Auto merge of rust-lang#132800 - matthiaskrgr:rollup-c1kkj56, r=matthiaskrgr

Rollup of 5 pull requests

Successful merges:

 - rust-lang#132552 (Add v9, v8plus, and leoncasa target feature to sparc and use v8plus in create_object_file)
 - rust-lang#132745 (pointee_info_at: fix logic for recursing into enums)
 - rust-lang#132777 (try_question_mark_nop: update test for LLVM 20)
 - rust-lang#132785 (rustc_target: more target string fixes for LLVM 20)
 - rust-lang#132794 (Use a separate dir for r-a builds consistently in helix config)

r? `@ghost`
`@rustbot` modify labels: rollup

Author: bors

compiler/rustc_abi/src/lib.rs

@@ -1743,15 +1743,23 @@ pub enum PointerKind {
     Box { unpin: bool, global: bool },
 }
 
-/// Note that this information is advisory only, and backends are free to ignore it.
-/// It can only be used to encode potential optimizations, but no critical information.
+/// Encodes extra information we have about a pointer.
+/// Note that this information is advisory only, and backends are free to ignore it:
+/// if the information is wrong, that can cause UB, but if the information is absent,
+/// that must always be okay.
 #[derive(Copy, Clone, Debug)]
 pub struct PointeeInfo {
-    pub size: Size,
-    pub align: Align,
     /// If this is `None`, then this is a raw pointer, so size and alignment are not guaranteed to
     /// be reliable.
     pub safe: Option<PointerKind>,
+    /// If `safe` is `Some`, then the pointer is either null or dereferenceable for this many bytes.
+    /// On a function argument, "dereferenceable" here means "dereferenceable for the entire duration
+    /// of this function call", i.e. it is UB for the memory that this pointer points to to be freed
+    /// while this function is still running.
+    /// The size can be zero if the pointer is not dereferenceable.
+    pub size: Size,
+    /// If `safe` is `Some`, then the pointer is aligned as indicated.
+    pub align: Align,
 }
 
 impl<FieldIdx: Idx, VariantIdx: Idx> LayoutData<FieldIdx, VariantIdx> {

compiler/rustc_codegen_llvm/src/llvm_util.rs

@@ -228,6 +228,8 @@ pub(crate) fn to_llvm_features<'a>(sess: &Session, s: &'a str) -> Option<LLVMFea
         "x86"
     } else if sess.target.arch == "arm64ec" {
         "aarch64"
+    } else if sess.target.arch == "sparc64" {
+        "sparc"
     } else {
         &*sess.target.arch
     };
@@ -280,6 +282,13 @@ pub(crate) fn to_llvm_features<'a>(sess: &Session, s: &'a str) -> Option<LLVMFea
         // Support for `wide-arithmetic` will first land in LLVM 20 as part of
         // llvm/llvm-project#111598
         ("wasm32" | "wasm64", "wide-arithmetic") if get_version() < (20, 0, 0) => None,
+        ("sparc", "leoncasa") => Some(LLVMFeature::new("hasleoncasa")),
+        // In LLVM 19, there is no `v8plus` feature and `v9` means "SPARC-V9 instruction available and SPARC-V8+ ABI used".
+        // https://github.com/llvm/llvm-project/blob/llvmorg-19.1.0/llvm/lib/Target/Sparc/MCTargetDesc/SparcELFObjectWriter.cpp#L27-L28
+        // Before LLVM 19, there is no `v8plus` feature and `v9` means "SPARC-V9 instruction available".
+        // https://github.com/llvm/llvm-project/blob/llvmorg-18.1.0/llvm/lib/Target/Sparc/MCTargetDesc/SparcELFObjectWriter.cpp#L26
+        ("sparc", "v8plus") if get_version().0 == 19 => Some(LLVMFeature::new("v9")),
+        ("sparc", "v8plus") if get_version().0 < 19 => None,
         (_, s) => Some(LLVMFeature::new(s)),
     }
 }
@@ -619,6 +628,8 @@ pub(crate) fn global_llvm_features(
             .features
             .split(',')
             .filter(|v| !v.is_empty() && backend_feature_name(sess, v).is_some())
+            // Drop +v8plus feature introduced in LLVM 20.
+            .filter(|v| *v != "+v8plus" || get_version() >= (20, 0, 0))
             .map(String::from),
     );
 

compiler/rustc_codegen_ssa/src/back/metadata.rs

@@ -212,7 +212,7 @@ pub(crate) fn create_object_file(sess: &Session) -> Option<write::Object<'static
         "riscv32" => (Architecture::Riscv32, None),
         "riscv64" => (Architecture::Riscv64, None),
         "sparc" => {
-            if sess.target.options.cpu == "v9" {
+            if sess.unstable_target_features.contains(&sym::v8plus) {
                 // Target uses V8+, aka EM_SPARC32PLUS, aka 64-bit V9 but in 32-bit mode
                 (Architecture::Sparc32Plus, None)
             } else {

compiler/rustc_feature/src/unstable.rs

@@ -336,6 +336,7 @@ declare_features! (
     (unstable, riscv_target_feature, "1.45.0", Some(44839)),
     (unstable, rtm_target_feature, "1.35.0", Some(44839)),
     (unstable, s390x_target_feature, "1.82.0", Some(44839)),
+    (unstable, sparc_target_feature, "CURRENT_RUSTC_VERSION", Some(132783)),
     (unstable, sse4a_target_feature, "1.27.0", Some(44839)),
     (unstable, tbm_target_feature, "1.27.0", Some(44839)),
     (unstable, wasm_target_feature, "1.30.0", Some(44839)),

compiler/rustc_middle/src/ty/layout.rs

@@ -1011,25 +1011,41 @@ where
             }
 
             _ => {
-                let mut data_variant = match this.variants {
+                let mut data_variant = match &this.variants {
                     // Within the discriminant field, only the niche itself is
                     // always initialized, so we only check for a pointer at its
                     // offset.
                     //
-                    // If the niche is a pointer, it's either valid (according
-                    // to its type), or null (which the niche field's scalar
-                    // validity range encodes). This allows using
-                    // `dereferenceable_or_null` for e.g., `Option<&T>`, and
-                    // this will continue to work as long as we don't start
-                    // using more niches than just null (e.g., the first page of
-                    // the address space, or unaligned pointers).
+                    // Our goal here is to check whether this represents a
+                    // "dereferenceable or null" pointer, so we need to ensure
+                    // that there is only one other variant, and it must be null.
+                    // Below, we will then check whether the pointer is indeed
+                    // dereferenceable.
                     Variants::Multiple {
-                        tag_encoding: TagEncoding::Niche { untagged_variant, .. },
+                        tag_encoding:
+                            TagEncoding::Niche { untagged_variant, niche_variants, niche_start },
                         tag_field,
+                        variants,
                         ..
-                    } if this.fields.offset(tag_field) == offset => {
-                        Some(this.for_variant(cx, untagged_variant))
+                    } if variants.len() == 2 && this.fields.offset(*tag_field) == offset => {
+                        let tagged_variant = if untagged_variant.as_u32() == 0 {
+                            VariantIdx::from_u32(1)
+                        } else {
+                            VariantIdx::from_u32(0)
+                        };
+                        assert_eq!(tagged_variant, *niche_variants.start());
+                        if *niche_start == 0 {
+                            // The other variant is encoded as "null", so we can recurse searching for
+                            // a pointer here. This relies on the fact that the codegen backend
+                            // only adds "dereferenceable" if there's also a "nonnull" proof,
+                            // and that null is aligned for all alignments so it's okay to forward
+                            // the pointer's alignment.
+                            Some(this.for_variant(cx, *untagged_variant))
+                        } else {
+                            None
+                        }
                     }
+                    Variants::Multiple { .. } => None,
                     _ => Some(this),
                 };
 

compiler/rustc_span/src/symbol.rs

@@ -1867,6 +1867,7 @@ symbols! {
         slice_patterns,
         slicing_syntax,
         soft,
+        sparc_target_feature,
         specialization,
         speed,
         spotlight,
@@ -2109,6 +2110,7 @@ symbols! {
         usize_legacy_fn_max_value,
         usize_legacy_fn_min_value,
         usize_legacy_mod,
+        v8plus,
         va_arg,
         va_copy,
         va_end,

compiler/rustc_target/src/callconv/mod.rs

@@ -143,7 +143,8 @@ pub struct ArgAttributes {
     pub regular: ArgAttribute,
     pub arg_ext: ArgExtension,
     /// The minimum size of the pointee, guaranteed to be valid for the duration of the whole call
-    /// (corresponding to LLVM's dereferenceable and dereferenceable_or_null attributes).
+    /// (corresponding to LLVM's dereferenceable_or_null attributes, i.e., it is okay for this to be
+    /// set on a null pointer, but all non-null pointers must be dereferenceable).
     pub pointee_size: Size,
     pub pointee_align: Option<Align>,
 }

compiler/rustc_target/src/spec/targets/aarch64_be_unknown_linux_gnu.rs

@@ -11,7 +11,7 @@ pub(crate) fn target() -> Target {
             std: Some(true),
         },
         pointer_width: 64,
-        data_layout: "E-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32".into(),
+        data_layout: "E-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32".into(),
         arch: "aarch64".into(),
         options: TargetOptions {
             features: "+v8a,+outline-atomics".into(),

compiler/rustc_target/src/spec/targets/aarch64_be_unknown_linux_gnu_ilp32.rs

@@ -14,7 +14,7 @@ pub(crate) fn target() -> Target {
             std: Some(true),
         },
         pointer_width: 32,
-        data_layout: "E-m:e-p:32:32-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32".into(),
+        data_layout: "E-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32".into(),
         arch: "aarch64".into(),
         options: TargetOptions {
             abi: "ilp32".into(),

compiler/rustc_target/src/spec/targets/aarch64_be_unknown_netbsd.rs

@@ -11,7 +11,7 @@ pub(crate) fn target() -> Target {
             std: Some(true),
         },
         pointer_width: 64,
-        data_layout: "E-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32".into(),
+        data_layout: "E-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32".into(),
         arch: "aarch64".into(),
         options: TargetOptions {
             mcount: "__mcount".into(),