[RFC,v3,07/26] um lkl: interrupt support

From: Octavian Purdila <tavi.purdila@gmail.com>

Add APIs that allows the host to reserve and free and interrupt number
and also to trigger an interrupt.

The trigger operation will simply store the interrupt data in
queue. The interrupt handler is run later, at the first opportunity it
has to avoid races with any kernel threads.

Currently, interrupts are run on the first interrupt enable operation
if interrupts are disabled and if we are not already in interrupt
context.

When triggering an interrupt, it uses GCC's built-in functions for
atomic memory access to synchronize and simple boolean flags.

Cc: Michael Zimmermann <sigmaepsilon92@gmail.com>
Signed-off-by: Hajime Tazaki <thehajime@gmail.com>
Signed-off-by: Octavian Purdila <tavi.purdila@gmail.com>
---
 arch/um/lkl/include/asm/irq.h             |  13 ++
 arch/um/lkl/include/uapi/asm/irq.h        |  36 ++++
 arch/um/lkl/include/uapi/asm/sigcontext.h |  16 ++
 arch/um/lkl/kernel/irq.c                  | 193 ++++++++++++++++++++++
 4 files changed, 258 insertions(+)
 create mode 100644 arch/um/lkl/include/asm/irq.h
 create mode 100644 arch/um/lkl/include/uapi/asm/irq.h
 create mode 100644 arch/um/lkl/include/uapi/asm/sigcontext.h
 create mode 100644 arch/um/lkl/kernel/irq.c

On 05/02/2020 07:30, Hajime Tazaki wrote:
> From: Octavian Purdila <tavi.purdila@gmail.com>
> 
> Add APIs that allows the host to reserve and free and interrupt number
> and also to trigger an interrupt.
> 
> The trigger operation will simply store the interrupt data in
> queue. The interrupt handler is run later, at the first opportunity it
> has to avoid races with any kernel threads.
> 
> Currently, interrupts are run on the first interrupt enable operation
> if interrupts are disabled and if we are not already in interrupt
> context.
> 
> When triggering an interrupt, it uses GCC's built-in functions for
> atomic memory access to synchronize and simple boolean flags.
> 
> Cc: Michael Zimmermann <sigmaepsilon92@gmail.com>
> Signed-off-by: Hajime Tazaki <thehajime@gmail.com>
> Signed-off-by: Octavian Purdila <tavi.purdila@gmail.com>
> ---
>   arch/um/lkl/include/asm/irq.h             |  13 ++
>   arch/um/lkl/include/uapi/asm/irq.h        |  36 ++++
>   arch/um/lkl/include/uapi/asm/sigcontext.h |  16 ++
>   arch/um/lkl/kernel/irq.c                  | 193 ++++++++++++++++++++++
>   4 files changed, 258 insertions(+)
>   create mode 100644 arch/um/lkl/include/asm/irq.h
>   create mode 100644 arch/um/lkl/include/uapi/asm/irq.h
>   create mode 100644 arch/um/lkl/include/uapi/asm/sigcontext.h
>   create mode 100644 arch/um/lkl/kernel/irq.c
> 
> diff --git a/arch/um/lkl/include/asm/irq.h b/arch/um/lkl/include/asm/irq.h
> new file mode 100644
> index 000000000000..948fc54cb76c
> --- /dev/null
> +++ b/arch/um/lkl/include/asm/irq.h
> @@ -0,0 +1,13 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef _ASM_LKL_IRQ_H
> +#define _ASM_LKL_IRQ_H
> +
> +#define IRQ_STATUS_BITS		(sizeof(long) * 8)
> +#define NR_IRQS			((int)(IRQ_STATUS_BITS * IRQ_STATUS_BITS))
> +
> +void run_irqs(void);
> +void set_irq_pending(int irq);
> +
> +#include <uapi/asm/irq.h>
> +
> +#endif
> diff --git a/arch/um/lkl/include/uapi/asm/irq.h b/arch/um/lkl/include/uapi/asm/irq.h
> new file mode 100644
> index 000000000000..666628b233eb
> --- /dev/null
> +++ b/arch/um/lkl/include/uapi/asm/irq.h
> @@ -0,0 +1,36 @@
> +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
> +#ifndef _ASM_UAPI_LKL_IRQ_H
> +#define _ASM_UAPI_LKL_IRQ_H
> +
> +/**
> + * lkl_trigger_irq - generate an interrupt
> + *
> + * This function is used by the device host side to signal its Linux counterpart
> + * that some event happened.
> + *
> + * @irq - the irq number to signal
> + */
> +int lkl_trigger_irq(int irq);
> +
> +/**
> + * lkl_get_free_irq - find and reserve a free IRQ number
> + *
> + * This function is called by the host device code to find an unused IRQ number
> + * and reserved it for its own use.
> + *
> + * @user - a string to identify the user
> + * @returns - and irq number that can be used by request_irq or an negative
> + * value in case of an error
> + */
> +int lkl_get_free_irq(const char *user);
> +
> +/**
> + * lkl_put_irq - release an IRQ number previously obtained with lkl_get_free_irq
> + *
> + * @irq - irq number to release
> + * @user - string identifying the user; should be the same as the one passed to
> + * lkl_get_free_irq when the irq number was obtained
> + */
> +void lkl_put_irq(int irq, const char *name);
> +
> +#endif
> diff --git a/arch/um/lkl/include/uapi/asm/sigcontext.h b/arch/um/lkl/include/uapi/asm/sigcontext.h
> new file mode 100644
> index 000000000000..2f4848843d1d
> --- /dev/null
> +++ b/arch/um/lkl/include/uapi/asm/sigcontext.h
> @@ -0,0 +1,16 @@
> +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
> +#ifndef _ASM_UAPI_LKL_SIGCONTEXT_H
> +#define _ASM_UAPI_LKL_SIGCONTEXT_H
> +
> +#include <asm/ptrace.h>
> +
> +struct pt_regs {
> +	void *irq_data;
> +};
> +
> +struct sigcontext {
> +	struct pt_regs regs;
> +	unsigned long oldmask;
> +};
> +
> +#endif
> diff --git a/arch/um/lkl/kernel/irq.c b/arch/um/lkl/kernel/irq.c
> new file mode 100644
> index 000000000000..e3b59e46ca50
> --- /dev/null
> +++ b/arch/um/lkl/kernel/irq.c
> @@ -0,0 +1,193 @@
> +// SPDX-License-Identifier: GPL-2.0
> +#include <linux/kernel.h>
> +#include <linux/init.h>
> +#include <linux/irq.h>
> +#include <linux/hardirq.h>
> +#include <asm/irq_regs.h>
> +#include <linux/sched.h>
> +#include <linux/seq_file.h>
> +#include <linux/tick.h>
> +#include <asm/irqflags.h>
> +#include <asm/host_ops.h>
> +#include <asm/cpu.h>
> +
> +/*
> + * To avoid much overhead we use an indirect approach: the irqs are marked using
> + * a bitmap (array of longs) and a summary of the modified bits is kept in a
> + * separate "index" long - one bit for each sizeof(long). Thus we can support
> + * 4096 irqs on 64bit platforms and 1024 irqs on 32bit platforms.
> + *
> + * Whenever an irq is trigger both the array and the index is updated. To find
> + * which irqs were triggered we first search the index and then the
> + * corresponding part of the arrary.
> + */
> +static unsigned long irq_status[NR_IRQS / IRQ_STATUS_BITS];
> +static unsigned long irq_index_status;
> +
> +static inline unsigned long test_and_clear_irq_index_status(void)
> +{
> +	if (!irq_index_status)
> +		return 0;
> +	return __sync_fetch_and_and(&irq_index_status, 0);
> +}
> +
> +static inline unsigned long test_and_clear_irq_status(int index)
> +{
> +	if (!&irq_status[index])
> +		return 0;
> +	return __sync_fetch_and_and(&irq_status[index], 0);
> +}
> +
> +void set_irq_pending(int irq)
> +{
> +	int index = irq / IRQ_STATUS_BITS;
> +	int bit = irq % IRQ_STATUS_BITS;
> +
> +	__sync_fetch_and_or(&irq_status[index], BIT(bit));
> +	__sync_fetch_and_or(&irq_index_status, BIT(index));
> +}
> +
> +static struct irq_info {
> +	const char *user;
> +} irqs[NR_IRQS];
> +
> +static bool irqs_enabled;
> +
> +static struct pt_regs dummy;
> +
> +static void run_irq(int irq)
> +{
> +	unsigned long flags;
> +	struct pt_regs *old_regs = set_irq_regs((struct pt_regs *)&dummy);
> +
> +	/* interrupt handlers need to run with interrupts disabled */
> +	local_irq_save(flags);
> +	irq_enter();
> +	generic_handle_irq(irq);
> +	irq_exit();
> +	set_irq_regs(old_regs);
> +	local_irq_restore(flags);
> +}
> +
> +/**
> + * This function can be called from arbitrary host threads, so do not
> + * issue any Linux calls (e.g. prink) if lkl_cpu_get() was not issued
> + * before.
> + */
> +int lkl_trigger_irq(int irq)
> +{
> +	int ret;
> +
> +	if (!irq || irq > NR_IRQS)
> +		return -EINVAL;
> +
> +	ret = lkl_cpu_try_run_irq(irq);
> +	if (ret <= 0)
> +		return ret;
> +
> +	/*
> +	 * Since this can be called from Linux context (e.g. lkl_trigger_irq ->
> +	 * IRQ -> softirq -> lkl_trigger_irq) make sure we are actually allowed
> +	 * to run irqs at this point
> +	 */
> +	if (!irqs_enabled) {
> +		set_irq_pending(irq);
> +		lkl_cpu_put();
> +		return 0;
> +	}
> +
> +	run_irq(irq);
> +
> +	lkl_cpu_put();
> +
> +	return 0;

Isn't that just:

	if (irqs_enabled)
		run_irq(irq);
	else
		set_irq_pending(irq);

	lkl_cpu_put();

	return 0;

> +}
> +
> +static inline void for_each_bit(unsigned long word, void (*f)(int, int), int j)
> +{
> +	int i = 0;
> +
> +	while (word) {
> +		if (word & 1)
> +			f(i, j);
> +		word >>= 1;
> +		i++;
> +	}
> +}
> +
> +static inline void deliver_irq(int bit, int index)
> +{
> +	run_irq(index * IRQ_STATUS_BITS + bit);
> +}
> +
> +static inline void check_irq_status(int i, int unused)
> +{
> +	for_each_bit(test_and_clear_irq_status(i), deliver_irq, i);
> +}
> +
> +void run_irqs(void)
> +{
> +	for_each_bit(test_and_clear_irq_index_status(), check_irq_status, 0);
> +}
> +
> +int show_interrupts(struct seq_file *p, void *v)
> +{
> +	return 0;
> +}
> +
> +int lkl_get_free_irq(const char *user)
> +{
> +	int i;
> +	int ret = -EBUSY;
> +
> +	/* 0 is not a valid IRQ */
> +	for (i = 1; i < NR_IRQS; i++) {
> +		if (!irqs[i].user) {
> +			irqs[i].user = user;
> +			irq_set_chip_and_handler(i, &dummy_irq_chip,
> +						 handle_simple_irq);
> +			ret = i;
> +			break;
> +		}
> +	}
> +
> +	return ret;
> +}
> +
> +void lkl_put_irq(int i, const char *user)
> +{
> +	if (!irqs[i].user || strcmp(irqs[i].user, user) != 0) {
> +		WARN("%s tried to release %s's irq %d", user, irqs[i].user, i);
> +		return;
> +	}
> +
> +	irqs[i].user = NULL;
> +}
> +
> +unsigned long arch_local_save_flags(void)
> +{
> +	return irqs_enabled;
> +}
> +
> +void arch_local_irq_restore(unsigned long flags)
> +{
> +	if (flags == ARCH_IRQ_ENABLED && irqs_enabled == ARCH_IRQ_DISABLED &&
> +	    !in_interrupt())
> +		run_irqs();
> +	irqs_enabled = flags;
> +}
> +
> +void init_IRQ(void)
> +{
> +	int i;
> +
> +	for (i = 0; i < NR_IRQS; i++)
> +		irq_set_chip_and_handler(i, &dummy_irq_chip, handle_simple_irq);
> +
> +	pr_info("lkl: irqs initialized\n");
> +}
> +
> +void cpu_yield_to_irqs(void)
> +{
> +	cpu_relax();
> +}
>

On Wed, 05 Feb 2020 19:47:36 +0900,
Anton Ivanov wrote:

> > +/**
> > + * This function can be called from arbitrary host threads, so do not
> > + * issue any Linux calls (e.g. prink) if lkl_cpu_get() was not issued
> > + * before.
> > + */
> > +int lkl_trigger_irq(int irq)
> > +{
> > +	int ret;
> > +
> > +	if (!irq || irq > NR_IRQS)
> > +		return -EINVAL;
> > +
> > +	ret = lkl_cpu_try_run_irq(irq);
> > +	if (ret <= 0)
> > +		return ret;
> > +
> > +	/*
> > +	 * Since this can be called from Linux context (e.g. lkl_trigger_irq ->
> > +	 * IRQ -> softirq -> lkl_trigger_irq) make sure we are actually allowed
> > +	 * to run irqs at this point
> > +	 */
> > +	if (!irqs_enabled) {
> > +		set_irq_pending(irq);
> > +		lkl_cpu_put();
> > +		return 0;
> > +	}
> > +
> > +	run_irq(irq);
> > +
> > +	lkl_cpu_put();
> > +
> > +	return 0;
> 
> Isn't that just:
> 
> 	if (irqs_enabled)
> 		run_irq(irq);
> 	else
> 		set_irq_pending(irq);
> 
> 	lkl_cpu_put();
> 
> 	return 0;

Thanks, this is much cleaner.  I will fix this in the next turn.

-- Hajime