target/arm: Allow cpu to configure GM blocksize

Previously we hard-coded the blocksize with GMID_EL1_BS.
But the value we choose for -cpu max does not match the
value that cortex-a710 uses.

Mirror the way we handle dcz_blocksize.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20230811214031.171020-3-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm/cpu.h

@@ -1075,6 +1075,8 @@ struct ArchCPU {
 
     /* DCZ blocksize, in log_2(words), ie low 4 bits of DCZID_EL0 */
     uint8_t dcz_blocksize;
+    /* GM blocksize, in log_2(words), ie low 4 bits of GMID_EL0 */
+    uint8_t gm_blocksize;
 
     uint64_t rvbar_prop; /* Property/input signals.  */
 

target/arm/helper.c

@@ -7748,10 +7748,6 @@ static const ARMCPRegInfo mte_reginfo[] = {
       .opc0 = 3, .opc1 = 0, .crn = 1, .crm = 0, .opc2 = 6,
       .access = PL1_RW, .accessfn = access_mte,
       .fieldoffset = offsetof(CPUARMState, cp15.gcr_el1) },
-    { .name = "GMID_EL1", .state = ARM_CP_STATE_AA64,
-      .opc0 = 3, .opc1 = 1, .crn = 0, .crm = 0, .opc2 = 4,
-      .access = PL1_R, .accessfn = access_aa64_tid5,
-      .type = ARM_CP_CONST, .resetvalue = GMID_EL1_BS },
     { .name = "TCO", .state = ARM_CP_STATE_AA64,
       .opc0 = 3, .opc1 = 3, .crn = 4, .crm = 2, .opc2 = 7,
       .type = ARM_CP_NO_RAW,
@@ -9342,6 +9338,13 @@ void register_cp_regs_for_features(ARMCPU *cpu)
      * then define only a RAZ/WI version of PSTATE.TCO.
      */
     if (cpu_isar_feature(aa64_mte, cpu)) {
+        ARMCPRegInfo gmid_reginfo = {
+            .name = "GMID_EL1", .state = ARM_CP_STATE_AA64,
+            .opc0 = 3, .opc1 = 1, .crn = 0, .crm = 0, .opc2 = 4,
+            .access = PL1_R, .accessfn = access_aa64_tid5,
+            .type = ARM_CP_CONST, .resetvalue = cpu->gm_blocksize,
+        };
+        define_one_arm_cp_reg(cpu, &gmid_reginfo);
         define_arm_cp_regs(cpu, mte_reginfo);
         define_arm_cp_regs(cpu, mte_el0_cacheop_reginfo);
     } else if (cpu_isar_feature(aa64_mte_insn_reg, cpu)) {

target/arm/internals.h

@@ -1246,12 +1246,6 @@ void arm_log_exception(CPUState *cs);
 
 #endif /* !CONFIG_USER_ONLY */
 
-/*
- * The log2 of the words in the tag block, for GMID_EL1.BS.
- * The is the maximum, 256 bytes, which manipulates 64-bits of tags.
- */
-#define GMID_EL1_BS  6
-
 /*
  * SVE predicates are 1/8 the size of SVE vectors, and cannot use
  * the same simd_desc() encoding due to restrictions on size.

target/arm/tcg/cpu64.c

@@ -868,6 +868,7 @@ void aarch64_max_tcg_initfn(Object *obj)
     cpu->ctr = 0x80038003; /* 32 byte I and D cacheline size, VIPT icache */
     cpu->dcz_blocksize = 7; /*  512 bytes */
 #endif
+    cpu->gm_blocksize = 6;  /*  256 bytes */
 
     cpu->sve_vq.supported = MAKE_64BIT_MASK(0, ARM_MAX_VQ);
     cpu->sme_vq.supported = SVE_VQ_POW2_MAP;

target/arm/tcg/mte_helper.c

@@ -421,46 +421,54 @@ void HELPER(st2g_stub)(CPUARMState *env, uint64_t ptr)
     }
 }
 
-#define LDGM_STGM_SIZE  (4 << GMID_EL1_BS)
-
 uint64_t HELPER(ldgm)(CPUARMState *env, uint64_t ptr)
 {
     int mmu_idx = cpu_mmu_index(env, false);
     uintptr_t ra = GETPC();
+    int gm_bs = env_archcpu(env)->gm_blocksize;
+    int gm_bs_bytes = 4 << gm_bs;
     void *tag_mem;
 
-    ptr = QEMU_ALIGN_DOWN(ptr, LDGM_STGM_SIZE);
+    ptr = QEMU_ALIGN_DOWN(ptr, gm_bs_bytes);
 
     /* Trap if accessing an invalid page.  */
     tag_mem = allocation_tag_mem(env, mmu_idx, ptr, MMU_DATA_LOAD,
-                                 LDGM_STGM_SIZE, MMU_DATA_LOAD,
+                                 gm_bs_bytes, MMU_DATA_LOAD,
-                                 LDGM_STGM_SIZE / (2 * TAG_GRANULE), ra);
+                                 gm_bs_bytes / (2 * TAG_GRANULE), ra);
 
     /* The tag is squashed to zero if the page does not support tags.  */
     if (!tag_mem) {
         return 0;
     }
 
-    QEMU_BUILD_BUG_ON(GMID_EL1_BS != 6);
     /*
-     * We are loading 64-bits worth of tags.  The ordering of elements
+     * The ordering of elements within the word corresponds to
-     * within the word corresponds to a 64-bit little-endian operation.
+     * a little-endian operation.
      */
-    return ldq_le_p(tag_mem);
+    switch (gm_bs) {
+    case 6:
+        /* 256 bytes -> 16 tags -> 64 result bits */
+        return ldq_le_p(tag_mem);
+    default:
+        /* cpu configured with unsupported gm blocksize. */
+        g_assert_not_reached();
+    }
 }
 
 void HELPER(stgm)(CPUARMState *env, uint64_t ptr, uint64_t val)
 {
     int mmu_idx = cpu_mmu_index(env, false);
     uintptr_t ra = GETPC();
+    int gm_bs = env_archcpu(env)->gm_blocksize;
+    int gm_bs_bytes = 4 << gm_bs;
     void *tag_mem;
 
-    ptr = QEMU_ALIGN_DOWN(ptr, LDGM_STGM_SIZE);
+    ptr = QEMU_ALIGN_DOWN(ptr, gm_bs_bytes);
 
     /* Trap if accessing an invalid page.  */
     tag_mem = allocation_tag_mem(env, mmu_idx, ptr, MMU_DATA_STORE,
-                                 LDGM_STGM_SIZE, MMU_DATA_LOAD,
+                                 gm_bs_bytes, MMU_DATA_LOAD,
-                                 LDGM_STGM_SIZE / (2 * TAG_GRANULE), ra);
+                                 gm_bs_bytes / (2 * TAG_GRANULE), ra);
 
     /*
      * Tag store only happens if the page support tags,
@@ -470,12 +478,18 @@ void HELPER(stgm)(CPUARMState *env, uint64_t ptr, uint64_t val)
         return;
     }
 
-    QEMU_BUILD_BUG_ON(GMID_EL1_BS != 6);
     /*
-     * We are storing 64-bits worth of tags.  The ordering of elements
+     * The ordering of elements within the word corresponds to
-     * within the word corresponds to a 64-bit little-endian operation.
+     * a little-endian operation.
      */
-    stq_le_p(tag_mem, val);
+    switch (gm_bs) {
+    case 6:
+        stq_le_p(tag_mem, val);
+        break;
+    default:
+        /* cpu configured with unsupported gm blocksize. */
+        g_assert_not_reached();
+    }
 }
 
 void HELPER(stzgm_tags)(CPUARMState *env, uint64_t ptr, uint64_t val)

target/arm/tcg/translate-a64.c

@@ -3786,7 +3786,7 @@ static bool trans_STGM(DisasContext *s, arg_ldst_tag *a)
         gen_helper_stgm(cpu_env, addr, tcg_rt);
     } else {
         MMUAccessType acc = MMU_DATA_STORE;
-        int size = 4 << GMID_EL1_BS;
+        int size = 4 << s->gm_blocksize;
 
         clean_addr = clean_data_tbi(s, addr);
         tcg_gen_andi_i64(clean_addr, clean_addr, -size);
@@ -3818,7 +3818,7 @@ static bool trans_LDGM(DisasContext *s, arg_ldst_tag *a)
         gen_helper_ldgm(tcg_rt, cpu_env, addr);
     } else {
         MMUAccessType acc = MMU_DATA_LOAD;
-        int size = 4 << GMID_EL1_BS;
+        int size = 4 << s->gm_blocksize;
 
         clean_addr = clean_data_tbi(s, addr);
         tcg_gen_andi_i64(clean_addr, clean_addr, -size);
@@ -13896,6 +13896,7 @@ static void aarch64_tr_init_disas_context(DisasContextBase *dcbase,
     dc->cp_regs = arm_cpu->cp_regs;
     dc->features = env->features;
     dc->dcz_blocksize = arm_cpu->dcz_blocksize;
+    dc->gm_blocksize = arm_cpu->gm_blocksize;
 
 #ifdef CONFIG_USER_ONLY
     /* In sve_probe_page, we assume TBI is enabled. */

target/arm/tcg/translate.h

@@ -151,6 +151,8 @@ typedef struct DisasContext {
     int8_t btype;
     /* A copy of cpu->dcz_blocksize. */
     uint8_t dcz_blocksize;
+    /* A copy of cpu->gm_blocksize. */
+    uint8_t gm_blocksize;
     /* True if this page is guarded.  */
     bool guarded_page;
     /* Bottom two bits of XScale c15_cpar coprocessor access control reg */