linux_dsm_epyc7002/drivers/gpu/drm/i915/gvt/mmio.c

316 lines
8.0 KiB
C
Raw Normal View History

/*
* Copyright(c) 2011-2016 Intel Corporation. All rights reserved.
*
* 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.
*
* Authors:
* Ke Yu
* Kevin Tian <kevin.tian@intel.com>
* Dexuan Cui
*
* Contributors:
* Tina Zhang <tina.zhang@intel.com>
* Min He <min.he@intel.com>
* Niu Bing <bing.niu@intel.com>
* Zhi Wang <zhi.a.wang@intel.com>
*
*/
#include "i915_drv.h"
#include "gvt.h"
/**
* intel_vgpu_gpa_to_mmio_offset - translate a GPA to MMIO offset
* @vgpu: a vGPU
* @gpa: guest physical address
*
* Returns:
* Zero on success, negative error code if failed
*/
int intel_vgpu_gpa_to_mmio_offset(struct intel_vgpu *vgpu, u64 gpa)
{
u64 gttmmio_gpa = intel_vgpu_get_bar_gpa(vgpu, PCI_BASE_ADDRESS_0);
return gpa - gttmmio_gpa;
}
#define reg_is_mmio(gvt, reg) \
(reg >= 0 && reg < gvt->device_info.mmio_size)
#define reg_is_gtt(gvt, reg) \
(reg >= gvt->device_info.gtt_start_offset \
&& reg < gvt->device_info.gtt_start_offset + gvt_ggtt_sz(gvt))
static void failsafe_emulate_mmio_rw(struct intel_vgpu *vgpu, u64 pa,
void *p_data, unsigned int bytes, bool read)
{
struct intel_gvt *gvt = NULL;
void *pt = NULL;
unsigned int offset = 0;
if (!vgpu || !p_data)
return;
gvt = vgpu->gvt;
drm/i915/gvt: Use vgpu_lock to protect per vgpu access The patch set splits out 2 small locks from the original big gvt lock: - vgpu_lock protects per-vGPU data and logic, especially the vGPU trap emulation path. - sched_lock protects gvt scheudler structure, context schedule logic and vGPU's schedule data. Use vgpu_lock to replace the gvt big lock. By doing this, the mmio read/write trap path, vgpu virtual event emulation and other vgpu related process, would be protected under per vgpu_lock. v9: - Change commit author since the patches are improved a lot compared with original version. Original author: Pei Zhang <pei.zhang@intel.com> - Rebase to latest gvt-staging. v8: - Correct coding and comment style. - Rebase to latest gvt-staging. v7: - Remove gtt_lock since already proteced by gvt_lock and vgpu_lock. - Fix a typo in intel_gvt_deactivate_vgpu, unlock the wrong lock. v6: - Rebase to latest gvt-staging. v5: - Rebase to latest gvt-staging. - intel_vgpu_page_track_handler should use vgpu_lock. v4: - Rebase to latest gvt-staging. - Protect vgpu->active access with vgpu_lock. - Do not wait gpu idle in vgpu_lock. v3: update to latest code base v2: add gvt->lock in function gvt_check_vblank_emulation Performance comparison on Kabylake platform. - Configuration: Host: Ubuntu 16.04. Guest 1 & 2: Ubuntu 16.04. glmark2 score comparison: - Configuration: Host: glxgears. Guests: glmark2. +--------------------------------+-----------------+ | Setup | glmark2 score | +--------------------------------+-----------------+ | unified lock, iommu=on | 58~62 (avg. 60) | +--------------------------------+-----------------+ | unified lock, iommu=igfx_off | 57~61 (avg. 59) | +--------------------------------+-----------------+ | per-logic lock, iommu=on | 60~68 (avg. 64) | +--------------------------------+-----------------+ | per-logic lock, iommu=igfx_off | 61~67 (avg. 64) | +--------------------------------+-----------------+ lock_stat comparison: - Configuration: Stop lock stat immediately after boot up. Boot 2 VM Guests. Run glmark2 in guests. Start perf lock_stat for 20 seconds and stop again. - Legend: c - contentions; w - waittime-avg +------------+-----------------+-----------+---------------+------------+ | | gvt_lock |sched_lock | vgpu_lock | gtt_lock | + lock type; +-----------------+-----------+---------------+------------+ | iommu set | c | w | c | w | c | w | c | w | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 20697 | 839 |N/A | N/A | N/A | N/A | N/A | N/A | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 21838 | 658.15 |N/A | N/A | N/A | N/A | N/A | N/A | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1553 | 1599.96 |9458|429.97| 5846 | 274.33 | 0 | 0.00 | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1911 | 1678.32 |8335|445.16| 5451 | 244.80 | 0 | 0.00 | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ Signed-off-by: Pei Zhang <pei.zhang@intel.com> Signed-off-by: Colin Xu <colin.xu@intel.com> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
2018-05-19 11:28:54 +07:00
mutex_lock(&vgpu->vgpu_lock);
offset = intel_vgpu_gpa_to_mmio_offset(vgpu, pa);
if (reg_is_mmio(gvt, offset)) {
if (read)
intel_vgpu_default_mmio_read(vgpu, offset, p_data,
bytes);
else
intel_vgpu_default_mmio_write(vgpu, offset, p_data,
bytes);
} else if (reg_is_gtt(gvt, offset)) {
offset -= gvt->device_info.gtt_start_offset;
pt = vgpu->gtt.ggtt_mm->ggtt_mm.virtual_ggtt + offset;
if (read)
memcpy(p_data, pt, bytes);
else
memcpy(pt, p_data, bytes);
}
drm/i915/gvt: Use vgpu_lock to protect per vgpu access The patch set splits out 2 small locks from the original big gvt lock: - vgpu_lock protects per-vGPU data and logic, especially the vGPU trap emulation path. - sched_lock protects gvt scheudler structure, context schedule logic and vGPU's schedule data. Use vgpu_lock to replace the gvt big lock. By doing this, the mmio read/write trap path, vgpu virtual event emulation and other vgpu related process, would be protected under per vgpu_lock. v9: - Change commit author since the patches are improved a lot compared with original version. Original author: Pei Zhang <pei.zhang@intel.com> - Rebase to latest gvt-staging. v8: - Correct coding and comment style. - Rebase to latest gvt-staging. v7: - Remove gtt_lock since already proteced by gvt_lock and vgpu_lock. - Fix a typo in intel_gvt_deactivate_vgpu, unlock the wrong lock. v6: - Rebase to latest gvt-staging. v5: - Rebase to latest gvt-staging. - intel_vgpu_page_track_handler should use vgpu_lock. v4: - Rebase to latest gvt-staging. - Protect vgpu->active access with vgpu_lock. - Do not wait gpu idle in vgpu_lock. v3: update to latest code base v2: add gvt->lock in function gvt_check_vblank_emulation Performance comparison on Kabylake platform. - Configuration: Host: Ubuntu 16.04. Guest 1 & 2: Ubuntu 16.04. glmark2 score comparison: - Configuration: Host: glxgears. Guests: glmark2. +--------------------------------+-----------------+ | Setup | glmark2 score | +--------------------------------+-----------------+ | unified lock, iommu=on | 58~62 (avg. 60) | +--------------------------------+-----------------+ | unified lock, iommu=igfx_off | 57~61 (avg. 59) | +--------------------------------+-----------------+ | per-logic lock, iommu=on | 60~68 (avg. 64) | +--------------------------------+-----------------+ | per-logic lock, iommu=igfx_off | 61~67 (avg. 64) | +--------------------------------+-----------------+ lock_stat comparison: - Configuration: Stop lock stat immediately after boot up. Boot 2 VM Guests. Run glmark2 in guests. Start perf lock_stat for 20 seconds and stop again. - Legend: c - contentions; w - waittime-avg +------------+-----------------+-----------+---------------+------------+ | | gvt_lock |sched_lock | vgpu_lock | gtt_lock | + lock type; +-----------------+-----------+---------------+------------+ | iommu set | c | w | c | w | c | w | c | w | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 20697 | 839 |N/A | N/A | N/A | N/A | N/A | N/A | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 21838 | 658.15 |N/A | N/A | N/A | N/A | N/A | N/A | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1553 | 1599.96 |9458|429.97| 5846 | 274.33 | 0 | 0.00 | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1911 | 1678.32 |8335|445.16| 5451 | 244.80 | 0 | 0.00 | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ Signed-off-by: Pei Zhang <pei.zhang@intel.com> Signed-off-by: Colin Xu <colin.xu@intel.com> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
2018-05-19 11:28:54 +07:00
mutex_unlock(&vgpu->vgpu_lock);
}
/**
* intel_vgpu_emulate_mmio_read - emulate MMIO read
* @vgpu: a vGPU
* @pa: guest physical address
* @p_data: data return buffer
* @bytes: access data length
*
* Returns:
* Zero on success, negative error code if failed
*/
int intel_vgpu_emulate_mmio_read(struct intel_vgpu *vgpu, u64 pa,
void *p_data, unsigned int bytes)
{
struct intel_gvt *gvt = vgpu->gvt;
unsigned int offset = 0;
int ret = -EINVAL;
if (vgpu->failsafe) {
failsafe_emulate_mmio_rw(vgpu, pa, p_data, bytes, true);
return 0;
}
drm/i915/gvt: Use vgpu_lock to protect per vgpu access The patch set splits out 2 small locks from the original big gvt lock: - vgpu_lock protects per-vGPU data and logic, especially the vGPU trap emulation path. - sched_lock protects gvt scheudler structure, context schedule logic and vGPU's schedule data. Use vgpu_lock to replace the gvt big lock. By doing this, the mmio read/write trap path, vgpu virtual event emulation and other vgpu related process, would be protected under per vgpu_lock. v9: - Change commit author since the patches are improved a lot compared with original version. Original author: Pei Zhang <pei.zhang@intel.com> - Rebase to latest gvt-staging. v8: - Correct coding and comment style. - Rebase to latest gvt-staging. v7: - Remove gtt_lock since already proteced by gvt_lock and vgpu_lock. - Fix a typo in intel_gvt_deactivate_vgpu, unlock the wrong lock. v6: - Rebase to latest gvt-staging. v5: - Rebase to latest gvt-staging. - intel_vgpu_page_track_handler should use vgpu_lock. v4: - Rebase to latest gvt-staging. - Protect vgpu->active access with vgpu_lock. - Do not wait gpu idle in vgpu_lock. v3: update to latest code base v2: add gvt->lock in function gvt_check_vblank_emulation Performance comparison on Kabylake platform. - Configuration: Host: Ubuntu 16.04. Guest 1 & 2: Ubuntu 16.04. glmark2 score comparison: - Configuration: Host: glxgears. Guests: glmark2. +--------------------------------+-----------------+ | Setup | glmark2 score | +--------------------------------+-----------------+ | unified lock, iommu=on | 58~62 (avg. 60) | +--------------------------------+-----------------+ | unified lock, iommu=igfx_off | 57~61 (avg. 59) | +--------------------------------+-----------------+ | per-logic lock, iommu=on | 60~68 (avg. 64) | +--------------------------------+-----------------+ | per-logic lock, iommu=igfx_off | 61~67 (avg. 64) | +--------------------------------+-----------------+ lock_stat comparison: - Configuration: Stop lock stat immediately after boot up. Boot 2 VM Guests. Run glmark2 in guests. Start perf lock_stat for 20 seconds and stop again. - Legend: c - contentions; w - waittime-avg +------------+-----------------+-----------+---------------+------------+ | | gvt_lock |sched_lock | vgpu_lock | gtt_lock | + lock type; +-----------------+-----------+---------------+------------+ | iommu set | c | w | c | w | c | w | c | w | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 20697 | 839 |N/A | N/A | N/A | N/A | N/A | N/A | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 21838 | 658.15 |N/A | N/A | N/A | N/A | N/A | N/A | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1553 | 1599.96 |9458|429.97| 5846 | 274.33 | 0 | 0.00 | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1911 | 1678.32 |8335|445.16| 5451 | 244.80 | 0 | 0.00 | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ Signed-off-by: Pei Zhang <pei.zhang@intel.com> Signed-off-by: Colin Xu <colin.xu@intel.com> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
2018-05-19 11:28:54 +07:00
mutex_lock(&vgpu->vgpu_lock);
offset = intel_vgpu_gpa_to_mmio_offset(vgpu, pa);
if (WARN_ON(bytes > 8))
goto err;
if (reg_is_gtt(gvt, offset)) {
if (WARN_ON(!IS_ALIGNED(offset, 4) && !IS_ALIGNED(offset, 8)))
goto err;
if (WARN_ON(bytes != 4 && bytes != 8))
goto err;
if (WARN_ON(!reg_is_gtt(gvt, offset + bytes - 1)))
goto err;
ret = intel_vgpu_emulate_ggtt_mmio_read(vgpu, offset,
p_data, bytes);
if (ret)
goto err;
goto out;
}
if (WARN_ON_ONCE(!reg_is_mmio(gvt, offset))) {
ret = intel_gvt_hypervisor_read_gpa(vgpu, pa, p_data, bytes);
goto out;
}
if (WARN_ON(!reg_is_mmio(gvt, offset + bytes - 1)))
goto err;
if (!intel_gvt_mmio_is_unalign(gvt, offset)) {
if (WARN_ON(!IS_ALIGNED(offset, bytes)))
goto err;
}
ret = intel_vgpu_mmio_reg_rw(vgpu, offset, p_data, bytes, true);
if (ret < 0)
goto err;
intel_gvt_mmio_set_accessed(gvt, offset);
ret = 0;
goto out;
err:
gvt_vgpu_err("fail to emulate MMIO read %08x len %d\n",
offset, bytes);
out:
drm/i915/gvt: Use vgpu_lock to protect per vgpu access The patch set splits out 2 small locks from the original big gvt lock: - vgpu_lock protects per-vGPU data and logic, especially the vGPU trap emulation path. - sched_lock protects gvt scheudler structure, context schedule logic and vGPU's schedule data. Use vgpu_lock to replace the gvt big lock. By doing this, the mmio read/write trap path, vgpu virtual event emulation and other vgpu related process, would be protected under per vgpu_lock. v9: - Change commit author since the patches are improved a lot compared with original version. Original author: Pei Zhang <pei.zhang@intel.com> - Rebase to latest gvt-staging. v8: - Correct coding and comment style. - Rebase to latest gvt-staging. v7: - Remove gtt_lock since already proteced by gvt_lock and vgpu_lock. - Fix a typo in intel_gvt_deactivate_vgpu, unlock the wrong lock. v6: - Rebase to latest gvt-staging. v5: - Rebase to latest gvt-staging. - intel_vgpu_page_track_handler should use vgpu_lock. v4: - Rebase to latest gvt-staging. - Protect vgpu->active access with vgpu_lock. - Do not wait gpu idle in vgpu_lock. v3: update to latest code base v2: add gvt->lock in function gvt_check_vblank_emulation Performance comparison on Kabylake platform. - Configuration: Host: Ubuntu 16.04. Guest 1 & 2: Ubuntu 16.04. glmark2 score comparison: - Configuration: Host: glxgears. Guests: glmark2. +--------------------------------+-----------------+ | Setup | glmark2 score | +--------------------------------+-----------------+ | unified lock, iommu=on | 58~62 (avg. 60) | +--------------------------------+-----------------+ | unified lock, iommu=igfx_off | 57~61 (avg. 59) | +--------------------------------+-----------------+ | per-logic lock, iommu=on | 60~68 (avg. 64) | +--------------------------------+-----------------+ | per-logic lock, iommu=igfx_off | 61~67 (avg. 64) | +--------------------------------+-----------------+ lock_stat comparison: - Configuration: Stop lock stat immediately after boot up. Boot 2 VM Guests. Run glmark2 in guests. Start perf lock_stat for 20 seconds and stop again. - Legend: c - contentions; w - waittime-avg +------------+-----------------+-----------+---------------+------------+ | | gvt_lock |sched_lock | vgpu_lock | gtt_lock | + lock type; +-----------------+-----------+---------------+------------+ | iommu set | c | w | c | w | c | w | c | w | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 20697 | 839 |N/A | N/A | N/A | N/A | N/A | N/A | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 21838 | 658.15 |N/A | N/A | N/A | N/A | N/A | N/A | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1553 | 1599.96 |9458|429.97| 5846 | 274.33 | 0 | 0.00 | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1911 | 1678.32 |8335|445.16| 5451 | 244.80 | 0 | 0.00 | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ Signed-off-by: Pei Zhang <pei.zhang@intel.com> Signed-off-by: Colin Xu <colin.xu@intel.com> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
2018-05-19 11:28:54 +07:00
mutex_unlock(&vgpu->vgpu_lock);
return ret;
}
/**
* intel_vgpu_emulate_mmio_write - emulate MMIO write
* @vgpu: a vGPU
* @pa: guest physical address
* @p_data: write data buffer
* @bytes: access data length
*
* Returns:
* Zero on success, negative error code if failed
*/
int intel_vgpu_emulate_mmio_write(struct intel_vgpu *vgpu, u64 pa,
void *p_data, unsigned int bytes)
{
struct intel_gvt *gvt = vgpu->gvt;
unsigned int offset = 0;
int ret = -EINVAL;
if (vgpu->failsafe) {
failsafe_emulate_mmio_rw(vgpu, pa, p_data, bytes, false);
return 0;
}
drm/i915/gvt: Use vgpu_lock to protect per vgpu access The patch set splits out 2 small locks from the original big gvt lock: - vgpu_lock protects per-vGPU data and logic, especially the vGPU trap emulation path. - sched_lock protects gvt scheudler structure, context schedule logic and vGPU's schedule data. Use vgpu_lock to replace the gvt big lock. By doing this, the mmio read/write trap path, vgpu virtual event emulation and other vgpu related process, would be protected under per vgpu_lock. v9: - Change commit author since the patches are improved a lot compared with original version. Original author: Pei Zhang <pei.zhang@intel.com> - Rebase to latest gvt-staging. v8: - Correct coding and comment style. - Rebase to latest gvt-staging. v7: - Remove gtt_lock since already proteced by gvt_lock and vgpu_lock. - Fix a typo in intel_gvt_deactivate_vgpu, unlock the wrong lock. v6: - Rebase to latest gvt-staging. v5: - Rebase to latest gvt-staging. - intel_vgpu_page_track_handler should use vgpu_lock. v4: - Rebase to latest gvt-staging. - Protect vgpu->active access with vgpu_lock. - Do not wait gpu idle in vgpu_lock. v3: update to latest code base v2: add gvt->lock in function gvt_check_vblank_emulation Performance comparison on Kabylake platform. - Configuration: Host: Ubuntu 16.04. Guest 1 & 2: Ubuntu 16.04. glmark2 score comparison: - Configuration: Host: glxgears. Guests: glmark2. +--------------------------------+-----------------+ | Setup | glmark2 score | +--------------------------------+-----------------+ | unified lock, iommu=on | 58~62 (avg. 60) | +--------------------------------+-----------------+ | unified lock, iommu=igfx_off | 57~61 (avg. 59) | +--------------------------------+-----------------+ | per-logic lock, iommu=on | 60~68 (avg. 64) | +--------------------------------+-----------------+ | per-logic lock, iommu=igfx_off | 61~67 (avg. 64) | +--------------------------------+-----------------+ lock_stat comparison: - Configuration: Stop lock stat immediately after boot up. Boot 2 VM Guests. Run glmark2 in guests. Start perf lock_stat for 20 seconds and stop again. - Legend: c - contentions; w - waittime-avg +------------+-----------------+-----------+---------------+------------+ | | gvt_lock |sched_lock | vgpu_lock | gtt_lock | + lock type; +-----------------+-----------+---------------+------------+ | iommu set | c | w | c | w | c | w | c | w | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 20697 | 839 |N/A | N/A | N/A | N/A | N/A | N/A | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 21838 | 658.15 |N/A | N/A | N/A | N/A | N/A | N/A | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1553 | 1599.96 |9458|429.97| 5846 | 274.33 | 0 | 0.00 | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1911 | 1678.32 |8335|445.16| 5451 | 244.80 | 0 | 0.00 | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ Signed-off-by: Pei Zhang <pei.zhang@intel.com> Signed-off-by: Colin Xu <colin.xu@intel.com> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
2018-05-19 11:28:54 +07:00
mutex_lock(&vgpu->vgpu_lock);
offset = intel_vgpu_gpa_to_mmio_offset(vgpu, pa);
if (WARN_ON(bytes > 8))
goto err;
if (reg_is_gtt(gvt, offset)) {
if (WARN_ON(!IS_ALIGNED(offset, 4) && !IS_ALIGNED(offset, 8)))
goto err;
if (WARN_ON(bytes != 4 && bytes != 8))
goto err;
if (WARN_ON(!reg_is_gtt(gvt, offset + bytes - 1)))
goto err;
ret = intel_vgpu_emulate_ggtt_mmio_write(vgpu, offset,
p_data, bytes);
if (ret)
goto err;
goto out;
}
if (WARN_ON_ONCE(!reg_is_mmio(gvt, offset))) {
ret = intel_gvt_hypervisor_write_gpa(vgpu, pa, p_data, bytes);
goto out;
}
ret = intel_vgpu_mmio_reg_rw(vgpu, offset, p_data, bytes, false);
if (ret < 0)
goto err;
intel_gvt_mmio_set_accessed(gvt, offset);
ret = 0;
goto out;
err:
gvt_vgpu_err("fail to emulate MMIO write %08x len %d\n", offset,
bytes);
out:
drm/i915/gvt: Use vgpu_lock to protect per vgpu access The patch set splits out 2 small locks from the original big gvt lock: - vgpu_lock protects per-vGPU data and logic, especially the vGPU trap emulation path. - sched_lock protects gvt scheudler structure, context schedule logic and vGPU's schedule data. Use vgpu_lock to replace the gvt big lock. By doing this, the mmio read/write trap path, vgpu virtual event emulation and other vgpu related process, would be protected under per vgpu_lock. v9: - Change commit author since the patches are improved a lot compared with original version. Original author: Pei Zhang <pei.zhang@intel.com> - Rebase to latest gvt-staging. v8: - Correct coding and comment style. - Rebase to latest gvt-staging. v7: - Remove gtt_lock since already proteced by gvt_lock and vgpu_lock. - Fix a typo in intel_gvt_deactivate_vgpu, unlock the wrong lock. v6: - Rebase to latest gvt-staging. v5: - Rebase to latest gvt-staging. - intel_vgpu_page_track_handler should use vgpu_lock. v4: - Rebase to latest gvt-staging. - Protect vgpu->active access with vgpu_lock. - Do not wait gpu idle in vgpu_lock. v3: update to latest code base v2: add gvt->lock in function gvt_check_vblank_emulation Performance comparison on Kabylake platform. - Configuration: Host: Ubuntu 16.04. Guest 1 & 2: Ubuntu 16.04. glmark2 score comparison: - Configuration: Host: glxgears. Guests: glmark2. +--------------------------------+-----------------+ | Setup | glmark2 score | +--------------------------------+-----------------+ | unified lock, iommu=on | 58~62 (avg. 60) | +--------------------------------+-----------------+ | unified lock, iommu=igfx_off | 57~61 (avg. 59) | +--------------------------------+-----------------+ | per-logic lock, iommu=on | 60~68 (avg. 64) | +--------------------------------+-----------------+ | per-logic lock, iommu=igfx_off | 61~67 (avg. 64) | +--------------------------------+-----------------+ lock_stat comparison: - Configuration: Stop lock stat immediately after boot up. Boot 2 VM Guests. Run glmark2 in guests. Start perf lock_stat for 20 seconds and stop again. - Legend: c - contentions; w - waittime-avg +------------+-----------------+-----------+---------------+------------+ | | gvt_lock |sched_lock | vgpu_lock | gtt_lock | + lock type; +-----------------+-----------+---------------+------------+ | iommu set | c | w | c | w | c | w | c | w | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 20697 | 839 |N/A | N/A | N/A | N/A | N/A | N/A | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | unified; | 21838 | 658.15 |N/A | N/A | N/A | N/A | N/A | N/A | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1553 | 1599.96 |9458|429.97| 5846 | 274.33 | 0 | 0.00 | | on | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ | per-logic; | 1911 | 1678.32 |8335|445.16| 5451 | 244.80 | 0 | 0.00 | | igfx_off | | | | | | | | | +------------+-------+---------+----+------+------+--------+-----+------+ Signed-off-by: Pei Zhang <pei.zhang@intel.com> Signed-off-by: Colin Xu <colin.xu@intel.com> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
2018-05-19 11:28:54 +07:00
mutex_unlock(&vgpu->vgpu_lock);
return ret;
}
/**
* intel_vgpu_reset_mmio - reset virtual MMIO space
* @vgpu: a vGPU
* @dmlr: whether this is device model level reset
*/
void intel_vgpu_reset_mmio(struct intel_vgpu *vgpu, bool dmlr)
{
struct intel_gvt *gvt = vgpu->gvt;
const struct intel_gvt_device_info *info = &gvt->device_info;
void *mmio = gvt->firmware.mmio;
if (dmlr) {
memcpy(vgpu->mmio.vreg, mmio, info->mmio_size);
vgpu_vreg_t(vgpu, GEN6_GT_THREAD_STATUS_REG) = 0;
/* set the bit 0:2(Core C-State ) to C0 */
vgpu_vreg_t(vgpu, GEN6_GT_CORE_STATUS) = 0;
drm/i915/gvt: Init PHY related registers for BXT Recent patch fixed the call trace "ERROR Port B enabled but PHY powered down? (PHY_CTL 00000000)". but introduced another similar call trace shown as: "ERROR Port C enabled but PHY powered down? (PHY_CTL 00000200)". The call trace will appear when host and guest enabled different ports, i.e. host using PORT C or neither PORT is enabled, while guest is always using PORT B as simulated by gvt. The issue is actually covered previously before the commit and reverals now when the commit do the right thing. On BXT, some PHY registers are initialized by vbios, before i915 loaded. Later i915 will re-program some, or skip some based on the implementation. The initialized mmio for guest i915 is done by gvt, based on the snapshot taken from host. If host and guest have different PORT enabled, some DPIO PHY mmios that gvt initialized for guest i915 will not match the simualted monitor for guest, which leads to guest i915 print the calltrace when it's trying to enable PHY and PORT. The solution is to init these DPIO PHY registers to default value, then guest i915 will program them to reasonable value based on the default powerwell table and enabled PORT. Together with the old patch, all similar call trace in guest kernel on BXT can be resolved. v2: Move PHY register init to intel_vgpu_reset_mmio (Min) v3: Do not delete empty line in issue fix patch. (zhenyu) Fixes: c8ab5ac30ccc ("drm/i915/gvt: Make correct handling to vreg BXT_PHY_CTL_FAMILY") Reviewed-by: He, Min <min.he@intel.com> Signed-off-by: Colin Xu <colin.xu@intel.com> Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
2018-09-17 11:19:03 +07:00
if (IS_BROXTON(vgpu->gvt->dev_priv)) {
vgpu_vreg_t(vgpu, BXT_P_CR_GT_DISP_PWRON) &=
~(BIT(0) | BIT(1));
vgpu_vreg_t(vgpu, BXT_PORT_CL1CM_DW0(DPIO_PHY0)) &=
~PHY_POWER_GOOD;
vgpu_vreg_t(vgpu, BXT_PORT_CL1CM_DW0(DPIO_PHY1)) &=
~PHY_POWER_GOOD;
vgpu_vreg_t(vgpu, BXT_PHY_CTL_FAMILY(DPIO_PHY0)) &=
~BIT(30);
vgpu_vreg_t(vgpu, BXT_PHY_CTL_FAMILY(DPIO_PHY1)) &=
~BIT(30);
vgpu_vreg_t(vgpu, BXT_PHY_CTL(PORT_A)) &=
~BXT_PHY_LANE_ENABLED;
vgpu_vreg_t(vgpu, BXT_PHY_CTL(PORT_A)) |=
BXT_PHY_CMNLANE_POWERDOWN_ACK |
BXT_PHY_LANE_POWERDOWN_ACK;
vgpu_vreg_t(vgpu, BXT_PHY_CTL(PORT_B)) &=
~BXT_PHY_LANE_ENABLED;
vgpu_vreg_t(vgpu, BXT_PHY_CTL(PORT_B)) |=
BXT_PHY_CMNLANE_POWERDOWN_ACK |
BXT_PHY_LANE_POWERDOWN_ACK;
vgpu_vreg_t(vgpu, BXT_PHY_CTL(PORT_C)) &=
~BXT_PHY_LANE_ENABLED;
vgpu_vreg_t(vgpu, BXT_PHY_CTL(PORT_C)) |=
BXT_PHY_CMNLANE_POWERDOWN_ACK |
BXT_PHY_LANE_POWERDOWN_ACK;
}
} else {
#define GVT_GEN8_MMIO_RESET_OFFSET (0x44200)
/* only reset the engine related, so starting with 0x44200
* interrupt include DE,display mmio related will not be
* touched
*/
memcpy(vgpu->mmio.vreg, mmio, GVT_GEN8_MMIO_RESET_OFFSET);
}
}
/**
* intel_vgpu_init_mmio - init MMIO space
* @vgpu: a vGPU
*
* Returns:
* Zero on success, negative error code if failed
*/
int intel_vgpu_init_mmio(struct intel_vgpu *vgpu)
{
const struct intel_gvt_device_info *info = &vgpu->gvt->device_info;
vgpu->mmio.vreg = vzalloc(info->mmio_size);
if (!vgpu->mmio.vreg)
return -ENOMEM;
intel_vgpu_reset_mmio(vgpu, true);
return 0;
}
/**
* intel_vgpu_clean_mmio - clean MMIO space
* @vgpu: a vGPU
*
*/
void intel_vgpu_clean_mmio(struct intel_vgpu *vgpu)
{
vfree(vgpu->mmio.vreg);
vgpu->mmio.vreg = NULL;
}