OpenCloudOS-Kernel/drivers/gpu/drm/i915/intel_uncore.c

1848 lines
52 KiB
C
Raw Normal View History

/*
* 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"
drm/i915: Introduce a PV INFO page structure for Intel GVT-g. Introduce a PV INFO structure, to facilitate the Intel GVT-g technology, which is a GPU virtualization solution with mediated pass-through. This page contains the shared information between i915 driver and the host emulator. For now, this structure utilizes an area of 4K bytes on HSW GPU's unused MMIO space. Future hardware will have the reserved window architecturally defined, and layout of the page will be added in future BSpec. The i915 driver load routine detects if it is running in a VM by reading the contents of this PV INFO page. Thereafter a flag, vgpu.active is set, and intel_vgpu_active() is used by checking this flag to conclude if GPU is virtualized with Intel GVT-g. By now, intel_vgpu_active() will return true, only when the driver is running as a guest in the Intel GVT-g enhanced environment on HSW platform. v2: take Chris' comments: - call the i915_check_vgpu() in intel_uncore_init() - sanitize i915_check_vgpu() by adding BUILD_BUG_ON() and debug info take Daniel's comments: - put the definition of PV INFO into a new header - i915_vgt_if.h other changes: - access mmio regs by readq/readw in i915_check_vgpu() v3: take Daniel's comments: - move the i915/vgt interfaces into a new i915_vgpu.c - update makefile - add kerneldoc to functions which are non-static - add a DOC: section describing some of the high-level design - update drm docbook other changes: - rename i915_vgt_if.h to i915_vgpu.h v4: take Tvrtko's comments: - fix a typo in commit message - add debug message when vgt version mismatches - rename low_gmadr/high_gmadr to mappable/non-mappable in PV INFO structure Signed-off-by: Yu Zhang <yu.c.zhang@linux.intel.com> Signed-off-by: Jike Song <jike.song@intel.com> Signed-off-by: Eddie Dong <eddie.dong@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-10 19:05:47 +08:00
#include "i915_vgpu.h"
drm/i915: Listen for PMIC bus access notifications Listen for PMIC bus access notifications and get FORCEWAKE_ALL while the bus is accessed to avoid needing to do any forcewakes, which need PMIC bus access, while the PMIC bus is busy: This fixes errors like these showing up in dmesg, usually followed by a gfx or system freeze: [drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request. [drm:fw_domains_get [i915]] *MEDIA* render: timed out waiting for forcewake ack request. i2c_designware 808622C1:06: punit semaphore timed out, resetting i2c_designware 808622C1:06: PUNIT SEM: 2 i2c_designware 808622C1:06: couldn't acquire bus ownership Downside of this approach is that it causes wakeups whenever the PMIC bus is accessed. Unfortunately we cannot simply wait for the PMIC bus to go idle when we hit a race, as forcewakes may be done from interrupt handlers where we cannot sleep to wait for the i2c PMIC bus access to finish. Note that the notifications and thus the wakeups will only happen on baytrail / cherrytrail devices using PMICs with a shared i2c bus for P-Unit and host PMIC access (i2c busses with a _SEM method in their APCI node), e.g. an axp288 PMIC. I plan to write some patches for drivers accessing the PMIC bus to limit their bus accesses to a bare minimum (e.g. cache registers, do not update battery level more often then 4 times a minute), to limit the amount of wakeups. BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=155241 Signed-off-by: Hans de Goede <hdegoede@redhat.com> Tested-by: tagorereddy <tagore.chandan@gmail.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> [danvet: Wiggle in conflicts.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2017-02-10 18:28:02 +08:00
#include <asm/iosf_mbi.h>
#include <linux/pm_runtime.h>
#define FORCEWAKE_ACK_TIMEOUT_MS 50
#define GT_FIFO_TIMEOUT_MS 10
#define __raw_posting_read(dev_priv__, reg__) (void)__raw_i915_read32((dev_priv__), (reg__))
static const char * const forcewake_domain_names[] = {
"render",
"blitter",
"media",
};
const char *
intel_uncore_forcewake_domain_to_str(const enum forcewake_domain_id id)
{
BUILD_BUG_ON(ARRAY_SIZE(forcewake_domain_names) != FW_DOMAIN_ID_COUNT);
if (id >= 0 && id < FW_DOMAIN_ID_COUNT)
return forcewake_domain_names[id];
WARN_ON(id);
return "unknown";
}
static inline void
fw_domain_reset(struct drm_i915_private *i915,
const struct intel_uncore_forcewake_domain *d)
{
__raw_i915_write32(i915, d->reg_set, i915->uncore.fw_reset);
}
static inline void
fw_domain_arm_timer(struct intel_uncore_forcewake_domain *d)
{
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
d->wake_count++;
hrtimer_start_range_ns(&d->timer,
NSEC_PER_MSEC,
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
NSEC_PER_MSEC,
HRTIMER_MODE_REL);
}
static inline void
fw_domain_wait_ack_clear(const struct drm_i915_private *i915,
const struct intel_uncore_forcewake_domain *d)
{
if (wait_for_atomic((__raw_i915_read32(i915, d->reg_ack) &
FORCEWAKE_KERNEL) == 0,
FORCEWAKE_ACK_TIMEOUT_MS))
DRM_ERROR("%s: timed out waiting for forcewake ack to clear.\n",
intel_uncore_forcewake_domain_to_str(d->id));
}
static inline void
fw_domain_get(struct drm_i915_private *i915,
const struct intel_uncore_forcewake_domain *d)
{
__raw_i915_write32(i915, d->reg_set, i915->uncore.fw_set);
}
static inline void
fw_domain_wait_ack(const struct drm_i915_private *i915,
const struct intel_uncore_forcewake_domain *d)
{
if (wait_for_atomic((__raw_i915_read32(i915, d->reg_ack) &
FORCEWAKE_KERNEL),
FORCEWAKE_ACK_TIMEOUT_MS))
DRM_ERROR("%s: timed out waiting for forcewake ack request.\n",
intel_uncore_forcewake_domain_to_str(d->id));
}
static inline void
fw_domain_put(const struct drm_i915_private *i915,
const struct intel_uncore_forcewake_domain *d)
{
__raw_i915_write32(i915, d->reg_set, i915->uncore.fw_clear);
}
static void
fw_domains_get(struct drm_i915_private *i915, enum forcewake_domains fw_domains)
{
struct intel_uncore_forcewake_domain *d;
unsigned int tmp;
GEM_BUG_ON(fw_domains & ~i915->uncore.fw_domains);
for_each_fw_domain_masked(d, fw_domains, i915, tmp) {
fw_domain_wait_ack_clear(i915, d);
fw_domain_get(i915, d);
}
for_each_fw_domain_masked(d, fw_domains, i915, tmp)
fw_domain_wait_ack(i915, d);
i915->uncore.fw_domains_active |= fw_domains;
}
static void
fw_domains_put(struct drm_i915_private *i915, enum forcewake_domains fw_domains)
{
struct intel_uncore_forcewake_domain *d;
unsigned int tmp;
GEM_BUG_ON(fw_domains & ~i915->uncore.fw_domains);
for_each_fw_domain_masked(d, fw_domains, i915, tmp)
fw_domain_put(i915, d);
i915->uncore.fw_domains_active &= ~fw_domains;
}
static void
fw_domains_reset(struct drm_i915_private *i915,
enum forcewake_domains fw_domains)
{
struct intel_uncore_forcewake_domain *d;
unsigned int tmp;
if (!fw_domains)
return;
GEM_BUG_ON(fw_domains & ~i915->uncore.fw_domains);
for_each_fw_domain_masked(d, fw_domains, i915, tmp)
fw_domain_reset(i915, d);
}
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,
enum forcewake_domains fw_domains)
{
fw_domains_get(dev_priv, fw_domains);
/* WaRsForcewakeWaitTC0:snb,ivb,hsw,bdw,vlv */
__gen6_gt_wait_for_thread_c0(dev_priv);
}
static inline u32 fifo_free_entries(struct drm_i915_private *dev_priv)
{
u32 count = __raw_i915_read32(dev_priv, GTFIFOCTL);
return count & GT_FIFO_FREE_ENTRIES_MASK;
}
static void __gen6_gt_wait_for_fifo(struct drm_i915_private *dev_priv)
{
u32 n;
/* 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))
n = fifo_free_entries(dev_priv);
else
n = dev_priv->uncore.fifo_count;
if (n <= GT_FIFO_NUM_RESERVED_ENTRIES) {
if (wait_for_atomic((n = fifo_free_entries(dev_priv)) >
GT_FIFO_NUM_RESERVED_ENTRIES,
GT_FIFO_TIMEOUT_MS)) {
DRM_DEBUG("GT_FIFO timeout, entries: %u\n", n);
return;
}
}
dev_priv->uncore.fifo_count = n - 1;
}
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
static enum hrtimer_restart
intel_uncore_fw_release_timer(struct hrtimer *timer)
{
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
struct intel_uncore_forcewake_domain *domain =
container_of(timer, struct intel_uncore_forcewake_domain, timer);
struct drm_i915_private *dev_priv =
container_of(domain, struct drm_i915_private, uncore.fw_domain[domain->id]);
unsigned long irqflags;
assert_rpm_device_not_suspended(dev_priv);
if (xchg(&domain->active, false))
return HRTIMER_RESTART;
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
if (WARN_ON(domain->wake_count == 0))
domain->wake_count++;
if (--domain->wake_count == 0)
dev_priv->uncore.funcs.force_wake_put(dev_priv, domain->mask);
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
return HRTIMER_NORESTART;
}
static void intel_uncore_forcewake_reset(struct drm_i915_private *dev_priv,
bool restore)
{
unsigned long irqflags;
struct intel_uncore_forcewake_domain *domain;
int retry_count = 100;
enum forcewake_domains fw, active_domains;
/* 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) {
unsigned int tmp;
active_domains = 0;
for_each_fw_domain(domain, dev_priv, tmp) {
smp_store_mb(domain->active, false);
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
if (hrtimer_cancel(&domain->timer) == 0)
continue;
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
intel_uncore_fw_release_timer(&domain->timer);
}
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
for_each_fw_domain(domain, dev_priv, tmp) {
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
if (hrtimer_active(&domain->timer))
active_domains |= domain->mask;
}
if (active_domains == 0)
break;
if (--retry_count == 0) {
DRM_ERROR("Timed out waiting for forcewake timers to finish\n");
break;
}
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
cond_resched();
}
WARN_ON(active_domains);
fw = dev_priv->uncore.fw_domains_active;
if (fw)
dev_priv->uncore.funcs.force_wake_put(dev_priv, fw);
fw_domains_reset(dev_priv, dev_priv->uncore.fw_domains);
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_priv) || IS_GEN7(dev_priv))
dev_priv->uncore.fifo_count =
fifo_free_entries(dev_priv);
}
if (!restore)
assert_forcewakes_inactive(dev_priv);
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
}
static u64 gen9_edram_size(struct drm_i915_private *dev_priv)
{
const unsigned int ways[8] = { 4, 8, 12, 16, 16, 16, 16, 16 };
const unsigned int sets[4] = { 1, 1, 2, 2 };
const u32 cap = dev_priv->edram_cap;
return EDRAM_NUM_BANKS(cap) *
ways[EDRAM_WAYS_IDX(cap)] *
sets[EDRAM_SETS_IDX(cap)] *
1024 * 1024;
}
u64 intel_uncore_edram_size(struct drm_i915_private *dev_priv)
{
if (!HAS_EDRAM(dev_priv))
return 0;
/* The needed capability bits for size calculation
* are not there with pre gen9 so return 128MB always.
*/
if (INTEL_GEN(dev_priv) < 9)
return 128 * 1024 * 1024;
return gen9_edram_size(dev_priv);
}
static void intel_uncore_edram_detect(struct drm_i915_private *dev_priv)
{
if (IS_HASWELL(dev_priv) ||
IS_BROADWELL(dev_priv) ||
INTEL_GEN(dev_priv) >= 9) {
dev_priv->edram_cap = __raw_i915_read32(dev_priv,
HSW_EDRAM_CAP);
/* NB: We can't write IDICR yet because we do not have gt funcs
* set up */
} else {
dev_priv->edram_cap = 0;
}
if (HAS_EDRAM(dev_priv))
DRM_INFO("Found %lluMB of eDRAM\n",
intel_uncore_edram_size(dev_priv) / (1024 * 1024));
}
static bool
fpga_check_for_unclaimed_mmio(struct drm_i915_private *dev_priv)
{
u32 dbg;
dbg = __raw_i915_read32(dev_priv, FPGA_DBG);
if (likely(!(dbg & FPGA_DBG_RM_NOCLAIM)))
return false;
__raw_i915_write32(dev_priv, FPGA_DBG, FPGA_DBG_RM_NOCLAIM);
return true;
}
static bool
vlv_check_for_unclaimed_mmio(struct drm_i915_private *dev_priv)
{
u32 cer;
cer = __raw_i915_read32(dev_priv, CLAIM_ER);
if (likely(!(cer & (CLAIM_ER_OVERFLOW | CLAIM_ER_CTR_MASK))))
return false;
__raw_i915_write32(dev_priv, CLAIM_ER, CLAIM_ER_CLR);
return true;
}
static bool
gen6_check_for_fifo_debug(struct drm_i915_private *dev_priv)
{
u32 fifodbg;
fifodbg = __raw_i915_read32(dev_priv, GTFIFODBG);
if (unlikely(fifodbg)) {
DRM_DEBUG_DRIVER("GTFIFODBG = 0x08%x\n", fifodbg);
__raw_i915_write32(dev_priv, GTFIFODBG, fifodbg);
}
return fifodbg;
}
static bool
check_for_unclaimed_mmio(struct drm_i915_private *dev_priv)
{
bool ret = false;
if (HAS_FPGA_DBG_UNCLAIMED(dev_priv))
ret |= fpga_check_for_unclaimed_mmio(dev_priv);
if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv))
ret |= vlv_check_for_unclaimed_mmio(dev_priv);
if (IS_GEN6(dev_priv) || IS_GEN7(dev_priv))
ret |= gen6_check_for_fifo_debug(dev_priv);
return ret;
}
static void __intel_uncore_early_sanitize(struct drm_i915_private *dev_priv,
bool restore_forcewake)
{
/* clear out unclaimed reg detection bit */
if (check_for_unclaimed_mmio(dev_priv))
DRM_DEBUG("unclaimed mmio detected on uncore init, clearing\n");
/* WaDisableShadowRegForCpd:chv */
if (IS_CHERRYVIEW(dev_priv)) {
__raw_i915_write32(dev_priv, GTFIFOCTL,
__raw_i915_read32(dev_priv, GTFIFOCTL) |
GT_FIFO_CTL_BLOCK_ALL_POLICY_STALL |
GT_FIFO_CTL_RC6_POLICY_STALL);
}
intel_uncore_forcewake_reset(dev_priv, restore_forcewake);
}
void intel_uncore_suspend(struct drm_i915_private *dev_priv)
{
drm/i915: Listen for PMIC bus access notifications Listen for PMIC bus access notifications and get FORCEWAKE_ALL while the bus is accessed to avoid needing to do any forcewakes, which need PMIC bus access, while the PMIC bus is busy: This fixes errors like these showing up in dmesg, usually followed by a gfx or system freeze: [drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request. [drm:fw_domains_get [i915]] *MEDIA* render: timed out waiting for forcewake ack request. i2c_designware 808622C1:06: punit semaphore timed out, resetting i2c_designware 808622C1:06: PUNIT SEM: 2 i2c_designware 808622C1:06: couldn't acquire bus ownership Downside of this approach is that it causes wakeups whenever the PMIC bus is accessed. Unfortunately we cannot simply wait for the PMIC bus to go idle when we hit a race, as forcewakes may be done from interrupt handlers where we cannot sleep to wait for the i2c PMIC bus access to finish. Note that the notifications and thus the wakeups will only happen on baytrail / cherrytrail devices using PMICs with a shared i2c bus for P-Unit and host PMIC access (i2c busses with a _SEM method in their APCI node), e.g. an axp288 PMIC. I plan to write some patches for drivers accessing the PMIC bus to limit their bus accesses to a bare minimum (e.g. cache registers, do not update battery level more often then 4 times a minute), to limit the amount of wakeups. BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=155241 Signed-off-by: Hans de Goede <hdegoede@redhat.com> Tested-by: tagorereddy <tagore.chandan@gmail.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> [danvet: Wiggle in conflicts.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2017-02-10 18:28:02 +08:00
iosf_mbi_unregister_pmic_bus_access_notifier(
&dev_priv->uncore.pmic_bus_access_nb);
intel_uncore_forcewake_reset(dev_priv, false);
}
void intel_uncore_resume_early(struct drm_i915_private *dev_priv)
{
__intel_uncore_early_sanitize(dev_priv, true);
drm/i915: Listen for PMIC bus access notifications Listen for PMIC bus access notifications and get FORCEWAKE_ALL while the bus is accessed to avoid needing to do any forcewakes, which need PMIC bus access, while the PMIC bus is busy: This fixes errors like these showing up in dmesg, usually followed by a gfx or system freeze: [drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request. [drm:fw_domains_get [i915]] *MEDIA* render: timed out waiting for forcewake ack request. i2c_designware 808622C1:06: punit semaphore timed out, resetting i2c_designware 808622C1:06: PUNIT SEM: 2 i2c_designware 808622C1:06: couldn't acquire bus ownership Downside of this approach is that it causes wakeups whenever the PMIC bus is accessed. Unfortunately we cannot simply wait for the PMIC bus to go idle when we hit a race, as forcewakes may be done from interrupt handlers where we cannot sleep to wait for the i2c PMIC bus access to finish. Note that the notifications and thus the wakeups will only happen on baytrail / cherrytrail devices using PMICs with a shared i2c bus for P-Unit and host PMIC access (i2c busses with a _SEM method in their APCI node), e.g. an axp288 PMIC. I plan to write some patches for drivers accessing the PMIC bus to limit their bus accesses to a bare minimum (e.g. cache registers, do not update battery level more often then 4 times a minute), to limit the amount of wakeups. BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=155241 Signed-off-by: Hans de Goede <hdegoede@redhat.com> Tested-by: tagorereddy <tagore.chandan@gmail.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> [danvet: Wiggle in conflicts.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2017-02-10 18:28:02 +08:00
iosf_mbi_register_pmic_bus_access_notifier(
&dev_priv->uncore.pmic_bus_access_nb);
i915_check_and_clear_faults(dev_priv);
}
void intel_uncore_sanitize(struct drm_i915_private *dev_priv)
{
i915.enable_rc6 = sanitize_rc6_option(dev_priv, i915.enable_rc6);
drm/i915/bxt: Check BIOS RC6 setup before enabling RC6 RC6 setup is shared between BIOS and Driver. BIOS sets up subset of RC6 setup registers. If those are not setup Driver should not enable RC6. For implementing this, driver can check RC_CTRL0 and RC_CTRL1 values to know if BIOS has enabled HW/SW RC6. This will also enable user to control RC6 using BIOS settings alone. RC6 related instability can be avoided by disabling via BIOS settings till driver fixes it. v2: Had placed logic in gen8 function by mistake. Fixed it. Ensuring RPM is not enabled in case BIOS disabled RC6. v3: Need to disable RPM if RC6 is disabled due to BIOS settings. (Daniel) Runtime PM enabling happens before gen9_enable_rc6. Moved the updation of enable_rc6 parameter in intel_uncore_sanitize. v4: Added elaborate check for BIOS RC6 setup. Prepared check_pctx for bxt. (Imre) v5: Caching reserved stolen base and size in the driver private data. Reorganized RC6 setup check. Moved from gen9_enable_rc6 to intel_uncore_sanitize. (Imre) v6: Rebasing on the patch submitted by Imre that moves gem_init_stolen earlier in the load. v7: Removed PWRCTX_MAXCNT_VCSUNIT1 check as it applies to SKL. (Imre) v8: Fixed formatting and checkpatch issues. Fixed functional issue where RC6 ctx size check was missing. (Imre) Cc: Imre Deak <imre.deak@intel.com> Signed-off-by: Sagar Arun Kamble <sagar.a.kamble@intel.com> Signed-off-by: Imre Deak <imre.deak@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1454697809-22113-1-git-send-email-sagar.a.kamble@intel.com
2016-02-06 02:43:29 +08:00
/* BIOS often leaves RC6 enabled, but disable it for hw init */
intel_sanitize_gt_powersave(dev_priv);
}
static void __intel_uncore_forcewake_get(struct drm_i915_private *dev_priv,
enum forcewake_domains fw_domains)
{
struct intel_uncore_forcewake_domain *domain;
unsigned int tmp;
fw_domains &= dev_priv->uncore.fw_domains;
for_each_fw_domain_masked(domain, fw_domains, dev_priv, tmp) {
if (domain->wake_count++) {
fw_domains &= ~domain->mask;
domain->active = true;
}
}
if (fw_domains)
dev_priv->uncore.funcs.force_wake_get(dev_priv, fw_domains);
}
/**
* 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.
*/
void intel_uncore_forcewake_get(struct drm_i915_private *dev_priv,
enum forcewake_domains fw_domains)
{
unsigned long irqflags;
if (!dev_priv->uncore.funcs.force_wake_get)
return;
assert_rpm_wakelock_held(dev_priv);
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
__intel_uncore_forcewake_get(dev_priv, fw_domains);
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
}
/**
* intel_uncore_forcewake_get__locked - grab forcewake domain references
* @dev_priv: i915 device instance
* @fw_domains: forcewake domains to get reference on
*
* See intel_uncore_forcewake_get(). This variant places the onus
* on the caller to explicitly handle the dev_priv->uncore.lock spinlock.
*/
void intel_uncore_forcewake_get__locked(struct drm_i915_private *dev_priv,
enum forcewake_domains fw_domains)
{
lockdep_assert_held(&dev_priv->uncore.lock);
if (!dev_priv->uncore.funcs.force_wake_get)
return;
__intel_uncore_forcewake_get(dev_priv, fw_domains);
}
static void __intel_uncore_forcewake_put(struct drm_i915_private *dev_priv,
enum forcewake_domains fw_domains)
{
struct intel_uncore_forcewake_domain *domain;
unsigned int tmp;
fw_domains &= dev_priv->uncore.fw_domains;
for_each_fw_domain_masked(domain, fw_domains, dev_priv, tmp) {
if (WARN_ON(domain->wake_count == 0))
continue;
if (--domain->wake_count) {
domain->active = true;
continue;
}
fw_domain_arm_timer(domain);
}
}
/**
* 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().
*/
void intel_uncore_forcewake_put(struct drm_i915_private *dev_priv,
enum forcewake_domains fw_domains)
{
unsigned long irqflags;
if (!dev_priv->uncore.funcs.force_wake_put)
return;
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags);
__intel_uncore_forcewake_put(dev_priv, fw_domains);
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
}
/**
* intel_uncore_forcewake_put__locked - grab forcewake domain references
* @dev_priv: i915 device instance
* @fw_domains: forcewake domains to get reference on
*
* See intel_uncore_forcewake_put(). This variant places the onus
* on the caller to explicitly handle the dev_priv->uncore.lock spinlock.
*/
void intel_uncore_forcewake_put__locked(struct drm_i915_private *dev_priv,
enum forcewake_domains fw_domains)
{
lockdep_assert_held(&dev_priv->uncore.lock);
if (!dev_priv->uncore.funcs.force_wake_put)
return;
__intel_uncore_forcewake_put(dev_priv, fw_domains);
}
void assert_forcewakes_inactive(struct drm_i915_private *dev_priv)
{
if (!dev_priv->uncore.funcs.force_wake_get)
return;
WARN_ON(dev_priv->uncore.fw_domains_active);
}
/* We give fast paths for the really cool registers */
#define NEEDS_FORCE_WAKE(reg) ((reg) < 0x40000)
#define __gen6_reg_read_fw_domains(offset) \
({ \
enum forcewake_domains __fwd; \
if (NEEDS_FORCE_WAKE(offset)) \
__fwd = FORCEWAKE_RENDER; \
else \
__fwd = 0; \
__fwd; \
})
static int fw_range_cmp(u32 offset, const struct intel_forcewake_range *entry)
{
if (offset < entry->start)
return -1;
else if (offset > entry->end)
return 1;
else
return 0;
}
/* Copied and "macroized" from lib/bsearch.c */
#define BSEARCH(key, base, num, cmp) ({ \
unsigned int start__ = 0, end__ = (num); \
typeof(base) result__ = NULL; \
while (start__ < end__) { \
unsigned int mid__ = start__ + (end__ - start__) / 2; \
int ret__ = (cmp)((key), (base) + mid__); \
if (ret__ < 0) { \
end__ = mid__; \
} else if (ret__ > 0) { \
start__ = mid__ + 1; \
} else { \
result__ = (base) + mid__; \
break; \
} \
} \
result__; \
})
static enum forcewake_domains
find_fw_domain(struct drm_i915_private *dev_priv, u32 offset)
{
const struct intel_forcewake_range *entry;
entry = BSEARCH(offset,
dev_priv->uncore.fw_domains_table,
dev_priv->uncore.fw_domains_table_entries,
fw_range_cmp);
if (!entry)
return 0;
WARN(entry->domains & ~dev_priv->uncore.fw_domains,
"Uninitialized forcewake domain(s) 0x%x accessed at 0x%x\n",
entry->domains & ~dev_priv->uncore.fw_domains, offset);
return entry->domains;
}
#define GEN_FW_RANGE(s, e, d) \
{ .start = (s), .end = (e), .domains = (d) }
#define HAS_FWTABLE(dev_priv) \
(INTEL_GEN(dev_priv) >= 9 || \
IS_CHERRYVIEW(dev_priv) || \
IS_VALLEYVIEW(dev_priv))
/* *Must* be sorted by offset ranges! See intel_fw_table_check(). */
static const struct intel_forcewake_range __vlv_fw_ranges[] = {
GEN_FW_RANGE(0x2000, 0x3fff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x5000, 0x7fff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0xb000, 0x11fff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x12000, 0x13fff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x22000, 0x23fff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x2e000, 0x2ffff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x30000, 0x3ffff, FORCEWAKE_MEDIA),
};
#define __fwtable_reg_read_fw_domains(offset) \
({ \
enum forcewake_domains __fwd = 0; \
if (NEEDS_FORCE_WAKE((offset))) \
__fwd = find_fw_domain(dev_priv, offset); \
__fwd; \
})
/* *Must* be sorted by offset! See intel_shadow_table_check(). */
static const i915_reg_t gen8_shadowed_regs[] = {
RING_TAIL(RENDER_RING_BASE), /* 0x2000 (base) */
GEN6_RPNSWREQ, /* 0xA008 */
GEN6_RC_VIDEO_FREQ, /* 0xA00C */
RING_TAIL(GEN6_BSD_RING_BASE), /* 0x12000 (base) */
RING_TAIL(VEBOX_RING_BASE), /* 0x1a000 (base) */
RING_TAIL(BLT_RING_BASE), /* 0x22000 (base) */
/* TODO: Other registers are not yet used */
};
static int mmio_reg_cmp(u32 key, const i915_reg_t *reg)
{
u32 offset = i915_mmio_reg_offset(*reg);
if (key < offset)
return -1;
else if (key > offset)
return 1;
else
return 0;
}
static bool is_gen8_shadowed(u32 offset)
{
const i915_reg_t *regs = gen8_shadowed_regs;
return BSEARCH(offset, regs, ARRAY_SIZE(gen8_shadowed_regs),
mmio_reg_cmp);
}
#define __gen8_reg_write_fw_domains(offset) \
({ \
enum forcewake_domains __fwd; \
if (NEEDS_FORCE_WAKE(offset) && !is_gen8_shadowed(offset)) \
__fwd = FORCEWAKE_RENDER; \
else \
__fwd = 0; \
__fwd; \
})
/* *Must* be sorted by offset ranges! See intel_fw_table_check(). */
static const struct intel_forcewake_range __chv_fw_ranges[] = {
GEN_FW_RANGE(0x2000, 0x3fff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x4000, 0x4fff, FORCEWAKE_RENDER | FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x5200, 0x7fff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x8000, 0x82ff, FORCEWAKE_RENDER | FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x8300, 0x84ff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x8500, 0x85ff, FORCEWAKE_RENDER | FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x8800, 0x88ff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x9000, 0xafff, FORCEWAKE_RENDER | FORCEWAKE_MEDIA),
GEN_FW_RANGE(0xb000, 0xb47f, FORCEWAKE_RENDER),
GEN_FW_RANGE(0xd000, 0xd7ff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0xe000, 0xe7ff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0xf000, 0xffff, FORCEWAKE_RENDER | FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x12000, 0x13fff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x1a000, 0x1bfff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x1e800, 0x1e9ff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x30000, 0x37fff, FORCEWAKE_MEDIA),
};
#define __fwtable_reg_write_fw_domains(offset) \
({ \
enum forcewake_domains __fwd = 0; \
if (NEEDS_FORCE_WAKE((offset)) && !is_gen8_shadowed(offset)) \
__fwd = find_fw_domain(dev_priv, offset); \
__fwd; \
})
/* *Must* be sorted by offset ranges! See intel_fw_table_check(). */
static const struct intel_forcewake_range __gen9_fw_ranges[] = {
GEN_FW_RANGE(0x0, 0xaff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0xb00, 0x1fff, 0), /* uncore range */
GEN_FW_RANGE(0x2000, 0x26ff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x2700, 0x2fff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x3000, 0x3fff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x4000, 0x51ff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x5200, 0x7fff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x8000, 0x812f, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x8130, 0x813f, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x8140, 0x815f, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x8160, 0x82ff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x8300, 0x84ff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x8500, 0x87ff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x8800, 0x89ff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x8a00, 0x8bff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x8c00, 0x8cff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x8d00, 0x93ff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x9400, 0x97ff, FORCEWAKE_RENDER | FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x9800, 0xafff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0xb000, 0xb47f, FORCEWAKE_RENDER),
GEN_FW_RANGE(0xb480, 0xcfff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0xd000, 0xd7ff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0xd800, 0xdfff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0xe000, 0xe8ff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0xe900, 0x11fff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x12000, 0x13fff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x14000, 0x19fff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x1a000, 0x1e9ff, FORCEWAKE_MEDIA),
GEN_FW_RANGE(0x1ea00, 0x243ff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x24400, 0x247ff, FORCEWAKE_RENDER),
GEN_FW_RANGE(0x24800, 0x2ffff, FORCEWAKE_BLITTER),
GEN_FW_RANGE(0x30000, 0x3ffff, FORCEWAKE_MEDIA),
};
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. */
__raw_i915_write32(dev_priv, MI_MODE, 0);
}
static void
__unclaimed_reg_debug(struct drm_i915_private *dev_priv,
const i915_reg_t reg,
const bool read,
const bool before)
{
if (WARN(check_for_unclaimed_mmio(dev_priv) && !before,
"Unclaimed %s register 0x%x\n",
read ? "read from" : "write to",
i915_mmio_reg_offset(reg)))
i915.mmio_debug--; /* Only report the first N failures */
}
static inline void
unclaimed_reg_debug(struct drm_i915_private *dev_priv,
const i915_reg_t reg,
const bool read,
const bool before)
{
if (likely(!i915.mmio_debug))
return;
__unclaimed_reg_debug(dev_priv, reg, read, before);
}
#define GEN2_READ_HEADER(x) \
u##x val = 0; \
assert_rpm_wakelock_held(dev_priv);
#define GEN2_READ_FOOTER \
trace_i915_reg_rw(false, reg, val, sizeof(val), trace); \
return val
#define __gen2_read(x) \
static u##x \
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
gen2_read##x(struct drm_i915_private *dev_priv, i915_reg_t reg, bool trace) { \
GEN2_READ_HEADER(x); \
val = __raw_i915_read##x(dev_priv, reg); \
GEN2_READ_FOOTER; \
}
#define __gen5_read(x) \
static u##x \
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
gen5_read##x(struct drm_i915_private *dev_priv, i915_reg_t reg, bool trace) { \
GEN2_READ_HEADER(x); \
ilk_dummy_write(dev_priv); \
val = __raw_i915_read##x(dev_priv, reg); \
GEN2_READ_FOOTER; \
}
__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) \
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
u32 offset = i915_mmio_reg_offset(reg); \
unsigned long irqflags; \
u##x val = 0; \
assert_rpm_wakelock_held(dev_priv); \
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags); \
unclaimed_reg_debug(dev_priv, reg, true, true)
#define GEN6_READ_FOOTER \
unclaimed_reg_debug(dev_priv, reg, true, false); \
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags); \
trace_i915_reg_rw(false, reg, val, sizeof(val), trace); \
return val
static noinline void ___force_wake_auto(struct drm_i915_private *dev_priv,
enum forcewake_domains fw_domains)
{
struct intel_uncore_forcewake_domain *domain;
unsigned int tmp;
GEM_BUG_ON(fw_domains & ~dev_priv->uncore.fw_domains);
for_each_fw_domain_masked(domain, fw_domains, dev_priv, tmp)
fw_domain_arm_timer(domain);
dev_priv->uncore.funcs.force_wake_get(dev_priv, fw_domains);
}
static inline void __force_wake_auto(struct drm_i915_private *dev_priv,
enum forcewake_domains fw_domains)
{
if (WARN_ON(!fw_domains))
return;
/* Turn on all requested but inactive supported forcewake domains. */
fw_domains &= dev_priv->uncore.fw_domains;
fw_domains &= ~dev_priv->uncore.fw_domains_active;
if (fw_domains)
___force_wake_auto(dev_priv, fw_domains);
}
#define __gen_read(func, x) \
static u##x \
func##_read##x(struct drm_i915_private *dev_priv, i915_reg_t reg, bool trace) { \
enum forcewake_domains fw_engine; \
GEN6_READ_HEADER(x); \
fw_engine = __##func##_reg_read_fw_domains(offset); \
if (fw_engine) \
__force_wake_auto(dev_priv, fw_engine); \
val = __raw_i915_read##x(dev_priv, reg); \
GEN6_READ_FOOTER; \
}
#define __gen6_read(x) __gen_read(gen6, x)
#define __fwtable_read(x) __gen_read(fwtable, x)
__fwtable_read(8)
__fwtable_read(16)
__fwtable_read(32)
__fwtable_read(64)
__gen6_read(8)
__gen6_read(16)
__gen6_read(32)
__gen6_read(64)
#undef __fwtable_read
#undef __gen6_read
#undef GEN6_READ_FOOTER
#undef GEN6_READ_HEADER
#define GEN2_WRITE_HEADER \
trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \
assert_rpm_wakelock_held(dev_priv); \
#define GEN2_WRITE_FOOTER
#define __gen2_write(x) \
static void \
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
gen2_write##x(struct drm_i915_private *dev_priv, i915_reg_t reg, u##x val, bool trace) { \
GEN2_WRITE_HEADER; \
__raw_i915_write##x(dev_priv, reg, val); \
GEN2_WRITE_FOOTER; \
}
#define __gen5_write(x) \
static void \
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
gen5_write##x(struct drm_i915_private *dev_priv, i915_reg_t reg, u##x val, bool trace) { \
GEN2_WRITE_HEADER; \
ilk_dummy_write(dev_priv); \
__raw_i915_write##x(dev_priv, reg, val); \
GEN2_WRITE_FOOTER; \
}
__gen5_write(8)
__gen5_write(16)
__gen5_write(32)
__gen2_write(8)
__gen2_write(16)
__gen2_write(32)
#undef __gen5_write
#undef __gen2_write
#undef GEN2_WRITE_FOOTER
#undef GEN2_WRITE_HEADER
#define GEN6_WRITE_HEADER \
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
u32 offset = i915_mmio_reg_offset(reg); \
unsigned long irqflags; \
trace_i915_reg_rw(true, reg, val, sizeof(val), trace); \
assert_rpm_wakelock_held(dev_priv); \
spin_lock_irqsave(&dev_priv->uncore.lock, irqflags); \
unclaimed_reg_debug(dev_priv, reg, false, true)
#define GEN6_WRITE_FOOTER \
unclaimed_reg_debug(dev_priv, reg, false, false); \
spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags)
#define __gen6_write(x) \
static void \
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
gen6_write##x(struct drm_i915_private *dev_priv, i915_reg_t reg, u##x val, bool trace) { \
GEN6_WRITE_HEADER; \
if (NEEDS_FORCE_WAKE(offset)) \
__gen6_gt_wait_for_fifo(dev_priv); \
__raw_i915_write##x(dev_priv, reg, val); \
GEN6_WRITE_FOOTER; \
}
#define __gen_write(func, x) \
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 void \
func##_write##x(struct drm_i915_private *dev_priv, i915_reg_t reg, u##x val, bool trace) { \
enum forcewake_domains fw_engine; \
GEN6_WRITE_HEADER; \
fw_engine = __##func##_reg_write_fw_domains(offset); \
if (fw_engine) \
__force_wake_auto(dev_priv, fw_engine); \
__raw_i915_write##x(dev_priv, reg, val); \
GEN6_WRITE_FOOTER; \
}
#define __gen8_write(x) __gen_write(gen8, x)
#define __fwtable_write(x) __gen_write(fwtable, x)
__fwtable_write(8)
__fwtable_write(16)
__fwtable_write(32)
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)
__gen6_write(8)
__gen6_write(16)
__gen6_write(32)
#undef __fwtable_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
#undef __gen6_write
#undef GEN6_WRITE_FOOTER
#undef GEN6_WRITE_HEADER
#define ASSIGN_WRITE_MMIO_VFUNCS(i915, x) \
do { \
(i915)->uncore.funcs.mmio_writeb = x##_write8; \
(i915)->uncore.funcs.mmio_writew = x##_write16; \
(i915)->uncore.funcs.mmio_writel = x##_write32; \
} while (0)
#define ASSIGN_READ_MMIO_VFUNCS(i915, x) \
do { \
(i915)->uncore.funcs.mmio_readb = x##_read8; \
(i915)->uncore.funcs.mmio_readw = x##_read16; \
(i915)->uncore.funcs.mmio_readl = x##_read32; \
(i915)->uncore.funcs.mmio_readq = x##_read64; \
} while (0)
static void fw_domain_init(struct drm_i915_private *dev_priv,
enum forcewake_domain_id domain_id,
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
i915_reg_t reg_set,
i915_reg_t reg_ack)
{
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);
WARN_ON(!i915_mmio_reg_valid(reg_set));
WARN_ON(!i915_mmio_reg_valid(reg_ack));
d->wake_count = 0;
d->reg_set = reg_set;
d->reg_ack = reg_ack;
d->id = domain_id;
BUILD_BUG_ON(FORCEWAKE_RENDER != (1 << FW_DOMAIN_ID_RENDER));
BUILD_BUG_ON(FORCEWAKE_BLITTER != (1 << FW_DOMAIN_ID_BLITTER));
BUILD_BUG_ON(FORCEWAKE_MEDIA != (1 << FW_DOMAIN_ID_MEDIA));
d->mask = BIT(domain_id);
drm/i915: Use consistent forcewake auto-release timeout across kernel configs Because it is based on jiffies, current implementation releases the forcewake at any time between straight away and between 1ms and 10ms, depending on the kernel configuration (CONFIG_HZ). This is probably not what has been desired, since the dynamics of keeping parts of the GPU awake should not be correlated with this kernel configuration parameter. Change the auto-release mechanism to use hrtimers and set the timeout to 1ms with a 1ms of slack. This should make the GPU power consistent across kernel configs, and timer slack should enable some timer coalescing where multiple force-wake domains exist, or with unrelated timers. For GlBench/T-Rex this decreases the number of forcewake releases from ~480 to ~300 per second, and for a heavy combined OGL/OCL test from ~670 to ~360 (HZ=1000 kernel). Even though this reduction can be attributed to the average release period extending from 0-1ms to 1-2ms, as discussed above, it will make the forcewake timeout consistent for different CONFIG_HZ values. Real life measurements with the above workload has shown that, with this patch, both manage to auto-release the forcewake between 2-4 times per 10ms, even though the number of forcewake gets is dramatically different. T-Rex requests between 5-10 explicit gets and 5-10 implict gets in each 10ms period, while the OGL/OCL test requests 250 and 380 times in the same period. The two data points together suggest that the nature of the forwake accesses is bursty and that further changes and potential timeout extensions, or moving the start of timeout from the first to the last automatic forcewake grab, should be carefully measured for power and performance effects. v2: * Commit spelling. (Dave Gordon) * More discussion on numbers in the commit. (Chris Wilson) Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
2016-04-08 00:04:32 +08:00
hrtimer_init(&d->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
d->timer.function = intel_uncore_fw_release_timer;
dev_priv->uncore.fw_domains |= BIT(domain_id);
fw_domain_reset(dev_priv, d);
}
static void intel_uncore_fw_domains_init(struct drm_i915_private *dev_priv)
{
if (INTEL_GEN(dev_priv) <= 5 || intel_vgpu_active(dev_priv))
return;
if (IS_GEN6(dev_priv)) {
dev_priv->uncore.fw_reset = 0;
dev_priv->uncore.fw_set = FORCEWAKE_KERNEL;
dev_priv->uncore.fw_clear = 0;
} else {
/* WaRsClearFWBitsAtReset:bdw,skl */
dev_priv->uncore.fw_reset = _MASKED_BIT_DISABLE(0xffff);
dev_priv->uncore.fw_set = _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL);
dev_priv->uncore.fw_clear = _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL);
}
if (INTEL_GEN(dev_priv) >= 9) {
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);
} else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
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_VLV, FORCEWAKE_ACK_VLV);
fw_domain_init(dev_priv, FW_DOMAIN_ID_MEDIA,
FORCEWAKE_MEDIA_VLV, FORCEWAKE_ACK_MEDIA_VLV);
} else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) {
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);
} else if (IS_IVYBRIDGE(dev_priv)) {
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.
*/
dev_priv->uncore.funcs.force_wake_get =
fw_domains_get_with_thread_status;
dev_priv->uncore.funcs.force_wake_put = fw_domains_put;
/* 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. In this stage we don't know which flavour this
* ivb is, so it is better to reset also the gen6 fw registers
* before the ecobus check.
*/
__raw_i915_write32(dev_priv, FORCEWAKE, 0);
__raw_posting_read(dev_priv, ECOBUS);
fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
FORCEWAKE_MT, FORCEWAKE_MT_ACK);
spin_lock_irq(&dev_priv->uncore.lock);
fw_domains_get_with_thread_status(dev_priv, FORCEWAKE_RENDER);
ecobus = __raw_i915_read32(dev_priv, ECOBUS);
fw_domains_put(dev_priv, FORCEWAKE_RENDER);
spin_unlock_irq(&dev_priv->uncore.lock);
if (!(ecobus & FORCEWAKE_MT_ENABLE)) {
DRM_INFO("No MT forcewake available on Ivybridge, this can result in issues\n");
DRM_INFO("when using vblank-synced partial screen updates.\n");
fw_domain_init(dev_priv, FW_DOMAIN_ID_RENDER,
FORCEWAKE, FORCEWAKE_ACK);
}
} else if (IS_GEN6(dev_priv)) {
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, FORCEWAKE_ACK);
}
/* All future platforms are expected to require complex power gating */
WARN_ON(dev_priv->uncore.fw_domains == 0);
}
#define ASSIGN_FW_DOMAINS_TABLE(d) \
{ \
dev_priv->uncore.fw_domains_table = \
(struct intel_forcewake_range *)(d); \
dev_priv->uncore.fw_domains_table_entries = ARRAY_SIZE((d)); \
}
drm/i915: Listen for PMIC bus access notifications Listen for PMIC bus access notifications and get FORCEWAKE_ALL while the bus is accessed to avoid needing to do any forcewakes, which need PMIC bus access, while the PMIC bus is busy: This fixes errors like these showing up in dmesg, usually followed by a gfx or system freeze: [drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request. [drm:fw_domains_get [i915]] *MEDIA* render: timed out waiting for forcewake ack request. i2c_designware 808622C1:06: punit semaphore timed out, resetting i2c_designware 808622C1:06: PUNIT SEM: 2 i2c_designware 808622C1:06: couldn't acquire bus ownership Downside of this approach is that it causes wakeups whenever the PMIC bus is accessed. Unfortunately we cannot simply wait for the PMIC bus to go idle when we hit a race, as forcewakes may be done from interrupt handlers where we cannot sleep to wait for the i2c PMIC bus access to finish. Note that the notifications and thus the wakeups will only happen on baytrail / cherrytrail devices using PMICs with a shared i2c bus for P-Unit and host PMIC access (i2c busses with a _SEM method in their APCI node), e.g. an axp288 PMIC. I plan to write some patches for drivers accessing the PMIC bus to limit their bus accesses to a bare minimum (e.g. cache registers, do not update battery level more often then 4 times a minute), to limit the amount of wakeups. BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=155241 Signed-off-by: Hans de Goede <hdegoede@redhat.com> Tested-by: tagorereddy <tagore.chandan@gmail.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> [danvet: Wiggle in conflicts.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2017-02-10 18:28:02 +08:00
static int i915_pmic_bus_access_notifier(struct notifier_block *nb,
unsigned long action, void *data)
{
struct drm_i915_private *dev_priv = container_of(nb,
struct drm_i915_private, uncore.pmic_bus_access_nb);
switch (action) {
case MBI_PMIC_BUS_ACCESS_BEGIN:
/*
* forcewake all now to make sure that we don't need to do a
* forcewake later which on systems where this notifier gets
* called requires the punit to access to the shared pmic i2c
* bus, which will be busy after this notification, leading to:
* "render: timed out waiting for forcewake ack request."
* errors.
*/
intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
break;
case MBI_PMIC_BUS_ACCESS_END:
intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
break;
}
return NOTIFY_OK;
}
void intel_uncore_init(struct drm_i915_private *dev_priv)
{
i915_check_vgpu(dev_priv);
drm/i915: Introduce a PV INFO page structure for Intel GVT-g. Introduce a PV INFO structure, to facilitate the Intel GVT-g technology, which is a GPU virtualization solution with mediated pass-through. This page contains the shared information between i915 driver and the host emulator. For now, this structure utilizes an area of 4K bytes on HSW GPU's unused MMIO space. Future hardware will have the reserved window architecturally defined, and layout of the page will be added in future BSpec. The i915 driver load routine detects if it is running in a VM by reading the contents of this PV INFO page. Thereafter a flag, vgpu.active is set, and intel_vgpu_active() is used by checking this flag to conclude if GPU is virtualized with Intel GVT-g. By now, intel_vgpu_active() will return true, only when the driver is running as a guest in the Intel GVT-g enhanced environment on HSW platform. v2: take Chris' comments: - call the i915_check_vgpu() in intel_uncore_init() - sanitize i915_check_vgpu() by adding BUILD_BUG_ON() and debug info take Daniel's comments: - put the definition of PV INFO into a new header - i915_vgt_if.h other changes: - access mmio regs by readq/readw in i915_check_vgpu() v3: take Daniel's comments: - move the i915/vgt interfaces into a new i915_vgpu.c - update makefile - add kerneldoc to functions which are non-static - add a DOC: section describing some of the high-level design - update drm docbook other changes: - rename i915_vgt_if.h to i915_vgpu.h v4: take Tvrtko's comments: - fix a typo in commit message - add debug message when vgt version mismatches - rename low_gmadr/high_gmadr to mappable/non-mappable in PV INFO structure Signed-off-by: Yu Zhang <yu.c.zhang@linux.intel.com> Signed-off-by: Jike Song <jike.song@intel.com> Signed-off-by: Eddie Dong <eddie.dong@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2015-02-10 19:05:47 +08:00
intel_uncore_edram_detect(dev_priv);
intel_uncore_fw_domains_init(dev_priv);
__intel_uncore_early_sanitize(dev_priv, false);
dev_priv->uncore.unclaimed_mmio_check = 1;
drm/i915: Listen for PMIC bus access notifications Listen for PMIC bus access notifications and get FORCEWAKE_ALL while the bus is accessed to avoid needing to do any forcewakes, which need PMIC bus access, while the PMIC bus is busy: This fixes errors like these showing up in dmesg, usually followed by a gfx or system freeze: [drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request. [drm:fw_domains_get [i915]] *MEDIA* render: timed out waiting for forcewake ack request. i2c_designware 808622C1:06: punit semaphore timed out, resetting i2c_designware 808622C1:06: PUNIT SEM: 2 i2c_designware 808622C1:06: couldn't acquire bus ownership Downside of this approach is that it causes wakeups whenever the PMIC bus is accessed. Unfortunately we cannot simply wait for the PMIC bus to go idle when we hit a race, as forcewakes may be done from interrupt handlers where we cannot sleep to wait for the i2c PMIC bus access to finish. Note that the notifications and thus the wakeups will only happen on baytrail / cherrytrail devices using PMICs with a shared i2c bus for P-Unit and host PMIC access (i2c busses with a _SEM method in their APCI node), e.g. an axp288 PMIC. I plan to write some patches for drivers accessing the PMIC bus to limit their bus accesses to a bare minimum (e.g. cache registers, do not update battery level more often then 4 times a minute), to limit the amount of wakeups. BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=155241 Signed-off-by: Hans de Goede <hdegoede@redhat.com> Tested-by: tagorereddy <tagore.chandan@gmail.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> [danvet: Wiggle in conflicts.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2017-02-10 18:28:02 +08:00
dev_priv->uncore.pmic_bus_access_nb.notifier_call =
i915_pmic_bus_access_notifier;
if (IS_GEN(dev_priv, 2, 4) || intel_vgpu_active(dev_priv)) {
ASSIGN_WRITE_MMIO_VFUNCS(dev_priv, gen2);
ASSIGN_READ_MMIO_VFUNCS(dev_priv, gen2);
} else if (IS_GEN5(dev_priv)) {
ASSIGN_WRITE_MMIO_VFUNCS(dev_priv, gen5);
ASSIGN_READ_MMIO_VFUNCS(dev_priv, gen5);
} else if (IS_GEN(dev_priv, 6, 7)) {
ASSIGN_WRITE_MMIO_VFUNCS(dev_priv, gen6);
if (IS_VALLEYVIEW(dev_priv)) {
ASSIGN_FW_DOMAINS_TABLE(__vlv_fw_ranges);
ASSIGN_READ_MMIO_VFUNCS(dev_priv, fwtable);
} else {
ASSIGN_READ_MMIO_VFUNCS(dev_priv, gen6);
drm/i915/bxt: Broxton decoupled MMIO Decoupled MMIO is an alternative way to access forcewake domain registers, which requires less cycles for a single read/write and avoids frequent software forcewake. This certainly gives advantage over the forcewake as this new mechanism “decouples” CPU cycles and allow them to complete even when GT is in a CPD (frequency change) or C6 state. This can co-exist with forcewake and we will continue to use forcewake as appropriate. E.g. 64-bit register writes to avoid writing 2 dwords separately and land into funny situations. v2: - Moved platform check out of the function and got rid of duplicate functions to find out decoupled power domain (Chris) - Added a check for forcewake already held and skipped decoupled access (Chris) - Skipped writing 64 bit registers through decoupled MMIO (Chris) v3: - Improved commit message with more info on decoupled mmio (Tvrtko) - Changed decoupled operation to enum and used u32 instead of uint_32 data type for register offset (Tvrtko) - Moved HAS_DECOUPLED_MMIO to device info (Tvrtko) - Added lookup table for converting fw_engine to pd_engine (Tvrtko) - Improved __gen9_decoupled_read and __gen9_decoupled_write routines (Tvrtko) v4: - Fixed alignment and variable names (Chris) - Write GEN9_DECOUPLED_REG0_DW1 register in just one go (Zhe Wang) v5: - Changed HAS_DECOUPLED_MMIO() argument name to dev_priv (Tvrtko) - Sanitize info->had_decoupled_mmio at init (Chris) Signed-off-by: Zhe Wang <zhe1.wang@intel.com> Signed-off-by: Praveen Paneri <praveen.paneri@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/1479230360-22395-1-git-send-email-praveen.paneri@intel.com
2016-11-16 01:19:20 +08:00
}
} else if (IS_GEN8(dev_priv)) {
if (IS_CHERRYVIEW(dev_priv)) {
ASSIGN_FW_DOMAINS_TABLE(__chv_fw_ranges);
ASSIGN_WRITE_MMIO_VFUNCS(dev_priv, fwtable);
ASSIGN_READ_MMIO_VFUNCS(dev_priv, fwtable);
} else {
ASSIGN_WRITE_MMIO_VFUNCS(dev_priv, gen8);
ASSIGN_READ_MMIO_VFUNCS(dev_priv, gen6);
}
} else {
ASSIGN_FW_DOMAINS_TABLE(__gen9_fw_ranges);
ASSIGN_WRITE_MMIO_VFUNCS(dev_priv, fwtable);
ASSIGN_READ_MMIO_VFUNCS(dev_priv, fwtable);
}
drm/i915: Listen for PMIC bus access notifications Listen for PMIC bus access notifications and get FORCEWAKE_ALL while the bus is accessed to avoid needing to do any forcewakes, which need PMIC bus access, while the PMIC bus is busy: This fixes errors like these showing up in dmesg, usually followed by a gfx or system freeze: [drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request. [drm:fw_domains_get [i915]] *MEDIA* render: timed out waiting for forcewake ack request. i2c_designware 808622C1:06: punit semaphore timed out, resetting i2c_designware 808622C1:06: PUNIT SEM: 2 i2c_designware 808622C1:06: couldn't acquire bus ownership Downside of this approach is that it causes wakeups whenever the PMIC bus is accessed. Unfortunately we cannot simply wait for the PMIC bus to go idle when we hit a race, as forcewakes may be done from interrupt handlers where we cannot sleep to wait for the i2c PMIC bus access to finish. Note that the notifications and thus the wakeups will only happen on baytrail / cherrytrail devices using PMICs with a shared i2c bus for P-Unit and host PMIC access (i2c busses with a _SEM method in their APCI node), e.g. an axp288 PMIC. I plan to write some patches for drivers accessing the PMIC bus to limit their bus accesses to a bare minimum (e.g. cache registers, do not update battery level more often then 4 times a minute), to limit the amount of wakeups. BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=155241 Signed-off-by: Hans de Goede <hdegoede@redhat.com> Tested-by: tagorereddy <tagore.chandan@gmail.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> [danvet: Wiggle in conflicts.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2017-02-10 18:28:02 +08:00
iosf_mbi_register_pmic_bus_access_notifier(
&dev_priv->uncore.pmic_bus_access_nb);
i915_check_and_clear_faults(dev_priv);
}
void intel_uncore_fini(struct drm_i915_private *dev_priv)
{
drm/i915: Listen for PMIC bus access notifications Listen for PMIC bus access notifications and get FORCEWAKE_ALL while the bus is accessed to avoid needing to do any forcewakes, which need PMIC bus access, while the PMIC bus is busy: This fixes errors like these showing up in dmesg, usually followed by a gfx or system freeze: [drm:fw_domains_get [i915]] *ERROR* render: timed out waiting for forcewake ack request. [drm:fw_domains_get [i915]] *MEDIA* render: timed out waiting for forcewake ack request. i2c_designware 808622C1:06: punit semaphore timed out, resetting i2c_designware 808622C1:06: PUNIT SEM: 2 i2c_designware 808622C1:06: couldn't acquire bus ownership Downside of this approach is that it causes wakeups whenever the PMIC bus is accessed. Unfortunately we cannot simply wait for the PMIC bus to go idle when we hit a race, as forcewakes may be done from interrupt handlers where we cannot sleep to wait for the i2c PMIC bus access to finish. Note that the notifications and thus the wakeups will only happen on baytrail / cherrytrail devices using PMICs with a shared i2c bus for P-Unit and host PMIC access (i2c busses with a _SEM method in their APCI node), e.g. an axp288 PMIC. I plan to write some patches for drivers accessing the PMIC bus to limit their bus accesses to a bare minimum (e.g. cache registers, do not update battery level more often then 4 times a minute), to limit the amount of wakeups. BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=155241 Signed-off-by: Hans de Goede <hdegoede@redhat.com> Tested-by: tagorereddy <tagore.chandan@gmail.com> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> [danvet: Wiggle in conflicts.] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
2017-02-10 18:28:02 +08:00
iosf_mbi_unregister_pmic_bus_access_notifier(
&dev_priv->uncore.pmic_bus_access_nb);
/* Paranoia: make sure we have disabled everything before we exit. */
intel_uncore_sanitize(dev_priv);
intel_uncore_forcewake_reset(dev_priv, false);
}
#define GEN_RANGE(l, h) GENMASK((h) - 1, (l) - 1)
static const struct register_whitelist {
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
i915_reg_t offset_ldw, offset_udw;
uint32_t size;
/* supported gens, 0x10 for 4, 0x30 for 4 and 5, etc. */
uint32_t gen_bitmask;
} whitelist[] = {
{ .offset_ldw = RING_TIMESTAMP(RENDER_RING_BASE),
.offset_udw = RING_TIMESTAMP_UDW(RENDER_RING_BASE),
.size = 8, .gen_bitmask = GEN_RANGE(4, 9) },
};
int i915_reg_read_ioctl(struct drm_device *dev,
void *data, struct drm_file *file)
{
struct drm_i915_private *dev_priv = to_i915(dev);
struct drm_i915_reg_read *reg = data;
struct register_whitelist const *entry = whitelist;
unsigned size;
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
i915_reg_t offset_ldw, offset_udw;
int i, ret = 0;
for (i = 0; i < ARRAY_SIZE(whitelist); i++, entry++) {
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
if (i915_mmio_reg_offset(entry->offset_ldw) == (reg->offset & -entry->size) &&
(INTEL_INFO(dev_priv)->gen_mask & entry->gen_bitmask))
break;
}
if (i == ARRAY_SIZE(whitelist))
return -EINVAL;
/* We use the low bits to encode extra flags as the register should
* be naturally aligned (and those that are not so aligned merely
* limit the available flags for that register).
*/
offset_ldw = entry->offset_ldw;
offset_udw = entry->offset_udw;
size = entry->size;
drm/i915: Type safe register read/write Make I915_READ and I915_WRITE more type safe by wrapping the register offset in a struct. This should eliminate most of the fumbles we've had with misplaced parens. This only takes care of normal mmio registers. We could extend the idea to other register types and define each with its own struct. That way you wouldn't be able to accidentally pass the wrong thing to a specific register access function. The gpio_reg setup is probably the ugliest thing left. But I figure I'd just leave it for now, and wait for some divine inspiration to strike before making it nice. As for the generated code, it's actually a bit better sometimes. Eg. looking at i915_irq_handler(), we can see the following change: lea 0x70024(%rdx,%rax,1),%r9d mov $0x1,%edx - movslq %r9d,%r9 - mov %r9,%rsi - mov %r9,-0x58(%rbp) - callq *0xd8(%rbx) + mov %r9d,%esi + mov %r9d,-0x48(%rbp) callq *0xd8(%rbx) So previously gcc thought the register offset might be signed and decided to sign extend it, just in case. The rest appears to be mostly just minor shuffling of instructions. v2: i915_mmio_reg_{offset,equal,valid}() helpers added s/_REG/_MMIO/ in the register defines mo more switch statements left to worry about ring_emit stuff got sorted in a prep patch cmd parser, lrc context and w/a batch buildup also in prep patch vgpu stuff cleaned up and moved to a prep patch all other unrelated changes split out v3: Rebased due to BXT DSI/BLC, MOCS, etc. v4: Rebased due to churn, s/i915_mmio_reg_t/i915_reg_t/ Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/1447853606-2751-1-git-send-email-ville.syrjala@linux.intel.com
2015-11-18 21:33:26 +08:00
size |= reg->offset ^ i915_mmio_reg_offset(offset_ldw);
intel_runtime_pm_get(dev_priv);
switch (size) {
case 8 | 1:
reg->val = I915_READ64_2x32(offset_ldw, offset_udw);
break;
case 8:
reg->val = I915_READ64(offset_ldw);
break;
case 4:
reg->val = I915_READ(offset_ldw);
break;
case 2:
reg->val = I915_READ16(offset_ldw);
break;
case 1:
reg->val = I915_READ8(offset_ldw);
break;
default:
ret = -EINVAL;
goto out;
}
out:
intel_runtime_pm_put(dev_priv);
return ret;
}
static void gen3_stop_rings(struct drm_i915_private *dev_priv)
{
struct intel_engine_cs *engine;
enum intel_engine_id id;
for_each_engine(engine, dev_priv, id) {
const u32 base = engine->mmio_base;
const i915_reg_t mode = RING_MI_MODE(base);
I915_WRITE_FW(mode, _MASKED_BIT_ENABLE(STOP_RING));
if (intel_wait_for_register_fw(dev_priv,
mode,
MODE_IDLE,
MODE_IDLE,
500))
DRM_DEBUG_DRIVER("%s: timed out on STOP_RING\n",
engine->name);
I915_WRITE_FW(RING_CTL(base), 0);
I915_WRITE_FW(RING_HEAD(base), 0);
I915_WRITE_FW(RING_TAIL(base), 0);
/* Check acts as a post */
if (I915_READ_FW(RING_HEAD(base)) != 0)
DRM_DEBUG_DRIVER("%s: ring head not parked\n",
engine->name);
}
}
static bool i915_reset_complete(struct pci_dev *pdev)
{
u8 gdrst;
pci_read_config_byte(pdev, I915_GDRST, &gdrst);
return (gdrst & GRDOM_RESET_STATUS) == 0;
}
static int i915_do_reset(struct drm_i915_private *dev_priv, unsigned engine_mask)
{
struct pci_dev *pdev = dev_priv->drm.pdev;
/* assert reset for at least 20 usec */
pci_write_config_byte(pdev, I915_GDRST, GRDOM_RESET_ENABLE);
usleep_range(50, 200);
pci_write_config_byte(pdev, I915_GDRST, 0);
return wait_for(i915_reset_complete(pdev), 500);
}
static bool g4x_reset_complete(struct pci_dev *pdev)
{
u8 gdrst;
pci_read_config_byte(pdev, I915_GDRST, &gdrst);
return (gdrst & GRDOM_RESET_ENABLE) == 0;
}
static int g33_do_reset(struct drm_i915_private *dev_priv, unsigned engine_mask)
{
struct pci_dev *pdev = dev_priv->drm.pdev;
/* Stop engines before we reset; see g4x_do_reset() below for why. */
gen3_stop_rings(dev_priv);
pci_write_config_byte(pdev, I915_GDRST, GRDOM_RESET_ENABLE);
return wait_for(g4x_reset_complete(pdev), 500);
}
static int g4x_do_reset(struct drm_i915_private *dev_priv, unsigned engine_mask)
{
struct pci_dev *pdev = dev_priv->drm.pdev;
int ret;
/* WaVcpClkGateDisableForMediaReset:ctg,elk */
I915_WRITE(VDECCLK_GATE_D,
I915_READ(VDECCLK_GATE_D) | VCP_UNIT_CLOCK_GATE_DISABLE);
POSTING_READ(VDECCLK_GATE_D);
/* We stop engines, otherwise we might get failed reset and a
* dead gpu (on elk).
* WaMediaResetMainRingCleanup:ctg,elk (presumably)
*/
gen3_stop_rings(dev_priv);
pci_write_config_byte(pdev, I915_GDRST,
GRDOM_MEDIA | GRDOM_RESET_ENABLE);
ret = wait_for(g4x_reset_complete(pdev), 500);
if (ret) {
DRM_DEBUG_DRIVER("Wait for media reset failed\n");
goto out;
}
pci_write_config_byte(pdev, I915_GDRST,
GRDOM_RENDER | GRDOM_RESET_ENABLE);
ret = wait_for(g4x_reset_complete(pdev), 500);
if (ret) {
DRM_DEBUG_DRIVER("Wait for render reset failed\n");
goto out;
}
out:
pci_write_config_byte(pdev, I915_GDRST, 0);
I915_WRITE(VDECCLK_GATE_D,
I915_READ(VDECCLK_GATE_D) & ~VCP_UNIT_CLOCK_GATE_DISABLE);
POSTING_READ(VDECCLK_GATE_D);
return ret;
}
static int ironlake_do_reset(struct drm_i915_private *dev_priv,
unsigned engine_mask)
{
int ret;
I915_WRITE(ILK_GDSR, ILK_GRDOM_RENDER | ILK_GRDOM_RESET_ENABLE);
ret = intel_wait_for_register(dev_priv,
ILK_GDSR, ILK_GRDOM_RESET_ENABLE, 0,
500);
if (ret) {
DRM_DEBUG_DRIVER("Wait for render reset failed\n");
goto out;
}
I915_WRITE(ILK_GDSR, ILK_GRDOM_MEDIA | ILK_GRDOM_RESET_ENABLE);
ret = intel_wait_for_register(dev_priv,
ILK_GDSR, ILK_GRDOM_RESET_ENABLE, 0,
500);
if (ret) {
DRM_DEBUG_DRIVER("Wait for media reset failed\n");
goto out;
}
out:
I915_WRITE(ILK_GDSR, 0);
POSTING_READ(ILK_GDSR);
return ret;
}
/* Reset the hardware domains (GENX_GRDOM_*) specified by mask */
static int gen6_hw_domain_reset(struct drm_i915_private *dev_priv,
u32 hw_domain_mask)
{
int err;
/* 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
*/
__raw_i915_write32(dev_priv, GEN6_GDRST, hw_domain_mask);
/* Wait for the device to ack the reset requests */
err = intel_wait_for_register_fw(dev_priv,
GEN6_GDRST, hw_domain_mask, 0,
500);
if (err)
DRM_DEBUG_DRIVER("Wait for 0x%08x engines reset failed\n",
hw_domain_mask);
return err;
}
/**
* gen6_reset_engines - reset individual engines
* @dev_priv: i915 device
* @engine_mask: mask of intel_ring_flag() engines or ALL_ENGINES for full reset
*
* This function will reset the individual engines that are set in engine_mask.
* If you provide ALL_ENGINES as mask, full global domain reset will be issued.
*
* Note: It is responsibility of the caller to handle the difference between
* asking full domain reset versus reset for all available individual engines.
*
* Returns 0 on success, nonzero on error.
*/
static int gen6_reset_engines(struct drm_i915_private *dev_priv,
unsigned engine_mask)
{
struct intel_engine_cs *engine;
const u32 hw_engine_mask[I915_NUM_ENGINES] = {
[RCS] = GEN6_GRDOM_RENDER,
[BCS] = GEN6_GRDOM_BLT,
[VCS] = GEN6_GRDOM_MEDIA,
[VCS2] = GEN8_GRDOM_MEDIA2,
[VECS] = GEN6_GRDOM_VECS,
};
u32 hw_mask;
if (engine_mask == ALL_ENGINES) {
hw_mask = GEN6_GRDOM_FULL;
} else {
unsigned int tmp;
hw_mask = 0;
for_each_engine_masked(engine, dev_priv, engine_mask, tmp)
hw_mask |= hw_engine_mask[engine->id];
}
drm/i915: Stop touching forcewake following a gen6+ engine reset Forcewake is not affected by the engine reset on gen6+. Indeed the reason why we added intel_uncore_forcewake_reset() to gen6_reset_engines() was to keep the bookkeeping intact because the reset did not touch the forcewake bit (yet we cancelled the forcewake consumers)! This was done in commit 521198a2e7095: Author: Mika Kuoppala <mika.kuoppala@linux.intel.com> Date: Fri Aug 23 16:52:30 2013 +0300 drm/i915: sanitize forcewake registers on reset In reset we try to restore the forcewake state to pre reset state, using forcewake_count. The reset doesn't seem to clear the forcewake bits so we get warn on forcewake ack register not clearing. That futzing of the forcewake bookkeeping was dropped in commit 0294ae7b44bb ("drm/i915: Consolidate forcewake resetting to a single function"), but it did not make the realisation that the remaining intel_uncore_forcewake_reset() was redundant. The new danger with using intel_uncore_forcewake_reset() with per-engine resets is that the driver and hw are still in an active state as we perform the reset. We may be using the forcewake to read protected registers elsewhere and those results may be clobbered by the concurrent dropping of forcewake. Reported-by: Michel Thierry <michel.thierry@intel.com> Fixes: 142bc7d99bcf ("drm/i915: Modify error handler for per engine hang recovery") Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Cc: Michel Thierry <michel.thierry@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20170817173229.20324-1-chris@chris-wilson.co.uk Reviewed-by: Michel Thierry <michel.thierry@intel.com> Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
2017-08-18 01:32:29 +08:00
return gen6_hw_domain_reset(dev_priv, hw_mask);
}
/**
* __intel_wait_for_register_fw - wait until register matches expected state
* @dev_priv: the i915 device
* @reg: the register to read
* @mask: mask to apply to register value
* @value: expected value
* @fast_timeout_us: fast timeout in microsecond for atomic/tight wait
* @slow_timeout_ms: slow timeout in millisecond
* @out_value: optional placeholder to hold registry value
*
* This routine waits until the target register @reg contains the expected
* @value after applying the @mask, i.e. it waits until ::
*
* (I915_READ_FW(reg) & mask) == value
*
* Otherwise, the wait will timeout after @slow_timeout_ms milliseconds.
* For atomic context @slow_timeout_ms must be zero and @fast_timeout_us
* must be not larger than 20,0000 microseconds.
*
* Note that this routine assumes the caller holds forcewake asserted, it is
* not suitable for very long waits. See intel_wait_for_register() if you
* wish to wait without holding forcewake for the duration (i.e. you expect
* the wait to be slow).
*
* Returns 0 if the register matches the desired condition, or -ETIMEOUT.
*/
int __intel_wait_for_register_fw(struct drm_i915_private *dev_priv,
i915_reg_t reg,
u32 mask,
u32 value,
unsigned int fast_timeout_us,
unsigned int slow_timeout_ms,
u32 *out_value)
{
u32 uninitialized_var(reg_value);
#define done (((reg_value = I915_READ_FW(reg)) & mask) == value)
int ret;
/* Catch any overuse of this function */
might_sleep_if(slow_timeout_ms);
GEM_BUG_ON(fast_timeout_us > 20000);
ret = -ETIMEDOUT;
if (fast_timeout_us && fast_timeout_us <= 20000)
ret = _wait_for_atomic(done, fast_timeout_us, 0);
if (ret && slow_timeout_ms)
ret = wait_for(done, slow_timeout_ms);
if (out_value)
*out_value = reg_value;
return ret;
#undef done
}
/**
* intel_wait_for_register - wait until register matches expected state
* @dev_priv: the i915 device
* @reg: the register to read
* @mask: mask to apply to register value
* @value: expected value
* @timeout_ms: timeout in millisecond
*
* This routine waits until the target register @reg contains the expected
* @value after applying the @mask, i.e. it waits until ::
*
* (I915_READ(reg) & mask) == value
*
* Otherwise, the wait will timeout after @timeout_ms milliseconds.
*
* Returns 0 if the register matches the desired condition, or -ETIMEOUT.
*/
int intel_wait_for_register(struct drm_i915_private *dev_priv,
i915_reg_t reg,
u32 mask,
u32 value,
unsigned int timeout_ms)
2015-06-18 17:51:40 +08:00
{
unsigned fw =
intel_uncore_forcewake_for_reg(dev_priv, reg, FW_REG_READ);
int ret;
might_sleep();
spin_lock_irq(&dev_priv->uncore.lock);
intel_uncore_forcewake_get__locked(dev_priv, fw);
ret = __intel_wait_for_register_fw(dev_priv,
reg, mask, value,
2, 0, NULL);
intel_uncore_forcewake_put__locked(dev_priv, fw);
spin_unlock_irq(&dev_priv->uncore.lock);
if (ret)
ret = wait_for((I915_READ_NOTRACE(reg) & mask) == value,
timeout_ms);
return ret;
}
static int gen8_reset_engine_start(struct intel_engine_cs *engine)
{
struct drm_i915_private *dev_priv = engine->i915;
int ret;
I915_WRITE_FW(RING_RESET_CTL(engine->mmio_base),
_MASKED_BIT_ENABLE(RESET_CTL_REQUEST_RESET));
ret = intel_wait_for_register_fw(dev_priv,
RING_RESET_CTL(engine->mmio_base),
RESET_CTL_READY_TO_RESET,
RESET_CTL_READY_TO_RESET,
700);
if (ret)
DRM_ERROR("%s: reset request timeout\n", engine->name);
return ret;
}
static void gen8_reset_engine_cancel(struct intel_engine_cs *engine)
{
struct drm_i915_private *dev_priv = engine->i915;
I915_WRITE_FW(RING_RESET_CTL(engine->mmio_base),
_MASKED_BIT_DISABLE(RESET_CTL_REQUEST_RESET));
2015-06-18 17:51:40 +08:00
}
static int gen8_reset_engines(struct drm_i915_private *dev_priv,
unsigned engine_mask)
2015-06-18 17:51:40 +08:00
{
struct intel_engine_cs *engine;
unsigned int tmp;
2015-06-18 17:51:40 +08:00
for_each_engine_masked(engine, dev_priv, engine_mask, tmp)
if (gen8_reset_engine_start(engine))
2015-06-18 17:51:40 +08:00
goto not_ready;
return gen6_reset_engines(dev_priv, engine_mask);
2015-06-18 17:51:40 +08:00
not_ready:
for_each_engine_masked(engine, dev_priv, engine_mask, tmp)
gen8_reset_engine_cancel(engine);
2015-06-18 17:51:40 +08:00
return -EIO;
}
typedef int (*reset_func)(struct drm_i915_private *, unsigned engine_mask);
static reset_func intel_get_gpu_reset(struct drm_i915_private *dev_priv)
{
if (!i915.reset)
return NULL;
if (INTEL_INFO(dev_priv)->gen >= 8)
return gen8_reset_engines;
else if (INTEL_INFO(dev_priv)->gen >= 6)
return gen6_reset_engines;
else if (IS_GEN5(dev_priv))
return ironlake_do_reset;
else if (IS_G4X(dev_priv))
return g4x_do_reset;
else if (IS_G33(dev_priv) || IS_PINEVIEW(dev_priv))
return g33_do_reset;
else if (INTEL_INFO(dev_priv)->gen >= 3)
return i915_do_reset;
else
return NULL;
}
int intel_gpu_reset(struct drm_i915_private *dev_priv, unsigned engine_mask)
{
reset_func reset;
int retry;
int ret;
might_sleep();
reset = intel_get_gpu_reset(dev_priv);
if (reset == NULL)
return -ENODEV;
/* If the power well sleeps during the reset, the reset
* request may be dropped and never completes (causing -EIO).
*/
intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
for (retry = 0; retry < 3; retry++) {
ret = reset(dev_priv, engine_mask);
if (ret != -ETIMEDOUT)
break;
cond_resched();
}
intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
return ret;
}
bool intel_has_gpu_reset(struct drm_i915_private *dev_priv)
{
return intel_get_gpu_reset(dev_priv) != NULL;
}
drm/i915: Modify error handler for per engine hang recovery This is a preparatory patch which modifies error handler to do per engine hang recovery. The actual patch which implements this sequence follows later in the series. The aim is to prepare existing recovery function to adapt to this new function where applicable (which fails at this point because core implementation is lacking) and continue recovery using legacy full gpu reset. A helper function is also added to query the availability of engine reset. A subsequent patch will add the capability to query which type of reset is present (engine -> full -> no-reset) via the get-param ioctl. It has been decided that the error events that are used to notify user of reset will only be sent in case if full chip reset. In case of just single (or multiple) engine resets, userspace won't be notified by these events. Note that this implementation of engine reset is for i915 directly submitting to the ELSP, where the driver manages the hang detection, recovery and resubmission. With GuC submission these tasks are shared between driver and firmware; i915 will still responsible for detecting a hang, and when it does it will have to request GuC to reset that Engine and remind the firmware about the outstanding submissions. This will be added in different patch. v2: rebase, advertise engine reset availability in platform definition, add note about GuC submission. v3: s/*engine_reset*/*reset_engine*/. (Chris) Handle reset as 2 level resets, by first going to engine only and fall backing to full/chip reset as needed, i.e. reset_engine will need the struct_mutex. v4: Pass the engine mask to i915_reset. (Chris) v5: Rebase, update selftests. v6: Rebase, prepare for mutex-less reset engine. v7: Pass reset_engine mask as a function parameter, and iterate over the engine mask for reset_engine. (Chris) v8: Use i915.reset >=2 in has_reset_engine; remove redundant reset logging; add a reset-engine-in-progress flag to prevent concurrent resets, and avoid dual purposing of reset-backoff. (Chris) v9: Support reset of different engines in parallel (Chris) v10: Handle reset-engine flag locking better (Chris) v11: Squash in reporting of per-engine-reset availability. Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com> Signed-off-by: Ian Lister <ian.lister@intel.com> Signed-off-by: Tomas Elf <tomas.elf@intel.com> Signed-off-by: Arun Siluvery <arun.siluvery@linux.intel.com> Signed-off-by: Michel Thierry <michel.thierry@intel.com> Link: http://patchwork.freedesktop.org/patch/msgid/20170615201828.23144-4-michel.thierry@intel.com Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Link: http://patchwork.freedesktop.org/patch/msgid/20170620095751.13127-5-chris@chris-wilson.co.uk
2017-06-20 17:57:46 +08:00
/*
* When GuC submission is enabled, GuC manages ELSP and can initiate the
* engine reset too. For now, fall back to full GPU reset if it is enabled.
*/
bool intel_has_reset_engine(struct drm_i915_private *dev_priv)
{
return (dev_priv->info.has_reset_engine &&
!dev_priv->guc.execbuf_client &&
i915.reset >= 2);
}
int intel_guc_reset(struct drm_i915_private *dev_priv)
{
int ret;
if (!HAS_GUC(dev_priv))
return -EINVAL;
intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
ret = gen6_hw_domain_reset(dev_priv, GEN9_GRDOM_GUC);
intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
return ret;
}
bool intel_uncore_unclaimed_mmio(struct drm_i915_private *dev_priv)
{
return check_for_unclaimed_mmio(dev_priv);
}
bool
intel_uncore_arm_unclaimed_mmio_detection(struct drm_i915_private *dev_priv)
{
if (unlikely(i915.mmio_debug ||
dev_priv->uncore.unclaimed_mmio_check <= 0))
return false;
if (unlikely(intel_uncore_unclaimed_mmio(dev_priv))) {
DRM_DEBUG("Unclaimed register detected, "
"enabling oneshot unclaimed register reporting. "
"Please use i915.mmio_debug=N for more information.\n");
i915.mmio_debug++;
dev_priv->uncore.unclaimed_mmio_check--;
return true;
}
return false;
}
static enum forcewake_domains
intel_uncore_forcewake_for_read(struct drm_i915_private *dev_priv,
i915_reg_t reg)
{
u32 offset = i915_mmio_reg_offset(reg);
enum forcewake_domains fw_domains;
if (HAS_FWTABLE(dev_priv)) {
fw_domains = __fwtable_reg_read_fw_domains(offset);
} else if (INTEL_GEN(dev_priv) >= 6) {
fw_domains = __gen6_reg_read_fw_domains(offset);
} else {
WARN_ON(!IS_GEN(dev_priv, 2, 5));
fw_domains = 0;
}
WARN_ON(fw_domains & ~dev_priv->uncore.fw_domains);
return fw_domains;
}
static enum forcewake_domains
intel_uncore_forcewake_for_write(struct drm_i915_private *dev_priv,
i915_reg_t reg)
{
u32 offset = i915_mmio_reg_offset(reg);
enum forcewake_domains fw_domains;
if (HAS_FWTABLE(dev_priv) && !IS_VALLEYVIEW(dev_priv)) {
fw_domains = __fwtable_reg_write_fw_domains(offset);
} else if (IS_GEN8(dev_priv)) {
fw_domains = __gen8_reg_write_fw_domains(offset);
} else if (IS_GEN(dev_priv, 6, 7)) {
fw_domains = FORCEWAKE_RENDER;
} else {
WARN_ON(!IS_GEN(dev_priv, 2, 5));
fw_domains = 0;
}
WARN_ON(fw_domains & ~dev_priv->uncore.fw_domains);
return fw_domains;
}
/**
* intel_uncore_forcewake_for_reg - which forcewake domains are needed to access
* a register
* @dev_priv: pointer to struct drm_i915_private
* @reg: register in question
* @op: operation bitmask of FW_REG_READ and/or FW_REG_WRITE
*
* Returns a set of forcewake domains required to be taken with for example
* intel_uncore_forcewake_get for the specified register to be accessible in the
* specified mode (read, write or read/write) with raw mmio accessors.
*
* NOTE: On Gen6 and Gen7 write forcewake domain (FORCEWAKE_RENDER) requires the
* callers to do FIFO management on their own or risk losing writes.
*/
enum forcewake_domains
intel_uncore_forcewake_for_reg(struct drm_i915_private *dev_priv,
i915_reg_t reg, unsigned int op)
{
enum forcewake_domains fw_domains = 0;
WARN_ON(!op);
if (intel_vgpu_active(dev_priv))
return 0;
if (op & FW_REG_READ)
fw_domains = intel_uncore_forcewake_for_read(dev_priv, reg);
if (op & FW_REG_WRITE)
fw_domains |= intel_uncore_forcewake_for_write(dev_priv, reg);
return fw_domains;
}
#if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
#include "selftests/mock_uncore.c"
#include "selftests/intel_uncore.c"
#endif