MIPS,prctl: add PR_[GS]ET_FP_MODE prctl options for MIPS

Userland code may be built using an ABI which permits linking to objects
that have more restrictive floating point requirements. For example,
userland code may be built to target the O32 FPXX ABI. Such code may be
linked with other FPXX code, or code built for either one of the more
restrictive FP32 or FP64. When linking with more restrictive code, the
overall requirement of the process becomes that of the more restrictive
code. The kernel has no way to know in advance which mode the process
will need to be executed in, and indeed it may need to change during
execution. The dynamic loader is the only code which will know the
overall required mode, and so it needs to have a means to instruct the
kernel to switch the FP mode of the process.

This patch introduces 2 new options to the prctl syscall which provide
such a capability. The FP mode of the process is represented as a
simple bitmask combining a number of mode bits mirroring those present
in the hardware. Userland can either retrieve the current FP mode of
the process:

  mode = prctl(PR_GET_FP_MODE);

or modify the current FP mode of the process:

  err = prctl(PR_SET_FP_MODE, new_mode);

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Matthew Fortune <matthew.fortune@imgtec.com>
Cc: Markos Chandras <markos.chandras@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/8899/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

2 files changed, 111 insertions, 0 deletions

diff --git a/arch/mips/kernel/process.c b/arch/mips/kernel/process.c
index eb76434..4677b4c 100644
--- a/arch/mips/kernel/process.c
+++ b/arch/mips/kernel/process.c
@@ -25,6 +25,7 @@
 #include <linux/completion.h>
 #include <linux/kallsyms.h>
 #include <linux/random.h>
+#include <linux/prctl.h>
 
 #include <asm/asm.h>
 #include <asm/bootinfo.h>
@@ -550,3 +551,94 @@ void arch_trigger_all_cpu_backtrace(bool include_self)
 {
 	smp_call_function(arch_dump_stack, NULL, 1);
 }
+
+int mips_get_process_fp_mode(struct task_struct *task)
+{
+	int value = 0;
+
+	if (!test_tsk_thread_flag(task, TIF_32BIT_FPREGS))
+		value |= PR_FP_MODE_FR;
+	if (test_tsk_thread_flag(task, TIF_HYBRID_FPREGS))
+		value |= PR_FP_MODE_FRE;
+
+	return value;
+}
+
+int mips_set_process_fp_mode(struct task_struct *task, unsigned int value)
+{
+	const unsigned int known_bits = PR_FP_MODE_FR | PR_FP_MODE_FRE;
+	unsigned long switch_count;
+	struct task_struct *t;
+
+	/* Check the value is valid */
+	if (value & ~known_bits)
+		return -EOPNOTSUPP;
+
+	/* Avoid inadvertently triggering emulation */
+	if ((value & PR_FP_MODE_FR) && cpu_has_fpu &&
+	    !(current_cpu_data.fpu_id & MIPS_FPIR_F64))
+		return -EOPNOTSUPP;
+	if ((value & PR_FP_MODE_FRE) && cpu_has_fpu && !cpu_has_fre)
+		return -EOPNOTSUPP;
+
+	/* Save FP & vector context, then disable FPU & MSA */
+	if (task->signal == current->signal)
+		lose_fpu(1);
+
+	/* Prevent any threads from obtaining live FP context */
+	atomic_set(&task->mm->context.fp_mode_switching, 1);
+	smp_mb__after_atomic();
+
+	/*
+	 * If there are multiple online CPUs then wait until all threads whose
+	 * FP mode is about to change have been context switched. This approach
+	 * allows us to only worry about whether an FP mode switch is in
+	 * progress when FP is first used in a tasks time slice. Pretty much all
+	 * of the mode switch overhead can thus be confined to cases where mode
+	 * switches are actually occuring. That is, to here. However for the
+	 * thread performing the mode switch it may take a while...
+	 */
+	if (num_online_cpus() > 1) {
+		spin_lock_irq(&task->sighand->siglock);
+
+		for_each_thread(task, t) {
+			if (t == current)
+				continue;
+
+			switch_count = t->nvcsw + t->nivcsw;
+
+			do {
+				spin_unlock_irq(&task->sighand->siglock);
+				cond_resched();
+				spin_lock_irq(&task->sighand->siglock);
+			} while ((t->nvcsw + t->nivcsw) == switch_count);
+		}
+
+		spin_unlock_irq(&task->sighand->siglock);
+	}
+
+	/*
+	 * There are now no threads of the process with live FP context, so it
+	 * is safe to proceed with the FP mode switch.
+	 */
+	for_each_thread(task, t) {
+		/* Update desired FP register width */
+		if (value & PR_FP_MODE_FR) {
+			clear_tsk_thread_flag(t, TIF_32BIT_FPREGS);
+		} else {
+			set_tsk_thread_flag(t, TIF_32BIT_FPREGS);
+			clear_tsk_thread_flag(t, TIF_MSA_CTX_LIVE);
+		}
+
+		/* Update desired FP single layout */
+		if (value & PR_FP_MODE_FRE)
+			set_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
+		else
+			clear_tsk_thread_flag(t, TIF_HYBRID_FPREGS);
+	}
+
+	/* Allow threads to use FP again */
+	atomic_set(&task->mm->context.fp_mode_switching, 0);
+
+	return 0;
+}
diff --git a/arch/mips/kernel/traps.c b/arch/mips/kernel/traps.c
index ad3d203..d5fbfb5 100644
--- a/arch/mips/kernel/traps.c
+++ b/arch/mips/kernel/traps.c
@@ -1134,10 +1134,29 @@ static int default_cu2_call(struct notifier_block *nfb, unsigned long action,
 	return NOTIFY_OK;
 }
 
+static int wait_on_fp_mode_switch(atomic_t *p)
+{
+	/*
+	 * The FP mode for this task is currently being switched. That may
+	 * involve modifications to the format of this tasks FP context which
+	 * make it unsafe to proceed with execution for the moment. Instead,
+	 * schedule some other task.
+	 */
+	schedule();
+	return 0;
+}
+
 static int enable_restore_fp_context(int msa)
 {
 	int err, was_fpu_owner, prior_msa;
 
+	/*
+	 * If an FP mode switch is currently underway, wait for it to
+	 * complete before proceeding.
+	 */
+	wait_on_atomic_t(&current->mm->context.fp_mode_switching,
+			 wait_on_fp_mode_switch, TASK_KILLABLE);
+
 	if (!used_math()) {
 		/* First time FP context user. */
 		preempt_disable();