[libc] Move the pointer to pin off the stack to the heap

[jhuber6]

Summary:
This may be problematic to pin a stack pointer. Allocate it via the OS
allocator instead as the documentation suggests.

For some reason, if you attempt to free this pointer after the memory
region has been unlocked, it will return an invalid pointer.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Joseph Huber (jhuber6)

Changes

Summary:
This may be problematic to pin a stack pointer. Allocate it via the OS
allocator instead as the documentation suggests.

For some reason, if you attempt to free this pointer after the memory
region has been unlocked, it will return an invalid pointer.

---

Full diff: https://github.com/llvm/llvm-project/pull/74118.diff

1 Files Affected:

• (modified) libc/utils/gpu/loader/amdgpu/Loader.cpp (+1-1)

diff --git a/libc/utils/gpu/loader/amdgpu/Loader.cpp b/libc/utils/gpu/loader/amdgpu/Loader.cpp
index a9a687656efcb39..d3d2c2b42080837 100644
--- a/libc/utils/gpu/loader/amdgpu/Loader.cpp
+++ b/libc/utils/gpu/loader/amdgpu/Loader.cpp
@@ -471,7 +471,7 @@ int load(int argc, char **argv, char **envp, void *image, size_t size,
     handle_error(err);
 
   // Pin some memory we can use to obtain the address of the rpc client.
-  void *rpc_client_storage = nullptr;
+  void **rpc_client_storage = new void *;
   void *rpc_client_host = nullptr;
   if (hsa_status_t err =
           hsa_amd_memory_lock(&rpc_client_storage, sizeof(void *),

[llvmbot]

@llvm/pr-subscribers-libc

Author: Joseph Huber (jhuber6)

Changes

Summary:
This may be problematic to pin a stack pointer. Allocate it via the OS
allocator instead as the documentation suggests.

For some reason, if you attempt to free this pointer after the memory
region has been unlocked, it will return an invalid pointer.

---

Full diff: https://github.com/llvm/llvm-project/pull/74118.diff

1 Files Affected:

• (modified) libc/utils/gpu/loader/amdgpu/Loader.cpp (+1-1)

diff --git a/libc/utils/gpu/loader/amdgpu/Loader.cpp b/libc/utils/gpu/loader/amdgpu/Loader.cpp
index a9a687656efcb39..d3d2c2b42080837 100644
--- a/libc/utils/gpu/loader/amdgpu/Loader.cpp
+++ b/libc/utils/gpu/loader/amdgpu/Loader.cpp
@@ -471,7 +471,7 @@ int load(int argc, char **argv, char **envp, void *image, size_t size,
     handle_error(err);
 
   // Pin some memory we can use to obtain the address of the rpc client.
-  void *rpc_client_storage = nullptr;
+  void **rpc_client_storage = new void *;
   void *rpc_client_host = nullptr;
   if (hsa_status_t err =
           hsa_amd_memory_lock(&rpc_client_storage, sizeof(void *),

[JonChesterfield]

probably wrong that it used to be a void* and is now a void**, could that be why delete was a problem?

[JonChesterfield]

This can't be right. If you allocated some memory, and then locked it, you still need to leave that memory allocated until you're finished with it. Free should be after the unlock. Nested lifetimes and all that.

[JonChesterfield]

Github has mangled this. If it's now a three line change - malloc, &, free - then we're good to go

[JonChesterfield]

Please don't land this as five separate commits