2013-07-20 03:36:52 +08:00
|
|
|
/*
|
|
|
|
* Copyright © 2013 Intel Corporation
|
|
|
|
*
|
|
|
|
* Permission is hereby granted, free of charge, to any person obtaining a
|
|
|
|
* copy of this software and associated documentation files (the "Software"),
|
|
|
|
* to deal in the Software without restriction, including without limitation
|
|
|
|
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
|
|
|
|
* and/or sell copies of the Software, and to permit persons to whom the
|
|
|
|
* Software is furnished to do so, subject to the following conditions:
|
|
|
|
*
|
|
|
|
* The above copyright notice and this permission notice (including the next
|
|
|
|
* paragraph) shall be included in all copies or substantial portions of the
|
|
|
|
* Software.
|
|
|
|
*
|
|
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
|
|
|
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
|
|
|
|
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
|
|
|
|
* IN THE SOFTWARE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "i915_drv.h"
|
|
|
|
#include "intel_drv.h"
|
|
|
|
|
2015-01-16 17:34:35 +08:00
|
|
|
#include <linux/pm_runtime.h>
|
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
#define FORCEWAKE_ACK_TIMEOUT_MS 2
|
|
|
|
|
2013-07-20 03:36:53 +08:00
|
|
|
#define __raw_i915_read8(dev_priv__, reg__) readb((dev_priv__)->regs + (reg__))
|
|
|
|
#define __raw_i915_write8(dev_priv__, reg__, val__) writeb(val__, (dev_priv__)->regs + (reg__))
|
|
|
|
|
|
|
|
#define __raw_i915_read16(dev_priv__, reg__) readw((dev_priv__)->regs + (reg__))
|
|
|
|
#define __raw_i915_write16(dev_priv__, reg__, val__) writew(val__, (dev_priv__)->regs + (reg__))
|
|
|
|
|
|
|
|
#define __raw_i915_read32(dev_priv__, reg__) readl((dev_priv__)->regs + (reg__))
|
|
|
|
#define __raw_i915_write32(dev_priv__, reg__, val__) writel(val__, (dev_priv__)->regs + (reg__))
|
|
|
|
|
|
|
|
#define __raw_i915_read64(dev_priv__, reg__) readq((dev_priv__)->regs + (reg__))
|
|
|
|
#define __raw_i915_write64(dev_priv__, reg__, val__) writeq(val__, (dev_priv__)->regs + (reg__))
|
|
|
|
|
|
|
|
#define __raw_posting_read(dev_priv__, reg__) (void)__raw_i915_read32(dev_priv__, reg__)
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static const char * const forcewake_domain_names[] = {
|
|
|
|
"render",
|
|
|
|
"blitter",
|
|
|
|
"media",
|
|
|
|
};
|
|
|
|
|
|
|
|
const char *
|
2015-01-16 17:34:41 +08:00
|
|
|
intel_uncore_forcewake_domain_to_str(const enum forcewake_domain_id id)
|
2015-01-19 22:20:43 +08:00
|
|
|
{
|
|
|
|
BUILD_BUG_ON((sizeof(forcewake_domain_names)/sizeof(const char *)) !=
|
|
|
|
FW_DOMAIN_ID_COUNT);
|
|
|
|
|
|
|
|
if (id >= 0 && id < FW_DOMAIN_ID_COUNT)
|
|
|
|
return forcewake_domain_names[id];
|
|
|
|
|
|
|
|
WARN_ON(id);
|
|
|
|
|
|
|
|
return "unknown";
|
|
|
|
}
|
|
|
|
|
2014-02-22 00:52:25 +08:00
|
|
|
static void
|
|
|
|
assert_device_not_suspended(struct drm_i915_private *dev_priv)
|
|
|
|
{
|
2014-11-24 16:03:12 +08:00
|
|
|
WARN_ONCE(HAS_RUNTIME_PM(dev_priv->dev) && dev_priv->pm.suspended,
|
|
|
|
"Device suspended\n");
|
2014-02-22 00:52:25 +08:00
|
|
|
}
|
2013-07-20 03:36:53 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static inline void
|
|
|
|
fw_domain_reset(const struct intel_uncore_forcewake_domain *d)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
2015-01-28 20:43:24 +08:00
|
|
|
WARN_ON(d->reg_set == 0);
|
2015-01-19 22:20:43 +08:00
|
|
|
__raw_i915_write32(d->i915, d->reg_set, d->val_reset);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static inline void
|
|
|
|
fw_domain_arm_timer(struct intel_uncore_forcewake_domain *d)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
2015-01-19 22:20:43 +08:00
|
|
|
mod_timer_pinned(&d->timer, jiffies + 1);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static inline void
|
|
|
|
fw_domain_wait_ack_clear(const struct intel_uncore_forcewake_domain *d)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
2015-01-19 22:20:43 +08:00
|
|
|
if (wait_for_atomic((__raw_i915_read32(d->i915, d->reg_ack) &
|
|
|
|
FORCEWAKE_KERNEL) == 0,
|
2013-07-20 03:36:52 +08:00
|
|
|
FORCEWAKE_ACK_TIMEOUT_MS))
|
2015-01-19 22:20:43 +08:00
|
|
|
DRM_ERROR("%s: timed out waiting for forcewake ack to clear.\n",
|
|
|
|
intel_uncore_forcewake_domain_to_str(d->id));
|
|
|
|
}
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static inline void
|
|
|
|
fw_domain_get(const struct intel_uncore_forcewake_domain *d)
|
|
|
|
{
|
|
|
|
__raw_i915_write32(d->i915, d->reg_set, d->val_set);
|
|
|
|
}
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static inline void
|
|
|
|
fw_domain_wait_ack(const struct intel_uncore_forcewake_domain *d)
|
|
|
|
{
|
|
|
|
if (wait_for_atomic((__raw_i915_read32(d->i915, d->reg_ack) &
|
|
|
|
FORCEWAKE_KERNEL),
|
2013-07-20 03:36:52 +08:00
|
|
|
FORCEWAKE_ACK_TIMEOUT_MS))
|
2015-01-19 22:20:43 +08:00
|
|
|
DRM_ERROR("%s: timed out waiting for forcewake ack request.\n",
|
|
|
|
intel_uncore_forcewake_domain_to_str(d->id));
|
|
|
|
}
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static inline void
|
|
|
|
fw_domain_put(const struct intel_uncore_forcewake_domain *d)
|
|
|
|
{
|
|
|
|
__raw_i915_write32(d->i915, d->reg_set, d->val_clear);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static inline void
|
|
|
|
fw_domain_posting_read(const struct intel_uncore_forcewake_domain *d)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
2015-01-19 22:20:43 +08:00
|
|
|
/* something from same cacheline, but not from the set register */
|
|
|
|
if (d->reg_post)
|
|
|
|
__raw_posting_read(d->i915, d->reg_post);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static void
|
2015-01-16 17:34:41 +08:00
|
|
|
fw_domains_get(struct drm_i915_private *dev_priv, enum forcewake_domains fw_domains)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
2015-01-19 22:20:43 +08:00
|
|
|
struct intel_uncore_forcewake_domain *d;
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id id;
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
for_each_fw_domain_mask(d, fw_domains, dev_priv, id) {
|
|
|
|
fw_domain_wait_ack_clear(d);
|
|
|
|
fw_domain_get(d);
|
|
|
|
fw_domain_wait_ack(d);
|
|
|
|
}
|
|
|
|
}
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static void
|
2015-01-16 17:34:41 +08:00
|
|
|
fw_domains_put(struct drm_i915_private *dev_priv, enum forcewake_domains fw_domains)
|
2015-01-19 22:20:43 +08:00
|
|
|
{
|
|
|
|
struct intel_uncore_forcewake_domain *d;
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id id;
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
for_each_fw_domain_mask(d, fw_domains, dev_priv, id) {
|
|
|
|
fw_domain_put(d);
|
|
|
|
fw_domain_posting_read(d);
|
|
|
|
}
|
|
|
|
}
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static void
|
|
|
|
fw_domains_posting_read(struct drm_i915_private *dev_priv)
|
|
|
|
{
|
|
|
|
struct intel_uncore_forcewake_domain *d;
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id id;
|
2015-01-19 22:20:43 +08:00
|
|
|
|
|
|
|
/* No need to do for all, just do for first found */
|
|
|
|
for_each_fw_domain(d, dev_priv, id) {
|
|
|
|
fw_domain_posting_read(d);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2015-01-16 17:34:41 +08:00
|
|
|
fw_domains_reset(struct drm_i915_private *dev_priv, enum forcewake_domains fw_domains)
|
2015-01-19 22:20:43 +08:00
|
|
|
{
|
|
|
|
struct intel_uncore_forcewake_domain *d;
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id id;
|
2015-01-19 22:20:43 +08:00
|
|
|
|
2015-01-28 20:43:24 +08:00
|
|
|
WARN_ON(dev_priv->uncore.fw_domains == 0);
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
for_each_fw_domain_mask(d, fw_domains, dev_priv, id)
|
|
|
|
fw_domain_reset(d);
|
|
|
|
|
|
|
|
fw_domains_posting_read(dev_priv);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void __gen6_gt_wait_for_thread_c0(struct drm_i915_private *dev_priv)
|
|
|
|
{
|
|
|
|
/* w/a for a sporadic read returning 0 by waiting for the GT
|
|
|
|
* thread to wake up.
|
|
|
|
*/
|
|
|
|
if (wait_for_atomic_us((__raw_i915_read32(dev_priv, GEN6_GT_THREAD_STATUS_REG) &
|
|
|
|
GEN6_GT_THREAD_STATUS_CORE_MASK) == 0, 500))
|
|
|
|
DRM_ERROR("GT thread status wait timed out\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
static void fw_domains_get_with_thread_status(struct drm_i915_private *dev_priv,
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domains fw_domains)
|
2015-01-19 22:20:43 +08:00
|
|
|
{
|
|
|
|
fw_domains_get(dev_priv, fw_domains);
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
/* WaRsForcewakeWaitTC0:snb,ivb,hsw,bdw,vlv */
|
2014-11-10 20:52:50 +08:00
|
|
|
__gen6_gt_wait_for_thread_c0(dev_priv);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void gen6_gt_check_fifodbg(struct drm_i915_private *dev_priv)
|
|
|
|
{
|
|
|
|
u32 gtfifodbg;
|
2013-07-20 03:36:53 +08:00
|
|
|
|
|
|
|
gtfifodbg = __raw_i915_read32(dev_priv, GTFIFODBG);
|
2013-11-14 07:59:59 +08:00
|
|
|
if (WARN(gtfifodbg, "GT wake FIFO error 0x%x\n", gtfifodbg))
|
|
|
|
__raw_i915_write32(dev_priv, GTFIFODBG, gtfifodbg);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
static void fw_domains_put_with_fifo(struct drm_i915_private *dev_priv,
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domains fw_domains)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domains_put(dev_priv, fw_domains);
|
2013-07-20 03:36:52 +08:00
|
|
|
gen6_gt_check_fifodbg(dev_priv);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int __gen6_gt_wait_for_fifo(struct drm_i915_private *dev_priv)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
|
drm/i915/vlv: Update Wait for FIFO and wait for 20 free entries. v3
On VLV, FIFO will be shared by both SW and HW. So, we read the
free entries through register and update dev_priv variable
and wait for only 20 entries to be free
From Deepak's follow-up mail explaining why vlv is special:
"On SB, Out of 64 FIFO Entries, 20 Entries will be used by HW and
remaining 44 will be used by the SW,. I think due to this reason, we
have a threshold of 20 Entries."
"On VLV, HW and SW can access all 64 fifo entries, I don't think
having a threshold of 20 Entries is mandatory on VLV. Also, since both
SW and HW can access all 64 Entries. I think on VLV, we need to update
the fifo_count before waiting for the FIFO."
v2: Apply mask when we read the number of free FIFO entries (Ville).
v3: Mask applied after reading the register (Deepak).
Signed-off-by: Deepak S <deepak.s@intel.com>
[danvet: Add further explanation from Deepak to commit message.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-11-29 18:26:30 +08:00
|
|
|
/* On VLV, FIFO will be shared by both SW and HW.
|
|
|
|
* So, we need to read the FREE_ENTRIES everytime */
|
|
|
|
if (IS_VALLEYVIEW(dev_priv->dev))
|
|
|
|
dev_priv->uncore.fifo_count =
|
|
|
|
__raw_i915_read32(dev_priv, GTFIFOCTL) &
|
|
|
|
GT_FIFO_FREE_ENTRIES_MASK;
|
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
if (dev_priv->uncore.fifo_count < GT_FIFO_NUM_RESERVED_ENTRIES) {
|
|
|
|
int loop = 500;
|
2013-11-14 08:00:00 +08:00
|
|
|
u32 fifo = __raw_i915_read32(dev_priv, GTFIFOCTL) & GT_FIFO_FREE_ENTRIES_MASK;
|
2013-07-20 03:36:52 +08:00
|
|
|
while (fifo <= GT_FIFO_NUM_RESERVED_ENTRIES && loop--) {
|
|
|
|
udelay(10);
|
2013-11-14 08:00:00 +08:00
|
|
|
fifo = __raw_i915_read32(dev_priv, GTFIFOCTL) & GT_FIFO_FREE_ENTRIES_MASK;
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
if (WARN_ON(loop < 0 && fifo <= GT_FIFO_NUM_RESERVED_ENTRIES))
|
|
|
|
++ret;
|
|
|
|
dev_priv->uncore.fifo_count = fifo;
|
|
|
|
}
|
|
|
|
dev_priv->uncore.fifo_count--;
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:40 +08:00
|
|
|
static void intel_uncore_fw_release_timer(unsigned long arg)
|
2014-11-05 01:07:04 +08:00
|
|
|
{
|
2015-01-16 17:34:37 +08:00
|
|
|
struct intel_uncore_forcewake_domain *domain = (void *)arg;
|
|
|
|
unsigned long irqflags;
|
2014-11-05 01:07:04 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
assert_device_not_suspended(domain->i915);
|
2014-11-05 01:07:04 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
spin_lock_irqsave(&domain->i915->uncore.lock, irqflags);
|
|
|
|
if (WARN_ON(domain->wake_count == 0))
|
|
|
|
domain->wake_count++;
|
|
|
|
|
|
|
|
if (--domain->wake_count == 0)
|
|
|
|
domain->i915->uncore.funcs.force_wake_put(domain->i915,
|
|
|
|
1 << domain->id);
|
|
|
|
|
|
|
|
spin_unlock_irqrestore(&domain->i915->uncore.lock, irqflags);
|
2014-11-05 01:07:04 +08:00
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
void intel_uncore_forcewake_reset(struct drm_device *dev, bool restore)
|
2014-11-05 01:07:04 +08:00
|
|
|
{
|
2015-01-16 17:34:37 +08:00
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
2015-01-16 17:34:41 +08:00
|
|
|
unsigned long irqflags;
|
2015-01-16 17:34:37 +08:00
|
|
|
struct intel_uncore_forcewake_domain *domain;
|
2015-01-16 17:34:41 +08:00
|
|
|
int retry_count = 100;
|
|
|
|
enum forcewake_domain_id id;
|
|
|
|
enum forcewake_domains fw = 0, active_domains;
|
2014-11-05 01:07:04 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
/* Hold uncore.lock across reset to prevent any register access
|
|
|
|
* with forcewake not set correctly. Wait until all pending
|
|
|
|
* timers are run before holding.
|
|
|
|
*/
|
|
|
|
while (1) {
|
|
|
|
active_domains = 0;
|
2014-11-05 01:07:04 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
for_each_fw_domain(domain, dev_priv, id) {
|
|
|
|
if (del_timer_sync(&domain->timer) == 0)
|
|
|
|
continue;
|
2014-11-05 01:07:04 +08:00
|
|
|
|
2015-01-16 17:34:40 +08:00
|
|
|
intel_uncore_fw_release_timer((unsigned long)domain);
|
2015-01-16 17:34:37 +08:00
|
|
|
}
|
2013-08-26 20:46:09 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
|
2014-02-22 00:52:25 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
for_each_fw_domain(domain, dev_priv, id) {
|
|
|
|
if (timer_pending(&domain->timer))
|
|
|
|
active_domains |= (1 << id);
|
|
|
|
}
|
2014-03-16 03:20:29 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
if (active_domains == 0)
|
|
|
|
break;
|
2013-08-26 20:46:09 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
if (--retry_count == 0) {
|
|
|
|
DRM_ERROR("Timed out waiting for forcewake timers to finish\n");
|
|
|
|
break;
|
|
|
|
}
|
2014-03-13 20:00:29 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
|
|
|
|
cond_resched();
|
|
|
|
}
|
2014-03-13 20:00:29 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
WARN_ON(active_domains);
|
|
|
|
|
|
|
|
for_each_fw_domain(domain, dev_priv, id)
|
|
|
|
if (domain->wake_count)
|
|
|
|
fw |= 1 << id;
|
|
|
|
|
|
|
|
if (fw)
|
|
|
|
dev_priv->uncore.funcs.force_wake_put(dev_priv, fw);
|
2013-11-16 23:00:09 +08:00
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domains_reset(dev_priv, FORCEWAKE_ALL);
|
2014-11-05 01:07:04 +08:00
|
|
|
|
2014-03-13 20:00:29 +08:00
|
|
|
if (restore) { /* If reset with a user forcewake, try to restore */
|
|
|
|
if (fw)
|
|
|
|
dev_priv->uncore.funcs.force_wake_get(dev_priv, fw);
|
|
|
|
|
|
|
|
if (IS_GEN6(dev) || IS_GEN7(dev))
|
|
|
|
dev_priv->uncore.fifo_count =
|
|
|
|
__raw_i915_read32(dev_priv, GTFIFOCTL) &
|
|
|
|
GT_FIFO_FREE_ENTRIES_MASK;
|
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
if (!restore)
|
2015-01-16 17:34:40 +08:00
|
|
|
assert_forcewakes_inactive(dev_priv);
|
2015-01-16 17:34:37 +08:00
|
|
|
|
2014-03-13 20:00:29 +08:00
|
|
|
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
|
2013-11-16 23:00:09 +08:00
|
|
|
}
|
|
|
|
|
2015-01-28 20:43:24 +08:00
|
|
|
static void intel_uncore_ellc_detect(struct drm_device *dev)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
|
2014-04-19 05:04:28 +08:00
|
|
|
if ((IS_HASWELL(dev) || IS_BROADWELL(dev)) &&
|
2013-10-05 12:22:50 +08:00
|
|
|
(__raw_i915_read32(dev_priv, HSW_EDRAM_PRESENT) == 1)) {
|
|
|
|
/* The docs do not explain exactly how the calculation can be
|
|
|
|
* made. It is somewhat guessable, but for now, it's always
|
|
|
|
* 128MB.
|
|
|
|
* NB: We can't write IDICR yet because we do not have gt funcs
|
|
|
|
* set up */
|
|
|
|
dev_priv->ellc_size = 128;
|
|
|
|
DRM_INFO("Found %zuMB of eLLC\n", dev_priv->ellc_size);
|
|
|
|
}
|
2015-01-28 20:43:24 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void __intel_uncore_early_sanitize(struct drm_device *dev,
|
|
|
|
bool restore_forcewake)
|
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
|
|
|
|
if (HAS_FPGA_DBG_UNCLAIMED(dev))
|
|
|
|
__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2013-12-03 17:30:09 +08:00
|
|
|
/* clear out old GT FIFO errors */
|
|
|
|
if (IS_GEN6(dev) || IS_GEN7(dev))
|
|
|
|
__raw_i915_write32(dev_priv, GTFIFODBG,
|
|
|
|
__raw_i915_read32(dev_priv, GTFIFODBG));
|
|
|
|
|
2014-06-06 17:59:39 +08:00
|
|
|
intel_uncore_forcewake_reset(dev, restore_forcewake);
|
2013-08-23 21:52:30 +08:00
|
|
|
}
|
|
|
|
|
2014-10-24 00:23:21 +08:00
|
|
|
void intel_uncore_early_sanitize(struct drm_device *dev, bool restore_forcewake)
|
|
|
|
{
|
|
|
|
__intel_uncore_early_sanitize(dev, restore_forcewake);
|
|
|
|
i915_check_and_clear_faults(dev);
|
|
|
|
}
|
|
|
|
|
2013-08-23 21:52:30 +08:00
|
|
|
void intel_uncore_sanitize(struct drm_device *dev)
|
|
|
|
{
|
2013-07-20 03:36:52 +08:00
|
|
|
/* BIOS often leaves RC6 enabled, but disable it for hw init */
|
|
|
|
intel_disable_gt_powersave(dev);
|
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:40 +08:00
|
|
|
/**
|
|
|
|
* intel_uncore_forcewake_get - grab forcewake domain references
|
|
|
|
* @dev_priv: i915 device instance
|
|
|
|
* @fw_domains: forcewake domains to get reference on
|
|
|
|
*
|
|
|
|
* This function can be used get GT's forcewake domain references.
|
|
|
|
* Normal register access will handle the forcewake domains automatically.
|
|
|
|
* However if some sequence requires the GT to not power down a particular
|
|
|
|
* forcewake domains this function should be called at the beginning of the
|
|
|
|
* sequence. And subsequently the reference should be dropped by symmetric
|
|
|
|
* call to intel_unforce_forcewake_put(). Usually caller wants all the domains
|
|
|
|
* to be kept awake so the @fw_domains would be then FORCEWAKE_ALL.
|
2013-07-20 03:36:52 +08:00
|
|
|
*/
|
2015-01-16 17:34:40 +08:00
|
|
|
void intel_uncore_forcewake_get(struct drm_i915_private *dev_priv,
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domains fw_domains)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
|
|
|
unsigned long irqflags;
|
2015-01-16 17:34:37 +08:00
|
|
|
struct intel_uncore_forcewake_domain *domain;
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id id;
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2013-10-06 08:57:11 +08:00
|
|
|
if (!dev_priv->uncore.funcs.force_wake_get)
|
|
|
|
return;
|
|
|
|
|
2015-01-16 17:34:35 +08:00
|
|
|
WARN_ON(dev_priv->pm.suspended);
|
2013-11-28 04:21:54 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
fw_domains &= dev_priv->uncore.fw_domains;
|
|
|
|
|
2015-01-16 17:34:35 +08:00
|
|
|
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
|
2014-11-05 01:07:04 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
for_each_fw_domain_mask(domain, fw_domains, dev_priv, id) {
|
|
|
|
if (domain->wake_count++)
|
|
|
|
fw_domains &= ~(1 << id);
|
2015-01-16 17:34:35 +08:00
|
|
|
}
|
2013-11-23 17:25:43 +08:00
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
if (fw_domains)
|
|
|
|
dev_priv->uncore.funcs.force_wake_get(dev_priv, fw_domains);
|
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
|
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:40 +08:00
|
|
|
/**
|
|
|
|
* intel_uncore_forcewake_put - release a forcewake domain reference
|
|
|
|
* @dev_priv: i915 device instance
|
|
|
|
* @fw_domains: forcewake domains to put references
|
|
|
|
*
|
|
|
|
* This function drops the device-level forcewakes for specified
|
|
|
|
* domains obtained by intel_uncore_forcewake_get().
|
2013-07-20 03:36:52 +08:00
|
|
|
*/
|
2015-01-16 17:34:40 +08:00
|
|
|
void intel_uncore_forcewake_put(struct drm_i915_private *dev_priv,
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domains fw_domains)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
|
|
|
unsigned long irqflags;
|
2015-01-16 17:34:37 +08:00
|
|
|
struct intel_uncore_forcewake_domain *domain;
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id id;
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2013-10-06 08:57:11 +08:00
|
|
|
if (!dev_priv->uncore.funcs.force_wake_put)
|
|
|
|
return;
|
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
fw_domains &= dev_priv->uncore.fw_domains;
|
|
|
|
|
2015-01-16 17:34:35 +08:00
|
|
|
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
|
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
for_each_fw_domain_mask(domain, fw_domains, dev_priv, id) {
|
|
|
|
if (WARN_ON(domain->wake_count == 0))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (--domain->wake_count)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
domain->wake_count++;
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domain_arm_timer(domain);
|
2013-08-26 20:46:09 +08:00
|
|
|
}
|
2015-01-16 17:34:34 +08:00
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
|
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:40 +08:00
|
|
|
void assert_forcewakes_inactive(struct drm_i915_private *dev_priv)
|
2014-02-22 00:52:26 +08:00
|
|
|
{
|
2015-01-16 17:34:37 +08:00
|
|
|
struct intel_uncore_forcewake_domain *domain;
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id id;
|
2015-01-16 17:34:37 +08:00
|
|
|
|
2014-02-22 00:52:26 +08:00
|
|
|
if (!dev_priv->uncore.funcs.force_wake_get)
|
|
|
|
return;
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
for_each_fw_domain(domain, dev_priv, id)
|
2015-01-16 17:34:37 +08:00
|
|
|
WARN_ON(domain->wake_count);
|
2014-02-22 00:52:26 +08:00
|
|
|
}
|
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
/* We give fast paths for the really cool registers */
|
|
|
|
#define NEEDS_FORCE_WAKE(dev_priv, reg) \
|
2013-10-06 08:57:11 +08:00
|
|
|
((reg) < 0x40000 && (reg) != FORCEWAKE)
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2014-05-23 23:30:16 +08:00
|
|
|
#define REG_RANGE(reg, start, end) ((reg) >= (start) && (reg) < (end))
|
2014-03-29 00:54:26 +08:00
|
|
|
|
2014-05-23 23:30:16 +08:00
|
|
|
#define FORCEWAKE_VLV_RENDER_RANGE_OFFSET(reg) \
|
|
|
|
(REG_RANGE((reg), 0x2000, 0x4000) || \
|
|
|
|
REG_RANGE((reg), 0x5000, 0x8000) || \
|
|
|
|
REG_RANGE((reg), 0xB000, 0x12000) || \
|
|
|
|
REG_RANGE((reg), 0x2E000, 0x30000))
|
|
|
|
|
|
|
|
#define FORCEWAKE_VLV_MEDIA_RANGE_OFFSET(reg) \
|
|
|
|
(REG_RANGE((reg), 0x12000, 0x14000) || \
|
|
|
|
REG_RANGE((reg), 0x22000, 0x24000) || \
|
|
|
|
REG_RANGE((reg), 0x30000, 0x40000))
|
|
|
|
|
|
|
|
#define FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg) \
|
|
|
|
(REG_RANGE((reg), 0x2000, 0x4000) || \
|
2014-12-12 00:12:49 +08:00
|
|
|
REG_RANGE((reg), 0x5200, 0x8000) || \
|
2014-05-23 23:30:16 +08:00
|
|
|
REG_RANGE((reg), 0x8300, 0x8500) || \
|
2014-12-12 00:12:49 +08:00
|
|
|
REG_RANGE((reg), 0xB000, 0xB480) || \
|
2014-05-23 23:30:16 +08:00
|
|
|
REG_RANGE((reg), 0xE000, 0xE800))
|
|
|
|
|
|
|
|
#define FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg) \
|
|
|
|
(REG_RANGE((reg), 0x8800, 0x8900) || \
|
|
|
|
REG_RANGE((reg), 0xD000, 0xD800) || \
|
|
|
|
REG_RANGE((reg), 0x12000, 0x14000) || \
|
|
|
|
REG_RANGE((reg), 0x1A000, 0x1C000) || \
|
|
|
|
REG_RANGE((reg), 0x1E800, 0x1EA00) || \
|
2014-12-12 00:12:49 +08:00
|
|
|
REG_RANGE((reg), 0x30000, 0x38000))
|
2014-05-23 23:30:16 +08:00
|
|
|
|
|
|
|
#define FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg) \
|
|
|
|
(REG_RANGE((reg), 0x4000, 0x5000) || \
|
|
|
|
REG_RANGE((reg), 0x8000, 0x8300) || \
|
|
|
|
REG_RANGE((reg), 0x8500, 0x8600) || \
|
|
|
|
REG_RANGE((reg), 0x9000, 0xB000) || \
|
2014-12-12 00:12:49 +08:00
|
|
|
REG_RANGE((reg), 0xF000, 0x10000))
|
2014-03-29 00:54:26 +08:00
|
|
|
|
2014-11-20 21:42:55 +08:00
|
|
|
#define FORCEWAKE_GEN9_UNCORE_RANGE_OFFSET(reg) \
|
2014-11-25 14:59:00 +08:00
|
|
|
REG_RANGE((reg), 0xB00, 0x2000)
|
2014-11-20 21:42:55 +08:00
|
|
|
|
|
|
|
#define FORCEWAKE_GEN9_RENDER_RANGE_OFFSET(reg) \
|
2014-11-25 14:59:00 +08:00
|
|
|
(REG_RANGE((reg), 0x2000, 0x2700) || \
|
|
|
|
REG_RANGE((reg), 0x3000, 0x4000) || \
|
2014-11-20 21:42:55 +08:00
|
|
|
REG_RANGE((reg), 0x5200, 0x8000) || \
|
2014-11-25 14:59:00 +08:00
|
|
|
REG_RANGE((reg), 0x8140, 0x8160) || \
|
2014-11-20 21:42:55 +08:00
|
|
|
REG_RANGE((reg), 0x8300, 0x8500) || \
|
|
|
|
REG_RANGE((reg), 0x8C00, 0x8D00) || \
|
|
|
|
REG_RANGE((reg), 0xB000, 0xB480) || \
|
2014-11-25 14:59:00 +08:00
|
|
|
REG_RANGE((reg), 0xE000, 0xE900) || \
|
|
|
|
REG_RANGE((reg), 0x24400, 0x24800))
|
2014-11-20 21:42:55 +08:00
|
|
|
|
|
|
|
#define FORCEWAKE_GEN9_MEDIA_RANGE_OFFSET(reg) \
|
2014-11-25 14:59:00 +08:00
|
|
|
(REG_RANGE((reg), 0x8130, 0x8140) || \
|
|
|
|
REG_RANGE((reg), 0x8800, 0x8A00) || \
|
2014-11-20 21:42:55 +08:00
|
|
|
REG_RANGE((reg), 0xD000, 0xD800) || \
|
|
|
|
REG_RANGE((reg), 0x12000, 0x14000) || \
|
|
|
|
REG_RANGE((reg), 0x1A000, 0x1EA00) || \
|
|
|
|
REG_RANGE((reg), 0x30000, 0x40000))
|
|
|
|
|
|
|
|
#define FORCEWAKE_GEN9_COMMON_RANGE_OFFSET(reg) \
|
|
|
|
REG_RANGE((reg), 0x9400, 0x9800)
|
|
|
|
|
|
|
|
#define FORCEWAKE_GEN9_BLITTER_RANGE_OFFSET(reg) \
|
|
|
|
((reg) < 0x40000 &&\
|
|
|
|
!FORCEWAKE_GEN9_UNCORE_RANGE_OFFSET(reg) && \
|
|
|
|
!FORCEWAKE_GEN9_RENDER_RANGE_OFFSET(reg) && \
|
|
|
|
!FORCEWAKE_GEN9_MEDIA_RANGE_OFFSET(reg) && \
|
|
|
|
!FORCEWAKE_GEN9_COMMON_RANGE_OFFSET(reg))
|
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
static void
|
|
|
|
ilk_dummy_write(struct drm_i915_private *dev_priv)
|
|
|
|
{
|
|
|
|
/* WaIssueDummyWriteToWakeupFromRC6:ilk Issue a dummy write to wake up
|
|
|
|
* the chip from rc6 before touching it for real. MI_MODE is masked,
|
|
|
|
* hence harmless to write 0 into. */
|
2013-07-20 03:36:53 +08:00
|
|
|
__raw_i915_write32(dev_priv, MI_MODE, 0);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
hsw_unclaimed_reg_debug(struct drm_i915_private *dev_priv, u32 reg, bool read,
|
|
|
|
bool before)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
const char *op = read ? "reading" : "writing to";
|
|
|
|
const char *when = before ? "before" : "after";
|
|
|
|
|
|
|
|
if (!i915.mmio_debug)
|
|
|
|
return;
|
|
|
|
|
2013-10-06 08:57:11 +08:00
|
|
|
if (__raw_i915_read32(dev_priv, FPGA_DBG) & FPGA_DBG_RM_NOCLAIM) {
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
WARN(1, "Unclaimed register detected %s %s register 0x%x\n",
|
|
|
|
when, op, reg);
|
2013-07-20 03:36:53 +08:00
|
|
|
__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
hsw_unclaimed_reg_detect(struct drm_i915_private *dev_priv)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
if (i915.mmio_debug)
|
|
|
|
return;
|
|
|
|
|
2013-10-06 08:57:11 +08:00
|
|
|
if (__raw_i915_read32(dev_priv, FPGA_DBG) & FPGA_DBG_RM_NOCLAIM) {
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
DRM_ERROR("Unclaimed register detected. Please use the i915.mmio_debug=1 to debug this problem.");
|
2013-07-20 03:36:53 +08:00
|
|
|
__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:36 +08:00
|
|
|
#define GEN2_READ_HEADER(x) \
|
2013-10-05 12:24:53 +08:00
|
|
|
u##x val = 0; \
|
2015-01-16 17:34:36 +08:00
|
|
|
assert_device_not_suspended(dev_priv);
|
2013-10-05 12:24:53 +08:00
|
|
|
|
2015-01-16 17:34:36 +08:00
|
|
|
#define GEN2_READ_FOOTER \
|
2013-10-05 12:24:53 +08:00
|
|
|
trace_i915_reg_rw(false, reg, val, sizeof(val), trace); \
|
|
|
|
return val
|
|
|
|
|
2015-01-16 17:34:36 +08:00
|
|
|
#define __gen2_read(x) \
|
2013-10-05 12:22:51 +08:00
|
|
|
static u##x \
|
2015-01-16 17:34:36 +08:00
|
|
|
gen2_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
|
|
|
|
GEN2_READ_HEADER(x); \
|
2013-10-05 12:22:53 +08:00
|
|
|
val = __raw_i915_read##x(dev_priv, reg); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN2_READ_FOOTER; \
|
2013-10-05 12:22:53 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
#define __gen5_read(x) \
|
|
|
|
static u##x \
|
|
|
|
gen5_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN2_READ_HEADER(x); \
|
2013-10-05 12:22:53 +08:00
|
|
|
ilk_dummy_write(dev_priv); \
|
|
|
|
val = __raw_i915_read##x(dev_priv, reg); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN2_READ_FOOTER; \
|
2013-10-05 12:22:53 +08:00
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:36 +08:00
|
|
|
__gen5_read(8)
|
|
|
|
__gen5_read(16)
|
|
|
|
__gen5_read(32)
|
|
|
|
__gen5_read(64)
|
|
|
|
__gen2_read(8)
|
|
|
|
__gen2_read(16)
|
|
|
|
__gen2_read(32)
|
|
|
|
__gen2_read(64)
|
|
|
|
|
|
|
|
#undef __gen5_read
|
|
|
|
#undef __gen2_read
|
|
|
|
|
|
|
|
#undef GEN2_READ_FOOTER
|
|
|
|
#undef GEN2_READ_HEADER
|
|
|
|
|
|
|
|
#define GEN6_READ_HEADER(x) \
|
|
|
|
unsigned long irqflags; \
|
|
|
|
u##x val = 0; \
|
|
|
|
assert_device_not_suspended(dev_priv); \
|
|
|
|
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)
|
|
|
|
|
|
|
|
#define GEN6_READ_FOOTER \
|
|
|
|
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags); \
|
|
|
|
trace_i915_reg_rw(false, reg, val, sizeof(val), trace); \
|
|
|
|
return val
|
|
|
|
|
2015-01-16 17:34:37 +08:00
|
|
|
static inline void __force_wake_get(struct drm_i915_private *dev_priv,
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domains fw_domains)
|
2015-01-16 17:34:37 +08:00
|
|
|
{
|
|
|
|
struct intel_uncore_forcewake_domain *domain;
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id id;
|
2015-01-16 17:34:37 +08:00
|
|
|
|
|
|
|
if (WARN_ON(!fw_domains))
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* Ideally GCC would be constant-fold and eliminate this loop */
|
2015-01-19 22:20:43 +08:00
|
|
|
for_each_fw_domain_mask(domain, fw_domains, dev_priv, id) {
|
2015-01-16 17:34:37 +08:00
|
|
|
if (domain->wake_count) {
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domains &= ~(1 << id);
|
2015-01-16 17:34:37 +08:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
domain->wake_count++;
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domain_arm_timer(domain);
|
2015-01-16 17:34:37 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (fw_domains)
|
|
|
|
dev_priv->uncore.funcs.force_wake_get(dev_priv, fw_domains);
|
|
|
|
}
|
|
|
|
|
2013-10-05 12:22:53 +08:00
|
|
|
#define __gen6_read(x) \
|
|
|
|
static u##x \
|
|
|
|
gen6_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_READ_HEADER(x); \
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
hsw_unclaimed_reg_debug(dev_priv, reg, true, true); \
|
2015-01-16 17:34:37 +08:00
|
|
|
if (NEEDS_FORCE_WAKE((dev_priv), (reg))) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
|
2015-01-16 17:34:34 +08:00
|
|
|
val = __raw_i915_read##x(dev_priv, reg); \
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
hsw_unclaimed_reg_debug(dev_priv, reg, true, false); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_READ_FOOTER; \
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2013-11-23 17:25:43 +08:00
|
|
|
#define __vlv_read(x) \
|
|
|
|
static u##x \
|
|
|
|
vlv_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_READ_HEADER(x); \
|
2015-01-16 17:34:37 +08:00
|
|
|
if (FORCEWAKE_VLV_RENDER_RANGE_OFFSET(reg)) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
|
|
|
|
else if (FORCEWAKE_VLV_MEDIA_RANGE_OFFSET(reg)) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_MEDIA); \
|
2014-02-28 04:07:21 +08:00
|
|
|
val = __raw_i915_read##x(dev_priv, reg); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_READ_FOOTER; \
|
2013-11-23 17:25:43 +08:00
|
|
|
}
|
|
|
|
|
2014-05-23 23:30:16 +08:00
|
|
|
#define __chv_read(x) \
|
|
|
|
static u##x \
|
|
|
|
chv_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_READ_HEADER(x); \
|
2015-01-16 17:34:37 +08:00
|
|
|
if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
|
|
|
|
else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_MEDIA); \
|
|
|
|
else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) \
|
|
|
|
__force_wake_get(dev_priv, \
|
|
|
|
FORCEWAKE_RENDER | FORCEWAKE_MEDIA); \
|
2014-05-23 23:30:16 +08:00
|
|
|
val = __raw_i915_read##x(dev_priv, reg); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_READ_FOOTER; \
|
2014-05-23 23:30:16 +08:00
|
|
|
}
|
2013-11-23 17:25:43 +08:00
|
|
|
|
2014-11-20 21:42:55 +08:00
|
|
|
#define SKL_NEEDS_FORCE_WAKE(dev_priv, reg) \
|
|
|
|
((reg) < 0x40000 && !FORCEWAKE_GEN9_UNCORE_RANGE_OFFSET(reg))
|
|
|
|
|
|
|
|
#define __gen9_read(x) \
|
|
|
|
static u##x \
|
|
|
|
gen9_read##x(struct drm_i915_private *dev_priv, off_t reg, bool trace) { \
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domains fw_engine; \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_READ_HEADER(x); \
|
2015-01-16 17:34:37 +08:00
|
|
|
if (!SKL_NEEDS_FORCE_WAKE((dev_priv), (reg))) \
|
|
|
|
fw_engine = 0; \
|
|
|
|
else if (FORCEWAKE_GEN9_RENDER_RANGE_OFFSET(reg)) \
|
|
|
|
fw_engine = FORCEWAKE_RENDER; \
|
|
|
|
else if (FORCEWAKE_GEN9_MEDIA_RANGE_OFFSET(reg)) \
|
|
|
|
fw_engine = FORCEWAKE_MEDIA; \
|
|
|
|
else if (FORCEWAKE_GEN9_COMMON_RANGE_OFFSET(reg)) \
|
|
|
|
fw_engine = FORCEWAKE_RENDER | FORCEWAKE_MEDIA; \
|
|
|
|
else \
|
|
|
|
fw_engine = FORCEWAKE_BLITTER; \
|
|
|
|
if (fw_engine) \
|
|
|
|
__force_wake_get(dev_priv, fw_engine); \
|
|
|
|
val = __raw_i915_read##x(dev_priv, reg); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_READ_FOOTER; \
|
2014-11-20 21:42:55 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
__gen9_read(8)
|
|
|
|
__gen9_read(16)
|
|
|
|
__gen9_read(32)
|
|
|
|
__gen9_read(64)
|
2014-05-23 23:30:16 +08:00
|
|
|
__chv_read(8)
|
|
|
|
__chv_read(16)
|
|
|
|
__chv_read(32)
|
|
|
|
__chv_read(64)
|
2013-11-23 17:25:43 +08:00
|
|
|
__vlv_read(8)
|
|
|
|
__vlv_read(16)
|
|
|
|
__vlv_read(32)
|
|
|
|
__vlv_read(64)
|
2013-10-05 12:22:53 +08:00
|
|
|
__gen6_read(8)
|
|
|
|
__gen6_read(16)
|
|
|
|
__gen6_read(32)
|
|
|
|
__gen6_read(64)
|
|
|
|
|
2014-11-20 21:42:55 +08:00
|
|
|
#undef __gen9_read
|
2014-05-23 23:30:16 +08:00
|
|
|
#undef __chv_read
|
2013-11-23 17:25:43 +08:00
|
|
|
#undef __vlv_read
|
2013-10-05 12:22:53 +08:00
|
|
|
#undef __gen6_read
|
2015-01-16 17:34:36 +08:00
|
|
|
#undef GEN6_READ_FOOTER
|
|
|
|
#undef GEN6_READ_HEADER
|
2013-10-05 12:24:53 +08:00
|
|
|
|
2015-01-16 17:34:36 +08:00
|
|
|
#define GEN2_WRITE_HEADER \
|
2013-10-05 12:24:53 +08:00
|
|
|
trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \
|
2014-02-22 00:52:28 +08:00
|
|
|
assert_device_not_suspended(dev_priv); \
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2015-01-16 17:34:36 +08:00
|
|
|
#define GEN2_WRITE_FOOTER
|
2013-12-02 20:23:02 +08:00
|
|
|
|
2015-01-16 17:34:36 +08:00
|
|
|
#define __gen2_write(x) \
|
2013-10-05 12:22:51 +08:00
|
|
|
static void \
|
2015-01-16 17:34:36 +08:00
|
|
|
gen2_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
|
|
|
|
GEN2_WRITE_HEADER; \
|
2013-10-05 12:22:54 +08:00
|
|
|
__raw_i915_write##x(dev_priv, reg, val); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN2_WRITE_FOOTER; \
|
2013-10-05 12:22:54 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
#define __gen5_write(x) \
|
|
|
|
static void \
|
|
|
|
gen5_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN2_WRITE_HEADER; \
|
2013-10-05 12:22:54 +08:00
|
|
|
ilk_dummy_write(dev_priv); \
|
|
|
|
__raw_i915_write##x(dev_priv, reg, val); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN2_WRITE_FOOTER; \
|
2013-10-05 12:22:54 +08:00
|
|
|
}
|
|
|
|
|
2015-01-16 17:34:36 +08:00
|
|
|
__gen5_write(8)
|
|
|
|
__gen5_write(16)
|
|
|
|
__gen5_write(32)
|
|
|
|
__gen5_write(64)
|
|
|
|
__gen2_write(8)
|
|
|
|
__gen2_write(16)
|
|
|
|
__gen2_write(32)
|
|
|
|
__gen2_write(64)
|
|
|
|
|
|
|
|
#undef __gen5_write
|
|
|
|
#undef __gen2_write
|
|
|
|
|
|
|
|
#undef GEN2_WRITE_FOOTER
|
|
|
|
#undef GEN2_WRITE_HEADER
|
|
|
|
|
|
|
|
#define GEN6_WRITE_HEADER \
|
|
|
|
unsigned long irqflags; \
|
|
|
|
trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \
|
|
|
|
assert_device_not_suspended(dev_priv); \
|
|
|
|
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags)
|
|
|
|
|
|
|
|
#define GEN6_WRITE_FOOTER \
|
|
|
|
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags)
|
|
|
|
|
2013-10-05 12:22:54 +08:00
|
|
|
#define __gen6_write(x) \
|
|
|
|
static void \
|
|
|
|
gen6_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
|
|
|
|
u32 __fifo_ret = 0; \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_HEADER; \
|
2013-10-05 12:22:54 +08:00
|
|
|
if (NEEDS_FORCE_WAKE((dev_priv), (reg))) { \
|
|
|
|
__fifo_ret = __gen6_gt_wait_for_fifo(dev_priv); \
|
|
|
|
} \
|
|
|
|
__raw_i915_write##x(dev_priv, reg, val); \
|
|
|
|
if (unlikely(__fifo_ret)) { \
|
|
|
|
gen6_gt_check_fifodbg(dev_priv); \
|
|
|
|
} \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_FOOTER; \
|
2013-10-05 12:22:54 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
#define __hsw_write(x) \
|
|
|
|
static void \
|
|
|
|
hsw_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
|
2013-07-20 03:36:52 +08:00
|
|
|
u32 __fifo_ret = 0; \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_HEADER; \
|
2013-07-20 03:36:52 +08:00
|
|
|
if (NEEDS_FORCE_WAKE((dev_priv), (reg))) { \
|
|
|
|
__fifo_ret = __gen6_gt_wait_for_fifo(dev_priv); \
|
|
|
|
} \
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
hsw_unclaimed_reg_debug(dev_priv, reg, false, true); \
|
2013-07-20 03:36:53 +08:00
|
|
|
__raw_i915_write##x(dev_priv, reg, val); \
|
2013-07-20 03:36:52 +08:00
|
|
|
if (unlikely(__fifo_ret)) { \
|
|
|
|
gen6_gt_check_fifodbg(dev_priv); \
|
|
|
|
} \
|
drm/i915: reorganize the unclaimed register detection code
The current code only runs when we do an I915_WRITE operation. It
checks if the unclaimed register flag is set before we do the
operation, and then it checks it again after we do the operation. This
double check allows us to find out if the I915_WRITE operation in
question is the bad one, or if some previous code is the bad one. When
it finds a problem, our code uses DRM_ERROR to signal it.
The good thing about the current code is that it detects the problem,
so at least we can know we did something wrong. The problem is that
even though we find the problem, we don't really have much information
to actually debug it. So whenever I see one of these DRM_ERROR
messages on my systems, the first thing I do is apply a patch to
change the DRM_ERROR to a WARN and also check for unclaimed registers
on I915_READ operations. This local patch makes things even slower,
but it usually helps a lot in finding the bad code.
The first point here is that since the current code is only useful to
detect whether we have a problem or not, but it is not really good to
find the cause of the problem, I don't think we should be checking
both before and after every I915_WRITE operation: just doing the check
once should be enough for us to quickly detect problems. With this
change, the code that runs by default for every single user will only
do 1 read operation for every single I915_WRITE, instead of 2. This
patch does this change.
The second point is that the local patch I have should be upstream,
but since it makes things slower it should be disabled by default. So
I added the i915.mmio_debug option to enable it.
So after this patch, this is what will happen:
- By default, we will try to detect unclaimed registers once after
every I915_WRITE operation. Previously we tried twice for every
I915_WRITE.
- When we find an unclaimed register we will still print a DRM_ERROR
message, but we will now tell the user to try again with
i915.mmio_debug=1.
- When we use i915.mmio_debug=1 we will try to find unclaimed
registers both before and after every I915_READ and I915_WRITE
operation, and we will print stack traces in case we find them.
This should really help locating the exact point of the bad code
(or at least finding out that i915.ko is not the problem).
This commit also opens space for really-slow register debugging
operations on other platforms. In theory we can now add lots and lots
of debug code behind i915.mmio_debug, enable this option on our tests,
and catch more problems.
v2: - Remove not-so-useful comments (Daniel)
- Fix the param definition macros (Rodrigo)
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-17 04:49:29 +08:00
|
|
|
hsw_unclaimed_reg_debug(dev_priv, reg, false, false); \
|
|
|
|
hsw_unclaimed_reg_detect(dev_priv); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_FOOTER; \
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
2013-10-05 12:22:53 +08:00
|
|
|
|
drm/i915/bdw: Handle forcewake for writes on gen8
GEN8 removes the GT FIFO which we've all come to know and love. Instead
it offers a wider range of optimized registers which always keep a
shadowed copy, and are fed to the GPU when it wakes.
How this is implemented in hardware is still somewhat of a mystery. As
far as I can tell, the basic design is as follows:
If the register is not optimized, you must use the old forcewake
mechanism to bring the GT out of sleep. [1]
If register is in the optimized list the write will signal that the
GT should begin to come out of whatever sleep state it is in.
While the GT is coming out of sleep, the requested write will be stored
in an intermediate shadow register.
Do to the fact that the implementation details are not clear, I see
several risks:
1. Order is not preserved as it is with GT FIFO. If we issue multiple
writes to optimized registers, where order matters, we may need to
serialize it with forcewake.
2. The optimized registers have only 1 shadowed slot, meaning if we
issue multiple writes to the same register, and those values need to
reach the GPU in order, forcewake will be required.
[1] We could implement a SW queue the way the GT FIFO used to work if
desired.
NOTE: Compile tested only until we get real silicon.
v2:
- Use a default case to make future platforms also work.
- Get rid of IS_BROADWELL since that's not yet defined, but we want to
MMIO as soon as possible.
v3: Apply suggestions from Mika's review:
- s/optimized/shadowed/
- invert the logic of the helper so that it does what it says (the
code itself was correct, just confusing to read).
v4:
- Squash in lost break.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-11-03 12:07:00 +08:00
|
|
|
static const u32 gen8_shadowed_regs[] = {
|
|
|
|
FORCEWAKE_MT,
|
|
|
|
GEN6_RPNSWREQ,
|
|
|
|
GEN6_RC_VIDEO_FREQ,
|
|
|
|
RING_TAIL(RENDER_RING_BASE),
|
|
|
|
RING_TAIL(GEN6_BSD_RING_BASE),
|
|
|
|
RING_TAIL(VEBOX_RING_BASE),
|
|
|
|
RING_TAIL(BLT_RING_BASE),
|
|
|
|
/* TODO: Other registers are not yet used */
|
|
|
|
};
|
|
|
|
|
|
|
|
static bool is_gen8_shadowed(struct drm_i915_private *dev_priv, u32 reg)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < ARRAY_SIZE(gen8_shadowed_regs); i++)
|
|
|
|
if (reg == gen8_shadowed_regs[i])
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define __gen8_write(x) \
|
|
|
|
static void \
|
|
|
|
gen8_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_HEADER; \
|
2014-07-17 04:49:30 +08:00
|
|
|
hsw_unclaimed_reg_debug(dev_priv, reg, false, true); \
|
2015-01-16 17:34:37 +08:00
|
|
|
if (reg < 0x40000 && !is_gen8_shadowed(dev_priv, reg)) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
|
|
|
|
__raw_i915_write##x(dev_priv, reg, val); \
|
2014-07-17 04:49:30 +08:00
|
|
|
hsw_unclaimed_reg_debug(dev_priv, reg, false, false); \
|
|
|
|
hsw_unclaimed_reg_detect(dev_priv); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_FOOTER; \
|
drm/i915/bdw: Handle forcewake for writes on gen8
GEN8 removes the GT FIFO which we've all come to know and love. Instead
it offers a wider range of optimized registers which always keep a
shadowed copy, and are fed to the GPU when it wakes.
How this is implemented in hardware is still somewhat of a mystery. As
far as I can tell, the basic design is as follows:
If the register is not optimized, you must use the old forcewake
mechanism to bring the GT out of sleep. [1]
If register is in the optimized list the write will signal that the
GT should begin to come out of whatever sleep state it is in.
While the GT is coming out of sleep, the requested write will be stored
in an intermediate shadow register.
Do to the fact that the implementation details are not clear, I see
several risks:
1. Order is not preserved as it is with GT FIFO. If we issue multiple
writes to optimized registers, where order matters, we may need to
serialize it with forcewake.
2. The optimized registers have only 1 shadowed slot, meaning if we
issue multiple writes to the same register, and those values need to
reach the GPU in order, forcewake will be required.
[1] We could implement a SW queue the way the GT FIFO used to work if
desired.
NOTE: Compile tested only until we get real silicon.
v2:
- Use a default case to make future platforms also work.
- Get rid of IS_BROADWELL since that's not yet defined, but we want to
MMIO as soon as possible.
v3: Apply suggestions from Mika's review:
- s/optimized/shadowed/
- invert the logic of the helper so that it does what it says (the
code itself was correct, just confusing to read).
v4:
- Squash in lost break.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-11-03 12:07:00 +08:00
|
|
|
}
|
|
|
|
|
2014-05-23 23:30:16 +08:00
|
|
|
#define __chv_write(x) \
|
|
|
|
static void \
|
|
|
|
chv_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, bool trace) { \
|
|
|
|
bool shadowed = is_gen8_shadowed(dev_priv, reg); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_HEADER; \
|
2014-05-23 23:30:16 +08:00
|
|
|
if (!shadowed) { \
|
2015-01-16 17:34:37 +08:00
|
|
|
if (FORCEWAKE_CHV_RENDER_RANGE_OFFSET(reg)) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_RENDER); \
|
|
|
|
else if (FORCEWAKE_CHV_MEDIA_RANGE_OFFSET(reg)) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_MEDIA); \
|
|
|
|
else if (FORCEWAKE_CHV_COMMON_RANGE_OFFSET(reg)) \
|
|
|
|
__force_wake_get(dev_priv, FORCEWAKE_RENDER | FORCEWAKE_MEDIA); \
|
2014-05-23 23:30:16 +08:00
|
|
|
} \
|
|
|
|
__raw_i915_write##x(dev_priv, reg, val); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_FOOTER; \
|
2014-05-23 23:30:16 +08:00
|
|
|
}
|
|
|
|
|
2014-11-20 21:42:56 +08:00
|
|
|
static const u32 gen9_shadowed_regs[] = {
|
|
|
|
RING_TAIL(RENDER_RING_BASE),
|
|
|
|
RING_TAIL(GEN6_BSD_RING_BASE),
|
|
|
|
RING_TAIL(VEBOX_RING_BASE),
|
|
|
|
RING_TAIL(BLT_RING_BASE),
|
|
|
|
FORCEWAKE_BLITTER_GEN9,
|
|
|
|
FORCEWAKE_RENDER_GEN9,
|
|
|
|
FORCEWAKE_MEDIA_GEN9,
|
|
|
|
GEN6_RPNSWREQ,
|
|
|
|
GEN6_RC_VIDEO_FREQ,
|
|
|
|
/* TODO: Other registers are not yet used */
|
|
|
|
};
|
|
|
|
|
|
|
|
static bool is_gen9_shadowed(struct drm_i915_private *dev_priv, u32 reg)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < ARRAY_SIZE(gen9_shadowed_regs); i++)
|
|
|
|
if (reg == gen9_shadowed_regs[i])
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2014-11-20 21:42:55 +08:00
|
|
|
#define __gen9_write(x) \
|
|
|
|
static void \
|
|
|
|
gen9_write##x(struct drm_i915_private *dev_priv, off_t reg, u##x val, \
|
|
|
|
bool trace) { \
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domains fw_engine; \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_HEADER; \
|
2015-01-16 17:34:37 +08:00
|
|
|
if (!SKL_NEEDS_FORCE_WAKE((dev_priv), (reg)) || \
|
|
|
|
is_gen9_shadowed(dev_priv, reg)) \
|
|
|
|
fw_engine = 0; \
|
|
|
|
else if (FORCEWAKE_GEN9_RENDER_RANGE_OFFSET(reg)) \
|
|
|
|
fw_engine = FORCEWAKE_RENDER; \
|
|
|
|
else if (FORCEWAKE_GEN9_MEDIA_RANGE_OFFSET(reg)) \
|
|
|
|
fw_engine = FORCEWAKE_MEDIA; \
|
|
|
|
else if (FORCEWAKE_GEN9_COMMON_RANGE_OFFSET(reg)) \
|
|
|
|
fw_engine = FORCEWAKE_RENDER | FORCEWAKE_MEDIA; \
|
|
|
|
else \
|
|
|
|
fw_engine = FORCEWAKE_BLITTER; \
|
|
|
|
if (fw_engine) \
|
|
|
|
__force_wake_get(dev_priv, fw_engine); \
|
|
|
|
__raw_i915_write##x(dev_priv, reg, val); \
|
2015-01-16 17:34:36 +08:00
|
|
|
GEN6_WRITE_FOOTER; \
|
2014-11-20 21:42:55 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
__gen9_write(8)
|
|
|
|
__gen9_write(16)
|
|
|
|
__gen9_write(32)
|
|
|
|
__gen9_write(64)
|
2014-05-23 23:30:16 +08:00
|
|
|
__chv_write(8)
|
|
|
|
__chv_write(16)
|
|
|
|
__chv_write(32)
|
|
|
|
__chv_write(64)
|
drm/i915/bdw: Handle forcewake for writes on gen8
GEN8 removes the GT FIFO which we've all come to know and love. Instead
it offers a wider range of optimized registers which always keep a
shadowed copy, and are fed to the GPU when it wakes.
How this is implemented in hardware is still somewhat of a mystery. As
far as I can tell, the basic design is as follows:
If the register is not optimized, you must use the old forcewake
mechanism to bring the GT out of sleep. [1]
If register is in the optimized list the write will signal that the
GT should begin to come out of whatever sleep state it is in.
While the GT is coming out of sleep, the requested write will be stored
in an intermediate shadow register.
Do to the fact that the implementation details are not clear, I see
several risks:
1. Order is not preserved as it is with GT FIFO. If we issue multiple
writes to optimized registers, where order matters, we may need to
serialize it with forcewake.
2. The optimized registers have only 1 shadowed slot, meaning if we
issue multiple writes to the same register, and those values need to
reach the GPU in order, forcewake will be required.
[1] We could implement a SW queue the way the GT FIFO used to work if
desired.
NOTE: Compile tested only until we get real silicon.
v2:
- Use a default case to make future platforms also work.
- Get rid of IS_BROADWELL since that's not yet defined, but we want to
MMIO as soon as possible.
v3: Apply suggestions from Mika's review:
- s/optimized/shadowed/
- invert the logic of the helper so that it does what it says (the
code itself was correct, just confusing to read).
v4:
- Squash in lost break.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-11-03 12:07:00 +08:00
|
|
|
__gen8_write(8)
|
|
|
|
__gen8_write(16)
|
|
|
|
__gen8_write(32)
|
|
|
|
__gen8_write(64)
|
2013-10-05 12:22:54 +08:00
|
|
|
__hsw_write(8)
|
|
|
|
__hsw_write(16)
|
|
|
|
__hsw_write(32)
|
|
|
|
__hsw_write(64)
|
|
|
|
__gen6_write(8)
|
|
|
|
__gen6_write(16)
|
|
|
|
__gen6_write(32)
|
|
|
|
__gen6_write(64)
|
|
|
|
|
2014-11-20 21:42:55 +08:00
|
|
|
#undef __gen9_write
|
2014-05-23 23:30:16 +08:00
|
|
|
#undef __chv_write
|
drm/i915/bdw: Handle forcewake for writes on gen8
GEN8 removes the GT FIFO which we've all come to know and love. Instead
it offers a wider range of optimized registers which always keep a
shadowed copy, and are fed to the GPU when it wakes.
How this is implemented in hardware is still somewhat of a mystery. As
far as I can tell, the basic design is as follows:
If the register is not optimized, you must use the old forcewake
mechanism to bring the GT out of sleep. [1]
If register is in the optimized list the write will signal that the
GT should begin to come out of whatever sleep state it is in.
While the GT is coming out of sleep, the requested write will be stored
in an intermediate shadow register.
Do to the fact that the implementation details are not clear, I see
several risks:
1. Order is not preserved as it is with GT FIFO. If we issue multiple
writes to optimized registers, where order matters, we may need to
serialize it with forcewake.
2. The optimized registers have only 1 shadowed slot, meaning if we
issue multiple writes to the same register, and those values need to
reach the GPU in order, forcewake will be required.
[1] We could implement a SW queue the way the GT FIFO used to work if
desired.
NOTE: Compile tested only until we get real silicon.
v2:
- Use a default case to make future platforms also work.
- Get rid of IS_BROADWELL since that's not yet defined, but we want to
MMIO as soon as possible.
v3: Apply suggestions from Mika's review:
- s/optimized/shadowed/
- invert the logic of the helper so that it does what it says (the
code itself was correct, just confusing to read).
v4:
- Squash in lost break.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-11-03 12:07:00 +08:00
|
|
|
#undef __gen8_write
|
2013-10-05 12:22:54 +08:00
|
|
|
#undef __hsw_write
|
|
|
|
#undef __gen6_write
|
2015-01-16 17:34:36 +08:00
|
|
|
#undef GEN6_WRITE_FOOTER
|
|
|
|
#undef GEN6_WRITE_HEADER
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2014-10-23 15:28:24 +08:00
|
|
|
#define ASSIGN_WRITE_MMIO_VFUNCS(x) \
|
|
|
|
do { \
|
|
|
|
dev_priv->uncore.funcs.mmio_writeb = x##_write8; \
|
|
|
|
dev_priv->uncore.funcs.mmio_writew = x##_write16; \
|
|
|
|
dev_priv->uncore.funcs.mmio_writel = x##_write32; \
|
|
|
|
dev_priv->uncore.funcs.mmio_writeq = x##_write64; \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
#define ASSIGN_READ_MMIO_VFUNCS(x) \
|
|
|
|
do { \
|
|
|
|
dev_priv->uncore.funcs.mmio_readb = x##_read8; \
|
|
|
|
dev_priv->uncore.funcs.mmio_readw = x##_read16; \
|
|
|
|
dev_priv->uncore.funcs.mmio_readl = x##_read32; \
|
|
|
|
dev_priv->uncore.funcs.mmio_readq = x##_read64; \
|
|
|
|
} while (0)
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
|
|
|
|
static void fw_domain_init(struct drm_i915_private *dev_priv,
|
2015-01-16 17:34:41 +08:00
|
|
|
enum forcewake_domain_id domain_id,
|
|
|
|
u32 reg_set, u32 reg_ack)
|
2015-01-19 22:20:43 +08:00
|
|
|
{
|
|
|
|
struct intel_uncore_forcewake_domain *d;
|
|
|
|
|
|
|
|
if (WARN_ON(domain_id >= FW_DOMAIN_ID_COUNT))
|
|
|
|
return;
|
|
|
|
|
|
|
|
d = &dev_priv->uncore.fw_domain[domain_id];
|
|
|
|
|
|
|
|
WARN_ON(d->wake_count);
|
|
|
|
|
|
|
|
d->wake_count = 0;
|
|
|
|
d->reg_set = reg_set;
|
|
|
|
d->reg_ack = reg_ack;
|
|
|
|
|
|
|
|
if (IS_GEN6(dev_priv)) {
|
|
|
|
d->val_reset = 0;
|
|
|
|
d->val_set = FORCEWAKE_KERNEL;
|
|
|
|
d->val_clear = 0;
|
|
|
|
} else {
|
|
|
|
d->val_reset = _MASKED_BIT_DISABLE(0xffff);
|
|
|
|
d->val_set = _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL);
|
|
|
|
d->val_clear = _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (IS_VALLEYVIEW(dev_priv))
|
|
|
|
d->reg_post = FORCEWAKE_ACK_VLV;
|
|
|
|
else if (IS_GEN6(dev_priv) || IS_GEN7(dev_priv) || IS_GEN8(dev_priv))
|
|
|
|
d->reg_post = ECOBUS;
|
|
|
|
else
|
|
|
|
d->reg_post = 0;
|
|
|
|
|
|
|
|
d->i915 = dev_priv;
|
|
|
|
d->id = domain_id;
|
|
|
|
|
2015-01-16 17:34:40 +08:00
|
|
|
setup_timer(&d->timer, intel_uncore_fw_release_timer, (unsigned long)d);
|
2015-01-19 22:20:43 +08:00
|
|
|
|
|
|
|
dev_priv->uncore.fw_domains |= (1 << domain_id);
|
2015-01-28 20:43:24 +08:00
|
|
|
|
|
|
|
fw_domain_reset(d);
|
2015-01-19 22:20:43 +08:00
|
|
|
}
|
|
|
|
|
2015-01-28 20:43:24 +08:00
|
|
|
static void intel_uncore_fw_domains_init(struct drm_device *dev)
|
2013-10-05 12:22:51 +08:00
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
|
2014-11-05 01:07:04 +08:00
|
|
|
if (IS_GEN9(dev)) {
|
2015-01-19 22:20:43 +08:00
|
|
|
dev_priv->uncore.funcs.force_wake_get = fw_domains_get;
|
|
|
|
dev_priv->uncore.funcs.force_wake_put = fw_domains_put;
|
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
|
|
|
|
FORCEWAKE_RENDER_GEN9,
|
|
|
|
FORCEWAKE_ACK_RENDER_GEN9);
|
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_BLITTER,
|
|
|
|
FORCEWAKE_BLITTER_GEN9,
|
|
|
|
FORCEWAKE_ACK_BLITTER_GEN9);
|
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_MEDIA,
|
|
|
|
FORCEWAKE_MEDIA_GEN9, FORCEWAKE_ACK_MEDIA_GEN9);
|
2014-11-05 01:07:04 +08:00
|
|
|
} else if (IS_VALLEYVIEW(dev)) {
|
2015-01-19 22:20:43 +08:00
|
|
|
dev_priv->uncore.funcs.force_wake_get = fw_domains_get;
|
2015-01-16 17:34:39 +08:00
|
|
|
if (!IS_CHERRYVIEW(dev))
|
|
|
|
dev_priv->uncore.funcs.force_wake_put =
|
|
|
|
fw_domains_put_with_fifo;
|
|
|
|
else
|
|
|
|
dev_priv->uncore.funcs.force_wake_put = fw_domains_put;
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
|
|
|
|
FORCEWAKE_VLV, FORCEWAKE_ACK_VLV);
|
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_MEDIA,
|
|
|
|
FORCEWAKE_MEDIA_VLV, FORCEWAKE_ACK_MEDIA_VLV);
|
2014-09-03 19:09:51 +08:00
|
|
|
} else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
|
2015-01-19 22:20:43 +08:00
|
|
|
dev_priv->uncore.funcs.force_wake_get =
|
|
|
|
fw_domains_get_with_thread_status;
|
|
|
|
dev_priv->uncore.funcs.force_wake_put = fw_domains_put;
|
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
|
|
|
|
FORCEWAKE_MT, FORCEWAKE_ACK_HSW);
|
2013-10-05 12:22:51 +08:00
|
|
|
} else if (IS_IVYBRIDGE(dev)) {
|
|
|
|
u32 ecobus;
|
|
|
|
|
|
|
|
/* IVB configs may use multi-threaded forcewake */
|
|
|
|
|
|
|
|
/* A small trick here - if the bios hasn't configured
|
|
|
|
* MT forcewake, and if the device is in RC6, then
|
|
|
|
* force_wake_mt_get will not wake the device and the
|
|
|
|
* ECOBUS read will return zero. Which will be
|
|
|
|
* (correctly) interpreted by the test below as MT
|
|
|
|
* forcewake being disabled.
|
|
|
|
*/
|
2015-01-19 22:20:43 +08:00
|
|
|
dev_priv->uncore.funcs.force_wake_get =
|
|
|
|
fw_domains_get_with_thread_status;
|
|
|
|
dev_priv->uncore.funcs.force_wake_put =
|
|
|
|
fw_domains_put_with_fifo;
|
|
|
|
|
2015-01-28 20:43:24 +08:00
|
|
|
/* We need to init first for ECOBUS access and then
|
|
|
|
* determine later if we want to reinit, in case of MT access is
|
|
|
|
* not working
|
|
|
|
*/
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
|
|
|
|
FORCEWAKE_MT, FORCEWAKE_MT_ACK);
|
2015-01-28 20:43:24 +08:00
|
|
|
|
2013-10-05 12:22:51 +08:00
|
|
|
mutex_lock(&dev->struct_mutex);
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domains_get_with_thread_status(dev_priv, FORCEWAKE_ALL);
|
2013-10-05 12:22:51 +08:00
|
|
|
ecobus = __raw_i915_read32(dev_priv, ECOBUS);
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domains_put_with_fifo(dev_priv, FORCEWAKE_ALL);
|
2013-10-05 12:22:51 +08:00
|
|
|
mutex_unlock(&dev->struct_mutex);
|
|
|
|
|
2015-01-19 22:20:43 +08:00
|
|
|
if (!(ecobus & FORCEWAKE_MT_ENABLE)) {
|
2013-10-05 12:22:51 +08:00
|
|
|
DRM_INFO("No MT forcewake available on Ivybridge, this can result in issues\n");
|
|
|
|
DRM_INFO("when using vblank-synced partial screen updates.\n");
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
|
|
|
|
FORCEWAKE, FORCEWAKE_ACK);
|
2013-10-05 12:22:51 +08:00
|
|
|
}
|
|
|
|
} else if (IS_GEN6(dev)) {
|
|
|
|
dev_priv->uncore.funcs.force_wake_get =
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domains_get_with_thread_status;
|
2013-10-05 12:22:51 +08:00
|
|
|
dev_priv->uncore.funcs.force_wake_put =
|
2015-01-19 22:20:43 +08:00
|
|
|
fw_domains_put_with_fifo;
|
|
|
|
fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
|
|
|
|
FORCEWAKE, FORCEWAKE_ACK);
|
2013-10-05 12:22:51 +08:00
|
|
|
}
|
2015-01-28 20:43:24 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
void intel_uncore_init(struct drm_device *dev)
|
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
|
|
|
|
intel_uncore_ellc_detect(dev);
|
|
|
|
intel_uncore_fw_domains_init(dev);
|
|
|
|
__intel_uncore_early_sanitize(dev, false);
|
2013-10-05 12:22:51 +08:00
|
|
|
|
2013-10-05 12:22:53 +08:00
|
|
|
switch (INTEL_INFO(dev)->gen) {
|
drm/i915/bdw: Handle forcewake for writes on gen8
GEN8 removes the GT FIFO which we've all come to know and love. Instead
it offers a wider range of optimized registers which always keep a
shadowed copy, and are fed to the GPU when it wakes.
How this is implemented in hardware is still somewhat of a mystery. As
far as I can tell, the basic design is as follows:
If the register is not optimized, you must use the old forcewake
mechanism to bring the GT out of sleep. [1]
If register is in the optimized list the write will signal that the
GT should begin to come out of whatever sleep state it is in.
While the GT is coming out of sleep, the requested write will be stored
in an intermediate shadow register.
Do to the fact that the implementation details are not clear, I see
several risks:
1. Order is not preserved as it is with GT FIFO. If we issue multiple
writes to optimized registers, where order matters, we may need to
serialize it with forcewake.
2. The optimized registers have only 1 shadowed slot, meaning if we
issue multiple writes to the same register, and those values need to
reach the GPU in order, forcewake will be required.
[1] We could implement a SW queue the way the GT FIFO used to work if
desired.
NOTE: Compile tested only until we get real silicon.
v2:
- Use a default case to make future platforms also work.
- Get rid of IS_BROADWELL since that's not yet defined, but we want to
MMIO as soon as possible.
v3: Apply suggestions from Mika's review:
- s/optimized/shadowed/
- invert the logic of the helper so that it does what it says (the
code itself was correct, just confusing to read).
v4:
- Squash in lost break.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-11-03 12:07:00 +08:00
|
|
|
default:
|
2014-12-08 23:40:10 +08:00
|
|
|
MISSING_CASE(INTEL_INFO(dev)->gen);
|
2014-11-20 21:42:55 +08:00
|
|
|
return;
|
|
|
|
case 9:
|
|
|
|
ASSIGN_WRITE_MMIO_VFUNCS(gen9);
|
|
|
|
ASSIGN_READ_MMIO_VFUNCS(gen9);
|
|
|
|
break;
|
|
|
|
case 8:
|
2014-05-23 23:30:16 +08:00
|
|
|
if (IS_CHERRYVIEW(dev)) {
|
2014-10-23 15:28:24 +08:00
|
|
|
ASSIGN_WRITE_MMIO_VFUNCS(chv);
|
|
|
|
ASSIGN_READ_MMIO_VFUNCS(chv);
|
2014-05-23 23:30:16 +08:00
|
|
|
|
|
|
|
} else {
|
2014-10-23 15:28:24 +08:00
|
|
|
ASSIGN_WRITE_MMIO_VFUNCS(gen8);
|
|
|
|
ASSIGN_READ_MMIO_VFUNCS(gen6);
|
2014-05-23 23:30:16 +08:00
|
|
|
}
|
drm/i915/bdw: Handle forcewake for writes on gen8
GEN8 removes the GT FIFO which we've all come to know and love. Instead
it offers a wider range of optimized registers which always keep a
shadowed copy, and are fed to the GPU when it wakes.
How this is implemented in hardware is still somewhat of a mystery. As
far as I can tell, the basic design is as follows:
If the register is not optimized, you must use the old forcewake
mechanism to bring the GT out of sleep. [1]
If register is in the optimized list the write will signal that the
GT should begin to come out of whatever sleep state it is in.
While the GT is coming out of sleep, the requested write will be stored
in an intermediate shadow register.
Do to the fact that the implementation details are not clear, I see
several risks:
1. Order is not preserved as it is with GT FIFO. If we issue multiple
writes to optimized registers, where order matters, we may need to
serialize it with forcewake.
2. The optimized registers have only 1 shadowed slot, meaning if we
issue multiple writes to the same register, and those values need to
reach the GPU in order, forcewake will be required.
[1] We could implement a SW queue the way the GT FIFO used to work if
desired.
NOTE: Compile tested only until we get real silicon.
v2:
- Use a default case to make future platforms also work.
- Get rid of IS_BROADWELL since that's not yet defined, but we want to
MMIO as soon as possible.
v3: Apply suggestions from Mika's review:
- s/optimized/shadowed/
- invert the logic of the helper so that it does what it says (the
code itself was correct, just confusing to read).
v4:
- Squash in lost break.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net> (v1)
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-11-03 12:07:00 +08:00
|
|
|
break;
|
2013-10-05 12:22:53 +08:00
|
|
|
case 7:
|
|
|
|
case 6:
|
2013-10-05 12:22:54 +08:00
|
|
|
if (IS_HASWELL(dev)) {
|
2014-10-23 15:28:24 +08:00
|
|
|
ASSIGN_WRITE_MMIO_VFUNCS(hsw);
|
2013-10-05 12:22:54 +08:00
|
|
|
} else {
|
2014-10-23 15:28:24 +08:00
|
|
|
ASSIGN_WRITE_MMIO_VFUNCS(gen6);
|
2013-10-05 12:22:54 +08:00
|
|
|
}
|
2013-11-23 17:25:43 +08:00
|
|
|
|
|
|
|
if (IS_VALLEYVIEW(dev)) {
|
2014-10-23 15:28:24 +08:00
|
|
|
ASSIGN_READ_MMIO_VFUNCS(vlv);
|
2013-11-23 17:25:43 +08:00
|
|
|
} else {
|
2014-10-23 15:28:24 +08:00
|
|
|
ASSIGN_READ_MMIO_VFUNCS(gen6);
|
2013-11-23 17:25:43 +08:00
|
|
|
}
|
2013-10-05 12:22:53 +08:00
|
|
|
break;
|
|
|
|
case 5:
|
2014-10-23 15:28:24 +08:00
|
|
|
ASSIGN_WRITE_MMIO_VFUNCS(gen5);
|
|
|
|
ASSIGN_READ_MMIO_VFUNCS(gen5);
|
2013-10-05 12:22:53 +08:00
|
|
|
break;
|
|
|
|
case 4:
|
|
|
|
case 3:
|
|
|
|
case 2:
|
2015-01-16 17:34:36 +08:00
|
|
|
ASSIGN_WRITE_MMIO_VFUNCS(gen2);
|
|
|
|
ASSIGN_READ_MMIO_VFUNCS(gen2);
|
2013-10-05 12:22:53 +08:00
|
|
|
break;
|
|
|
|
}
|
2014-10-24 00:23:21 +08:00
|
|
|
|
|
|
|
i915_check_and_clear_faults(dev);
|
2013-10-05 12:22:51 +08:00
|
|
|
}
|
2014-10-23 15:28:24 +08:00
|
|
|
#undef ASSIGN_WRITE_MMIO_VFUNCS
|
|
|
|
#undef ASSIGN_READ_MMIO_VFUNCS
|
2013-10-05 12:22:51 +08:00
|
|
|
|
|
|
|
void intel_uncore_fini(struct drm_device *dev)
|
|
|
|
{
|
|
|
|
/* Paranoia: make sure we have disabled everything before we exit. */
|
|
|
|
intel_uncore_sanitize(dev);
|
2014-03-13 20:00:29 +08:00
|
|
|
intel_uncore_forcewake_reset(dev, false);
|
2013-10-05 12:22:51 +08:00
|
|
|
}
|
|
|
|
|
drm/i915: Use a macro to express the range of valid gens for reg_read
The reg_read whitelist has a gen bitmask to code the gens we're allowing
the register to be read on. Until now, it was a literal, but we can be
a bit more expressive.
To ease the review, a small test program:
$ cat bit-range.c
#include <stdio.h>
#include <stdint.h>
#define U32_C(x) x ## U
#define GENMASK(h, l) (((U32_C(1) << ((h) - (l) + 1)) - 1) << (l))
#define GEN_RANGE(l, h) GENMASK(h, l)
int main(int argc, char **argv)
{
printf("0x%08x\n", GEN_RANGE(1, 1));
printf("0x%08x\n", GEN_RANGE(1, 2));
printf("0x%08x\n", GEN_RANGE(4, 4));
printf("0x%08x\n", GEN_RANGE(4, 5));
printf("0x%08x\n", GEN_RANGE(1, 31));
printf("0x%08x\n", GEN_RANGE(4, 8));
return 0;
}
$ ./bit-range
0x00000002
0x00000006
0x00000010
0x00000030
0xfffffffe
0x000001f0
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-03-31 18:24:08 +08:00
|
|
|
#define GEN_RANGE(l, h) GENMASK(h, l)
|
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
static const struct register_whitelist {
|
|
|
|
uint64_t offset;
|
|
|
|
uint32_t size;
|
drm/i915: Use a macro to express the range of valid gens for reg_read
The reg_read whitelist has a gen bitmask to code the gens we're allowing
the register to be read on. Until now, it was a literal, but we can be
a bit more expressive.
To ease the review, a small test program:
$ cat bit-range.c
#include <stdio.h>
#include <stdint.h>
#define U32_C(x) x ## U
#define GENMASK(h, l) (((U32_C(1) << ((h) - (l) + 1)) - 1) << (l))
#define GEN_RANGE(l, h) GENMASK(h, l)
int main(int argc, char **argv)
{
printf("0x%08x\n", GEN_RANGE(1, 1));
printf("0x%08x\n", GEN_RANGE(1, 2));
printf("0x%08x\n", GEN_RANGE(4, 4));
printf("0x%08x\n", GEN_RANGE(4, 5));
printf("0x%08x\n", GEN_RANGE(1, 31));
printf("0x%08x\n", GEN_RANGE(4, 8));
return 0;
}
$ ./bit-range
0x00000002
0x00000006
0x00000010
0x00000030
0xfffffffe
0x000001f0
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-03-31 18:24:08 +08:00
|
|
|
/* supported gens, 0x10 for 4, 0x30 for 4 and 5, etc. */
|
|
|
|
uint32_t gen_bitmask;
|
2013-07-20 03:36:52 +08:00
|
|
|
} whitelist[] = {
|
2014-03-30 23:28:23 +08:00
|
|
|
{ RING_TIMESTAMP(RENDER_RING_BASE), 8, GEN_RANGE(4, 9) },
|
2013-07-20 03:36:52 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
int i915_reg_read_ioctl(struct drm_device *dev,
|
|
|
|
void *data, struct drm_file *file)
|
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
struct drm_i915_reg_read *reg = data;
|
|
|
|
struct register_whitelist const *entry = whitelist;
|
2014-04-02 01:55:08 +08:00
|
|
|
int i, ret = 0;
|
2013-07-20 03:36:52 +08:00
|
|
|
|
|
|
|
for (i = 0; i < ARRAY_SIZE(whitelist); i++, entry++) {
|
|
|
|
if (entry->offset == reg->offset &&
|
|
|
|
(1 << INTEL_INFO(dev)->gen & entry->gen_bitmask))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (i == ARRAY_SIZE(whitelist))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2014-04-02 01:55:08 +08:00
|
|
|
intel_runtime_pm_get(dev_priv);
|
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
switch (entry->size) {
|
|
|
|
case 8:
|
|
|
|
reg->val = I915_READ64(reg->offset);
|
|
|
|
break;
|
|
|
|
case 4:
|
|
|
|
reg->val = I915_READ(reg->offset);
|
|
|
|
break;
|
|
|
|
case 2:
|
|
|
|
reg->val = I915_READ16(reg->offset);
|
|
|
|
break;
|
|
|
|
case 1:
|
|
|
|
reg->val = I915_READ8(reg->offset);
|
|
|
|
break;
|
|
|
|
default:
|
2014-12-08 23:40:10 +08:00
|
|
|
MISSING_CASE(entry->size);
|
2014-04-02 01:55:08 +08:00
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2014-04-02 01:55:08 +08:00
|
|
|
out:
|
|
|
|
intel_runtime_pm_put(dev_priv);
|
|
|
|
return ret;
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2013-10-30 21:44:16 +08:00
|
|
|
int i915_get_reset_stats_ioctl(struct drm_device *dev,
|
|
|
|
void *data, struct drm_file *file)
|
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
struct drm_i915_reset_stats *args = data;
|
|
|
|
struct i915_ctx_hang_stats *hs;
|
2014-05-22 21:13:37 +08:00
|
|
|
struct intel_context *ctx;
|
2013-10-30 21:44:16 +08:00
|
|
|
int ret;
|
|
|
|
|
2013-11-13 01:49:35 +08:00
|
|
|
if (args->flags || args->pad)
|
|
|
|
return -EINVAL;
|
|
|
|
|
drm/i915: Emphasize that ctx->id is merely a user handle
This is an Execlists preparatory patch, since they make context ID become an
overloaded term:
- In the software, it was used to distinguish which context userspace was
trying to use.
- In the BSpec, the term is used to describe the 20-bits long field the
hardware uses to it to discriminate the contexts that are submitted to
the ELSP and inform the driver about their current status (via Context
Switch Interrupts and Context Status Buffers).
Initially, I tried to make the different meanings converge, but it proved
impossible:
- The software ctx->id is per-filp, while the hardware one needs to be
globally unique.
- Also, we multiplex several backing states objects per intel_context,
and all of them need unique HW IDs.
- I tried adding a per-filp ID and then composing the HW context ID as:
ctx->id + file_priv->id + ring->id, but the fact that the hardware only
uses 20-bits means we have to artificially limit the number of filps or
contexts the userspace can create.
The ctx->user_handle renaming bits are done with this Cocci patch (plus
manual frobbing of the struct declaration):
@@
struct intel_context c;
@@
- (c).id
+ c.user_handle
@@
struct intel_context *c;
@@
- (c)->id
+ c->user_handle
Also, while we are at it, s/DEFAULT_CONTEXT_ID/DEFAULT_CONTEXT_HANDLE and
change the type to unsigned 32 bits.
v2: s/handle/user_handle and change the type to uint32_t as suggested by
Chris Wilson.
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v1)
Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2014-07-03 23:28:00 +08:00
|
|
|
if (args->ctx_id == DEFAULT_CONTEXT_HANDLE && !capable(CAP_SYS_ADMIN))
|
2013-10-30 21:44:16 +08:00
|
|
|
return -EPERM;
|
|
|
|
|
|
|
|
ret = mutex_lock_interruptible(&dev->struct_mutex);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
drm/i915: Get context early in execbuf
We need to have the address space when reserving space for the objects.
Since the address space and context are tied together, and reserve
occurs before context switch (for good reason), we must lookup our
context earlier in the process.
This leaves some room for optimizations where we no longer need to use
ctx_id in certain places. This will be addressed in a subsequent patch.
Important tricky bit:
Because slow relocations during execbuffer drop struct_mutex
Perhaps it would be best to acquire the reference when we get the
context, but I'll save that for another day (note I have written the
patch before, and I found the changes required to be uglier than this).
Note that since we currently access everything via context id, and not
the data structure this is fine, though not desirable. The next change
attempts to get the context only once via the context ID idr lookup, and
as such, the following can happen:
CTX-A is created, refcount = 1
CTX-A execbuf, mutex dropped
close IOCTL called on CTX-A, refcount = 0
CTX-A resumes in execbuf.
v2: Rebased on top of
commit b6359918b885da7c7b58c050674278dbd06020ab
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date: Wed Oct 30 15:44:16 2013 +0200
drm/i915: add i915_get_reset_stats_ioctl
v3: Rebased on top of
commit 25b3dfc87bff80317d67ddd2cd4cfb91e6fe7d79
Author: Mika Westerberg <mika.westerberg@linux.intel.com>
Date: Tue Nov 12 11:57:30 2013 +0200
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date: Tue Nov 26 16:14:33 2013 +0200
drm/i915: check context reset stats before relocations
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-07 06:11:21 +08:00
|
|
|
ctx = i915_gem_context_get(file->driver_priv, args->ctx_id);
|
|
|
|
if (IS_ERR(ctx)) {
|
2013-10-30 21:44:16 +08:00
|
|
|
mutex_unlock(&dev->struct_mutex);
|
drm/i915: Get context early in execbuf
We need to have the address space when reserving space for the objects.
Since the address space and context are tied together, and reserve
occurs before context switch (for good reason), we must lookup our
context earlier in the process.
This leaves some room for optimizations where we no longer need to use
ctx_id in certain places. This will be addressed in a subsequent patch.
Important tricky bit:
Because slow relocations during execbuffer drop struct_mutex
Perhaps it would be best to acquire the reference when we get the
context, but I'll save that for another day (note I have written the
patch before, and I found the changes required to be uglier than this).
Note that since we currently access everything via context id, and not
the data structure this is fine, though not desirable. The next change
attempts to get the context only once via the context ID idr lookup, and
as such, the following can happen:
CTX-A is created, refcount = 1
CTX-A execbuf, mutex dropped
close IOCTL called on CTX-A, refcount = 0
CTX-A resumes in execbuf.
v2: Rebased on top of
commit b6359918b885da7c7b58c050674278dbd06020ab
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date: Wed Oct 30 15:44:16 2013 +0200
drm/i915: add i915_get_reset_stats_ioctl
v3: Rebased on top of
commit 25b3dfc87bff80317d67ddd2cd4cfb91e6fe7d79
Author: Mika Westerberg <mika.westerberg@linux.intel.com>
Date: Tue Nov 12 11:57:30 2013 +0200
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date: Tue Nov 26 16:14:33 2013 +0200
drm/i915: check context reset stats before relocations
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-07 06:11:21 +08:00
|
|
|
return PTR_ERR(ctx);
|
2013-10-30 21:44:16 +08:00
|
|
|
}
|
drm/i915: Get context early in execbuf
We need to have the address space when reserving space for the objects.
Since the address space and context are tied together, and reserve
occurs before context switch (for good reason), we must lookup our
context earlier in the process.
This leaves some room for optimizations where we no longer need to use
ctx_id in certain places. This will be addressed in a subsequent patch.
Important tricky bit:
Because slow relocations during execbuffer drop struct_mutex
Perhaps it would be best to acquire the reference when we get the
context, but I'll save that for another day (note I have written the
patch before, and I found the changes required to be uglier than this).
Note that since we currently access everything via context id, and not
the data structure this is fine, though not desirable. The next change
attempts to get the context only once via the context ID idr lookup, and
as such, the following can happen:
CTX-A is created, refcount = 1
CTX-A execbuf, mutex dropped
close IOCTL called on CTX-A, refcount = 0
CTX-A resumes in execbuf.
v2: Rebased on top of
commit b6359918b885da7c7b58c050674278dbd06020ab
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date: Wed Oct 30 15:44:16 2013 +0200
drm/i915: add i915_get_reset_stats_ioctl
v3: Rebased on top of
commit 25b3dfc87bff80317d67ddd2cd4cfb91e6fe7d79
Author: Mika Westerberg <mika.westerberg@linux.intel.com>
Date: Tue Nov 12 11:57:30 2013 +0200
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date: Tue Nov 26 16:14:33 2013 +0200
drm/i915: check context reset stats before relocations
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2013-12-07 06:11:21 +08:00
|
|
|
hs = &ctx->hang_stats;
|
2013-10-30 21:44:16 +08:00
|
|
|
|
|
|
|
if (capable(CAP_SYS_ADMIN))
|
|
|
|
args->reset_count = i915_reset_count(&dev_priv->gpu_error);
|
|
|
|
else
|
|
|
|
args->reset_count = 0;
|
|
|
|
|
|
|
|
args->batch_active = hs->batch_active;
|
|
|
|
args->batch_pending = hs->batch_pending;
|
|
|
|
|
|
|
|
mutex_unlock(&dev->struct_mutex);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2014-11-22 03:54:27 +08:00
|
|
|
static int i915_reset_complete(struct drm_device *dev)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
|
|
|
u8 gdrst;
|
2014-11-22 03:54:27 +08:00
|
|
|
pci_read_config_byte(dev->pdev, I915_GDRST, &gdrst);
|
2014-11-22 03:54:25 +08:00
|
|
|
return (gdrst & GRDOM_RESET_STATUS) == 0;
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2014-11-22 03:54:27 +08:00
|
|
|
static int i915_do_reset(struct drm_device *dev)
|
2013-07-20 03:36:52 +08:00
|
|
|
{
|
2014-11-22 03:54:25 +08:00
|
|
|
/* assert reset for at least 20 usec */
|
2014-11-22 03:54:27 +08:00
|
|
|
pci_write_config_byte(dev->pdev, I915_GDRST, GRDOM_RESET_ENABLE);
|
2014-11-22 03:54:25 +08:00
|
|
|
udelay(20);
|
2014-11-22 03:54:27 +08:00
|
|
|
pci_write_config_byte(dev->pdev, I915_GDRST, 0);
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2014-11-22 03:54:27 +08:00
|
|
|
return wait_for(i915_reset_complete(dev), 500);
|
2014-11-22 03:54:25 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int g4x_reset_complete(struct drm_device *dev)
|
|
|
|
{
|
|
|
|
u8 gdrst;
|
2014-11-22 03:54:27 +08:00
|
|
|
pci_read_config_byte(dev->pdev, I915_GDRST, &gdrst);
|
2014-11-22 03:54:25 +08:00
|
|
|
return (gdrst & GRDOM_RESET_ENABLE) == 0;
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
2014-11-22 03:54:28 +08:00
|
|
|
static int g33_do_reset(struct drm_device *dev)
|
|
|
|
{
|
|
|
|
pci_write_config_byte(dev->pdev, I915_GDRST, GRDOM_RESET_ENABLE);
|
|
|
|
return wait_for(g4x_reset_complete(dev), 500);
|
|
|
|
}
|
|
|
|
|
2014-05-20 00:23:27 +08:00
|
|
|
static int g4x_do_reset(struct drm_device *dev)
|
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
int ret;
|
|
|
|
|
2014-11-22 03:54:27 +08:00
|
|
|
pci_write_config_byte(dev->pdev, I915_GDRST,
|
2014-05-20 00:23:27 +08:00
|
|
|
GRDOM_RENDER | GRDOM_RESET_ENABLE);
|
2014-11-22 03:54:25 +08:00
|
|
|
ret = wait_for(g4x_reset_complete(dev), 500);
|
2014-05-20 00:23:27 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
/* WaVcpClkGateDisableForMediaReset:ctg,elk */
|
|
|
|
I915_WRITE(VDECCLK_GATE_D, I915_READ(VDECCLK_GATE_D) | VCP_UNIT_CLOCK_GATE_DISABLE);
|
|
|
|
POSTING_READ(VDECCLK_GATE_D);
|
|
|
|
|
2014-11-22 03:54:27 +08:00
|
|
|
pci_write_config_byte(dev->pdev, I915_GDRST,
|
2014-05-20 00:23:27 +08:00
|
|
|
GRDOM_MEDIA | GRDOM_RESET_ENABLE);
|
2014-11-22 03:54:25 +08:00
|
|
|
ret = wait_for(g4x_reset_complete(dev), 500);
|
2014-05-20 00:23:27 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
/* WaVcpClkGateDisableForMediaReset:ctg,elk */
|
|
|
|
I915_WRITE(VDECCLK_GATE_D, I915_READ(VDECCLK_GATE_D) & ~VCP_UNIT_CLOCK_GATE_DISABLE);
|
|
|
|
POSTING_READ(VDECCLK_GATE_D);
|
|
|
|
|
2014-11-22 03:54:27 +08:00
|
|
|
pci_write_config_byte(dev->pdev, I915_GDRST, 0);
|
2014-05-20 00:23:27 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
static int ironlake_do_reset(struct drm_device *dev)
|
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
I915_WRITE(MCHBAR_MIRROR_BASE + ILK_GDSR,
|
2014-05-20 00:23:25 +08:00
|
|
|
ILK_GRDOM_RENDER | ILK_GRDOM_RESET_ENABLE);
|
2014-05-20 00:23:23 +08:00
|
|
|
ret = wait_for((I915_READ(MCHBAR_MIRROR_BASE + ILK_GDSR) &
|
2014-05-20 00:23:24 +08:00
|
|
|
ILK_GRDOM_RESET_ENABLE) == 0, 500);
|
2013-07-20 03:36:52 +08:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
I915_WRITE(MCHBAR_MIRROR_BASE + ILK_GDSR,
|
2014-05-20 00:23:25 +08:00
|
|
|
ILK_GRDOM_MEDIA | ILK_GRDOM_RESET_ENABLE);
|
2014-05-20 00:23:26 +08:00
|
|
|
ret = wait_for((I915_READ(MCHBAR_MIRROR_BASE + ILK_GDSR) &
|
|
|
|
ILK_GRDOM_RESET_ENABLE) == 0, 500);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
I915_WRITE(MCHBAR_MIRROR_BASE + ILK_GDSR, 0);
|
|
|
|
|
|
|
|
return 0;
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int gen6_do_reset(struct drm_device *dev)
|
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
/* Reset the chip */
|
|
|
|
|
|
|
|
/* GEN6_GDRST is not in the gt power well, no need to check
|
|
|
|
* for fifo space for the write or forcewake the chip for
|
|
|
|
* the read
|
|
|
|
*/
|
2013-07-20 03:36:53 +08:00
|
|
|
__raw_i915_write32(dev_priv, GEN6_GDRST, GEN6_GRDOM_FULL);
|
2013-07-20 03:36:52 +08:00
|
|
|
|
|
|
|
/* Spin waiting for the device to ack the reset request */
|
2013-07-20 03:36:53 +08:00
|
|
|
ret = wait_for((__raw_i915_read32(dev_priv, GEN6_GDRST) & GEN6_GRDOM_FULL) == 0, 500);
|
2013-07-20 03:36:52 +08:00
|
|
|
|
2014-03-13 20:00:29 +08:00
|
|
|
intel_uncore_forcewake_reset(dev, true);
|
2014-03-06 00:08:18 +08:00
|
|
|
|
2013-07-20 03:36:52 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
int intel_gpu_reset(struct drm_device *dev)
|
|
|
|
{
|
2014-06-03 16:08:26 +08:00
|
|
|
if (INTEL_INFO(dev)->gen >= 6)
|
|
|
|
return gen6_do_reset(dev);
|
|
|
|
else if (IS_GEN5(dev))
|
|
|
|
return ironlake_do_reset(dev);
|
|
|
|
else if (IS_G4X(dev))
|
|
|
|
return g4x_do_reset(dev);
|
2014-11-22 03:54:28 +08:00
|
|
|
else if (IS_G33(dev))
|
|
|
|
return g33_do_reset(dev);
|
|
|
|
else if (INTEL_INFO(dev)->gen >= 3)
|
2014-11-22 03:54:27 +08:00
|
|
|
return i915_do_reset(dev);
|
2014-06-03 16:08:26 +08:00
|
|
|
else
|
|
|
|
return -ENODEV;
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
void intel_uncore_check_errors(struct drm_device *dev)
|
|
|
|
{
|
|
|
|
struct drm_i915_private *dev_priv = dev->dev_private;
|
|
|
|
|
|
|
|
if (HAS_FPGA_DBG_UNCLAIMED(dev) &&
|
2013-07-20 03:36:53 +08:00
|
|
|
(__raw_i915_read32(dev_priv, FPGA_DBG) & FPGA_DBG_RM_NOCLAIM)) {
|
2013-07-20 03:36:52 +08:00
|
|
|
DRM_ERROR("Unclaimed register before interrupt\n");
|
2013-07-20 03:36:53 +08:00
|
|
|
__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
|
2013-07-20 03:36:52 +08:00
|
|
|
}
|
|
|
|
}
|