Modified: trunk/Source/WTF/ChangeLog (231759 => 231760)
--- trunk/Source/WTF/ChangeLog 2018-05-14 16:18:36 UTC (rev 231759)
+++ trunk/Source/WTF/ChangeLog 2018-05-14 16:45:09 UTC (rev 231760)
@@ -1,3 +1,20 @@
+2018-05-13 Geoffrey Garen <gga...@apple.com>
+
+ Simplified Mach exception handling
+ https://bugs.webkit.org/show_bug.cgi?id=185595
+
+ Reviewed by Keith Miller.
+
+ * wtf/threads/Signals.cpp:
+ (WTF::startMachExceptionHandlerThread): Use mach_msg_server_once instead
+ of duplicating its functionality. Separate error handling logic from
+ program logic to help program logic stand out. Use
+ DISPATCH_TARGET_QUEUE_* instead of explicitly fetching a queue.
+
+ Also, we don't need the high priority queue. The kernel donates a
+ priority voucher from the exception thread to the receiver thread, and
+ mach_msg_server_once takes care to forward that voucher.
+
2018-05-14 Zan Dobersek <zdober...@igalia.com>
[GTK] REGRESSION(r231170) Build broken with Clang 5.0
Modified: trunk/Source/WTF/wtf/threads/Signals.cpp (231759 => 231760)
--- trunk/Source/WTF/wtf/threads/Signals.cpp 2018-05-14 16:18:36 UTC (rev 231759)
+++ trunk/Source/WTF/wtf/threads/Signals.cpp 2018-05-14 16:45:09 UTC (rev 231760)
@@ -73,43 +73,25 @@
{
static std::once_flag once;
std::call_once(once, [] {
- if (mach_port_allocate(mach_task_self(), MACH_PORT_RIGHT_RECEIVE, &exceptionPort) != KERN_SUCCESS)
- CRASH();
+ kern_return_t kr = mach_port_allocate(mach_task_self(), MACH_PORT_RIGHT_RECEIVE, &exceptionPort);
+ RELEASE_ASSERT(kr == KERN_SUCCESS);
+ kr = mach_port_insert_right(mach_task_self(), exceptionPort, exceptionPort, MACH_MSG_TYPE_MAKE_SEND);
+ RELEASE_ASSERT(kr == KERN_SUCCESS);
- if (mach_port_insert_right(mach_task_self(), exceptionPort, exceptionPort, MACH_MSG_TYPE_MAKE_SEND) != KERN_SUCCESS)
- CRASH();
+ dispatch_source_t source = dispatch_source_create(
+ DISPATCH_SOURCE_TYPE_MACH_RECV, exceptionPort, 0, DISPATCH_TARGET_QUEUE_DEFAULT);
+ RELEASE_ASSERT(source);
- // It's not clear that this needs to be the high priority queue but it should be rare and it might be
- // handling exceptions from high priority threads. Anyway, our handlers should be very fast anyway so it's
- // probably not the end of the world if we handle a low priority exception on a high priority queue.
- dispatch_queue_t queue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_HIGH, 0);
- dispatch_source_t source = dispatch_source_create(DISPATCH_SOURCE_TYPE_MACH_RECV, exceptionPort, 0, queue);
- RELEASE_ASSERT_WITH_MESSAGE(source, "We need to ensure our source was created.");
-
- // We should never cancel our handler since it's a permanent thing so we don't add a cancel handler.
dispatch_source_set_event_handler(source, ^{
- // the leaks tool will get mad at us if we don't pretend to watch the source.
- UNUSED_PARAM(source);
- union Message {
- mach_msg_header_t header;
- char data[maxMessageSize];
- };
- Message messageHeaderIn;
- Message messageHeaderOut;
+ UNUSED_PARAM(source); // Capture a pointer to source in user space to silence the leaks tool.
- kern_return_t messageResult = mach_msg(&messageHeaderIn.header, MACH_RCV_MSG, 0, maxMessageSize, exceptionPort, MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
- if (messageResult == KERN_SUCCESS) {
- if (!mach_exc_server(&messageHeaderIn.header, &messageHeaderOut.header))
- CRASH();
-
- messageResult = mach_msg(&messageHeaderOut.header, MACH_SEND_MSG, messageHeaderOut.header.msgh_size, 0, messageHeaderOut.header.msgh_local_port, 0, MACH_PORT_NULL);
- RELEASE_ASSERT(messageResult == KERN_SUCCESS);
- } else {
- dataLogLn("Failed to receive mach message due to ", mach_error_string(messageResult));
- RELEASE_ASSERT_NOT_REACHED();
- }
+ kern_return_t kr = mach_msg_server_once(
+ mach_exc_server, maxMessageSize, exceptionPort, MACH_MSG_TIMEOUT_NONE);
+ RELEASE_ASSERT(kr == KERN_SUCCESS);
});
+ // No need for a cancel handler because we never destroy exceptionPort.
+
dispatch_resume(source);
});
}
