TheRock/debug-tools/rocgdb/source/gdb/testsuite/gdb.rocm/resume-exception.exp

# Copyright 2024 Free Software Foundation, Inc.
# Copyright (C) 2024 Advanced Micro Devices, Inc. All rights reserved.

# This file is part of GDB.

# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.

# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.

# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.

# Test that after an exception, a wave remains halted when GDB resumes.
# For testing purposes, the runtime thread is stopped, so it will not
# terminate the process (call abort) as the result of an exception
# (signal) being raised.

load_lib rocm.exp

require allow_hip_tests

standard_testfile .cpp

if {[build_executable "failed to prepare $testfile" $testfile $srcfile \
	{debug hip}]} {
    return
}

# Imagine a sample code that looks like:
#
#   c = 1.0 / 0.0;             // exception (becomes GDB_SIGNAL_FPE)
#   __builtin_debugtrap ();    // a trap in its true sense
#
# The idea is to run the inferior in non-stop mode.  When an exception
# is raised, the GPU thread is stopped while the other threads, like the
# runtime thread, are still running.  At this point, the test will
# interrupt other threads as well and only let the GPU thread run in the
# background.  If everything is in order and the wave is halted, this
# should lead to no execution of the next instruction (s_trap 3).  On
# the other hand, if the trap signal is received by GDB, that'd be a
# failure of the test.  Stopping other threads, specially the runtime
# thread, is necessary.  Else, the GPU kernel could be terminated before
# the test could inspect the corresponding wave's status.  This abrupt
# termination would be triggered by a SIGABRT signal raised by the
# runtime as a response to EC_QUEUE_WAVE_MATH_ERROR notification of the
# exception.
#
# Remarks:
#
# - Why non-stop mode?
#   Because it allows us to control other threads, stop them at will, and
#   makes it possible to _only_ run the selected one, while having other
#   actors, specifically the runtime thread, stopped; lest it terminate
#   the kernel with a SIGABRT.
#
# - Quirks of non-stop mode:
#   When GDB reports a breakpoint, signal, etc., by default, it will
#   have Thread 1 as the chosen one.  It's incumbent upon us to switch
#   to the GPU thread (and know its number).  For that, we will discover
#   it by setting a breakpoint on a GPU kernel and record the thread
#   number when GDB reports hitting of that breakpoint.
#
# - Why an ALU exception and not something like SIGSEGV?
#   As of now, SIGSEGV is not a good option because when the driver
#   sends a notification about SIGSEGV to the runtime, it also evicts the
#   process from the GPU, effectively preventing the wave from running
#   irrespective of its HALT bit value.  Therefore, robbing us from the
#   opportunity of detecting any abnormalities in the wave's status.

# Confine non-stop mode only to this test.
save_vars { ::GDBFLAGS } {
    append ::GDBFLAGS " -ex \"set non-stop on\""
    clean_restart $::binfile
}

# Based on the output of "info threads", return the number of running threads.
# (Inspired by gdb.threads/linux-dp.exp)

proc get_nr_of_running_threads { } {
    set thread_count 0
    gdb_test_multiple "info threads" "count running threads" -lbl {
	-re "\r\n *$::decimal *Thread \[^\n\]* \\(running\\)" {
	    incr thread_count
	    exp_continue
	}
	-re -wrap "" {
	}
    }
    return $thread_count
}

# Stop other running threads.
# (Inspired by gdb.multi/multi-target-continue.exp)

proc interrupt_running_threads { } {
    set count [get_nr_of_running_threads]
    set ok 0

    gdb_test_multiple "interrupt -a" "stop all threads" {
	-re "$::gdb_prompt " {
	    for {set i 0} {$i < $count} {incr i} {
		gdb_test_multiple "" $gdb_test_name {
		    -re "Thread\[^\r\n\]*stopped\\." {
			set ok 1
		    }
		}
		if {!$ok} {
		    break
		}
	    }
	    gdb_assert $ok $gdb_test_name
	}
    }
}

with_rocm_gpu_lock {
    # Run to GPU thread.
    gdb_breakpoint "raise_fpe" "allow-pending"
    gdb_test_multiple "run" "" {
	-re -wrap ".*Thread ($::decimal) \[^\r\n]* hit Breakpoint .*" {
	    set gpu_thread $expect_out(1,string)
	    pass $gdb_test_name
	}
    }
    gdb_test "thread $gpu_thread" "raise_fpe.*" "select GPU thread"

    enable_alu_exception "float_div0"

    # "continue" the GPU kernel and expect a SIGFPE.
    gdb_test "cont" "Thread $gpu_thread.*received signal SIGFPE.*" "SIGFPE"

    interrupt_running_threads

    # "continue" the GPU thread in the background.
    gdb_test -no-prompt-anchor "continue&" ""
    # After a period of "no news", then the test will pass.
    gdb_test_multiple "" "check GPU wave is halted" {
	-re ".*received signal SIGTRAP, " {
	    fail $gdb_test_name
	}
	-timeout 2
	timeout {
	    pass $gdb_test_name
	}
    }
}