTEX_WAIT:
+        val = tswap32(val);
+        break;
      case FUTEX_WAIT_BITSET:
+        val3 = tswap32(val3);
          val = tswap32(val);
          break;

I see why we need to swap the "val" for these operations:
it is the value that the host kernel will be comparing against the in-memory value. It's currently in host order but it will be in memory in guest order, so we must swap it.

Quoted 3 lines

But why do we need to swap the bitmask in val3 ? That is in host order now and I would have thought the host kernel also wants it in host order.

Seems it was a brain-I/O error on my side.
I'll drop that patch.

Thanks!
Helge

#6 lore raw

PATCH 3/3 linux-user: Fix MADV_XXX constants on hppa target

In reply to [email protected]

From: Helge Deller <[email protected]>

Older hppa binaries may still use the old MADV_XXX constants for the madivise() syscall. Fix it up in the same manner as it's done in the Linux kernel in the parisc_madvise() function.

---
 linux-user/syscall.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
linux-user/syscall.c +14 -0
index 7cb9de97e9..80f7355ccd 100644
@@ -12929,6 +12929,20 @@ static abi_long do_syscall1(CPUArchState *cpu_env, int num, abi_long arg1,
 
 #ifdef TARGET_NR_madvise
     case TARGET_NR_madvise:
+#ifdef TARGET_HPPA
+        /* Emulate old hppa mapping of MADV_xxx constants. */
+        switch (arg3) {
+        case 65: arg3 = MADV_MERGEABLE;     break;
+        case 66: arg3 = MADV_UNMERGEABLE;   break;
+        case 67: arg3 = MADV_HUGEPAGE;      break;
+        case 68: arg3 = MADV_NOHUGEPAGE;    break;
+        case 69: arg3 = MADV_DONTDUMP;      break;
+        case 70: arg3 = MADV_DODUMP;        break;
+        case 71: arg3 = MADV_WIPEONFORK;    break;
+        case 72: arg3 = MADV_KEEPONFORK;    break;
+        case 73: arg3 = MADV_COLLAPSE;      break;
+        }
+#endif /* TARGET_HPPA */
         return target_madvise(arg1, arg2, arg3);
 #endif
 #ifdef TARGET_NR_fcntl64
Signature · 1 line
2.52.0
Reviewed-by on v1 3/3
#7 lore raw

Re: [PATCH 3/3] linux-user: Fix MADV_XXX constants on hppa target

Quoted 6 lines · On Fri, 23 Jan 2026 at 15:11, <[email protected]> wrote:

From: Helge Deller <[email protected]>

Older hppa binaries may still use the old MADV_XXX constants for the madivise() syscall. Fix it up in the same manner as it's done in the Linux kernel in the parisc_madvise() function.

"Remove this wrapper in year 2025" it says :-)

Quoted 1 line

thanks
-- PMM

#8 lore raw

Re: [PATCH 3/3] linux-user: Fix MADV_XXX constants on hppa target

Quoted 9 lines · Peter Maydell
Quoted 6 lines · On Fri, 23 Jan 2026 at 15:11, <[email protected]> wrote:

From: Helge Deller <[email protected]>

Older hppa binaries may still use the old MADV_XXX constants for the madivise() syscall. Fix it up in the same manner as it's done in the Linux kernel in the parisc_madvise() function.

"Remove this wrapper in year 2025" it says :-)

Yes, I think I was too optimistic when I added that code back then. I found such programs yesterday....

Quoted 3 lines
Quoted 1 line

Thanks!
Helge