Skip to contentSkip to conversation

target/arm: Remove access_el3_aa32ns_aa64any()

PATCH Applied Edgar E. Iglesias 2 participants 3 messages May 5, 2020 – May 11, 2020

lore
Share & tools
lore
#0 lore raw

v3 cover target/arm: Remove access_el3_aa32ns_aa64any()

From: "Edgar E. Iglesias" <[email protected]>

Hi,

Laurent reported hitting the assert in access_el3_aa32ns() when accessing 32-bit versions of some of the virtualization regs when EL3 is 64-bit.

I think we got this wrong back then and it seems to me like we should merge access_el3_aa32ns and access_el3_aa32ns_aa64_any() and always call the merged function to handle both aa32-only cases and mixed aa32/aa64.

Cheers,
Edgar

ChangeLog:

v2 -> v3:
* Update commit message and cover letter to reflect that access_el3_aa32ns_aa64any is now being removed

v1 -> v2:
* Keep access_el3_aa32ns in favor of access_el3_aa32ns_aa64any
* Simplify description of access_el3_aa32ns
* Tweak secure aa32-el3 check in access_el3_aa32ns as suggested by Peter

Edgar E. Iglesias (1):
target/arm: Drop access_el3_aa32ns_aa64any()

 target/arm/helper.c | 30 +++++++-----------------------
 1 file changed, 7 insertions(+), 23 deletions(-)
Signature · 1 line
2.20.1
#1 lore raw

PATCH v3 1/1 target/arm: Drop access_el3_aa32ns_aa64any()

From: "Edgar E. Iglesias" <[email protected]>

Calling access_el3_aa32ns() works for AArch32 only cores but it does not handle 32-bit EL2 on top of 64-bit EL3 for mixed 32/64-bit cores.

Merge access_el3_aa32ns_aa64any() into access_el3_aa32ns() and only use the latter.

Fixes 68e9c2fe65 ("target-arm: Add VTCR_EL2")
---
 target/arm/helper.c | 30 +++++++-----------------------
 1 file changed, 7 insertions(+), 23 deletions(-)
target/arm/helper.c +7 -23
index dfefb9b3d9..7d21bf1cc7 100644
@@ -502,35 +502,19 @@ void init_cpreg_list(ARMCPU *cpu)
 }
 
 /*
- * Some registers are not accessible if EL3.NS=0 and EL3 is using AArch32 but
- * they are accessible when EL3 is using AArch64 regardless of EL3.NS.
- *
- * access_el3_aa32ns: Used to check AArch32 register views.
- * access_el3_aa32ns_aa64any: Used to check both AArch32/64 register views.
+ * Some registers are not accessible from AArch32 EL3 if SCR.NS == 0.
  */
 static CPAccessResult access_el3_aa32ns(CPUARMState *env,
                                         const ARMCPRegInfo *ri,
                                         bool isread)
 {
-    bool secure = arm_is_secure_below_el3(env);
-
-    assert(!arm_el_is_aa64(env, 3));
-    if (secure) {
+    if (!is_a64(env) && arm_current_el(env) == 3 &&
+        arm_is_secure_below_el3(env)) {
         return CP_ACCESS_TRAP_UNCATEGORIZED;
     }
     return CP_ACCESS_OK;
 }
 
-static CPAccessResult access_el3_aa32ns_aa64any(CPUARMState *env,
-                                                const ARMCPRegInfo *ri,
-                                                bool isread)
-{
-    if (!arm_el_is_aa64(env, 3)) {
-        return access_el3_aa32ns(env, ri, isread);
-    }
-    return CP_ACCESS_OK;
-}
-
 /* Some secure-only AArch32 registers trap to EL3 if used from
  * Secure EL1 (but are just ordinary UNDEF in other non-EL3 contexts).
  * Note that an access from Secure EL1 can only happen if EL3 is AArch64.
@@ -5236,7 +5220,7 @@ static const ARMCPRegInfo el3_no_el2_cp_reginfo[] = {
       .access = PL2_RW, .type = ARM_CP_CONST, .resetvalue = 0 },
     { .name = "VTCR_EL2", .state = ARM_CP_STATE_BOTH,
       .opc0 = 3, .opc1 = 4, .crn = 2, .crm = 1, .opc2 = 2,
-      .access = PL2_RW, .accessfn = access_el3_aa32ns_aa64any,
+      .access = PL2_RW, .accessfn = access_el3_aa32ns,
       .type = ARM_CP_CONST, .resetvalue = 0 },
     { .name = "VTTBR", .state = ARM_CP_STATE_AA32,
       .cp = 15, .opc1 = 6, .crm = 2,
@@ -5284,7 +5268,7 @@ static const ARMCPRegInfo el3_no_el2_cp_reginfo[] = {
       .type = ARM_CP_CONST, .resetvalue = 0 },
     { .name = "HPFAR_EL2", .state = ARM_CP_STATE_BOTH,
       .opc0 = 3, .opc1 = 4, .crn = 6, .crm = 0, .opc2 = 4,
-      .access = PL2_RW, .accessfn = access_el3_aa32ns_aa64any,
+      .access = PL2_RW, .accessfn = access_el3_aa32ns,
       .type = ARM_CP_CONST, .resetvalue = 0 },
     { .name = "HSTR_EL2", .state = ARM_CP_STATE_BOTH,
       .opc0 = 3, .opc1 = 4, .crn = 1, .crm = 1, .opc2 = 3,
@@ -7626,12 +7610,12 @@ void register_cp_regs_for_features(ARMCPU *cpu)
             ARMCPRegInfo vpidr_regs[] = {
                 { .name = "VPIDR_EL2", .state = ARM_CP_STATE_BOTH,
                   .opc0 = 3, .opc1 = 4, .crn = 0, .crm = 0, .opc2 = 0,
-                  .access = PL2_RW, .accessfn = access_el3_aa32ns_aa64any,
+                  .access = PL2_RW, .accessfn = access_el3_aa32ns,
                   .type = ARM_CP_CONST, .resetvalue = cpu->midr,
                   .fieldoffset = offsetof(CPUARMState, cp15.vpidr_el2) },
                 { .name = "VMPIDR_EL2", .state = ARM_CP_STATE_BOTH,
                   .opc0 = 3, .opc1 = 4, .crn = 0, .crm = 0, .opc2 = 5,
-                  .access = PL2_RW, .accessfn = access_el3_aa32ns_aa64any,
+                  .access = PL2_RW, .accessfn = access_el3_aa32ns,
                   .type = ARM_CP_NO_RAW,
                   .writefn = arm_cp_write_ignore, .readfn = mpidr_read },
                 REGINFO_SENTINEL
Signature · 1 line
2.20.1
#2 lore raw
In reply to Edgar E. Iglesias
Quoted 5 lines · Edgar E. Iglesias

From: "Edgar E. Iglesias" <[email protected]>

Hi,

Quoted 8 lines

Laurent reported hitting the assert in access_el3_aa32ns() when accessing 32-bit versions of some of the virtualization regs when EL3 is 64-bit.

I think we got this wrong back then and it seems to me like we should merge access_el3_aa32ns and access_el3_aa32ns_aa64_any() and always call the merged function to handle both aa32-only cases and mixed aa32/aa64.

Applied to target-arm.next, thanks.

-- PMM