[llvm-branch-commits] [lld] [lld] Fix handling of RISCV TLSDESC relocations (PR #85817)

[Fangrui Song via llvm-branch-commits]

================
@@ -136,28 +140,30 @@
 
 # IE32:       .got     00000008 00012248
 
-# IE32-LABEL: <.text>:
+# IE32-LABEL: <.Ltlsdesc_hi0>:
 ## st_value(a) = 8
 # IE32-NEXT:         addi    zero, zero, 0x0
 # IE32-NEXT:         addi    zero, zero, 0x0
 # IE32-NEXT:         addi    zero, zero, 0x0
 # IE32-NEXT:         addi    a0, zero, 0x8
 # IE32-NEXT:         add     a0, a0, tp
 ## st_value(b) = 2047
+# IE32-LABEL: <.Ltlsdesc_hi1>:
 # IE32-NEXT:         addi    zero, zero, 0x0
 # IE32-NEXT:         addi    zero, zero, 0x0
 # IE32-NEXT:         addi    zero, zero, 0x0
 # IE32-NEXT:         addi    a0, zero, 0x7ff
 # IE32-NEXT:         add     a0, a0, tp
 ## &.got[c]-. = 0x12248+4 - 0x111cc = 0x1080
+# IE32-LABEL: <.Ltlsdesc_hi2>:
 # IE32-NEXT:         addi    zero, zero, 0x0
 # IE32-NEXT:         addi    zero, zero, 0x0
 # IE32-NEXT:  111cc: auipc   a0, 0x1
 # IE32-NEXT:         lw      a0, 0x80(a0)
 # IE32-NEXT:         add     a0, a0, tp
 
-## FIXME This should not pass, but the code MC layer needs a fix to prevent this.
-# BADTLSLABEL: error: d.{{.*}}.o has an STT_TLS symbol but doesn't have an SHF_TLS section
+# At one point the local TLSDESC labels would be marked STT_TLS, so make sure we don't regress
+# BADTLSLABEL-NOT: error: d.{{.*}}.o has an STT_TLS symbol but doesn't have an SHF_TLS section
----------------
MaskRay wrote:

Delete the NOT pattern. If the diagnostic changes, the NOT pattern may likely not get updated.

`# ` => `## Before #85817`

https://github.com/llvm/llvm-project/pull/85817