drm/i915/vgpu: Disallow loading on old vGPU hosts
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 30 Nov 2018 12:59:54 +0000 (12:59 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Mon, 3 Dec 2018 16:08:26 +0000 (16:08 +0000)
Since commit fd8526e50902 ("drm/i915/execlists: Trust the CSB") we
actually broke the force-mmio mode for our execlists implementation. No
one noticed, so ergo no one is actually using an old vGPU host (where we
required the older method) and so can simply remove the broken support.

v2: csb_read can go as well (Mika)

Reported-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Fixes: fd8526e50902 ("drm/i915/execlists: Trust the CSB")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181130125954.11924-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/i915_drv.c
drivers/gpu/drm/i915/intel_lrc.c
drivers/gpu/drm/i915/intel_ringbuffer.h

index e3901671346465e1ca97eb53c44e492ac2bcfa1a..3e5e2efce670bd88d334f36a23fbeb6ab9a83ee4 100644 (file)
@@ -1384,6 +1384,20 @@ static int i915_driver_init_hw(struct drm_i915_private *dev_priv)
                }
        }
 
+       if (HAS_EXECLISTS(dev_priv)) {
+               /*
+                * Older GVT emulation depends upon intercepting CSB mmio,
+                * which we no longer use, preferring to use the HWSP cache
+                * instead.
+                */
+               if (intel_vgpu_active(dev_priv) &&
+                   !intel_vgpu_has_hwsp_emulation(dev_priv)) {
+                       i915_report_error(dev_priv,
+                                         "old vGPU host found, support for HWSP emulation required\n");
+                       return -ENXIO;
+               }
+       }
+
        intel_sanitize_options(dev_priv);
 
        i915_perf_init(dev_priv);
index 11f4e6148557a855727fe346914d6c846d362e8f..1f004683b777dd8134287e4894e7f07ea72dac66 100644 (file)
@@ -767,6 +767,8 @@ execlists_cancel_port_requests(struct intel_engine_execlists * const execlists)
 
 static void reset_csb_pointers(struct intel_engine_execlists *execlists)
 {
+       const unsigned int reset_value = GEN8_CSB_ENTRIES - 1;
+
        /*
         * After a reset, the HW starts writing into CSB entry [0]. We
         * therefore have to set our HEAD pointer back one entry so that
@@ -776,8 +778,8 @@ static void reset_csb_pointers(struct intel_engine_execlists *execlists)
         * inline comparison of our cached head position against the last HW
         * write works even before the first interrupt.
         */
-       execlists->csb_head = execlists->csb_write_reset;
-       WRITE_ONCE(*execlists->csb_write, execlists->csb_write_reset);
+       execlists->csb_head = reset_value;
+       WRITE_ONCE(*execlists->csb_write, reset_value);
 }
 
 static void nop_submission_tasklet(unsigned long data)
@@ -2217,12 +2219,6 @@ logical_ring_setup(struct intel_engine_cs *engine)
        logical_ring_default_irqs(engine);
 }
 
-static bool csb_force_mmio(struct drm_i915_private *i915)
-{
-       /* Older GVT emulation depends upon intercepting CSB mmio */
-       return intel_vgpu_active(i915) && !intel_vgpu_has_hwsp_emulation(i915);
-}
-
 static int logical_ring_init(struct intel_engine_cs *engine)
 {
        struct drm_i915_private *i915 = engine->i915;
@@ -2252,24 +2248,12 @@ static int logical_ring_init(struct intel_engine_cs *engine)
                        upper_32_bits(ce->lrc_desc);
        }
 
-       execlists->csb_read =
-               i915->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_PTR(engine));
-       if (csb_force_mmio(i915)) {
-               execlists->csb_status = (u32 __force *)
-                       (i915->regs + i915_mmio_reg_offset(RING_CONTEXT_STATUS_BUF_LO(engine, 0)));
+       execlists->csb_status =
+               &engine->status_page.page_addr[I915_HWS_CSB_BUF0_INDEX];
 
-               execlists->csb_write = (u32 __force *)execlists->csb_read;
-               execlists->csb_write_reset =
-                       _MASKED_FIELD(GEN8_CSB_WRITE_PTR_MASK,
-                                     GEN8_CSB_ENTRIES - 1);
-       } else {
-               execlists->csb_status =
-                       &engine->status_page.page_addr[I915_HWS_CSB_BUF0_INDEX];
+       execlists->csb_write =
+               &engine->status_page.page_addr[intel_hws_csb_write_index(i915)];
 
-               execlists->csb_write =
-                       &engine->status_page.page_addr[intel_hws_csb_write_index(i915)];
-               execlists->csb_write_reset = GEN8_CSB_ENTRIES - 1;
-       }
        reset_csb_pointers(execlists);
 
        return 0;
index 8a2270b209b0c5da57864beec724f0bfd9bbc348..f4988f7bc756033f23ca83740681e11a0ff81fe8 100644 (file)
@@ -312,13 +312,6 @@ struct intel_engine_execlists {
         */
        struct rb_root_cached queue;
 
-       /**
-        * @csb_read: control register for Context Switch buffer
-        *
-        * Note this register is always in mmio.
-        */
-       u32 __iomem *csb_read;
-
        /**
         * @csb_write: control register for Context Switch buffer
         *
@@ -338,15 +331,6 @@ struct intel_engine_execlists {
         */
        u32 preempt_complete_status;
 
-       /**
-        * @csb_write_reset: reset value for CSB write pointer
-        *
-        * As the CSB write pointer maybe either in HWSP or as a field
-        * inside an mmio register, we want to reprogram it slightly
-        * differently to avoid later confusion.
-        */
-       u32 csb_write_reset;
-
        /**
         * @csb_head: context status buffer head
         */