[xen staging] x86/bitops: Account for POPCNT errata on earlier Intel CPUs

[patchbot@xxxxxxx]

commit 38adc2d7879c9a68b21a1dddb09d9c9f34d15ee4
Author:     Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
AuthorDate: Tue Mar 25 18:02:03 2025 +0000
Commit:     Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
CommitDate: Wed Mar 26 11:54:59 2025 +0000

    x86/bitops: Account for POPCNT errata on earlier Intel CPUs
    
    Manually break the false dependency for the benefit of cases such as
    bitmap_weight() which is a reasonable hotpath.
    
    Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
    Acked-by: Jan Beulich <jbeulich@xxxxxxxx>
---
 xen/arch/x86/include/asm/bitops.h | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/xen/arch/x86/include/asm/bitops.h 
b/xen/arch/x86/include/asm/bitops.h
index bb9d756460..87eac7782f 100644
--- a/xen/arch/x86/include/asm/bitops.h
+++ b/xen/arch/x86/include/asm/bitops.h
@@ -488,10 +488,16 @@ static always_inline unsigned int arch_hweightl(unsigned 
long x)
      *
      * This limits the POPCNT instruction to using the same ABI as a function
      * call (input in %rdi, output in %eax) but that's fine.
+     *
+     * On Intel CPUs prior to Cannon Lake, the POPCNT instruction has a false
+     * input dependency on it's destination register (errata HSD146, SKL029
+     * amongst others), impacting loops such as bitmap_weight().  Insert an
+     * XOR to manually break the dependency.
      */
     alternative_io("call arch_generic_hweightl",
+                   "xor %k[res], %k[res]\n\t"
                    "popcnt %[val], %q[res]", X86_FEATURE_POPCNT,
-                   ASM_OUTPUT2([res] "=a" (r) ASM_CALL_CONSTRAINT),
+                   ASM_OUTPUT2([res] "=&a" (r) ASM_CALL_CONSTRAINT),
                    [val] "D" (x));
 
     return r;
--
generated by git-patchbot for /home/xen/git/xen.git#staging