Skip to content

Add ppc64 ELFv2 support to bthread and butil #3696

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
pkubaj wants to merge 10 commits into
base: master
Choose a base branch
from
Open

Add ppc64 ELFv2 support to bthread and butil #3696

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
11a4e3f
Add ppc64 ELFv2 support to bthread and butil
pkubaj Oct 27, 2026
3df0773
Add a ppc64 ELFv2 test to bthread_context_unittest
pkubaj Nov 1, 2026
a8e33fa
Fix the first frame and the exit path of the ppc64 context switch
pkubaj Nov 2, 2026
c9dac71
Return the value observed by the CAS in atomicops_internals_gcc.h
pkubaj Nov 2, 2026
73421c2
Preserve the caller's FPRs and VRs in the ppc64 context test helper
pkubaj Nov 2, 2026
dd5a5a5
Use __sync_add_and_fetch in atomicops_internals_gcc.h
pkubaj Nov 2, 2026
410f275
Handle the Mongo wire format on big-endian hosts
pkubaj Nov 2, 2026
0c13bae
Select the ppc64 ELFv2 context switch independently of the OS
pkubaj Nov 7, 2026
5a1a9f0
Save v20-v31 in the ppc64 context switch unconditionally
pkubaj Nov 7, 2026
1154a4c
Restore the previous section after the ppc64 asm blocks
pkubaj Nov 7, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 9 additions & 6 deletions src/brpc/mongo_head.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,13 +62,16 @@ struct mongo_head_t {
int32_t op_code; // request type, see MongoOpCode.

void make_host_endian() {
if (!ARCH_CPU_LITTLE_ENDIAN) {
message_length = butil::ByteSwap((uint32_t)message_length);
request_id = butil::ByteSwap((uint32_t)request_id);
response_to = butil::ByteSwap((uint32_t)response_to);
op_code = butil::ByteSwap((uint32_t)op_code);
}
#if !defined(ARCH_CPU_LITTLE_ENDIAN)
message_length = butil::ByteSwap((uint32_t)message_length);
request_id = butil::ByteSwap((uint32_t)request_id);
response_to = butil::ByteSwap((uint32_t)response_to);
op_code = butil::ByteSwap((uint32_t)op_code);
Comment thread
pkubaj marked this conversation as resolved.
Resolved
#endif
}
// The wire format is little-endian, so the conversion between it and the
// host byte order is the same swap in both directions.
void make_wire_endian() { make_host_endian(); }
};
#pragma pack()

Expand Down
2 changes: 2 additions & 0 deletions src/brpc/mongo_service_adaptor.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,6 +50,8 @@ class MongoServiceAdaptor {

// Make an error msg when the cntl fails. If cntl fails, we must send mongo client a msg not
// only to indicate the error, but also to finish the round trip.
// The msg must be in the wire format, which is little-endian on every
// host (mongo_head_t::make_wire_endian(), butil::ByteSwapToLE32()).
virtual void SerializeError(int response_to, butil::IOBuf* out_buf) const = 0;

// Create a custom context which is attached to socket. This func is called only when the first
Expand Down
30 changes: 17 additions & 13 deletions src/brpc/policy/mongo_protocol.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@

#include "butil/iobuf.h" // butil::IOBuf
#include "butil/strings/string_util.h"
#include "butil/sys_byteorder.h"
#include "butil/time.h"

#include "brpc/controller.h" // Controller
Expand Down Expand Up @@ -86,11 +87,13 @@ void SendMongoResponse::Run() {
res.header().response_to(),
res.header().op_code()
};
header.make_wire_endian();
res_buf.append(static_cast<const void*>(&header), sizeof(mongo_head_t));
int32_t response_flags = res.response_flags();
int64_t cursor_id = res.cursor_id();
int32_t starting_from = res.starting_from();
int32_t number_returned = res.number_returned();
// The wire format is little-endian.
const uint32_t response_flags = butil::ByteSwapToLE32(res.response_flags());
const uint64_t cursor_id = butil::ByteSwapToLE64(res.cursor_id());
const uint32_t starting_from = butil::ByteSwapToLE32(res.starting_from());
const uint32_t number_returned = butil::ByteSwapToLE32(res.number_returned());
res_buf.append(&response_flags, sizeof(response_flags));
res_buf.append(&cursor_id, sizeof(cursor_id));
res_buf.append(&starting_from, sizeof(starting_from));
Expand Down Expand Up @@ -200,7 +203,8 @@ void ProcessMongoRequest(InputMessageBase* msg_base) {

char buf[sizeof(mongo_head_t)];
const char *p = (const char *)msg->meta.fetch(buf, sizeof(buf));
const mongo_head_t *header = (const mongo_head_t*)p;
mongo_head_t header = *(const mongo_head_t*)p;
header.make_host_endian();

const google::protobuf::ServiceDescriptor* srv_des = MongoService::descriptor();
if (1 != srv_des->method_count()) {
Expand Down Expand Up @@ -280,20 +284,20 @@ void ProcessMongoRequest(InputMessageBase* msg_base) {
}
}

if (!MongoOp_IsValid(header->op_code)) {
mongo_done->cntl.SetFailed(EREQUEST, "Unknown op_code:%d", header->op_code);
if (!MongoOp_IsValid(header.op_code)) {
mongo_done->cntl.SetFailed(EREQUEST, "Unknown op_code:%d", header.op_code);
break;
}

mongo_done->cntl.set_log_id(header->request_id);
mongo_done->cntl.set_log_id(header.request_id);
const std::string &body_str = msg->payload.to_string();
mongo_done->req.set_message(body_str.c_str(), body_str.size());
mongo_done->req.mutable_header()->set_message_length(header->message_length);
mongo_done->req.mutable_header()->set_request_id(header->request_id);
mongo_done->req.mutable_header()->set_response_to(header->response_to);
mongo_done->req.mutable_header()->set_message_length(header.message_length);
mongo_done->req.mutable_header()->set_request_id(header.request_id);
mongo_done->req.mutable_header()->set_response_to(header.response_to);
mongo_done->req.mutable_header()->set_op_code(
static_cast<MongoOp>(header->op_code));
mongo_done->res.mutable_header()->set_response_to(header->request_id);
static_cast<MongoOp>(header.op_code));
mongo_done->res.mutable_header()->set_response_to(header.request_id);
mongo_done->received_us = msg->received_us();

google::protobuf::Service* svc = mp->service;
Expand Down
252 changes: 252 additions & 0 deletions src/bthread/context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1012,3 +1012,255 @@ __asm (
" call _exit\n"
);
#endif

#if defined(BTHREAD_CONTEXT_PLATFORM_ppc64_elfv2) && defined(BTHREAD_CONTEXT_COMPILER_gcc)
__asm (
".text\n"
".align 2\n"
".machine push\n"
".machine altivec\n"
".globl bthread_jump_fcontext\n"
".type bthread_jump_fcontext, @function\n"
"bthread_jump_fcontext:\n"
" addis %r2, %r12, .TOC.-bthread_jump_fcontext@ha\n"
" addi %r2, %r2, .TOC.-bthread_jump_fcontext@l\n"
Comment thread
pkubaj marked this conversation as resolved.
Resolved
" .localentry bthread_jump_fcontext, . - bthread_jump_fcontext\n"
" # reserve space on stack\n"
" subi %r1, %r1, 528\n"
" std %r14, 352(%r1)\n"
" std %r15, 360(%r1)\n"
" std %r16, 368(%r1)\n"
" std %r17, 376(%r1)\n"
" std %r18, 384(%r1)\n"
" std %r19, 392(%r1)\n"
" std %r20, 400(%r1)\n"
" std %r21, 408(%r1)\n"
" std %r22, 416(%r1)\n"
" std %r23, 424(%r1)\n"
" std %r24, 432(%r1)\n"
" std %r25, 440(%r1)\n"
" std %r26, 448(%r1)\n"
" std %r27, 456(%r1)\n"
" std %r28, 464(%r1)\n"
" std %r29, 472(%r1)\n"
" std %r30, 480(%r1)\n"
" std %r31, 488(%r1)\n"
" # save CR\n"
" mfcr %r0\n"
" std %r0, 496(%r1)\n"
" # save LR, and LR as PC\n"
" mflr %r0\n"
" std %r0, 504(%r1)\n"
" std %r0, 512(%r1)\n"
" # the fourth argument (preserve_fpu) is ignored: the compiler may keep\n"
" # any value in the non-volatile FPRs/VRs across the call\n"
" stfd %f14, 200(%r1)\n"
" stfd %f15, 208(%r1)\n"
" stfd %f16, 216(%r1)\n"
" stfd %f17, 224(%r1)\n"
" stfd %f18, 232(%r1)\n"
" stfd %f19, 240(%r1)\n"
" stfd %f20, 248(%r1)\n"
" stfd %f21, 256(%r1)\n"
" stfd %f22, 264(%r1)\n"
" stfd %f23, 272(%r1)\n"
" stfd %f24, 280(%r1)\n"
" stfd %f25, 288(%r1)\n"
" stfd %f26, 296(%r1)\n"
" stfd %f27, 304(%r1)\n"
" stfd %f28, 312(%r1)\n"
" stfd %f29, 320(%r1)\n"
" stfd %f30, 328(%r1)\n"
" stfd %f31, 336(%r1)\n"
" mffs %f0\n"
" stfd %f0, 344(%r1)\n"
" mfspr %r0, 256\n"
" std %r0, 192(%r1)\n"
" li %r11, 0\n"
" stvx %v20, %r1, %r11\n"
" li %r11, 16\n"
" stvx %v21, %r1, %r11\n"
" li %r11, 32\n"
" stvx %v22, %r1, %r11\n"
" li %r11, 48\n"
" stvx %v23, %r1, %r11\n"
" li %r11, 64\n"
" stvx %v24, %r1, %r11\n"
" li %r11, 80\n"
" stvx %v25, %r1, %r11\n"
" li %r11, 96\n"
" stvx %v26, %r1, %r11\n"
" li %r11, 112\n"
" stvx %v27, %r1, %r11\n"
" li %r11, 128\n"
" stvx %v28, %r1, %r11\n"
" li %r11, 144\n"
" stvx %v29, %r1, %r11\n"
" li %r11, 160\n"
" stvx %v30, %r1, %r11\n"
" li %r11, 176\n"
" stvx %v31, %r1, %r11\n"
" # store RSP (pointing to context-data) in R3\n"
" std %r1, 0(%r3)\n"
" # restore RSP (pointing to context-data) from R4\n"
" mr %r1, %r4\n"
" li %r11, 0\n"
" lvx %v20, %r1, %r11\n"
" li %r11, 16\n"
" lvx %v21, %r1, %r11\n"
" li %r11, 32\n"
" lvx %v22, %r1, %r11\n"
" li %r11, 48\n"
" lvx %v23, %r1, %r11\n"
" li %r11, 64\n"
" lvx %v24, %r1, %r11\n"
" li %r11, 80\n"
" lvx %v25, %r1, %r11\n"
" li %r11, 96\n"
" lvx %v26, %r1, %r11\n"
" li %r11, 112\n"
" lvx %v27, %r1, %r11\n"
" li %r11, 128\n"
" lvx %v28, %r1, %r11\n"
" li %r11, 144\n"
" lvx %v29, %r1, %r11\n"
" li %r11, 160\n"
" lvx %v30, %r1, %r11\n"
" li %r11, 176\n"
" lvx %v31, %r1, %r11\n"
" ld %r0, 192(%r1)\n"
" mtspr 256, %r0\n"
" lfd %f0, 344(%r1)\n"
" mtfsf 0xff, %f0\n"
" lfd %f14, 200(%r1)\n"
" lfd %f15, 208(%r1)\n"
" lfd %f16, 216(%r1)\n"
" lfd %f17, 224(%r1)\n"
" lfd %f18, 232(%r1)\n"
" lfd %f19, 240(%r1)\n"
" lfd %f20, 248(%r1)\n"
" lfd %f21, 256(%r1)\n"
" lfd %f22, 264(%r1)\n"
" lfd %f23, 272(%r1)\n"
" lfd %f24, 280(%r1)\n"
" lfd %f25, 288(%r1)\n"
" lfd %f26, 296(%r1)\n"
" lfd %f27, 304(%r1)\n"
" lfd %f28, 312(%r1)\n"
" lfd %f29, 320(%r1)\n"
" lfd %f30, 328(%r1)\n"
" lfd %f31, 336(%r1)\n"
" ld %r14, 352(%r1)\n"
" ld %r15, 360(%r1)\n"
" ld %r16, 368(%r1)\n"
" ld %r17, 376(%r1)\n"
" ld %r18, 384(%r1)\n"
" ld %r19, 392(%r1)\n"
" ld %r20, 400(%r1)\n"
" ld %r21, 408(%r1)\n"
" ld %r22, 416(%r1)\n"
" ld %r23, 424(%r1)\n"
" ld %r24, 432(%r1)\n"
" ld %r25, 440(%r1)\n"
" ld %r26, 448(%r1)\n"
" ld %r27, 456(%r1)\n"
" ld %r28, 464(%r1)\n"
" ld %r29, 472(%r1)\n"
" ld %r30, 480(%r1)\n"
" ld %r31, 488(%r1)\n"
" # restore CR\n"
" ld %r0, 496(%r1)\n"
" mtcr %r0\n"
" # restore LR\n"
" ld %r0, 504(%r1)\n"
" mtlr %r0\n"
" # load PC into CTR (and R12 for the callee's TOC setup)\n"
" ld %r12, 512(%r1)\n"
" mtctr %r12\n"
" # adjust stack\n"
" addi %r1, %r1, 528\n"
" # use third arg as return value after jump\n"
" # and as first arg in context function\n"
" mr %r3, %r5\n"
" # jump to context\n"
" bctr\n"
".machine pop\n"
".size bthread_jump_fcontext, .-bthread_jump_fcontext\n"
".section .note.GNU-stack,\"\",%progbits\n"
Comment thread
pkubaj marked this conversation as resolved.
Resolved
".previous\n"
);

#endif

#if defined(BTHREAD_CONTEXT_PLATFORM_ppc64_elfv2) && defined(BTHREAD_CONTEXT_COMPILER_gcc)
__asm (
".text\n"
".align 2\n"
".machine push\n"
".machine altivec\n"
".globl bthread_make_fcontext\n"
".type bthread_make_fcontext, @function\n"
"bthread_make_fcontext:\n"
" addis %r2, %r12, .TOC.-bthread_make_fcontext@ha\n"
" addi %r2, %r2, .TOC.-bthread_make_fcontext@l\n"
Comment thread
pkubaj marked this conversation as resolved.
Resolved
" .localentry bthread_make_fcontext, . - bthread_make_fcontext\n"
" # save return address into R6\n"
" mflr %r6\n"
" # first arg of bthread_make_fcontext() == top address of context-stack\n"
" # shift address in R3 to lower 16 byte boundary\n"
" clrrdi %r3, %r3, 4\n"
" # reserve space for context-data on context-stack\n"
" # including 64 byte of linkage + parameter area (R1 % 16 == 0)\n"
" subi %r3, %r3, 592\n"
" # third arg of bthread_make_fcontext() == address of context-function\n"
" # save address of context-function entry point\n"
" std %r5, 512(%r3)\n"
" # set the back chain of the context's first frame to zero\n"
" li %r0, 0\n"
" std %r0, 528(%r3)\n"
" # initial CR, FPSCR and VRSAVE of the context: the current ones\n"
" mfcr %r0\n"
" std %r0, 496(%r3)\n"
" mffs %f0\n"
" stfd %f0, 344(%r3)\n"
" mfspr %r0, 256\n"
" std %r0, 192(%r3)\n"
" # load LR\n"
" mflr %r0\n"
" # jump to label 1\n"
" bl 1f\n"
"1:\n"
" # load LR into R4\n"
" mflr %r4\n"
" # compute abs address of label finish\n"
" addi %r4, %r4, finish - 1b\n"
" # restore LR\n"
" mtlr %r0\n"
" # save address of finish as return-address for context-function\n"
" # will be entered after context-function returns\n"
" std %r4, 504(%r3)\n"
" # restore return address from R6\n"
" mtlr %r6\n"
" blr # return pointer to context-data\n"
"finish:\n"
" # LR holds the address of finish: derive this module's TOC from it for\n"
" # the _exit call, the context function may live in another DSO\n"
" mflr %r12\n"
" addis %r2, %r12, .TOC.-finish@ha\n"
" addi %r2, %r2, .TOC.-finish@l\n"
" # save return address on stack, set up stack frame\n"
" std %r12, 16(%r1)\n"
" # allocate stack space, R1 % 16 == 0\n"
" stdu %r1, -32(%r1)\n"
" # exit code is zero\n"
" li %r3, 0\n"
" # exit application\n"
" bl _exit\n"
" nop\n"
".machine pop\n"
".size bthread_make_fcontext, .-bthread_make_fcontext\n"
".section .note.GNU-stack,\"\",%progbits\n"
".previous\n"
);

#endif
11 changes: 11 additions & 0 deletions src/bthread/context.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,17 @@
#endif
#endif

// The 64-bit PowerPC ELFv2 ABI is the same on every OS that uses it:
// ppc64le everywhere, big-endian ppc64 on FreeBSD and musl.
#if defined(__powerpc64__)
#if defined(_CALL_ELF) && _CALL_ELF == 2
#define BTHREAD_CONTEXT_PLATFORM_ppc64_elfv2
#define BTHREAD_CONTEXT_CALL_CONVENTION
#else
#error "64-bit PowerPC is supported with the ELFv2 ABI only"
#endif
#endif

#endif

#if defined(_WIN32_WCE)
Expand Down
2 changes: 2 additions & 0 deletions src/butil/atomicops.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,6 +159,8 @@ Atomic64 Release_Load(volatile const Atomic64* ptr);
#include "butil/atomicops_internals_loongarch64_gcc.h"
#elif defined(COMPILER_GCC) && defined(ARCH_CPU_RISCV_FAMILY)
#include "butil/atomicops_internals_riscv_gcc.h"
#elif defined(COMPILER_GCC) && defined(ARCH_CPU_PPC64_FAMILY)
#include "butil/atomicops_internals_gcc.h"
#else
#error "Atomic operations are not supported on your platform"
#endif
Expand Down
Loading
Loading