All of lore.kernel.org
 help / color / mirror / Atom feed
From: Mika Kuoppala <mika.kuoppala@linux.intel.com>
To: Chris Wilson <chris@chris-wilson.co.uk>, intel-gfx@lists.freedesktop.org
Cc: Hans de Goede <hdegoede@redhat.com>
Subject: Re: [PATCH v2 01/11] drm/i915: Disable preemption and sleeping while using the punit sideband
Date: Mon, 15 Jan 2018 14:04:40 +0200	[thread overview]
Message-ID: <87zi5fiazr.fsf@gaia.fi.intel.com> (raw)
In-Reply-To: <20180115084723.2221-2-chris@chris-wilson.co.uk>

Chris Wilson <chris@chris-wilson.co.uk> writes:

> While we talk to the punit over its sideband, we need to prevent the cpu
> from sleeping in order to prevent a potential machine hang.
>
> Note that by itself, it appears that pm_qos_update_request (via
> intel_idle) doesn't provide a sufficient barrier to ensure that all core
> are indeed awake (out of Cstate) and that the package is awake. To do so,
> we need to supplement the pm_qos with a manual ping on_each_cpu.
>
> v2: Restrict the heavy-weight wakeup to just the ISOF_PORT_PUNIT, there
> is insufficient evidence to implicate a wider problem atm. Similarly,
> restrict the w/a to Valleyview, as Cherryview doesn't have an angry cadre
> of users.
>

One datapoint about the v1 of this patch with the cpu ping in it.
The j1900 byt did end up with system hang during weekend with
drm-tip + v1.

:(
-Mika

> The working theory, courtesy of Ville and Hans, is the issue lies within
> the power delivery and so is likely to be unit and board specific and
> occurs when both the unit/fw require extra power at the same time as the
> cpu package is changing its own power state.
>
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=109051
> References: https://bugs.freedesktop.org/show_bug.cgi?id=102657
> References: https://bugzilla.kernel.org/show_bug.cgi?id=195255
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
> Cc: Hans de Goede <hdegoede@redhat.com>
> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
> ---
>  drivers/gpu/drm/i915/i915_drv.c       |  6 +++
>  drivers/gpu/drm/i915/i915_drv.h       |  1 +
>  drivers/gpu/drm/i915/intel_sideband.c | 89 +++++++++++++++++++++++++++--------
>  3 files changed, 77 insertions(+), 19 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
> index 6c8da9d20c33..d4b90cc0130b 100644
> --- a/drivers/gpu/drm/i915/i915_drv.c
> +++ b/drivers/gpu/drm/i915/i915_drv.c
> @@ -902,6 +902,9 @@ static int i915_driver_init_early(struct drm_i915_private *dev_priv,
>  	spin_lock_init(&dev_priv->uncore.lock);
>  
>  	mutex_init(&dev_priv->sb_lock);
> +	pm_qos_add_request(&dev_priv->sb_qos,
> +			   PM_QOS_CPU_DMA_LATENCY, PM_QOS_DEFAULT_VALUE);
> +
>  	mutex_init(&dev_priv->modeset_restore_lock);
>  	mutex_init(&dev_priv->av_mutex);
>  	mutex_init(&dev_priv->wm.wm_mutex);
> @@ -953,6 +956,9 @@ static void i915_driver_cleanup_early(struct drm_i915_private *dev_priv)
>  	intel_irq_fini(dev_priv);
>  	i915_workqueues_cleanup(dev_priv);
>  	i915_engines_cleanup(dev_priv);
> +
> +	pm_qos_remove_request(&dev_priv->sb_qos);
> +	mutex_destroy(&dev_priv->sb_lock);
>  }
>  
>  static int i915_mmio_setup(struct drm_i915_private *dev_priv)
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index c42015b05b47..d95d8c3d04aa 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1886,6 +1886,7 @@ struct drm_i915_private {
>  
>  	/* Sideband mailbox protection */
>  	struct mutex sb_lock;
> +	struct pm_qos_request sb_qos;
>  
>  	/** Cached value of IMR to avoid reads in updating the bitfield */
>  	union {
> diff --git a/drivers/gpu/drm/i915/intel_sideband.c b/drivers/gpu/drm/i915/intel_sideband.c
> index 75c872bb8cc9..d56eda33734e 100644
> --- a/drivers/gpu/drm/i915/intel_sideband.c
> +++ b/drivers/gpu/drm/i915/intel_sideband.c
> @@ -22,6 +22,8 @@
>   *
>   */
>  
> +#include <asm/iosf_mbi.h>
> +
>  #include "i915_drv.h"
>  #include "intel_drv.h"
>  
> @@ -39,18 +41,48 @@
>  /* Private register write, double-word addressing, non-posted */
>  #define SB_CRWRDA_NP	0x07
>  
> -static int vlv_sideband_rw(struct drm_i915_private *dev_priv, u32 devfn,
> -			   u32 port, u32 opcode, u32 addr, u32 *val)
> +static void ping(void *info)
>  {
> -	u32 cmd, be = 0xf, bar = 0;
> -	bool is_read = (opcode == SB_MRD_NP || opcode == SB_CRRDDA_NP);
> +}
>  
> -	cmd = (devfn << IOSF_DEVFN_SHIFT) | (opcode << IOSF_OPCODE_SHIFT) |
> -		(port << IOSF_PORT_SHIFT) | (be << IOSF_BYTE_ENABLES_SHIFT) |
> -		(bar << IOSF_BAR_SHIFT);
> +static void __vlv_punit_get(struct drm_i915_private *dev_priv)
> +{
> +	iosf_mbi_punit_acquire();
>  
> -	WARN_ON(!mutex_is_locked(&dev_priv->sb_lock));
> +	/*
> +	 * Prevent the cpu from sleeping while we use this sideband, otherwise
> +	 * the punit may cause a machine hang. The issue appears to be isolated
> +	 * with changing the power state of the CPU package while changing
> +	 * the power state via the punit, and we have only observed it
> +	 * reliably on 4-core Baytail systems suggesting the issue is in the
> +	 * power delivery mechanism and likely to be be board/function
> +	 * specific. Hence we presume the workaround needs only be applied
> +	 * to the Valleyview P-unit and not all sideband communications.
> +	 */
> +	if (IS_VALLEYVIEW(dev_priv)) {
> +		pm_qos_update_request(&dev_priv->sb_qos, 0);
> +		on_each_cpu(ping, NULL, 1);
> +	}
> +}
> +
> +static void __vlv_punit_put(struct drm_i915_private *dev_priv)
> +{
> +	if (IS_VALLEYVIEW(dev_priv))
> +		pm_qos_update_request(&dev_priv->sb_qos, PM_QOS_DEFAULT_VALUE);
>  
> +	iosf_mbi_punit_release();
> +}
> +
> +static int vlv_sideband_rw(struct drm_i915_private *dev_priv,
> +			   u32 devfn, u32 port, u32 opcode,
> +			   u32 addr, u32 *val)
> +{
> +	const bool is_read = (opcode == SB_MRD_NP || opcode == SB_CRRDDA_NP);
> +	int err;
> +
> +	lockdep_assert_held(&dev_priv->sb_lock);
> +
> +	/* Flush the previous comms, just in case it failed last time. */
>  	if (intel_wait_for_register(dev_priv,
>  				    VLV_IOSF_DOORBELL_REQ, IOSF_SB_BUSY, 0,
>  				    5)) {
> @@ -59,22 +91,33 @@ static int vlv_sideband_rw(struct drm_i915_private *dev_priv, u32 devfn,
>  		return -EAGAIN;
>  	}
>  
> -	I915_WRITE(VLV_IOSF_ADDR, addr);
> -	I915_WRITE(VLV_IOSF_DATA, is_read ? 0 : *val);
> -	I915_WRITE(VLV_IOSF_DOORBELL_REQ, cmd);
> -
> -	if (intel_wait_for_register(dev_priv,
> -				    VLV_IOSF_DOORBELL_REQ, IOSF_SB_BUSY, 0,
> -				    5)) {
> +	preempt_disable();
> +
> +	I915_WRITE_FW(VLV_IOSF_ADDR, addr);
> +	I915_WRITE_FW(VLV_IOSF_DATA, is_read ? 0 : *val);
> +	I915_WRITE_FW(VLV_IOSF_DOORBELL_REQ,
> +		      (devfn << IOSF_DEVFN_SHIFT) |
> +		      (opcode << IOSF_OPCODE_SHIFT) |
> +		      (port << IOSF_PORT_SHIFT) |
> +		      (0xf << IOSF_BYTE_ENABLES_SHIFT) |
> +		      (0 << IOSF_BAR_SHIFT) |
> +		      IOSF_SB_BUSY);
> +
> +	if (__intel_wait_for_register_fw(dev_priv,
> +					 VLV_IOSF_DOORBELL_REQ, IOSF_SB_BUSY, 0,
> +					 10000, 0, NULL) == 0) {
> +		if (is_read)
> +			*val = I915_READ_FW(VLV_IOSF_DATA);
> +		err = 0;
> +	} else {
>  		DRM_DEBUG_DRIVER("IOSF sideband finish wait (%s) timed out\n",
>  				 is_read ? "read" : "write");
> -		return -ETIMEDOUT;
> +		err = -ETIMEDOUT;
>  	}
>  
> -	if (is_read)
> -		*val = I915_READ(VLV_IOSF_DATA);
> +	preempt_enable();
>  
> -	return 0;
> +	return err;
>  }
>  
>  u32 vlv_punit_read(struct drm_i915_private *dev_priv, u32 addr)
> @@ -84,8 +127,12 @@ u32 vlv_punit_read(struct drm_i915_private *dev_priv, u32 addr)
>  	WARN_ON(!mutex_is_locked(&dev_priv->pcu_lock));
>  
>  	mutex_lock(&dev_priv->sb_lock);
> +	__vlv_punit_get(dev_priv);
> +
>  	vlv_sideband_rw(dev_priv, PCI_DEVFN(0, 0), IOSF_PORT_PUNIT,
>  			SB_CRRDDA_NP, addr, &val);
> +
> +	__vlv_punit_put(dev_priv);
>  	mutex_unlock(&dev_priv->sb_lock);
>  
>  	return val;
> @@ -98,8 +145,12 @@ int vlv_punit_write(struct drm_i915_private *dev_priv, u32 addr, u32 val)
>  	WARN_ON(!mutex_is_locked(&dev_priv->pcu_lock));
>  
>  	mutex_lock(&dev_priv->sb_lock);
> +	__vlv_punit_get(dev_priv);
> +
>  	err = vlv_sideband_rw(dev_priv, PCI_DEVFN(0, 0), IOSF_PORT_PUNIT,
>  			      SB_CRWRDA_NP, addr, &val);
> +
> +	__vlv_punit_put(dev_priv);
>  	mutex_unlock(&dev_priv->sb_lock);
>  
>  	return err;
> -- 
> 2.15.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

  reply	other threads:[~2018-01-15 12:05 UTC|newest]

Thread overview: 26+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-01-15  8:47 Vlv punit w/a (take two) Chris Wilson
2018-01-15  8:47 ` [PATCH v2 01/11] drm/i915: Disable preemption and sleeping while using the punit sideband Chris Wilson
2018-01-15 12:04   ` Mika Kuoppala [this message]
2018-01-15 12:13     ` Chris Wilson
2018-01-16 15:27       ` Mika Kuoppala
2018-01-15 12:21     ` Chris Wilson
2018-01-15 13:07       ` Hans de Goede
2018-01-15 13:43       ` Mika Kuoppala
2018-01-22 12:54     ` Chris Wilson
2018-01-15  8:47 ` [PATCH v2 02/11] drm/i915: Lift acquiring the vlv punit magic to a common sb-get Chris Wilson
2018-01-15 10:36   ` Mika Kuoppala
2018-01-15  8:47 ` [PATCH v2 03/11] drm/i915: Lift sideband locking for vlv_punit_(read|write) Chris Wilson
2018-01-15  8:47 ` [PATCH v2 04/11] drm/i915: Reduce RPS update frequency on Valleyview/Cherryview Chris Wilson
2018-01-15  8:47 ` [PATCH v2 05/11] Revert "drm/i915: Avoid tweaking evaluation thresholds on Baytrail v3" Chris Wilson
2018-01-15  8:47 ` [PATCH v2 06/11] drm/i915: Replace pcu_lock with sb_lock Chris Wilson
2018-01-15  8:47 ` [PATCH v2 07/11] drm/i915: Separate sideband declarations to intel_sideband.h Chris Wilson
2018-01-15  8:47 ` [PATCH v2 08/11] drm/i915: Merge sbi read/write into a single accessor Chris Wilson
2018-01-15  8:47 ` [PATCH v2 09/11] drm/i915: Merge sandybride_pcode_(read|write) Chris Wilson
2018-01-15  8:47 ` [PATCH v2 10/11] drm/i915: Move sandybride pcode access to intel_sideband.c Chris Wilson
2018-01-19  8:21   ` Mika Kuoppala
2018-01-19  8:33     ` Chris Wilson
2018-01-15  8:47 ` [PATCH v2 11/11] drm/i915: Avoid waitboosting on the active request Chris Wilson
2018-01-18  9:49   ` Joonas Lahtinen
2018-01-15  9:19 ` ✓ Fi.CI.BAT: success for series starting with [v2,01/11] drm/i915: Disable preemption and sleeping while using the punit sideband Patchwork
2018-01-15 10:23 ` ✗ Fi.CI.IGT: failure " Patchwork
2018-01-15 10:27   ` Chris Wilson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=87zi5fiazr.fsf@gaia.fi.intel.com \
    --to=mika.kuoppala@linux.intel.com \
    --cc=chris@chris-wilson.co.uk \
    --cc=hdegoede@redhat.com \
    --cc=intel-gfx@lists.freedesktop.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.