| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20220127180256.811271517@linuxfoundation.org>
Date: Thu, 27 Jan 2022 19:08:56 +0100
From: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
To: linux-kernel@...r.kernel.org
Cc: Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
stable@...r.kernel.org, Tvrtko Ursulin <tvrtko.ursulin@...el.com>,
Sushma Venkatesh Reddy <sushma.venkatesh.reddy@...el.com>,
Daniel Vetter <daniel.vetter@...ll.ch>,
Dave Airlie <airlied@...hat.com>,
Jon Bloomfield <jon.bloomfield@...el.com>,
Joonas Lahtinen <joonas.lahtinen@...ux.intel.com>,
Jani Nikula <jani.nikula@...el.com>,
Linus Torvalds <torvalds@...ux-foundation.org>
Subject: [PATCH 4.14 1/2] drm/i915: Flush TLBs before releasing backing store
From: Tvrtko Ursulin <tvrtko.ursulin@...el.com>
commit 7938d61591d33394a21bdd7797a245b65428f44c upstream.
We need to flush TLBs before releasing backing store otherwise userspace
is able to encounter stale entries if a) it is not declaring access to
certain buffers and b) it races with the backing store release from a
such undeclared execution already executing on the GPU in parallel.
The approach taken is to mark any buffer objects which were ever bound
to the GPU and to trigger a serialized TLB flush when their backing
store is released.
Alternatively the flushing could be done on VMA unbind, at which point
we would be able to ascertain whether there is potential a parallel GPU
execution (which could race), but essentially it boils down to paying
the cost of TLB flushes potentially needlessly at VMA unbind time (when
the backing store is not known to be going away so not needed for
safety), versus potentially needlessly at backing store relase time
(since we at that point cannot tell whether there is anything executing
on the GPU which uses that object).
Thereforce simplicity of implementation has been chosen for now with
scope to benchmark and refine later as required.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@...el.com>
Reported-by: Sushma Venkatesh Reddy <sushma.venkatesh.reddy@...el.com>
Reviewed-by: Daniel Vetter <daniel.vetter@...ll.ch>
Acked-by: Dave Airlie <airlied@...hat.com>
Cc: Daniel Vetter <daniel.vetter@...ll.ch>
Cc: Jon Bloomfield <jon.bloomfield@...el.com>
Cc: Joonas Lahtinen <joonas.lahtinen@...ux.intel.com>
Cc: Jani Nikula <jani.nikula@...el.com>
Cc: stable@...r.kernel.org
Signed-off-by: Linus Torvalds <torvalds@...ux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
---
drivers/gpu/drm/i915/i915_drv.h | 2
drivers/gpu/drm/i915/i915_gem.c | 84 ++++++++++++++++++++++++++++++++-
drivers/gpu/drm/i915/i915_gem_object.h | 1
drivers/gpu/drm/i915/i915_reg.h | 6 ++
drivers/gpu/drm/i915/i915_vma.c | 4 +
5 files changed, 96 insertions(+), 1 deletion(-)
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2166,6 +2166,8 @@ struct drm_i915_private {
struct intel_uncore uncore;
+ struct mutex tlb_invalidate_lock;
+
struct i915_virtual_gpu vgpu;
struct intel_gvt *gvt;
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2220,6 +2220,76 @@ static void __i915_gem_object_reset_page
rcu_read_unlock();
}
+struct reg_and_bit {
+ i915_reg_t reg;
+ u32 bit;
+};
+
+static struct reg_and_bit
+get_reg_and_bit(const struct intel_engine_cs *engine,
+ const i915_reg_t *regs, const unsigned int num)
+{
+ const unsigned int class = engine->class;
+ struct reg_and_bit rb = { .bit = 1 };
+
+ if (WARN_ON_ONCE(class >= num || !regs[class].reg))
+ return rb;
+
+ rb.reg = regs[class];
+ if (class == VIDEO_DECODE_CLASS)
+ rb.reg.reg += 4 * engine->instance; /* GEN8_M2TCR */
+
+ return rb;
+}
+
+static void invalidate_tlbs(struct drm_i915_private *dev_priv)
+{
+ static const i915_reg_t gen8_regs[] = {
+ [RENDER_CLASS] = GEN8_RTCR,
+ [VIDEO_DECODE_CLASS] = GEN8_M1TCR, /* , GEN8_M2TCR */
+ [VIDEO_ENHANCEMENT_CLASS] = GEN8_VTCR,
+ [COPY_ENGINE_CLASS] = GEN8_BTCR,
+ };
+ const unsigned int num = ARRAY_SIZE(gen8_regs);
+ const i915_reg_t *regs = gen8_regs;
+ struct intel_engine_cs *engine;
+ enum intel_engine_id id;
+
+ if (INTEL_GEN(dev_priv) < 8)
+ return;
+
+ assert_rpm_wakelock_held(dev_priv);
+
+ mutex_lock(&dev_priv->tlb_invalidate_lock);
+ intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
+
+ for_each_engine(engine, dev_priv, id) {
+ /*
+ * HW architecture suggest typical invalidation time at 40us,
+ * with pessimistic cases up to 100us and a recommendation to
+ * cap at 1ms. We go a bit higher just in case.
+ */
+ const unsigned int timeout_us = 100;
+ const unsigned int timeout_ms = 4;
+ struct reg_and_bit rb;
+
+ rb = get_reg_and_bit(engine, regs, num);
+ if (!i915_mmio_reg_offset(rb.reg))
+ continue;
+
+ I915_WRITE_FW(rb.reg, rb.bit);
+ if (__intel_wait_for_register_fw(dev_priv,
+ rb.reg, rb.bit, 0,
+ timeout_us, timeout_ms,
+ NULL))
+ DRM_ERROR_RATELIMITED("%s TLB invalidation did not complete in %ums!\n",
+ engine->name, timeout_ms);
+ }
+
+ intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
+ mutex_unlock(&dev_priv->tlb_invalidate_lock);
+}
+
void __i915_gem_object_put_pages(struct drm_i915_gem_object *obj,
enum i915_mm_subclass subclass)
{
@@ -2257,8 +2327,18 @@ void __i915_gem_object_put_pages(struct
__i915_gem_object_reset_page_iter(obj);
- if (!IS_ERR(pages))
+ if (!IS_ERR(pages)) {
+ if (test_and_clear_bit(I915_BO_WAS_BOUND_BIT, &obj->flags)) {
+ struct drm_i915_private *i915 = to_i915(obj->base.dev);
+
+ if (intel_runtime_pm_get_if_in_use(i915)) {
+ invalidate_tlbs(i915);
+ intel_runtime_pm_put(i915);
+ }
+ }
+
obj->ops->put_pages(obj, pages);
+ }
unlock:
mutex_unlock(&obj->mm.lock);
@@ -4972,6 +5052,8 @@ i915_gem_load_init(struct drm_i915_priva
spin_lock_init(&dev_priv->fb_tracking.lock);
+ mutex_init(&dev_priv->tlb_invalidate_lock);
+
return 0;
err_priorities:
--- a/drivers/gpu/drm/i915/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/i915_gem_object.h
@@ -135,6 +135,7 @@ struct drm_i915_gem_object {
* activity?
*/
#define I915_BO_ACTIVE_REF 0
+#define I915_BO_WAS_BOUND_BIT 1
/*
* Is the object to be mapped as read-only to the GPU
--- a/drivers/gpu/drm/i915/i915_reg.h
+++ b/drivers/gpu/drm/i915/i915_reg.h
@@ -2380,6 +2380,12 @@ enum i915_power_well_id {
#define GAMT_CHKN_BIT_REG _MMIO(0x4ab8)
#define GAMT_CHKN_DISABLE_DYNAMIC_CREDIT_SHARING (1<<28)
+#define GEN8_RTCR _MMIO(0x4260)
+#define GEN8_M1TCR _MMIO(0x4264)
+#define GEN8_M2TCR _MMIO(0x4268)
+#define GEN8_BTCR _MMIO(0x426c)
+#define GEN8_VTCR _MMIO(0x4270)
+
#if 0
#define PRB0_TAIL _MMIO(0x2030)
#define PRB0_HEAD _MMIO(0x2034)
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -272,6 +272,10 @@ int i915_vma_bind(struct i915_vma *vma,
return ret;
vma->flags |= bind_flags;
+
+ if (vma->obj)
+ set_bit(I915_BO_WAS_BOUND_BIT, &vma->obj->flags);
+
return 0;
}
Powered by blists - more mailing lists