| 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
| ||
|
Date: Mon, 31 Oct 2016 16:32:24 +0000
From: Karl Beldan <karl.beldan@...il.com>
To: Bartosz Golaszewski <bgolaszewski@...libre.com>
Cc: Jyri Sarha <jsarha@...com>, Tomi Valkeinen <tomi.valkeinen@...com>,
David Airlie <airlied@...ux.ie>,
Kevin Hilman <khilman@...libre.com>,
Michael Turquette <mturquette@...libre.com>,
Sekhar Nori <nsekhar@...com>,
Peter Ujfalusi <peter.ujfalusi@...com>,
LKML <linux-kernel@...r.kernel.org>,
linux-drm <dri-devel@...ts.freedesktop.org>,
arm-soc <linux-arm-kernel@...ts.infradead.org>
Subject: Re: [PATCH v5 2/2] drm: tilcdc: clear the sync lost bit in crtc isr
Hi,
On Mon, Oct 31, 2016 at 2:19 PM, Bartosz Golaszewski
<bgolaszewski@...libre.com> wrote:
> The frame synchronization error happens when the DMA engine attempts
> to read what it believes to be the first word of the video buffer but
> it cannot be recognized as such or when the LCDC is starved of data
> due to insufficient bandwidth of the system interconnect.
>
This also happens when the framebuffer boundaries are "incorrect", eg
when flipping while the crtc is enabled. The driver doesn't handle it yet,
so working around it is made via toggling LCDC_RASTER_ENABLE as
you put in this change, it is worth noting because that's how modetest
for example can "seem" to work with v1 when handling SYNC_LOST
and is partly why the "vsynced page flipping" of modetest won't work
either with v1 ATM. I haven't looked at the different trees since early
september but my guess would be that nobody has reworked this yet.
> On some SoCs (notably: da850) the memory settings do not meet the
> LCDC throughput requirements even after increasing the memory
> controller command re-ordering and the LDCD master peripheral
> priority, sometimes causing the sync lost error (typically when
> changing the resolution).
>
> When the sync lost error occurs, simply reset the input FIFO in the
> DMA controller unless a sync lost flood is detected in which case
> disable the interrupt.
>
> Signed-off-by: Bartosz Golaszewski <bgolaszewski@...libre.com>
> ---
> drivers/gpu/drm/tilcdc/tilcdc_crtc.c | 50 ++++++++++++++++++++++++++----------
> drivers/gpu/drm/tilcdc/tilcdc_regs.h | 1 +
> 2 files changed, 37 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/tilcdc/tilcdc_crtc.c b/drivers/gpu/drm/tilcdc/tilcdc_crtc.c
> index 937697d..c4c6323 100644
> --- a/drivers/gpu/drm/tilcdc/tilcdc_crtc.c
> +++ b/drivers/gpu/drm/tilcdc/tilcdc_crtc.c
> @@ -163,7 +163,7 @@ static void tilcdc_crtc_enable_irqs(struct drm_device *dev)
>
> if (priv->rev == 1) {
> tilcdc_set(dev, LCDC_RASTER_CTRL_REG,
> - LCDC_V1_UNDERFLOW_INT_ENA);
> + LCDC_V1_UNDERFLOW_INT_ENA | LCDC_V1_SYNC_LOST_ENA);
> tilcdc_set(dev, LCDC_DMA_CTRL_REG,
> LCDC_V1_END_OF_FRAME_INT_ENA);
> } else {
> @@ -181,7 +181,9 @@ static void tilcdc_crtc_disable_irqs(struct drm_device *dev)
> /* disable irqs that we might have enabled: */
> if (priv->rev == 1) {
> tilcdc_clear(dev, LCDC_RASTER_CTRL_REG,
> - LCDC_V1_UNDERFLOW_INT_ENA | LCDC_V1_PL_INT_ENA);
> + LCDC_V1_UNDERFLOW_INT_ENA |
> + LCDC_V1_PL_INT_ENA |
> + LCDC_V1_SYNC_LOST_ENA);
> tilcdc_clear(dev, LCDC_DMA_CTRL_REG,
> LCDC_V1_END_OF_FRAME_INT_ENA);
> } else {
> @@ -885,24 +887,44 @@ irqreturn_t tilcdc_crtc_irq(struct drm_crtc *crtc)
> wake_up(&tilcdc_crtc->frame_done_wq);
> }
>
> - if (stat & LCDC_SYNC_LOST) {
> - dev_err_ratelimited(dev->dev, "%s(0x%08x): Sync lost",
> - __func__, stat);
> - tilcdc_crtc->frame_intact = false;
> - if (tilcdc_crtc->sync_lost_count++ >
> - SYNC_LOST_COUNT_LIMIT) {
> - dev_err(dev->dev, "%s(0x%08x): Sync lost flood detected, disabling the interrupt", __func__, stat);
> - tilcdc_write(dev, LCDC_INT_ENABLE_CLR_REG,
> - LCDC_SYNC_LOST);
> - }
> - }
> -
> /* Indicate to LCDC that the interrupt service routine has
> * completed, see 13.3.6.1.6 in AM335x TRM.
> */
> tilcdc_write(dev, LCDC_END_OF_INT_IND_REG, 0);
> }
>
> + if (stat & LCDC_SYNC_LOST) {
> + dev_err_ratelimited(dev->dev, "%s(0x%08x): Sync lost",
> + __func__, stat);
> +
> + tilcdc_crtc->frame_intact = false;
> + if (tilcdc_crtc->sync_lost_count++ > SYNC_LOST_COUNT_LIMIT) {
> + dev_err(dev->dev,
> + "%s(0x%08x): Sync lost flood detected, disabling the interrupt",
> + __func__, stat);
> + if (priv->rev == 2)
> + tilcdc_write(dev, LCDC_INT_ENABLE_CLR_REG,
> + LCDC_SYNC_LOST);
> + else if (priv->rev == 1)
> + tilcdc_clear(dev, LCDC_RASTER_CTRL_REG,
> + LCDC_V1_SYNC_LOST_ENA);
> + }
> +
> + if (priv->rev == 2) {
> + /*
> + * Indicate to LCDC that the interrupt service routine
> + * has completed, see 13.3.6.1.6 in AM335x TRM.
> + */
> + tilcdc_write(dev, LCDC_END_OF_INT_IND_REG, 0);
> + } else if (priv->rev == 1) {
> + /* Reset the input FIFO in the DMA controller. */
> + tilcdc_clear(dev,
> + LCDC_RASTER_CTRL_REG, LCDC_RASTER_ENABLE);
> + tilcdc_set(dev,
> + LCDC_RASTER_CTRL_REG, LCDC_RASTER_ENABLE);
Here you are leaving the RASTER_ENABLE bit set unconditionnally.
It is disabled in tilcdc_crtc_disable with the irqs on (for v2 to handle
the last vblank), if the controller loses sync in the window between
disabling it and disabling the irqs in tilcdc_crtc_disable it is likely to
be problematic for the next call to tilcdc_crtc_enable.
Regards,
Karl
Powered by blists - more mailing lists