linux-kernel - Re: [PATCH 07/23] drm: omapdrm: crtc: switch pending variable to atomic bitset

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <10595698.j3sjVXcOKr@avalon>
Date:	Sat, 26 Mar 2016 18:30:02 +0200
From:	Laurent Pinchart <laurent.pinchart@...asonboard.com>
To:	Sebastian Reichel <sre@...nel.org>
Cc:	Tony Lindgren <tony@...mide.com>,
	Aaro Koskinen <aaro.koskinen@....fi>,
	Tomi Valkeinen <tomi.valkeinen@...com>,
	David Airlie <airlied@...ux.ie>, linux-omap@...r.kernel.org,
	dri-devel@...ts.freedesktop.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 07/23] drm: omapdrm: crtc: switch pending variable to atomic bitset

Hi Sebastian,

Thank you for the patch.

On Tuesday 08 Mar 2016 17:39:39 Sebastian Reichel wrote:
> Having the pending variable available as atomic bit helps
> with the later addition of manually updated display support.
> 
> Signed-off-by: Sebastian Reichel <sre@...nel.org>
> ---
>  drivers/gpu/drm/omapdrm/omap_crtc.c | 31 +++++++++++++++++--------------
>  1 file changed, 17 insertions(+), 14 deletions(-)
> 
> diff --git a/drivers/gpu/drm/omapdrm/omap_crtc.c
> b/drivers/gpu/drm/omapdrm/omap_crtc.c index 2ed0754ed19e..5ef27664bcfa
> 100644
> --- a/drivers/gpu/drm/omapdrm/omap_crtc.c
> +++ b/drivers/gpu/drm/omapdrm/omap_crtc.c
> @@ -28,6 +28,11 @@
> 
>  #define to_omap_crtc(x) container_of(x, struct omap_crtc, base)
> 
> +enum omap_crtc_state {

I find that using an enum and calling it omap_crtc_state is confusing, it 
seems to imply that the CRTC state will be one of the enumerated values, while 
the values are actually non-exclusive bits in a bitmask.

> +	crtc_enabled	= 0,

You don't seem to be using this bit in this patch, you can define it in patch 
08/23.

> +	crtc_pending	= 1

Please name this OMAP_CRTC_STATE_PENDING.

> +};

Please add a short description of each bit in a comment above the enum.

> +
>  struct omap_crtc {
>  	struct drm_crtc base;
> 
> @@ -49,7 +54,7 @@ struct omap_crtc {
> 
>  	bool ignore_digit_sync_lost;
> 
> -	bool pending;
> +	unsigned long state;
>  	wait_queue_head_t pending_wait;
>  };
> 
> @@ -81,7 +86,7 @@ int omap_crtc_wait_pending(struct drm_crtc *crtc)
>  	struct omap_crtc *omap_crtc = to_omap_crtc(crtc);
> 
>  	return wait_event_timeout(omap_crtc->pending_wait,
> -				  !omap_crtc->pending,
> +				  !test_bit(crtc_pending, &omap_crtc->state),
>  				  msecs_to_jiffies(50));
>  }
> 
> @@ -311,10 +316,8 @@ static void omap_crtc_vblank_irq(struct omap_drm_irq
> *irq, uint32_t irqstatus)
> 
>  	__omap_irq_unregister(dev, &omap_crtc->vblank_irq);
> 
> -	rmb();
> -	WARN_ON(!omap_crtc->pending);
> -	omap_crtc->pending = false;
> -	wmb();
> +	if (!test_and_clear_bit(crtc_pending, &omap_crtc->state))
> +		dev_warn(dev->dev, "pending bit was not set in vblank irq");

Documentation/atomic_ops.txt states that the atomic bit ops must be atomic and 
include memory barriers, but I'm confused by the ARM implementation.

The constant bit number version ____atomic_test_and_clear_bit() uses 
raw_local_irq_save() and raw_low_irq_restore() for synchronization, which 
expand to arch_local_irq_save() and arch_local_irq_restore(). On ARMv6 and 
newer, the functions are defined as

static inline unsigned long arch_local_irq_save(void)
{
        unsigned long flags;

        asm volatile(
                "       mrs     %0, " IRQMASK_REG_NAME_R "      @ 
arch_local_irq_save\n"
                "       cpsid   i"
                : "=r" (flags) : : "memory", "cc");
        return flags;
}

and

static inline void arch_local_irq_restore(unsigned long flags)
{
        asm volatile(
                "       msr     " IRQMASK_REG_NAME_W ", %0      @ 
local_irq_restore"
                :
                : "r" (flags)
                : "memory", "cc");
}

I'm probably missing something obvious, but I don't see the memory barriers 
:-/

>  	/* wake up userspace */
>  	omap_crtc_complete_page_flip(&omap_crtc->base);
> @@ -351,13 +354,12 @@ static bool omap_crtc_mode_fixup(struct drm_crtc
> *crtc, static void omap_crtc_enable(struct drm_crtc *crtc)
>  {
>  	struct omap_crtc *omap_crtc = to_omap_crtc(crtc);
> +	struct drm_device *dev = crtc->dev;
> 
>  	DBG("%s", omap_crtc->name);
> 
> -	rmb();
> -	WARN_ON(omap_crtc->pending);
> -	omap_crtc->pending = true;
> -	wmb();
> +	if (test_and_set_bit(crtc_pending, &omap_crtc->state))
> +		dev_warn(dev->dev, "crtc enable while pending bit set!");
> 
>  	omap_irq_register(crtc->dev, &omap_crtc->vblank_irq);
> 
> @@ -397,6 +399,7 @@ static void omap_crtc_atomic_flush(struct drm_crtc
> *crtc, struct drm_crtc_state *old_crtc_state) {
>  	struct omap_crtc *omap_crtc = to_omap_crtc(crtc);
> +	struct drm_device *dev = crtc->dev;
> 
>  	WARN_ON(omap_crtc->vblank_irq.registered);
> 
> @@ -404,10 +407,8 @@ static void omap_crtc_atomic_flush(struct drm_crtc
> *crtc,
> 
>  		DBG("%s: GO", omap_crtc->name);
> 
> -		rmb();
> -		WARN_ON(omap_crtc->pending);
> -		omap_crtc->pending = true;
> -		wmb();
> +		if (test_and_set_bit(crtc_pending, &omap_crtc->state))
> +			dev_warn(dev->dev, "atomic flush while pending bit set!");
> 
>  		dispc_mgr_go(omap_crtc->channel);
>  		omap_irq_register(crtc->dev, &omap_crtc->vblank_irq);
> @@ -509,6 +510,8 @@ struct drm_crtc *omap_crtc_init(struct drm_device *dev,
> 
>  	init_waitqueue_head(&omap_crtc->pending_wait);
> 
> +	omap_crtc->state = 0;
> +
>  	omap_crtc->channel = channel;
>  	omap_crtc->name = channel_names[channel];

-- 
Regards,

Laurent Pinchart