standalone_systests: fix tlbp64 to probe HSV39 extended TLB entries

The previous tlbp64 split the 64-bit lookup key into two uint32_t
halves and probed each separately.  Each probe used `"r"(<uint32_t>)`
as the inline-asm constraint, which made the compiler allocate a
single 32-bit register.  The assembler's `tlbp` mnemonic is
operand-width-driven: a single 32-bit register selects the legacy
Y2_tlbp instruction (which only searches legacy TLB entries at
indices 0-511); a 64-bit register pair (Rss) selects Y2_tlbpp
(which searches the full TLB including HSV39 extended entries at
indices 512+).

The hsv39_tlb test writes its TLB entries at indices >=512.  The
old tlbp64 always emitted Y2_tlbp and so never found those entries,
making the test fail with bogus TLB_NOT_FOUND results.

Pass the lookup key as a single uint64_t.  This forces a register
pair and makes the assembler emit Y2_tlbpp.

Signed-off-by: Marco Liebel <marco.liebel@oss.qualcomm.com>

Author: quic-mliebel

standalone_systests/include/hsv39.h

@@ -87,16 +87,9 @@ static inline void remove_hsv39_trans(int index)
 
 static inline uint32_t tlbp64(uint32_t asid, uint64_t VA)
 {
-    uint32_t lookup_hi = ((VA >> 32) & 0xfffff) | ((asid & 0x7f) << 20);
-    uint32_t lookup_lo = (VA >> 12) & 0xfffff;
+    uint64_t lookup = (asid & 0x7f) | (uint64_t)(VA & 0xfffffffffffff000);
     uint32_t ret;
-
-    /* For 64-bit TLB lookup, we need to handle the upper bits appropriately */
-    /* This is a simplified implementation - actual HW may differ */
-    asm volatile ("%0 = tlbp(%1)\n\t" : "=r"(ret) : "r"(lookup_hi));
-    if (ret == TLB_NOT_FOUND) {
-        asm volatile ("%0 = tlbp(%1)\n\t" : "=r"(ret) : "r"(lookup_lo));
-    }
+    asm volatile ("%0 = tlbp(%1)\n\t" : "=r"(ret) : "r"(lookup));
     return ret;
 }