diff options
| author | Linus Torvalds | 2015-06-26 22:18:51 +0200 |
|---|---|---|
| committer | Linus Torvalds | 2015-06-26 22:18:51 +0200 |
| commit | 099bfbfc7fbbe22356c02f0caf709ac32e1126ea (patch) | |
| tree | c2dfe2f9445255d866e9203cff9e9f865ef93513 /drivers/gpu/drm/amd/amdkfd | |
| parent | Merge tag 'dm-4.2-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/dev... (diff) | |
| parent | drm/nouveau: Pause between setting gpu to D3hot and cutting the power (diff) | |
| download | kernel-qcow2-linux-099bfbfc7fbbe22356c02f0caf709ac32e1126ea.tar.gz kernel-qcow2-linux-099bfbfc7fbbe22356c02f0caf709ac32e1126ea.tar.xz kernel-qcow2-linux-099bfbfc7fbbe22356c02f0caf709ac32e1126ea.zip | |
Merge branch 'drm-next' of git://people.freedesktop.org/~airlied/linux
Pull drm updates from Dave Airlie:
"This is the main drm pull request for v4.2.
I've one other new driver from freescale on my radar, it's been posted
and reviewed, I'd just like to get someone to give it a last look, so
maybe I'll send it or maybe I'll leave it.
There is no major nouveau changes in here, Ben was working on
something big, and we agreed it was a bit late, there wasn't anything
else he considered urgent to merge.
There might be another msm pull for some bits that are waiting on
arm-soc, I'll see how we time it.
This touches some "of" stuff, acks are in place except for the fixes
to the build in various configs,t hat I just applied.
Summary:
New drivers:
- virtio-gpu:
KMS only pieces of driver for virtio-gpu in qemu.
This is just the first part of this driver, enough to run
unaccelerated userspace on. As qemu merges more we'll start
adding the 3D features for the virgl 3d work.
- amdgpu:
a new driver from AMD to driver their newer GPUs. (VI+)
It contains a new cleaner userspace API, and is a clean
break from radeon moving forward, that AMD are going to
concentrate on. It also contains a set of register headers
auto generated from AMD internal database.
core:
- atomic modesetting API completed, enabled by default now.
- Add support for mode_id blob to atomic ioctl to complete interface.
- bunch of Displayport MST fixes
- lots of misc fixes.
panel:
- new simple panels
- fix some long-standing build issues with bridge drivers
radeon:
- VCE1 support
- add a GPU reset counter for userspace
- lots of fixes.
amdkfd:
- H/W debugger support module
- static user-mode queues
- support killing all the waves when a process terminates
- use standard DECLARE_BITMAP
i915:
- Add Broxton support
- S3, rotation support for Skylake
- RPS booting tuning
- CPT modeset sequence fixes
- ns2501 dither support
- enable cmd parser on haswell
- cdclk handling fixes
- gen8 dynamic pte allocation
- lots of atomic conversion work
exynos:
- Add atomic modesetting support
- Add iommu support
- Consolidate drm driver initialization
- and MIC, DECON and MIPI-DSI support for exynos5433
omapdrm:
- atomic modesetting support (fixes lots of things in rewrite)
tegra:
- DP aux transaction fixes
- iommu support fix
msm:
- adreno a306 support
- various dsi bits
- various 64-bit fixes
- NV12MT support
rcar-du:
- atomic and misc fixes
sti:
- fix HDMI timing complaince
tilcdc:
- use drm component API to access tda998x driver
- fix module unloading
qxl:
- stability fixes"
* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (872 commits)
drm/nouveau: Pause between setting gpu to D3hot and cutting the power
drm/dp/mst: close deadlock in connector destruction.
drm: Always enable atomic API
drm/vgem: Set unique to "vgem"
of: fix a build error to of_graph_get_endpoint_by_regs function
drm/dp/mst: take lock around looking up the branch device on hpd irq
drm/dp/mst: make sure mst_primary mstb is valid in work function
of: add EXPORT_SYMBOL for of_graph_get_endpoint_by_regs
ARM: dts: rename the clock of MIPI DSI 'pll_clk' to 'sclk_mipi'
drm/atomic: Don't set crtc_state->enable manually
drm/exynos: dsi: do not set TE GPIO direction by input
drm/exynos: dsi: add support for MIC driver as a bridge
drm/exynos: dsi: add support for Exynos5433
drm/exynos: dsi: make use of array for clock access
drm/exynos: dsi: make use of driver data for static values
drm/exynos: dsi: add macros for register access
drm/exynos: dsi: rename pll_clk to sclk_clk
drm/exynos: mic: add MIC driver
of: add helper for getting endpoint node of specific identifiers
drm/exynos: add Exynos5433 decon driver
...
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
26 files changed, 3969 insertions, 228 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile b/drivers/gpu/drm/amd/amdkfd/Makefile index 0f4960148126..28551153ec6d 100644 --- a/drivers/gpu/drm/amd/amdkfd/Makefile +++ b/drivers/gpu/drm/amd/amdkfd/Makefile @@ -12,5 +12,7 @@ amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \ kfd_kernel_queue_vi.o kfd_packet_manager.o \ kfd_process_queue_manager.o kfd_device_queue_manager.o \ kfd_device_queue_manager_cik.o kfd_device_queue_manager_vi.o \ + kfd_interrupt.o kfd_events.o cik_event_interrupt.o \ + kfd_dbgdev.o kfd_dbgmgr.o obj-$(CONFIG_HSA_AMD) += amdkfd.o diff --git a/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c new file mode 100644 index 000000000000..211fc48697fa --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c @@ -0,0 +1,66 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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 "kfd_priv.h" +#include "kfd_events.h" +#include "cik_int.h" + +static bool cik_event_interrupt_isr(struct kfd_dev *dev, + const uint32_t *ih_ring_entry) +{ + unsigned int pasid; + const struct cik_ih_ring_entry *ihre = + (const struct cik_ih_ring_entry *)ih_ring_entry; + + pasid = (ihre->ring_id & 0xffff0000) >> 16; + + /* Do not process in ISR, just request it to be forwarded to WQ. */ + return (pasid != 0) && + (ihre->source_id == CIK_INTSRC_CP_END_OF_PIPE || + ihre->source_id == CIK_INTSRC_SQ_INTERRUPT_MSG || + ihre->source_id == CIK_INTSRC_CP_BAD_OPCODE); +} + +static void cik_event_interrupt_wq(struct kfd_dev *dev, + const uint32_t *ih_ring_entry) +{ + unsigned int pasid; + const struct cik_ih_ring_entry *ihre = + (const struct cik_ih_ring_entry *)ih_ring_entry; + + pasid = (ihre->ring_id & 0xffff0000) >> 16; + + if (pasid == 0) + return; + + if (ihre->source_id == CIK_INTSRC_CP_END_OF_PIPE) + kfd_signal_event_interrupt(pasid, 0, 0); + else if (ihre->source_id == CIK_INTSRC_SQ_INTERRUPT_MSG) + kfd_signal_event_interrupt(pasid, ihre->data & 0xFF, 8); + else if (ihre->source_id == CIK_INTSRC_CP_BAD_OPCODE) + kfd_signal_hw_exception_event(pasid); +} + +const struct kfd_event_interrupt_class event_interrupt_class_cik = { + .interrupt_isr = cik_event_interrupt_isr, + .interrupt_wq = cik_event_interrupt_wq, +}; diff --git a/drivers/gpu/drm/amd/amdkfd/cik_int.h b/drivers/gpu/drm/amd/amdkfd/cik_int.h new file mode 100644 index 000000000000..79a16d24c1b8 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/cik_int.h @@ -0,0 +1,41 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + */ + +#ifndef HSA_RADEON_CIK_INT_H_INCLUDED +#define HSA_RADEON_CIK_INT_H_INCLUDED + +#include <linux/types.h> + +struct cik_ih_ring_entry { + uint32_t source_id; + uint32_t data; + uint32_t ring_id; + uint32_t reserved; +}; + +#define CIK_INTSRC_DEQUEUE_COMPLETE 0xC6 +#define CIK_INTSRC_CP_END_OF_PIPE 0xB5 +#define CIK_INTSRC_CP_BAD_OPCODE 0xB7 +#define CIK_INTSRC_SQ_INTERRUPT_MSG 0xEF + +#endif + diff --git a/drivers/gpu/drm/amd/amdkfd/cik_regs.h b/drivers/gpu/drm/amd/amdkfd/cik_regs.h index 01ff332fabd4..183be5b8414f 100644 --- a/drivers/gpu/drm/amd/amdkfd/cik_regs.h +++ b/drivers/gpu/drm/amd/amdkfd/cik_regs.h @@ -23,33 +23,11 @@ #ifndef CIK_REGS_H #define CIK_REGS_H -#define IH_VMID_0_LUT 0x3D40u - -#define BIF_DOORBELL_CNTL 0x530Cu - -#define SRBM_GFX_CNTL 0xE44 -#define PIPEID(x) ((x) << 0) -#define MEID(x) ((x) << 2) -#define VMID(x) ((x) << 4) -#define QUEUEID(x) ((x) << 8) - -#define SQ_CONFIG 0x8C00 - -#define SH_MEM_BASES 0x8C28 /* if PTR32, these are the bases for scratch and lds */ #define PRIVATE_BASE(x) ((x) << 0) /* scratch */ #define SHARED_BASE(x) ((x) << 16) /* LDS */ -#define SH_MEM_APE1_BASE 0x8C2C -/* if PTR32, this is the base location of GPUVM */ -#define SH_MEM_APE1_LIMIT 0x8C30 -/* if PTR32, this is the upper limit of GPUVM */ -#define SH_MEM_CONFIG 0x8C34 #define PTR32 (1 << 0) -#define PRIVATE_ATC (1 << 1) #define ALIGNMENT_MODE(x) ((x) << 2) -#define SH_MEM_ALIGNMENT_MODE_DWORD 0 -#define SH_MEM_ALIGNMENT_MODE_DWORD_STRICT 1 -#define SH_MEM_ALIGNMENT_MODE_STRICT 2 #define SH_MEM_ALIGNMENT_MODE_UNALIGNED 3 #define DEFAULT_MTYPE(x) ((x) << 4) #define APE1_MTYPE(x) ((x) << 7) @@ -58,137 +36,34 @@ #define MTYPE_CACHED 0 #define MTYPE_NONCACHED 3 - -#define SH_STATIC_MEM_CONFIG 0x9604u - -#define TC_CFG_L1_LOAD_POLICY0 0xAC68 -#define TC_CFG_L1_LOAD_POLICY1 0xAC6C -#define TC_CFG_L1_STORE_POLICY 0xAC70 -#define TC_CFG_L2_LOAD_POLICY0 0xAC74 -#define TC_CFG_L2_LOAD_POLICY1 0xAC78 -#define TC_CFG_L2_STORE_POLICY0 0xAC7C -#define TC_CFG_L2_STORE_POLICY1 0xAC80 -#define TC_CFG_L2_ATOMIC_POLICY 0xAC84 -#define TC_CFG_L1_VOLATILE 0xAC88 -#define TC_CFG_L2_VOLATILE 0xAC8C - -#define CP_PQ_WPTR_POLL_CNTL 0xC20C -#define WPTR_POLL_EN (1 << 31) - -#define CPC_INT_CNTL 0xC2D0 -#define CP_ME1_PIPE0_INT_CNTL 0xC214 -#define CP_ME1_PIPE1_INT_CNTL 0xC218 -#define CP_ME1_PIPE2_INT_CNTL 0xC21C -#define CP_ME1_PIPE3_INT_CNTL 0xC220 -#define CP_ME2_PIPE0_INT_CNTL 0xC224 -#define CP_ME2_PIPE1_INT_CNTL 0xC228 -#define CP_ME2_PIPE2_INT_CNTL 0xC22C -#define CP_ME2_PIPE3_INT_CNTL 0xC230 -#define DEQUEUE_REQUEST_INT_ENABLE (1 << 13) -#define WRM_POLL_TIMEOUT_INT_ENABLE (1 << 17) -#define PRIV_REG_INT_ENABLE (1 << 23) -#define TIME_STAMP_INT_ENABLE (1 << 26) -#define GENERIC2_INT_ENABLE (1 << 29) -#define GENERIC1_INT_ENABLE (1 << 30) -#define GENERIC0_INT_ENABLE (1 << 31) -#define CP_ME1_PIPE0_INT_STATUS 0xC214 -#define CP_ME1_PIPE1_INT_STATUS 0xC218 -#define CP_ME1_PIPE2_INT_STATUS 0xC21C -#define CP_ME1_PIPE3_INT_STATUS 0xC220 -#define CP_ME2_PIPE0_INT_STATUS 0xC224 -#define CP_ME2_PIPE1_INT_STATUS 0xC228 -#define CP_ME2_PIPE2_INT_STATUS 0xC22C -#define CP_ME2_PIPE3_INT_STATUS 0xC230 -#define DEQUEUE_REQUEST_INT_STATUS (1 << 13) -#define WRM_POLL_TIMEOUT_INT_STATUS (1 << 17) -#define PRIV_REG_INT_STATUS (1 << 23) -#define TIME_STAMP_INT_STATUS (1 << 26) -#define GENERIC2_INT_STATUS (1 << 29) -#define GENERIC1_INT_STATUS (1 << 30) -#define GENERIC0_INT_STATUS (1 << 31) - -#define CP_HPD_EOP_BASE_ADDR 0xC904 -#define CP_HPD_EOP_BASE_ADDR_HI 0xC908 -#define CP_HPD_EOP_VMID 0xC90C -#define CP_HPD_EOP_CONTROL 0xC910 -#define EOP_SIZE(x) ((x) << 0) -#define EOP_SIZE_MASK (0x3f << 0) -#define CP_MQD_BASE_ADDR 0xC914 -#define CP_MQD_BASE_ADDR_HI 0xC918 -#define CP_HQD_ACTIVE 0xC91C -#define CP_HQD_VMID 0xC920 - -#define CP_HQD_PERSISTENT_STATE 0xC924u #define DEFAULT_CP_HQD_PERSISTENT_STATE (0x33U << 8) #define PRELOAD_REQ (1 << 0) -#define CP_HQD_PIPE_PRIORITY 0xC928u -#define CP_HQD_QUEUE_PRIORITY 0xC92Cu -#define CP_HQD_QUANTUM 0xC930u +#define MQD_CONTROL_PRIV_STATE_EN (1U << 8) + +#define DEFAULT_MIN_IB_AVAIL_SIZE (3U << 20) + +#define IB_ATC_EN (1U << 23) + #define QUANTUM_EN 1U #define QUANTUM_SCALE_1MS (1U << 4) #define QUANTUM_DURATION(x) ((x) << 8) -#define CP_HQD_PQ_BASE 0xC934 -#define CP_HQD_PQ_BASE_HI 0xC938 -#define CP_HQD_PQ_RPTR 0xC93C -#define CP_HQD_PQ_RPTR_REPORT_ADDR 0xC940 -#define CP_HQD_PQ_RPTR_REPORT_ADDR_HI 0xC944 -#define CP_HQD_PQ_WPTR_POLL_ADDR 0xC948 -#define CP_HQD_PQ_WPTR_POLL_ADDR_HI 0xC94C -#define CP_HQD_PQ_DOORBELL_CONTROL 0xC950 -#define DOORBELL_OFFSET(x) ((x) << 2) -#define DOORBELL_OFFSET_MASK (0x1fffff << 2) -#define DOORBELL_SOURCE (1 << 28) -#define DOORBELL_SCHD_HIT (1 << 29) -#define DOORBELL_EN (1 << 30) -#define DOORBELL_HIT (1 << 31) -#define CP_HQD_PQ_WPTR 0xC954 -#define CP_HQD_PQ_CONTROL 0xC958 -#define QUEUE_SIZE(x) ((x) << 0) -#define QUEUE_SIZE_MASK (0x3f << 0) #define RPTR_BLOCK_SIZE(x) ((x) << 8) -#define RPTR_BLOCK_SIZE_MASK (0x3f << 8) #define MIN_AVAIL_SIZE(x) ((x) << 20) -#define PQ_ATC_EN (1 << 23) -#define PQ_VOLATILE (1 << 26) -#define NO_UPDATE_RPTR (1 << 27) -#define UNORD_DISPATCH (1 << 28) -#define ROQ_PQ_IB_FLIP (1 << 29) -#define PRIV_STATE (1 << 30) -#define KMD_QUEUE (1 << 31) - #define DEFAULT_RPTR_BLOCK_SIZE RPTR_BLOCK_SIZE(5) #define DEFAULT_MIN_AVAIL_SIZE MIN_AVAIL_SIZE(3) -#define CP_HQD_IB_BASE_ADDR 0xC95Cu -#define CP_HQD_IB_BASE_ADDR_HI 0xC960u -#define CP_HQD_IB_RPTR 0xC964u -#define CP_HQD_IB_CONTROL 0xC968u -#define IB_ATC_EN (1U << 23) -#define DEFAULT_MIN_IB_AVAIL_SIZE (3U << 20) - -#define AQL_ENABLE 1 - -#define CP_HQD_DEQUEUE_REQUEST 0xC974 -#define DEQUEUE_REQUEST_DRAIN 1 -#define DEQUEUE_REQUEST_RESET 2 -#define DEQUEUE_INT (1U << 8) +#define PQ_ATC_EN (1 << 23) +#define NO_UPDATE_RPTR (1 << 27) -#define CP_HQD_SEMA_CMD 0xC97Cu -#define CP_HQD_MSG_TYPE 0xC980u -#define CP_HQD_ATOMIC0_PREOP_LO 0xC984u -#define CP_HQD_ATOMIC0_PREOP_HI 0xC988u -#define CP_HQD_ATOMIC1_PREOP_LO 0xC98Cu -#define CP_HQD_ATOMIC1_PREOP_HI 0xC990u -#define CP_HQD_HQ_SCHEDULER0 0xC994u -#define CP_HQD_HQ_SCHEDULER1 0xC998u +#define DOORBELL_OFFSET(x) ((x) << 2) +#define DOORBELL_EN (1 << 30) +#define PRIV_STATE (1 << 30) +#define KMD_QUEUE (1 << 31) -#define CP_MQD_CONTROL 0xC99C -#define MQD_VMID(x) ((x) << 0) -#define MQD_VMID_MASK (0xf << 0) -#define MQD_CONTROL_PRIV_STATE_EN (1U << 8) +#define AQL_ENABLE 1 #define SDMA_RB_VMID(x) (x << 24) #define SDMA_RB_ENABLE (1 << 0) @@ -202,33 +77,7 @@ #define SDMA_VA_SHARED_BASE(x) (x << 8) #define GRBM_GFX_INDEX 0x30800 -#define INSTANCE_INDEX(x) ((x) << 0) -#define SH_INDEX(x) ((x) << 8) -#define SE_INDEX(x) ((x) << 16) -#define SH_BROADCAST_WRITES (1 << 29) -#define INSTANCE_BROADCAST_WRITES (1 << 30) -#define SE_BROADCAST_WRITES (1 << 31) - -#define SQC_CACHES 0x30d20 -#define SQC_POLICY 0x8C38u -#define SQC_VOLATILE 0x8C3Cu -#define CP_PERFMON_CNTL 0x36020 - -#define ATC_VMID0_PASID_MAPPING 0x339Cu -#define ATC_VMID_PASID_MAPPING_UPDATE_STATUS 0x3398u #define ATC_VMID_PASID_MAPPING_VALID (1U << 31) -#define ATC_VM_APERTURE0_CNTL 0x3310u -#define ATS_ACCESS_MODE_NEVER 0 -#define ATS_ACCESS_MODE_ALWAYS 1 - -#define ATC_VM_APERTURE0_CNTL2 0x3318u -#define ATC_VM_APERTURE0_HIGH_ADDR 0x3308u -#define ATC_VM_APERTURE0_LOW_ADDR 0x3300u -#define ATC_VM_APERTURE1_CNTL 0x3314u -#define ATC_VM_APERTURE1_CNTL2 0x331Cu -#define ATC_VM_APERTURE1_HIGH_ADDR 0x330Cu -#define ATC_VM_APERTURE1_LOW_ADDR 0x3304u - #endif diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 19a4fba46e4e..c991973019d0 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -35,6 +35,7 @@ #include <asm/processor.h> #include "kfd_priv.h" #include "kfd_device_queue_manager.h" +#include "kfd_dbgmgr.h" static long kfd_ioctl(struct file *, unsigned int, unsigned long); static int kfd_open(struct inode *, struct file *); @@ -289,8 +290,10 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, args->queue_id = queue_id; + /* Return gpu_id as doorbell offset for mmap usage */ - args->doorbell_offset = args->gpu_id << PAGE_SHIFT; + args->doorbell_offset = (KFD_MMAP_DOORBELL_MASK | args->gpu_id); + args->doorbell_offset <<= PAGE_SHIFT; mutex_unlock(&p->mutex); @@ -430,6 +433,301 @@ out: return err; } +static int kfd_ioctl_dbg_register(struct file *filep, + struct kfd_process *p, void *data) +{ + struct kfd_ioctl_dbg_register_args *args = data; + struct kfd_dev *dev; + struct kfd_dbgmgr *dbgmgr_ptr; + struct kfd_process_device *pdd; + bool create_ok; + long status = 0; + + dev = kfd_device_by_id(args->gpu_id); + if (dev == NULL) + return -EINVAL; + + if (dev->device_info->asic_family == CHIP_CARRIZO) { + pr_debug("kfd_ioctl_dbg_register not supported on CZ\n"); + return -EINVAL; + } + + mutex_lock(kfd_get_dbgmgr_mutex()); + mutex_lock(&p->mutex); + + /* + * make sure that we have pdd, if this the first queue created for + * this process + */ + pdd = kfd_bind_process_to_device(dev, p); + if (IS_ERR(pdd)) { + mutex_unlock(&p->mutex); + mutex_unlock(kfd_get_dbgmgr_mutex()); + return PTR_ERR(pdd); + } + + if (dev->dbgmgr == NULL) { + /* In case of a legal call, we have no dbgmgr yet */ + create_ok = kfd_dbgmgr_create(&dbgmgr_ptr, dev); + if (create_ok) { + status = kfd_dbgmgr_register(dbgmgr_ptr, p); + if (status != 0) + kfd_dbgmgr_destroy(dbgmgr_ptr); + else + dev->dbgmgr = dbgmgr_ptr; + } + } else { + pr_debug("debugger already registered\n"); + status = -EINVAL; + } + + mutex_unlock(&p->mutex); + mutex_unlock(kfd_get_dbgmgr_mutex()); + + return status; +} + +static int kfd_ioctl_dbg_unrgesiter(struct file *filep, + struct kfd_process *p, void *data) +{ + struct kfd_ioctl_dbg_unregister_args *args = data; + struct kfd_dev *dev; + long status; + + dev = kfd_device_by_id(args->gpu_id); + if (dev == NULL) + return -EINVAL; + + if (dev->device_info->asic_family == CHIP_CARRIZO) { + pr_debug("kfd_ioctl_dbg_unrgesiter not supported on CZ\n"); + return -EINVAL; + } + + mutex_lock(kfd_get_dbgmgr_mutex()); + + status = kfd_dbgmgr_unregister(dev->dbgmgr, p); + if (status == 0) { + kfd_dbgmgr_destroy(dev->dbgmgr); + dev->dbgmgr = NULL; + } + + mutex_unlock(kfd_get_dbgmgr_mutex()); + + return status; +} + +/* + * Parse and generate variable size data structure for address watch. + * Total size of the buffer and # watch points is limited in order + * to prevent kernel abuse. (no bearing to the much smaller HW limitation + * which is enforced by dbgdev module) + * please also note that the watch address itself are not "copied from user", + * since it be set into the HW in user mode values. + * + */ +static int kfd_ioctl_dbg_address_watch(struct file *filep, + struct kfd_process *p, void *data) +{ + struct kfd_ioctl_dbg_address_watch_args *args = data; + struct kfd_dev *dev; + struct dbg_address_watch_info aw_info; + unsigned char *args_buff; + long status; + void __user *cmd_from_user; + uint64_t watch_mask_value = 0; + unsigned int args_idx = 0; + + memset((void *) &aw_info, 0, sizeof(struct dbg_address_watch_info)); + + dev = kfd_device_by_id(args->gpu_id); + if (dev == NULL) + return -EINVAL; + + if (dev->device_info->asic_family == CHIP_CARRIZO) { + pr_debug("kfd_ioctl_dbg_wave_control not supported on CZ\n"); + return -EINVAL; + } + + cmd_from_user = (void __user *) args->content_ptr; + + /* Validate arguments */ + + if ((args->buf_size_in_bytes > MAX_ALLOWED_AW_BUFF_SIZE) || + (args->buf_size_in_bytes <= sizeof(*args) + sizeof(int) * 2) || + (cmd_from_user == NULL)) + return -EINVAL; + + /* this is the actual buffer to work with */ + + args_buff = kmalloc(args->buf_size_in_bytes - + sizeof(*args), GFP_KERNEL); + if (args_buff == NULL) + return -ENOMEM; + + status = copy_from_user(args_buff, cmd_from_user, + args->buf_size_in_bytes - sizeof(*args)); + + if (status != 0) { + pr_debug("Failed to copy address watch user data\n"); + kfree(args_buff); + return -EINVAL; + } + + aw_info.process = p; + + aw_info.num_watch_points = *((uint32_t *)(&args_buff[args_idx])); + args_idx += sizeof(aw_info.num_watch_points); + + aw_info.watch_mode = (enum HSA_DBG_WATCH_MODE *) &args_buff[args_idx]; + args_idx += sizeof(enum HSA_DBG_WATCH_MODE) * aw_info.num_watch_points; + + /* + * set watch address base pointer to point on the array base + * within args_buff + */ + aw_info.watch_address = (uint64_t *) &args_buff[args_idx]; + + /* skip over the addresses buffer */ + args_idx += sizeof(aw_info.watch_address) * aw_info.num_watch_points; + + if (args_idx >= args->buf_size_in_bytes - sizeof(*args)) { + kfree(args_buff); + return -EINVAL; + } + + watch_mask_value = (uint64_t) args_buff[args_idx]; + + if (watch_mask_value > 0) { + /* + * There is an array of masks. + * set watch mask base pointer to point on the array base + * within args_buff + */ + aw_info.watch_mask = (uint64_t *) &args_buff[args_idx]; + + /* skip over the masks buffer */ + args_idx += sizeof(aw_info.watch_mask) * + aw_info.num_watch_points; + } else { + /* just the NULL mask, set to NULL and skip over it */ + aw_info.watch_mask = NULL; + args_idx += sizeof(aw_info.watch_mask); + } + + if (args_idx >= args->buf_size_in_bytes - sizeof(args)) { + kfree(args_buff); + return -EINVAL; + } + + /* Currently HSA Event is not supported for DBG */ + aw_info.watch_event = NULL; + + mutex_lock(kfd_get_dbgmgr_mutex()); + + status = kfd_dbgmgr_address_watch(dev->dbgmgr, &aw_info); + + mutex_unlock(kfd_get_dbgmgr_mutex()); + + kfree(args_buff); + + return status; +} + +/* Parse and generate fixed size data structure for wave control */ +static int kfd_ioctl_dbg_wave_control(struct file *filep, + struct kfd_process *p, void *data) +{ + struct kfd_ioctl_dbg_wave_control_args *args = data; + struct kfd_dev *dev; + struct dbg_wave_control_info wac_info; + unsigned char *args_buff; + uint32_t computed_buff_size; + long status; + void __user *cmd_from_user; + unsigned int args_idx = 0; + + memset((void *) &wac_info, 0, sizeof(struct dbg_wave_control_info)); + + /* we use compact form, independent of the packing attribute value */ + computed_buff_size = sizeof(*args) + + sizeof(wac_info.mode) + + sizeof(wac_info.operand) + + sizeof(wac_info.dbgWave_msg.DbgWaveMsg) + + sizeof(wac_info.dbgWave_msg.MemoryVA) + + sizeof(wac_info.trapId); + + dev = kfd_device_by_id(args->gpu_id); + if (dev == NULL) + return -EINVAL; + + if (dev->device_info->asic_family == CHIP_CARRIZO) { + pr_debug("kfd_ioctl_dbg_wave_control not supported on CZ\n"); + return -EINVAL; + } + + /* input size must match the computed "compact" size */ + if (args->buf_size_in_bytes != computed_buff_size) { + pr_debug("size mismatch, computed : actual %u : %u\n", + args->buf_size_in_bytes, computed_buff_size); + return -EINVAL; + } + + cmd_from_user = (void __user *) args->content_ptr; + + if (cmd_from_user == NULL) + return -EINVAL; + + /* this is the actual buffer to work with */ + + args_buff = kmalloc(args->buf_size_in_bytes - sizeof(*args), + GFP_KERNEL); + + if (args_buff == NULL) + return -ENOMEM; + + /* Now copy the entire buffer from user */ + status = copy_from_user(args_buff, cmd_from_user, + args->buf_size_in_bytes - sizeof(*args)); + if (status != 0) { + pr_debug("Failed to copy wave control user data\n"); + kfree(args_buff); + return -EINVAL; + } + + /* move ptr to the start of the "pay-load" area */ + wac_info.process = p; + + wac_info.operand = *((enum HSA_DBG_WAVEOP *)(&args_buff[args_idx])); + args_idx += sizeof(wac_info.operand); + + wac_info.mode = *((enum HSA_DBG_WAVEMODE *)(&args_buff[args_idx])); + args_idx += sizeof(wac_info.mode); + + wac_info.trapId = *((uint32_t *)(&args_buff[args_idx])); + args_idx += sizeof(wac_info.trapId); + + wac_info.dbgWave_msg.DbgWaveMsg.WaveMsgInfoGen2.Value = + *((uint32_t *)(&args_buff[args_idx])); + wac_info.dbgWave_msg.MemoryVA = NULL; + + mutex_lock(kfd_get_dbgmgr_mutex()); + + pr_debug("Calling dbg manager process %p, operand %u, mode %u, trapId %u, message %u\n", + wac_info.process, wac_info.operand, + wac_info.mode, wac_info.trapId, + wac_info.dbgWave_msg.DbgWaveMsg.WaveMsgInfoGen2.Value); + + status = kfd_dbgmgr_wave_control(dev->dbgmgr, &wac_info); + + pr_debug("Returned status of dbg manager is %ld\n", status); + + mutex_unlock(kfd_get_dbgmgr_mutex()); + + kfree(args_buff); + + return status; +} + static int kfd_ioctl_get_clock_counters(struct file *filep, struct kfd_process *p, void *data) { @@ -514,6 +812,62 @@ static int kfd_ioctl_get_process_apertures(struct file *filp, return 0; } +static int kfd_ioctl_create_event(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_create_event_args *args = data; + int err; + + err = kfd_event_create(filp, p, args->event_type, + args->auto_reset != 0, args->node_id, + &args->event_id, &args->event_trigger_data, + &args->event_page_offset, + &args->event_slot_index); + + return err; +} + +static int kfd_ioctl_destroy_event(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_destroy_event_args *args = data; + + return kfd_event_destroy(p, args->event_id); +} + +static int kfd_ioctl_set_event(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_set_event_args *args = data; + + return kfd_set_event(p, args->event_id); +} + +static int kfd_ioctl_reset_event(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_reset_event_args *args = data; + + return kfd_reset_event(p, args->event_id); +} + +static int kfd_ioctl_wait_events(struct file *filp, struct kfd_process *p, + void *data) +{ + struct kfd_ioctl_wait_events_args *args = data; + enum kfd_event_wait_result wait_result; + int err; + + err = kfd_wait_on_events(p, args->num_events, + (void __user *)args->events_ptr, + (args->wait_for_all != 0), + args->timeout, &wait_result); + + args->wait_result = wait_result; + + return err; +} + #define AMDKFD_IOCTL_DEF(ioctl, _func, _flags) \ [_IOC_NR(ioctl)] = {.cmd = ioctl, .func = _func, .flags = _flags, .cmd_drv = 0, .name = #ioctl} @@ -539,6 +893,33 @@ static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = { AMDKFD_IOCTL_DEF(AMDKFD_IOC_UPDATE_QUEUE, kfd_ioctl_update_queue, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_CREATE_EVENT, + kfd_ioctl_create_event, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_DESTROY_EVENT, + kfd_ioctl_destroy_event, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_SET_EVENT, + kfd_ioctl_set_event, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_RESET_EVENT, + kfd_ioctl_reset_event, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_WAIT_EVENTS, + kfd_ioctl_wait_events, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_DBG_REGISTER, + kfd_ioctl_dbg_register, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_DBG_UNREGISTER, + kfd_ioctl_dbg_unrgesiter, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_DBG_ADDRESS_WATCH, + kfd_ioctl_dbg_address_watch, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_DBG_WAVE_CONTROL, + kfd_ioctl_dbg_wave_control, 0), }; #define AMDKFD_CORE_IOCTL_COUNT ARRAY_SIZE(amdkfd_ioctls) @@ -639,5 +1020,15 @@ static int kfd_mmap(struct file *filp, struct vm_area_struct *vma) if (IS_ERR(process)) return PTR_ERR(process); - return kfd_doorbell_mmap(process, vma); + if ((vma->vm_pgoff & KFD_MMAP_DOORBELL_MASK) == + KFD_MMAP_DOORBELL_MASK) { + vma->vm_pgoff = vma->vm_pgoff ^ KFD_MMAP_DOORBELL_MASK; + return kfd_doorbell_mmap(process, vma); + } else if ((vma->vm_pgoff & KFD_MMAP_EVENTS_MASK) == + KFD_MMAP_EVENTS_MASK) { + vma->vm_pgoff = vma->vm_pgoff ^ KFD_MMAP_EVENTS_MASK; + return kfd_event_mmap(process, vma); + } + + return -EFAULT; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c new file mode 100644 index 000000000000..c34c393e9aea --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c @@ -0,0 +1,886 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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 <linux/types.h> +#include <linux/kernel.h> +#include <linux/log2.h> +#include <linux/sched.h> +#include <linux/slab.h> +#include <linux/mutex.h> +#include <linux/device.h> + +#include "kfd_pm4_headers.h" +#include "kfd_pm4_headers_diq.h" +#include "kfd_kernel_queue.h" +#include "kfd_priv.h" +#include "kfd_pm4_opcodes.h" +#include "cik_regs.h" +#include "kfd_dbgmgr.h" +#include "kfd_dbgdev.h" +#include "kfd_device_queue_manager.h" +#include "../../radeon/cik_reg.h" + +static void dbgdev_address_watch_disable_nodiq(struct kfd_dev *dev) +{ + BUG_ON(!dev || !dev->kfd2kgd); + + dev->kfd2kgd->address_watch_disable(dev->kgd); +} + +static int dbgdev_diq_submit_ib(struct kfd_dbgdev *dbgdev, + unsigned int pasid, uint64_t vmid0_address, + uint32_t *packet_buff, size_t size_in_bytes) +{ + struct pm4__release_mem *rm_packet; + struct pm4__indirect_buffer_pasid *ib_packet; + struct kfd_mem_obj *mem_obj; + size_t pq_packets_size_in_bytes; + union ULARGE_INTEGER *largep; + union ULARGE_INTEGER addr; + struct kernel_queue *kq; + uint64_t *rm_state; + unsigned int *ib_packet_buff; + int status; + + BUG_ON(!dbgdev || !dbgdev->kq || !packet_buff || !size_in_bytes); + + kq = dbgdev->kq; + + pq_packets_size_in_bytes = sizeof(struct pm4__release_mem) + + sizeof(struct pm4__indirect_buffer_pasid); + + /* + * We acquire a buffer from DIQ + * The receive packet buff will be sitting on the Indirect Buffer + * and in the PQ we put the IB packet + sync packet(s). + */ + status = kq->ops.acquire_packet_buffer(kq, + pq_packets_size_in_bytes / sizeof(uint32_t), + &ib_packet_buff); + if (status != 0) { + pr_err("amdkfd: acquire_packet_buffer failed\n"); + return status; + } + + memset(ib_packet_buff, 0, pq_packets_size_in_bytes); + + ib_packet = (struct pm4__indirect_buffer_pasid *) (ib_packet_buff); + + ib_packet->header.count = 3; + ib_packet->header.opcode = IT_INDIRECT_BUFFER_PASID; + ib_packet->header.type = PM4_TYPE_3; + + largep = (union ULARGE_INTEGER *) &vmid0_address; + + ib_packet->bitfields2.ib_base_lo = largep->u.low_part >> 2; + ib_packet->bitfields3.ib_base_hi = largep->u.high_part; + + ib_packet->control = (1 << 23) | (1 << 31) | + ((size_in_bytes / sizeof(uint32_t)) & 0xfffff); + + ib_packet->bitfields5.pasid = pasid; + + /* + * for now we use release mem for GPU-CPU synchronization + * Consider WaitRegMem + WriteData as a better alternative + * we get a GART allocations ( gpu/cpu mapping), + * for the sync variable, and wait until: + * (a) Sync with HW + * (b) Sync var is written by CP to mem. + */ + rm_packet = (struct pm4__release_mem *) (ib_packet_buff + + (sizeof(struct pm4__indirect_buffer_pasid) / + sizeof(unsigned int))); + + status = kfd_gtt_sa_allocate(dbgdev->dev, sizeof(uint64_t), + &mem_obj); + + if (status != 0) { + pr_err("amdkfd: Failed to allocate GART memory\n"); + kq->ops.rollback_packet(kq); + return status; + } + + rm_state = (uint64_t *) mem_obj->cpu_ptr; + + *rm_state = QUEUESTATE__ACTIVE_COMPLETION_PENDING; + + rm_packet->header.opcode = IT_RELEASE_MEM; + rm_packet->header.type = PM4_TYPE_3; + rm_packet->header.count = sizeof(struct pm4__release_mem) / + sizeof(unsigned int) - 2; + + rm_packet->bitfields2.event_type = CACHE_FLUSH_AND_INV_TS_EVENT; + rm_packet->bitfields2.event_index = + event_index___release_mem__end_of_pipe; + + rm_packet->bitfields2.cache_policy = cache_policy___release_mem__lru; + rm_packet->bitfields2.atc = 0; + rm_packet->bitfields2.tc_wb_action_ena = 1; + + addr.quad_part = mem_obj->gpu_addr; + + rm_packet->bitfields4.address_lo_32b = addr.u.low_part >> 2; + rm_packet->address_hi = addr.u.high_part; + + rm_packet->bitfields3.data_sel = + data_sel___release_mem__send_64_bit_data; + + rm_packet->bitfields3.int_sel = + int_sel___release_mem__send_data_after_write_confirm; + + rm_packet->bitfields3.dst_sel = + dst_sel___release_mem__memory_controller; + + rm_packet->data_lo = QUEUESTATE__ACTIVE; + + kq->ops.submit_packet(kq); + + /* Wait till CP writes sync code: */ + status = amdkfd_fence_wait_timeout( + (unsigned int *) rm_state, + QUEUESTATE__ACTIVE, 1500); + + kfd_gtt_sa_free(dbgdev->dev, mem_obj); + + return status; +} + +static int dbgdev_register_nodiq(struct kfd_dbgdev *dbgdev) +{ + BUG_ON(!dbgdev); + + /* + * no action is needed in this case, + * just make sure diq will not be used + */ + + dbgdev->kq = NULL; + + return 0; +} + +static int dbgdev_register_diq(struct kfd_dbgdev *dbgdev) +{ + struct queue_properties properties; + unsigned int qid; + struct kernel_queue *kq = NULL; + int status; + + BUG_ON(!dbgdev || !dbgdev->pqm || !dbgdev->dev); + + status = pqm_create_queue(dbgdev->pqm, dbgdev->dev, NULL, + &properties, 0, KFD_QUEUE_TYPE_DIQ, + &qid); + + if (status) { + pr_err("amdkfd: Failed to create DIQ\n"); + return status; + } + + pr_debug("DIQ Created with queue id: %d\n", qid); + + kq = pqm_get_kernel_queue(dbgdev->pqm, qid); + + if (kq == NULL) { + pr_err("amdkfd: Error getting DIQ\n"); + pqm_destroy_queue(dbgdev->pqm, qid); + return -EFAULT; + } + + dbgdev->kq = kq; + + return status; +} + +static int dbgdev_unregister_nodiq(struct kfd_dbgdev *dbgdev) +{ + BUG_ON(!dbgdev || !dbgdev->dev); + + /* disable watch address */ + dbgdev_address_watch_disable_nodiq(dbgdev->dev); + return 0; +} + +static int dbgdev_unregister_diq(struct kfd_dbgdev *dbgdev) +{ + /* todo - disable address watch */ + int status; + + BUG_ON(!dbgdev || !dbgdev->pqm || !dbgdev->kq); + + status = pqm_destroy_queue(dbgdev->pqm, + dbgdev->kq->queue->properties.queue_id); + dbgdev->kq = NULL; + + return status; +} + +static void dbgdev_address_watch_set_registers( + const struct dbg_address_watch_info *adw_info, + union TCP_WATCH_ADDR_H_BITS *addrHi, + union TCP_WATCH_ADDR_L_BITS *addrLo, + union TCP_WATCH_CNTL_BITS *cntl, + unsigned int index, unsigned int vmid) +{ + union ULARGE_INTEGER addr; + + BUG_ON(!adw_info || !addrHi || !addrLo || !cntl); + + addr.quad_part = 0; + addrHi->u32All = 0; + addrLo->u32All = 0; + cntl->u32All = 0; + + if (adw_info->watch_mask != NULL) + cntl->bitfields.mask = + (uint32_t) (adw_info->watch_mask[index] & + ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK); + else + cntl->bitfields.mask = ADDRESS_WATCH_REG_CNTL_DEFAULT_MASK; + + addr.quad_part = (unsigned long long) adw_info->watch_address[index]; + + addrHi->bitfields.addr = addr.u.high_part & + ADDRESS_WATCH_REG_ADDHIGH_MASK; + addrLo->bitfields.addr = + (addr.u.low_part >> ADDRESS_WATCH_REG_ADDLOW_SHIFT); + + cntl->bitfields.mode = adw_info->watch_mode[index]; + cntl->bitfields.vmid = (uint32_t) vmid; + /* for now assume it is an ATC address */ + cntl->u32All |= ADDRESS_WATCH_REG_CNTL_ATC_BIT; + + pr_debug("\t\t%20s %08x\n", "set reg mask :", cntl->bitfields.mask); + pr_debug("\t\t%20s %08x\n", "set reg add high :", + addrHi->bitfields.addr); + pr_debug("\t\t%20s %08x\n", "set reg add low :", + addrLo->bitfields.addr); +} + +static int dbgdev_address_watch_nodiq(struct kfd_dbgdev *dbgdev, + struct dbg_address_watch_info *adw_info) +{ + union TCP_WATCH_ADDR_H_BITS addrHi; + union TCP_WATCH_ADDR_L_BITS addrLo; + union TCP_WATCH_CNTL_BITS cntl; + struct kfd_process_device *pdd; + unsigned int i; + + BUG_ON(!dbgdev || !dbgdev->dev || !adw_info); + + /* taking the vmid for that process on the safe way using pdd */ + pdd = kfd_get_process_device_data(dbgdev->dev, + adw_info->process); + if (!pdd) { + pr_err("amdkfd: Failed to get pdd for wave control no DIQ\n"); + return -EFAULT; + } + + addrHi.u32All = 0; + addrLo.u32All = 0; + cntl.u32All = 0; + + if ((adw_info->num_watch_points > MAX_WATCH_ADDRESSES) || + (adw_info->num_watch_points == 0)) { + pr_err("amdkfd: num_watch_points is invalid\n"); + return -EINVAL; + } + + if ((adw_info->watch_mode == NULL) || + (adw_info->watch_address == NULL)) { + pr_err("amdkfd: adw_info fields are not valid\n"); + return -EINVAL; + } + + for (i = 0 ; i < adw_info->num_watch_points ; i++) { + dbgdev_address_watch_set_registers(adw_info, &addrHi, &addrLo, + &cntl, i, pdd->qpd.vmid); + + pr_debug("\t\t%30s\n", "* * * * * * * * * * * * * * * * * *"); + pr_debug("\t\t%20s %08x\n", "register index :", i); + pr_debug("\t\t%20s %08x\n", "vmid is :", pdd->qpd.vmid); + pr_debug("\t\t%20s %08x\n", "Address Low is :", + addrLo.bitfields.addr); + pr_debug("\t\t%20s %08x\n", "Address high is :", + addrHi.bitfields.addr); + pr_debug("\t\t%20s %08x\n", "Address high is :", + addrHi.bitfields.addr); + pr_debug("\t\t%20s %08x\n", "Control Mask is :", + cntl.bitfields.mask); + pr_debug("\t\t%20s %08x\n", "Control Mode is :", + cntl.bitfields.mode); + pr_debug("\t\t%20s %08x\n", "Control Vmid is :", + cntl.bitfields.vmid); + pr_debug("\t\t%20s %08x\n", "Control atc is :", + cntl.bitfields.atc); + pr_debug("\t\t%30s\n", "* * * * * * * * * * * * * * * * * *"); + + pdd->dev->kfd2kgd->address_watch_execute( + dbgdev->dev->kgd, + i, + cntl.u32All, + addrHi.u32All, + addrLo.u32All); + } + + return 0; +} + +static int dbgdev_address_watch_diq(struct kfd_dbgdev *dbgdev, + struct dbg_address_watch_info *adw_info) +{ + struct pm4__set_config_reg *packets_vec; + union TCP_WATCH_ADDR_H_BITS addrHi; + union TCP_WATCH_ADDR_L_BITS addrLo; + union TCP_WATCH_CNTL_BITS cntl; + struct kfd_mem_obj *mem_obj; + unsigned int aw_reg_add_dword; + uint32_t *packet_buff_uint; + unsigned int i; + int status; + size_t ib_size = sizeof(struct pm4__set_config_reg) * 4; + /* we do not control the vmid in DIQ mode, just a place holder */ + unsigned int vmid = 0; + + BUG_ON(!dbgdev || !dbgdev->dev || !adw_info); + + addrHi.u32All = 0; + addrLo.u32All = 0; + cntl.u32All = 0; + + if ((adw_info->num_watch_points > MAX_WATCH_ADDRESSES) || + (adw_info->num_watch_points == 0)) { + pr_err("amdkfd: num_watch_points is invalid\n"); + return -EINVAL; + } + + if ((NULL == adw_info->watch_mode) || + (NULL == adw_info->watch_address)) { + pr_err("amdkfd: adw_info fields are not valid\n"); + return -EINVAL; + } + + status = kfd_gtt_sa_allocate(dbgdev->dev, ib_size, &mem_obj); + + if (status != 0) { + pr_err("amdkfd: Failed to allocate GART memory\n"); + return status; + } + + packet_buff_uint = mem_obj->cpu_ptr; + + memset(packet_buff_uint, 0, ib_size); + + packets_vec = (struct pm4__set_config_reg *) (packet_buff_uint); + + packets_vec[0].header.count = 1; + packets_vec[0].header.opcode = IT_SET_CONFIG_REG; + packets_vec[0].header.type = PM4_TYPE_3; + packets_vec[0].bitfields2.vmid_shift = ADDRESS_WATCH_CNTL_OFFSET; + packets_vec[0].bitfields2.insert_vmid = 1; + packets_vec[1].ordinal1 = packets_vec[0].ordinal1; + packets_vec[1].bitfields2.insert_vmid = 0; + packets_vec[2].ordinal1 = packets_vec[0].ordinal1; + packets_vec[2].bitfields2.insert_vmid = 0; + packets_vec[3].ordinal1 = packets_vec[0].ordinal1; + packets_vec[3].bitfields2.vmid_shift = ADDRESS_WATCH_CNTL_OFFSET; + packets_vec[3].bitfields2.insert_vmid = 1; + + for (i = 0; i < adw_info->num_watch_points; i++) { + dbgdev_address_watch_set_registers(adw_info, + &addrHi, + &addrLo, + &cntl, + i, + vmid); + + pr_debug("\t\t%30s\n", "* * * * * * * * * * * * * * * * * *"); + pr_debug("\t\t%20s %08x\n", "register index :", i); + pr_debug("\t\t%20s %08x\n", "vmid is :", vmid); + pr_debug("\t\t%20s %p\n", "Add ptr is :", + adw_info->watch_address); + pr_debug("\t\t%20s %08llx\n", "Add is :", + adw_info->watch_address[i]); + pr_debug("\t\t%20s %08x\n", "Address Low is :", + addrLo.bitfields.addr); + pr_debug("\t\t%20s %08x\n", "Address high is :", + addrHi.bitfields.addr); + pr_debug("\t\t%20s %08x\n", "Control Mask is :", + cntl.bitfields.mask); + pr_debug("\t\t%20s %08x\n", "Control Mode is :", + cntl.bitfields.mode); + pr_debug("\t\t%20s %08x\n", "Control Vmid is :", + cntl.bitfields.vmid); + pr_debug("\t\t%20s %08x\n", "Control atc is :", + cntl.bitfields.atc); + pr_debug("\t\t%30s\n", "* * * * * * * * * * * * * * * * * *"); + + aw_reg_add_dword = + dbgdev->dev->kfd2kgd->address_watch_get_offset( + dbgdev->dev->kgd, + i, + ADDRESS_WATCH_REG_CNTL); + + aw_reg_add_dword /= sizeof(uint32_t); + + packets_vec[0].bitfields2.reg_offset = + aw_reg_add_dword - AMD_CONFIG_REG_BASE; + + packets_vec[0].reg_data[0] = cntl.u32All; + + aw_reg_add_dword = + dbgdev->dev->kfd2kgd->address_watch_get_offset( + dbgdev->dev->kgd, + i, + ADDRESS_WATCH_REG_ADDR_HI); + + aw_reg_add_dword /= sizeof(uint32_t); + + packets_vec[1].bitfields2.reg_offset = + aw_reg_add_dword - AMD_CONFIG_REG_BASE; + packets_vec[1].reg_data[0] = addrHi.u32All; + + aw_reg_add_dword = + dbgdev->dev->kfd2kgd->address_watch_get_offset( + dbgdev->dev->kgd, + i, + ADDRESS_WATCH_REG_ADDR_LO); + + aw_reg_add_dword /= sizeof(uint32_t); + + packets_vec[2].bitfields2.reg_offset = + aw_reg_add_dword - AMD_CONFIG_REG_BASE; + packets_vec[2].reg_data[0] = addrLo.u32All; + + /* enable watch flag if address is not zero*/ + if (adw_info->watch_address[i] > 0) + cntl.bitfields.valid = 1; + else + cntl.bitfields.valid = 0; + + aw_reg_add_dword = + dbgdev->dev->kfd2kgd->address_watch_get_offset( + dbgdev->dev->kgd, + i, + ADDRESS_WATCH_REG_CNTL); + + aw_reg_add_dword /= sizeof(uint32_t); + + packets_vec[3].bitfields2.reg_offset = + aw_reg_add_dword - AMD_CONFIG_REG_BASE; + packets_vec[3].reg_data[0] = cntl.u32All; + + status = dbgdev_diq_submit_ib( + dbgdev, + adw_info->process->pasid, + mem_obj->gpu_addr, + packet_buff_uint, + ib_size); + + if (status != 0) { + pr_err("amdkfd: Failed to submit IB to DIQ\n"); + break; + } + } + + kfd_gtt_sa_free(dbgdev->dev, mem_obj); + return status; +} + +static int dbgdev_wave_control_set_registers( + struct dbg_wave_control_info *wac_info, + union SQ_CMD_BITS *in_reg_sq_cmd, + union GRBM_GFX_INDEX_BITS *in_reg_gfx_index) +{ + int status; + union SQ_CMD_BITS reg_sq_cmd; + union GRBM_GFX_INDEX_BITS reg_gfx_index; + struct HsaDbgWaveMsgAMDGen2 *pMsg; + + BUG_ON(!wac_info || !in_reg_sq_cmd || !in_reg_gfx_index); + + reg_sq_cmd.u32All = 0; + reg_gfx_index.u32All = 0; + pMsg = &wac_info->dbgWave_msg.DbgWaveMsg.WaveMsgInfoGen2; + + switch (wac_info->mode) { + /* Send command to single wave */ + case HSA_DBG_WAVEMODE_SINGLE: + /* + * Limit access to the process waves only, + * by setting vmid check + */ + reg_sq_cmd.bits.check_vmid = 1; + reg_sq_cmd.bits.simd_id = pMsg->ui32.SIMD; + reg_sq_cmd.bits.wave_id = pMsg->ui32.WaveId; + reg_sq_cmd.bits.mode = SQ_IND_CMD_MODE_SINGLE; + + reg_gfx_index.bits.sh_index = pMsg->ui32.ShaderArray; + reg_gfx_index.bits.se_index = pMsg->ui32.ShaderEngine; + reg_gfx_index.bits.instance_index = pMsg->ui32.HSACU; + + break; + + /* Send command to all waves with matching VMID */ + case HSA_DBG_WAVEMODE_BROADCAST_PROCESS: + + reg_gfx_index.bits.sh_broadcast_writes = 1; + reg_gfx_index.bits.se_broadcast_writes = 1; + reg_gfx_index.bits.instance_broadcast_writes = 1; + + reg_sq_cmd.bits.mode = SQ_IND_CMD_MODE_BROADCAST; + + break; + + /* Send command to all CU waves with matching VMID */ + case HSA_DBG_WAVEMODE_BROADCAST_PROCESS_CU: + + reg_sq_cmd.bits.check_vmid = 1; + reg_sq_cmd.bits.mode = SQ_IND_CMD_MODE_BROADCAST; + + reg_gfx_index.bits.sh_index = pMsg->ui32.ShaderArray; + reg_gfx_index.bits.se_index = pMsg->ui32.ShaderEngine; + reg_gfx_index.bits.instance_index = pMsg->ui32.HSACU; + + break; + + default: + return -EINVAL; + } + + switch (wac_info->operand) { + case HSA_DBG_WAVEOP_HALT: + reg_sq_cmd.bits.cmd = SQ_IND_CMD_CMD_HALT; + break; + + case HSA_DBG_WAVEOP_RESUME: + reg_sq_cmd.bits.cmd = SQ_IND_CMD_CMD_RESUME; + break; + + case HSA_DBG_WAVEOP_KILL: + reg_sq_cmd.bits.cmd = SQ_IND_CMD_CMD_KILL; + break; + + case HSA_DBG_WAVEOP_DEBUG: + reg_sq_cmd.bits.cmd = SQ_IND_CMD_CMD_DEBUG; + break; + + case HSA_DBG_WAVEOP_TRAP: + if (wac_info->trapId < MAX_TRAPID) { + reg_sq_cmd.bits.cmd = SQ_IND_CMD_CMD_TRAP; + reg_sq_cmd.bits.trap_id = wac_info->trapId; + } else { + status = -EINVAL; + } + break; + + default: + status = -EINVAL; + break; + } + + if (status == 0) { + *in_reg_sq_cmd = reg_sq_cmd; + *in_reg_gfx_index = reg_gfx_index; + } + + return status; +} + +static int dbgdev_wave_control_diq(struct kfd_dbgdev *dbgdev, + struct dbg_wave_control_info *wac_info) +{ + + int status; + union SQ_CMD_BITS reg_sq_cmd; + union GRBM_GFX_INDEX_BITS reg_gfx_index; + struct kfd_mem_obj *mem_obj; + uint32_t *packet_buff_uint; + struct pm4__set_config_reg *packets_vec; + size_t ib_size = sizeof(struct pm4__set_config_reg) * 3; + + BUG_ON(!dbgdev || !wac_info); + + reg_sq_cmd.u32All = 0; + + status = dbgdev_wave_control_set_registers(wac_info, ®_sq_cmd, + ®_gfx_index); + if (status) { + pr_err("amdkfd: Failed to set wave control registers\n"); + return status; + } + + /* we do not control the VMID in DIQ,so reset it to a known value */ + reg_sq_cmd.bits.vm_id = 0; + + pr_debug("\t\t %30s\n", "* * * * * * * * * * * * * * * * * *"); + + pr_debug("\t\t mode is: %u\n", wac_info->mode); + pr_debug("\t\t operand is: %u\n", wac_info->operand); + pr_debug("\t\t trap id is: %u\n", wac_info->trapId); + pr_debug("\t\t msg value is: %u\n", + wac_info->dbgWave_msg.DbgWaveMsg.WaveMsgInfoGen2.Value); + pr_debug("\t\t vmid is: N/A\n"); + + pr_debug("\t\t chk_vmid is : %u\n", reg_sq_cmd.bitfields.check_vmid); + pr_debug("\t\t command is : %u\n", reg_sq_cmd.bitfields.cmd); + pr_debug("\t\t queue id is : %u\n", reg_sq_cmd.bitfields.queue_id); + pr_debug("\t\t simd id is : %u\n", reg_sq_cmd.bitfields.simd_id); + pr_debug("\t\t mode is : %u\n", reg_sq_cmd.bitfields.mode); + pr_debug("\t\t vm_id is : %u\n", reg_sq_cmd.bitfields.vm_id); + pr_debug("\t\t wave_id is : %u\n", reg_sq_cmd.bitfields.wave_id); + + pr_debug("\t\t ibw is : %u\n", + reg_gfx_index.bitfields.instance_broadcast_writes); + pr_debug("\t\t ii is : %u\n", + reg_gfx_index.bitfields.instance_index); + pr_debug("\t\t sebw is : %u\n", + reg_gfx_index.bitfields.se_broadcast_writes); + pr_debug("\t\t se_ind is : %u\n", reg_gfx_index.bitfields.se_index); + pr_debug("\t\t sh_ind is : %u\n", reg_gfx_index.bitfields.sh_index); + pr_debug("\t\t sbw is : %u\n", + reg_gfx_index.bitfields.sh_broadcast_writes); + + pr_debug("\t\t %30s\n", "* * * * * * * * * * * * * * * * * *"); + + status = kfd_gtt_sa_allocate(dbgdev->dev, ib_size, &mem_obj); + + if (status != 0) { + pr_err("amdkfd: Failed to allocate GART memory\n"); + return status; + } + + packet_buff_uint = mem_obj->cpu_ptr; + + memset(packet_buff_uint, 0, ib_size); + + packets_vec = (struct pm4__set_config_reg *) packet_buff_uint; + packets_vec[0].header.count = 1; + packets_vec[0].header.opcode = IT_SET_UCONFIG_REG; + packets_vec[0].header.type = PM4_TYPE_3; + packets_vec[0].bitfields2.reg_offset = + GRBM_GFX_INDEX / (sizeof(uint32_t)) - + USERCONFIG_REG_BASE; + + packets_vec[0].bitfields2.insert_vmid = 0; + packets_vec[0].reg_data[0] = reg_gfx_index.u32All; + + packets_vec[1].header.count = 1; + packets_vec[1].header.opcode = IT_SET_CONFIG_REG; + packets_vec[1].header.type = PM4_TYPE_3; + packets_vec[1].bitfields2.reg_offset = SQ_CMD / (sizeof(uint32_t)) - + AMD_CONFIG_REG_BASE; + + packets_vec[1].bitfields2.vmid_shift = SQ_CMD_VMID_OFFSET; + packets_vec[1].bitfields2.insert_vmid = 1; + packets_vec[1].reg_data[0] = reg_sq_cmd.u32All; + + /* Restore the GRBM_GFX_INDEX register */ + + reg_gfx_index.u32All = 0; + reg_gfx_index.bits.sh_broadcast_writes = 1; + reg_gfx_index.bits.instance_broadcast_writes = 1; + reg_gfx_index.bits.se_broadcast_writes = 1; + + + packets_vec[2].ordinal1 = packets_vec[0].ordinal1; + packets_vec[2].bitfields2.reg_offset = + GRBM_GFX_INDEX / (sizeof(uint32_t)) - + USERCONFIG_REG_BASE; + + packets_vec[2].bitfields2.insert_vmid = 0; + packets_vec[2].reg_data[0] = reg_gfx_index.u32All; + + status = dbgdev_diq_submit_ib( + dbgdev, + wac_info->process->pasid, + mem_obj->gpu_addr, + packet_buff_uint, + ib_size); + + if (status != 0) + pr_err("amdkfd: Failed to submit IB to DIQ\n"); + + kfd_gtt_sa_free(dbgdev->dev, mem_obj); + + return status; +} + +static int dbgdev_wave_control_nodiq(struct kfd_dbgdev *dbgdev, + struct dbg_wave_control_info *wac_info) +{ + int status; + union SQ_CMD_BITS reg_sq_cmd; + union GRBM_GFX_INDEX_BITS reg_gfx_index; + struct kfd_process_device *pdd; + + BUG_ON(!dbgdev || !dbgdev->dev || !wac_info); + + reg_sq_cmd.u32All = 0; + + /* taking the VMID for that process on the safe way using PDD */ + pdd = kfd_get_process_device_data(dbgdev->dev, wac_info->process); + + if (!pdd) { + pr_err("amdkfd: Failed to get pdd for wave control no DIQ\n"); + return -EFAULT; + } + status = dbgdev_wave_control_set_registers(wac_info, ®_sq_cmd, + ®_gfx_index); + if (status) { + pr_err("amdkfd: Failed to set wave control registers\n"); + return status; + } + + /* for non DIQ we need to patch the VMID: */ + + reg_sq_cmd.bits.vm_id = pdd->qpd.vmid; + + pr_debug("\t\t %30s\n", "* * * * * * * * * * * * * * * * * *"); + + pr_debug("\t\t mode is: %u\n", wac_info->mode); + pr_debug("\t\t operand is: %u\n", wac_info->operand); + pr_debug("\t\t trap id is: %u\n", wac_info->trapId); + pr_debug("\t\t msg value is: %u\n", + wac_info->dbgWave_msg.DbgWaveMsg.WaveMsgInfoGen2.Value); + pr_debug("\t\t vmid is: %u\n", pdd->qpd.vmid); + + pr_debug("\t\t chk_vmid is : %u\n", reg_sq_cmd.bitfields.check_vmid); + pr_debug("\t\t command is : %u\n", reg_sq_cmd.bitfields.cmd); + pr_debug("\t\t queue id is : %u\n", reg_sq_cmd.bitfields.queue_id); + pr_debug("\t\t simd id is : %u\n", reg_sq_cmd.bitfields.simd_id); + pr_debug("\t\t mode is : %u\n", reg_sq_cmd.bitfields.mode); + pr_debug("\t\t vm_id is : %u\n", reg_sq_cmd.bitfields.vm_id); + pr_debug("\t\t wave_id is : %u\n", reg_sq_cmd.bitfields.wave_id); + + pr_debug("\t\t ibw is : %u\n", + reg_gfx_index.bitfields.instance_broadcast_writes); + pr_debug("\t\t ii is : %u\n", + reg_gfx_index.bitfields.instance_index); + pr_debug("\t\t sebw is : %u\n", + reg_gfx_index.bitfields.se_broadcast_writes); + pr_debug("\t\t se_ind is : %u\n", reg_gfx_index.bitfields.se_index); + pr_debug("\t\t sh_ind is : %u\n", reg_gfx_index.bitfields.sh_index); + pr_debug("\t\t sbw is : %u\n", + reg_gfx_index.bitfields.sh_broadcast_writes); + + pr_debug("\t\t %30s\n", "* * * * * * * * * * * * * * * * * *"); + + return dbgdev->dev->kfd2kgd->wave_control_execute(dbgdev->dev->kgd, + reg_gfx_index.u32All, + reg_sq_cmd.u32All); +} + +int dbgdev_wave_reset_wavefronts(struct kfd_dev *dev, struct kfd_process *p) +{ + int status = 0; + unsigned int vmid; + union SQ_CMD_BITS reg_sq_cmd; + union GRBM_GFX_INDEX_BITS reg_gfx_index; + struct kfd_process_device *pdd; + struct dbg_wave_control_info wac_info; + int temp; + int first_vmid_to_scan = 8; + int last_vmid_to_scan = 15; + + first_vmid_to_scan = ffs(dev->shared_resources.compute_vmid_bitmap) - 1; + temp = dev->shared_resources.compute_vmid_bitmap >> first_vmid_to_scan; + last_vmid_to_scan = first_vmid_to_scan + ffz(temp); + + reg_sq_cmd.u32All = 0; + status = 0; + + wac_info.mode = HSA_DBG_WAVEMODE_BROADCAST_PROCESS; + wac_info.operand = HSA_DBG_WAVEOP_KILL; + + pr_debug("Killing all process wavefronts\n"); + + /* Scan all registers in the range ATC_VMID8_PASID_MAPPING .. + * ATC_VMID15_PASID_MAPPING + * to check which VMID the current process is mapped to. */ + + for (vmid = first_vmid_to_scan; vmid <= last_vmid_to_scan; vmid++) { + if (dev->kfd2kgd->get_atc_vmid_pasid_mapping_valid + (dev->kgd, vmid)) { + if (dev->kfd2kgd->get_atc_vmid_pasid_mapping_valid + (dev->kgd, vmid) == p->pasid) { + pr_debug("Killing wave fronts of vmid %d and pasid %d\n", + vmid, p->pasid); + break; + } + } + } + + if (vmid > last_vmid_to_scan) { + pr_err("amdkfd: didn't found vmid for pasid (%d)\n", p->pasid); + return -EFAULT; + } + + /* taking the VMID for that process on the safe way using PDD */ + pdd = kfd_get_process_device_data(dev, p); + if (!pdd) + return -EFAULT; + + status = dbgdev_wave_control_set_registers(&wac_info, ®_sq_cmd, + ®_gfx_index); + if (status != 0) + return -EINVAL; + + /* for non DIQ we need to patch the VMID: */ + reg_sq_cmd.bits.vm_id = vmid; + + dev->kfd2kgd->wave_control_execute(dev->kgd, + reg_gfx_index.u32All, + reg_sq_cmd.u32All); + + return 0; +} + +void kfd_dbgdev_init(struct kfd_dbgdev *pdbgdev, struct kfd_dev *pdev, + enum DBGDEV_TYPE type) +{ + BUG_ON(!pdbgdev || !pdev); + + pdbgdev->dev = pdev; + pdbgdev->kq = NULL; + pdbgdev->type = type; + pdbgdev->pqm = NULL; + + switch (type) { + case DBGDEV_TYPE_NODIQ: + pdbgdev->dbgdev_register = dbgdev_register_nodiq; + pdbgdev->dbgdev_unregister = dbgdev_unregister_nodiq; + pdbgdev->dbgdev_wave_control = dbgdev_wave_control_nodiq; + pdbgdev->dbgdev_address_watch = dbgdev_address_watch_nodiq; + break; + case DBGDEV_TYPE_DIQ: + default: + pdbgdev->dbgdev_register = dbgdev_register_diq; + pdbgdev->dbgdev_unregister = dbgdev_unregister_diq; + pdbgdev->dbgdev_wave_control = dbgdev_wave_control_diq; + pdbgdev->dbgdev_address_watch = dbgdev_address_watch_diq; + break; + } + +} diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.h b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.h new file mode 100644 index 000000000000..03424c20920c --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.h @@ -0,0 +1,193 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + */ + +#ifndef KFD_DBGDEV_H_ +#define KFD_DBGDEV_H_ + +enum { + SQ_CMD_VMID_OFFSET = 28, + ADDRESS_WATCH_CNTL_OFFSET = 24 +}; + +enum { + PRIV_QUEUE_SYNC_TIME_MS = 200 +}; + +/* CONTEXT reg space definition */ +enum { + CONTEXT_REG_BASE = 0xA000, + CONTEXT_REG_END = 0xA400, + CONTEXT_REG_SIZE = CONTEXT_REG_END - CONTEXT_REG_BASE +}; + +/* USER CONFIG reg space definition */ +enum { + USERCONFIG_REG_BASE = 0xC000, + USERCONFIG_REG_END = 0x10000, + USERCONFIG_REG_SIZE = USERCONFIG_REG_END - USERCONFIG_REG_BASE +}; + +/* CONFIG reg space definition */ +enum { + AMD_CONFIG_REG_BASE = 0x2000, /* in dwords */ + AMD_CONFIG_REG_END = 0x2B00, + AMD_CONFIG_REG_SIZE = AMD_CONFIG_REG_END - AMD_CONFIG_REG_BASE +}; + +/* SH reg space definition */ +enum { + SH_REG_BASE = 0x2C00, + SH_REG_END = 0x3000, + SH_REG_SIZE = SH_REG_END - SH_REG_BASE +}; + +enum SQ_IND_CMD_CMD { + SQ_IND_CMD_CMD_NULL = 0x00000000, + SQ_IND_CMD_CMD_HALT = 0x00000001, + SQ_IND_CMD_CMD_RESUME = 0x00000002, + SQ_IND_CMD_CMD_KILL = 0x00000003, + SQ_IND_CMD_CMD_DEBUG = 0x00000004, + SQ_IND_CMD_CMD_TRAP = 0x00000005, +}; + +enum SQ_IND_CMD_MODE { + SQ_IND_CMD_MODE_SINGLE = 0x00000000, + SQ_IND_CMD_MODE_BROADCAST = 0x00000001, + SQ_IND_CMD_MODE_BROADCAST_QUEUE = 0x00000002, + SQ_IND_CMD_MODE_BROADCAST_PIPE = 0x00000003, + SQ_IND_CMD_MODE_BROADCAST_ME = 0x00000004, +}; + +union SQ_IND_INDEX_BITS { + struct { + uint32_t wave_id:4; + uint32_t simd_id:2; + uint32_t thread_id:6; + uint32_t:1; + uint32_t force_read:1; + uint32_t read_timeout:1; + uint32_t unindexed:1; + uint32_t index:16; + + } bitfields, bits; + uint32_t u32All; + signed int i32All; + float f32All; +}; + +union SQ_IND_CMD_BITS { + struct { + uint32_t data:32; + } bitfields, bits; + uint32_t u32All; + signed int i32All; + float f32All; +}; + +union SQ_CMD_BITS { + struct { + uint32_t cmd:3; + uint32_t:1; + uint32_t mode:3; + uint32_t check_vmid:1; + uint32_t trap_id:3; + uint32_t:5; + uint32_t wave_id:4; + uint32_t simd_id:2; + uint32_t:2; + uint32_t queue_id:3; + uint32_t:1; + uint32_t vm_id:4; + } bitfields, bits; + uint32_t u32All; + signed int i32All; + float f32All; +}; + +union SQ_IND_DATA_BITS { + struct { + uint32_t data:32; + } bitfields, bits; + uint32_t u32All; + signed int i32All; + float f32All; +}; + +union GRBM_GFX_INDEX_BITS { + struct { + uint32_t instance_index:8; + uint32_t sh_index:8; + uint32_t se_index:8; + uint32_t:5; + uint32_t sh_broadcast_writes:1; + uint32_t instance_broadcast_writes:1; + uint32_t se_broadcast_writes:1; + } bitfields, bits; + uint32_t u32All; + signed int i32All; + float f32All; +}; + +union TCP_WATCH_ADDR_H_BITS { + struct { + uint32_t addr:16; + uint32_t:16; + + } bitfields, bits; + uint32_t u32All; + signed int i32All; + float f32All; +}; + +union TCP_WATCH_ADDR_L_BITS { + struct { + uint32_t:6; + uint32_t addr:26; + } bitfields, bits; + uint32_t u32All; + signed int i32All; + float f32All; +}; + +enum { + QUEUESTATE__INVALID = 0, /* so by default we'll get invalid state */ + QUEUESTATE__ACTIVE_COMPLETION_PENDING, + QUEUESTATE__ACTIVE +}; + +union ULARGE_INTEGER { + struct { + uint32_t low_part; + uint32_t high_part; + } u; + unsigned long long quad_part; +}; + + +#define KFD_CIK_VMID_START_OFFSET (8) +#define KFD_CIK_VMID_END_OFFSET (KFD_CIK_VMID_START_OFFSET + (8)) + + +void kfd_dbgdev_init(struct kfd_dbgdev *pdbgdev, struct kfd_dev *pdev, + enum DBGDEV_TYPE type); + +#endif /* KFD_DBGDEV_H_ */ diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c b/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c new file mode 100644 index 000000000000..56d676396342 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.c @@ -0,0 +1,168 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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 <linux/types.h> +#include <linux/kernel.h> +#include <linux/log2.h> +#include <linux/sched.h> +#include <linux/slab.h> +#include <linux/device.h> + +#include "kfd_priv.h" +#include "cik_regs.h" +#include "kfd_pm4_headers.h" +#include "kfd_pm4_headers_diq.h" +#include "kfd_dbgmgr.h" +#include "kfd_dbgdev.h" + +static DEFINE_MUTEX(kfd_dbgmgr_mutex); + +struct mutex *kfd_get_dbgmgr_mutex(void) +{ + return &kfd_dbgmgr_mutex; +} + + +static void kfd_dbgmgr_uninitialize(struct kfd_dbgmgr *pmgr) +{ + BUG_ON(!pmgr); + + kfree(pmgr->dbgdev); + + pmgr->dbgdev = NULL; + pmgr->pasid = 0; + pmgr->dev = NULL; +} + +void kfd_dbgmgr_destroy(struct kfd_dbgmgr *pmgr) +{ + if (pmgr != NULL) { + kfd_dbgmgr_uninitialize(pmgr); + kfree(pmgr); + } +} + +bool kfd_dbgmgr_create(struct kfd_dbgmgr **ppmgr, struct kfd_dev *pdev) +{ + enum DBGDEV_TYPE type = DBGDEV_TYPE_DIQ; + struct kfd_dbgmgr *new_buff; + + BUG_ON(pdev == NULL); + BUG_ON(!pdev->init_complete); + + new_buff = kfd_alloc_struct(new_buff); + if (!new_buff) { + pr_err("amdkfd: Failed to allocate dbgmgr instance\n"); + return false; + } + + new_buff->pasid = 0; + new_buff->dev = pdev; + new_buff->dbgdev = kfd_alloc_struct(new_buff->dbgdev); + if (!new_buff->dbgdev) { + pr_err("amdkfd: Failed to allocate dbgdev instance\n"); + kfree(new_buff); + return false; + } + + /* get actual type of DBGDevice cpsch or not */ + if (sched_policy == KFD_SCHED_POLICY_NO_HWS) + type = DBGDEV_TYPE_NODIQ; + + kfd_dbgdev_init(new_buff->dbgdev, pdev, type); + *ppmgr = new_buff; + + return true; +} + +long kfd_dbgmgr_register(struct kfd_dbgmgr *pmgr, struct kfd_process *p) +{ + BUG_ON(!p || !pmgr || !pmgr->dbgdev); + + if (pmgr->pasid != 0) { + pr_debug("H/W debugger is already active using pasid %d\n", + pmgr->pasid); + return -EBUSY; + } + + /* remember pasid */ + pmgr->pasid = p->pasid; + + /* provide the pqm for diq generation */ + pmgr->dbgdev->pqm = &p->pqm; + + /* activate the actual registering */ + pmgr->dbgdev->dbgdev_register(pmgr->dbgdev); + + return 0; +} + +long kfd_dbgmgr_unregister(struct kfd_dbgmgr *pmgr, struct kfd_process *p) +{ + BUG_ON(!p || !pmgr || !pmgr->dbgdev); + + /* Is the requests coming from the already registered process? */ + if (pmgr->pasid != p->pasid) { + pr_debug("H/W debugger is not registered by calling pasid %d\n", + p->pasid); + return -EINVAL; + } + + pmgr->dbgdev->dbgdev_unregister(pmgr->dbgdev); + + pmgr->pasid = 0; + + return 0; +} + +long kfd_dbgmgr_wave_control(struct kfd_dbgmgr *pmgr, + struct dbg_wave_control_info *wac_info) +{ + BUG_ON(!pmgr || !pmgr->dbgdev || !wac_info); + + /* Is the requests coming from the already registered process? */ + if (pmgr->pasid != wac_info->process->pasid) { + pr_debug("H/W debugger support was not registered for requester pasid %d\n", + wac_info->process->pasid); + return -EINVAL; + } + + return (long) pmgr->dbgdev->dbgdev_wave_control(pmgr->dbgdev, wac_info); +} + +long kfd_dbgmgr_address_watch(struct kfd_dbgmgr *pmgr, + struct dbg_address_watch_info *adw_info) +{ + BUG_ON(!pmgr || !pmgr->dbgdev || !adw_info); + + + /* Is the requests coming from the already registered process? */ + if (pmgr->pasid != adw_info->process->pasid) { + pr_debug("H/W debugger support was not registered for requester pasid %d\n", + adw_info->process->pasid); + return -EINVAL; + } + + return (long) pmgr->dbgdev->dbgdev_address_watch(pmgr->dbgdev, + adw_info); +} + diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h b/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h new file mode 100644 index 000000000000..257a745ad0b5 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgmgr.h @@ -0,0 +1,294 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + * + */ + +#ifndef KFD_DBGMGR_H_ +#define KFD_DBGMGR_H_ + +#include "kfd_priv.h" + +/* must align with hsakmttypes definition */ +#pragma pack(push, 4) + +enum HSA_DBG_WAVEOP { + HSA_DBG_WAVEOP_HALT = 1, /* Halts a wavefront */ + HSA_DBG_WAVEOP_RESUME = 2, /* Resumes a wavefront */ + HSA_DBG_WAVEOP_KILL = 3, /* Kills a wavefront */ + HSA_DBG_WAVEOP_DEBUG = 4, /* Causes wavefront to enter + debug mode */ + HSA_DBG_WAVEOP_TRAP = 5, /* Causes wavefront to take + a trap */ + HSA_DBG_NUM_WAVEOP = 5, + HSA_DBG_MAX_WAVEOP = 0xFFFFFFFF +}; + +enum HSA_DBG_WAVEMODE { + /* send command to a single wave */ + HSA_DBG_WAVEMODE_SINGLE = 0, + /* + * Broadcast to all wavefronts of all processes is not + * supported for HSA user mode + */ + + /* send to waves within current process */ + HSA_DBG_WAVEMODE_BROADCAST_PROCESS = 2, + /* send to waves within current process on CU */ + HSA_DBG_WAVEMODE_BROADCAST_PROCESS_CU = 3, + HSA_DBG_NUM_WAVEMODE = 3, + HSA_DBG_MAX_WAVEMODE = 0xFFFFFFFF +}; + +enum HSA_DBG_WAVEMSG_TYPE { + HSA_DBG_WAVEMSG_AUTO = 0, + HSA_DBG_WAVEMSG_USER = 1, + HSA_DBG_WAVEMSG_ERROR = 2, + HSA_DBG_NUM_WAVEMSG, + HSA_DBG_MAX_WAVEMSG = 0xFFFFFFFF +}; + +enum HSA_DBG_WATCH_MODE { + HSA_DBG_WATCH_READ = 0, /* Read operations only */ + HSA_DBG_WATCH_NONREAD = 1, /* Write or Atomic operations only */ + HSA_DBG_WATCH_ATOMIC = 2, /* Atomic Operations only */ + HSA_DBG_WATCH_ALL = 3, /* Read, Write or Atomic operations */ + HSA_DBG_WATCH_NUM, + HSA_DBG_WATCH_SIZE = 0xFFFFFFFF +}; + +/* This structure is hardware specific and may change in the future */ +struct HsaDbgWaveMsgAMDGen2 { + union { + struct ui32 { + uint32_t UserData:8; /* user data */ + uint32_t ShaderArray:1; /* Shader array */ + uint32_t Priv:1; /* Privileged */ + uint32_t Reserved0:4; /* This field is reserved, + should be 0 */ + uint32_t WaveId:4; /* wave id */ + uint32_t SIMD:2; /* SIMD id */ + uint32_t HSACU:4; /* Compute unit */ + uint32_t ShaderEngine:2;/* Shader engine */ + uint32_t MessageType:2; /* see HSA_DBG_WAVEMSG_TYPE */ + uint32_t Reserved1:4; /* This field is reserved, + should be 0 */ + } ui32; + uint32_t Value; + }; + uint32_t Reserved2; +}; + +union HsaDbgWaveMessageAMD { + struct HsaDbgWaveMsgAMDGen2 WaveMsgInfoGen2; + /* for future HsaDbgWaveMsgAMDGen3; */ +}; + +struct HsaDbgWaveMessage { + void *MemoryVA; /* ptr to associated host-accessible data */ + union HsaDbgWaveMessageAMD DbgWaveMsg; +}; + +/* + * TODO: This definitions to be MOVED to kfd_event, once it is implemented. + * + * HSA sync primitive, Event and HW Exception notification API definitions. + * The API functions allow the runtime to define a so-called sync-primitive, + * a SW object combining a user-mode provided "syncvar" and a scheduler event + * that can be signaled through a defined GPU interrupt. A syncvar is + * a process virtual memory location of a certain size that can be accessed + * by CPU and GPU shader code within the process to set and query the content + * within that memory. The definition of the content is determined by the HSA + * runtime and potentially GPU shader code interfacing with the HSA runtime. + * The syncvar values may be commonly written through an PM4 WRITE_DATA packet + * in the user mode instruction stream. The OS scheduler event is typically + * associated and signaled by an interrupt issued by the GPU, but other HSA + * system interrupt conditions from other HW (e.g. IOMMUv2) may be surfaced + * by the KFD by this mechanism, too. */ + +/* these are the new definitions for events */ +enum HSA_EVENTTYPE { + HSA_EVENTTYPE_SIGNAL = 0, /* user-mode generated GPU signal */ + HSA_EVENTTYPE_NODECHANGE = 1, /* HSA node change (attach/detach) */ + HSA_EVENTTYPE_DEVICESTATECHANGE = 2, /* HSA device state change + (start/stop) */ + HSA_EVENTTYPE_HW_EXCEPTION = 3, /* GPU shader exception event */ + HSA_EVENTTYPE_SYSTEM_EVENT = 4, /* GPU SYSCALL with parameter info */ + HSA_EVENTTYPE_DEBUG_EVENT = 5, /* GPU signal for debugging */ + HSA_EVENTTYPE_PROFILE_EVENT = 6,/* GPU signal for profiling */ + HSA_EVENTTYPE_QUEUE_EVENT = 7, /* GPU signal queue idle state + (EOP pm4) */ + /* ... */ + HSA_EVENTTYPE_MAXID, + HSA_EVENTTYPE_TYPE_SIZE = 0xFFFFFFFF +}; + +/* Sub-definitions for various event types: Syncvar */ +struct HsaSyncVar { + union SyncVar { + void *UserData; /* pointer to user mode data */ + uint64_t UserDataPtrValue; /* 64bit compatibility of value */ + } SyncVar; + uint64_t SyncVarSize; +}; + +/* Sub-definitions for various event types: NodeChange */ + +enum HSA_EVENTTYPE_NODECHANGE_FLAGS { + HSA_EVENTTYPE_NODECHANGE_ADD = 0, + HSA_EVENTTYPE_NODECHANGE_REMOVE = 1, + HSA_EVENTTYPE_NODECHANGE_SIZE = 0xFFFFFFFF +}; + +struct HsaNodeChange { + /* HSA node added/removed on the platform */ + enum HSA_EVENTTYPE_NODECHANGE_FLAGS Flags; +}; + +/* Sub-definitions for various event types: DeviceStateChange */ +enum HSA_EVENTTYPE_DEVICESTATECHANGE_FLAGS { + /* device started (and available) */ + HSA_EVENTTYPE_DEVICESTATUSCHANGE_START = 0, + /* device stopped (i.e. unavailable) */ + HSA_EVENTTYPE_DEVICESTATUSCHANGE_STOP = 1, + HSA_EVENTTYPE_DEVICESTATUSCHANGE_SIZE = 0xFFFFFFFF +}; + +enum HSA_DEVICE { + HSA_DEVICE_CPU = 0, + HSA_DEVICE_GPU = 1, + MAX_HSA_DEVICE = 2 +}; + +struct HsaDeviceStateChange { + uint32_t NodeId; /* F-NUMA node that contains the device */ + enum HSA_DEVICE Device; /* device type: GPU or CPU */ + enum HSA_EVENTTYPE_DEVICESTATECHANGE_FLAGS Flags; /* event flags */ +}; + +struct HsaEventData { + enum HSA_EVENTTYPE EventType; /* event type */ + union EventData { + /* + * return data associated with HSA_EVENTTYPE_SIGNAL + * and other events + */ + struct HsaSyncVar SyncVar; + + /* data associated with HSA_EVENTTYPE_NODE_CHANGE */ + struct HsaNodeChange NodeChangeState; + + /* data associated with HSA_EVENTTYPE_DEVICE_STATE_CHANGE */ + struct HsaDeviceStateChange DeviceState; + } EventData; + + /* the following data entries are internal to the KFD & thunk itself */ + + /* internal thunk store for Event data (OsEventHandle) */ + uint64_t HWData1; + /* internal thunk store for Event data (HWAddress) */ + uint64_t HWData2; + /* internal thunk store for Event data (HWData) */ + uint32_t HWData3; +}; + +struct HsaEventDescriptor { + /* event type to allocate */ + enum HSA_EVENTTYPE EventType; + /* H-NUMA node containing GPU device that is event source */ + uint32_t NodeId; + /* pointer to user mode syncvar data, syncvar->UserDataPtrValue + * may be NULL + */ + struct HsaSyncVar SyncVar; +}; + +struct HsaEvent { + uint32_t EventId; + struct HsaEventData EventData; +}; + +#pragma pack(pop) + +enum DBGDEV_TYPE { + DBGDEV_TYPE_ILLEGAL = 0, + DBGDEV_TYPE_NODIQ = 1, + DBGDEV_TYPE_DIQ = 2, + DBGDEV_TYPE_TEST = 3 +}; + +struct dbg_address_watch_info { + struct kfd_process *process; + enum HSA_DBG_WATCH_MODE *watch_mode; + uint64_t *watch_address; + uint64_t *watch_mask; + struct HsaEvent *watch_event; + uint32_t num_watch_points; +}; + +struct dbg_wave_control_info { + struct kfd_process *process; + uint32_t trapId; + enum HSA_DBG_WAVEOP operand; + enum HSA_DBG_WAVEMODE mode; + struct HsaDbgWaveMessage dbgWave_msg; +}; + +struct kfd_dbgdev { + + /* The device that owns this data. */ + struct kfd_dev *dev; + + /* kernel queue for DIQ */ + struct kernel_queue *kq; + + /* a pointer to the pqm of the calling process */ + struct process_queue_manager *pqm; + + /* type of debug device ( DIQ, non DIQ, etc. ) */ + enum DBGDEV_TYPE type; + + /* virtualized function pointers to device dbg */ + int (*dbgdev_register)(struct kfd_dbgdev *dbgdev); + int (*dbgdev_unregister)(struct kfd_dbgdev *dbgdev); + int (*dbgdev_address_watch)(struct kfd_dbgdev *dbgdev, + struct dbg_address_watch_info *adw_info); + int (*dbgdev_wave_control)(struct kfd_dbgdev *dbgdev, + struct dbg_wave_control_info *wac_info); + +}; + +struct kfd_dbgmgr { + unsigned int pasid; + struct kfd_dev *dev; + struct kfd_dbgdev *dbgdev; +}; + +/* prototypes for debug manager functions */ +struct mutex *kfd_get_dbgmgr_mutex(void); +void kfd_dbgmgr_destroy(struct kfd_dbgmgr *pmgr); +bool kfd_dbgmgr_create(struct kfd_dbgmgr **ppmgr, struct kfd_dev *pdev); +long kfd_dbgmgr_register(struct kfd_dbgmgr *pmgr, struct kfd_process *p); +long kfd_dbgmgr_unregister(struct kfd_dbgmgr *pmgr, struct kfd_process *p); +long kfd_dbgmgr_wave_control(struct kfd_dbgmgr *pmgr, + struct dbg_wave_control_info *wac_info); +long kfd_dbgmgr_address_watch(struct kfd_dbgmgr *pmgr, + struct dbg_address_watch_info *adw_info); +#endif /* KFD_DBGMGR_H_ */ diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index ca7f2d3af2ff..75312c82969f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -33,14 +33,21 @@ static const struct kfd_device_info kaveri_device_info = { .asic_family = CHIP_KAVERI, .max_pasid_bits = 16, + /* max num of queues for KV.TODO should be a dynamic value */ + .max_no_of_hqd = 24, .ih_ring_entry_size = 4 * sizeof(uint32_t), + .event_interrupt_class = &event_interrupt_class_cik, + .num_of_watch_points = 4, .mqd_size_aligned = MQD_SIZE_ALIGNED }; static const struct kfd_device_info carrizo_device_info = { .asic_family = CHIP_CARRIZO, .max_pasid_bits = 16, + /* max num of queues for CZ.TODO should be a dynamic value */ + .max_no_of_hqd = 24, .ih_ring_entry_size = 4 * sizeof(uint32_t), + .event_interrupt_class = &event_interrupt_class_cik, .num_of_watch_points = 4, .mqd_size_aligned = MQD_SIZE_ALIGNED }; @@ -181,6 +188,32 @@ static void iommu_pasid_shutdown_callback(struct pci_dev *pdev, int pasid) kfd_unbind_process_from_device(dev, pasid); } +/* + * This function called by IOMMU driver on PPR failure + */ +static int iommu_invalid_ppr_cb(struct pci_dev *pdev, int pasid, + unsigned long address, u16 flags) +{ + struct kfd_dev *dev; + + dev_warn(kfd_device, + "Invalid PPR device %x:%x.%x pasid %d address 0x%lX flags 0x%X", + PCI_BUS_NUM(pdev->devfn), + PCI_SLOT(pdev->devfn), + PCI_FUNC(pdev->devfn), + pasid, + address, + flags); + + dev = kfd_device_by_pci_dev(pdev); + BUG_ON(dev == NULL); + + kfd_signal_iommu_event(dev, pasid, address, + flags & PPR_FAULT_WRITE, flags & PPR_FAULT_EXEC); + + return AMD_IOMMU_INV_PRI_RSP_INVALID; +} + bool kgd2kfd_device_init(struct kfd_dev *kfd, const struct kgd2kfd_shared_resources *gpu_resources) { @@ -235,6 +268,13 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, goto kfd_topology_add_device_error; } + if (kfd_interrupt_init(kfd)) { + dev_err(kfd_device, + "Error initializing interrupts for device (%x:%x)\n", + kfd->pdev->vendor, kfd->pdev->device); + goto kfd_interrupt_error; + } + if (!device_iommu_pasid_init(kfd)) { dev_err(kfd_device, "Error initializing iommuv2 for device (%x:%x)\n", @@ -243,6 +283,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, } amd_iommu_set_invalidate_ctx_cb(kfd->pdev, iommu_pasid_shutdown_callback); + amd_iommu_set_invalid_ppr_cb(kfd->pdev, iommu_invalid_ppr_cb); kfd->dqm = device_queue_manager_init(kfd); if (!kfd->dqm) { @@ -259,6 +300,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, goto dqm_start_error; } + kfd->dbgmgr = NULL; + kfd->init_complete = true; dev_info(kfd_device, "added device (%x:%x)\n", kfd->pdev->vendor, kfd->pdev->device); @@ -273,6 +316,8 @@ dqm_start_error: device_queue_manager_error: amd_iommu_free_device(kfd->pdev); device_iommu_pasid_error: + kfd_interrupt_exit(kfd); +kfd_interrupt_error: kfd_topology_remove_device(kfd); kfd_topology_add_device_error: kfd_gtt_sa_fini(kfd); @@ -290,6 +335,7 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd) if (kfd->init_complete) { device_queue_manager_uninit(kfd->dqm); amd_iommu_free_device(kfd->pdev); + kfd_interrupt_exit(kfd); kfd_topology_remove_device(kfd); kfd_gtt_sa_fini(kfd); kfd->kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem); @@ -305,6 +351,7 @@ void kgd2kfd_suspend(struct kfd_dev *kfd) if (kfd->init_complete) { kfd->dqm->ops.stop(kfd->dqm); amd_iommu_set_invalidate_ctx_cb(kfd->pdev, NULL); + amd_iommu_set_invalid_ppr_cb(kfd->pdev, NULL); amd_iommu_free_device(kfd->pdev); } } @@ -324,6 +371,7 @@ int kgd2kfd_resume(struct kfd_dev *kfd) return -ENXIO; amd_iommu_set_invalidate_ctx_cb(kfd->pdev, iommu_pasid_shutdown_callback); + amd_iommu_set_invalid_ppr_cb(kfd->pdev, iommu_invalid_ppr_cb); kfd->dqm->ops.start(kfd->dqm); } @@ -333,7 +381,17 @@ int kgd2kfd_resume(struct kfd_dev *kfd) /* This is called directly from KGD at ISR. */ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry) { - /* Process interrupts / schedule work as necessary */ + if (!kfd->init_complete) + return; + + spin_lock(&kfd->interrupt_lock); + + if (kfd->interrupts_active + && interrupt_is_wanted(kfd, ih_ring_entry) + && enqueue_ih_ring_entry(kfd, ih_ring_entry)) + schedule_work(&kfd->interrupt_work); + + spin_unlock(&kfd->interrupt_lock); } static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 596ee5cd3b84..4bb7f4223762 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -45,7 +45,8 @@ static int create_compute_queue_nocpsch(struct device_queue_manager *dqm, struct qcm_process_device *qpd); static int execute_queues_cpsch(struct device_queue_manager *dqm, bool lock); -static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock); +static int destroy_queues_cpsch(struct device_queue_manager *dqm, + bool preempt_static_queues, bool lock); static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, struct queue *q, @@ -523,6 +524,17 @@ int init_pipelines(struct device_queue_manager *dqm, return 0; } +static void init_interrupts(struct device_queue_manager *dqm) +{ + unsigned int i; + + BUG_ON(dqm == NULL); + + for (i = 0 ; i < get_pipes_num(dqm) ; i++) + dqm->dev->kfd2kgd->init_interrupts(dqm->dev->kgd, + i + get_first_pipe(dqm)); +} + static int init_scheduler(struct device_queue_manager *dqm) { int retval; @@ -582,6 +594,7 @@ static void uninitialize_nocpsch(struct device_queue_manager *dqm) static int start_nocpsch(struct device_queue_manager *dqm) { + init_interrupts(dqm); return 0; } @@ -615,19 +628,6 @@ static void deallocate_sdma_queue(struct device_queue_manager *dqm, set_bit(sdma_queue_id, (unsigned long *)&dqm->sdma_bitmap); } -static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, - struct qcm_process_device *qpd) -{ - uint32_t value = SDMA_ATC; - - if (q->process->is_32bit_user_mode) - value |= SDMA_VA_PTR32 | get_sh_mem_bases_32(qpd_to_pdd(qpd)); - else - value |= SDMA_VA_SHARED_BASE(get_sh_mem_bases_nybble_64( - qpd_to_pdd(qpd))); - q->properties.sdma_vm_addr = value; -} - static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, struct queue *q, struct qcm_process_device *qpd) @@ -650,7 +650,7 @@ static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, pr_debug(" sdma queue id: %d\n", q->properties.sdma_queue_id); pr_debug(" sdma engine id: %d\n", q->properties.sdma_engine_id); - init_sdma_vm(dqm, q, qpd); + dqm->ops_asic_specific.init_sdma_vm(dqm, q, qpd); retval = mqd->init_mqd(mqd, &q->mqd, &q->mqd_mem_obj, &q->gart_mqd_addr, &q->properties); if (retval != 0) { @@ -751,6 +751,9 @@ static int start_cpsch(struct device_queue_manager *dqm) dqm->fence_addr = dqm->fence_mem->cpu_ptr; dqm->fence_gpu_addr = dqm->fence_mem->gpu_addr; + + init_interrupts(dqm); + list_for_each_entry(node, &dqm->queues, list) if (node->qpd->pqm->process && dqm->dev) kfd_bind_process_to_device(dqm->dev, @@ -773,7 +776,7 @@ static int stop_cpsch(struct device_queue_manager *dqm) BUG_ON(!dqm); - destroy_queues_cpsch(dqm, true); + destroy_queues_cpsch(dqm, true, true); list_for_each_entry(node, &dqm->queues, list) { pdd = qpd_to_pdd(node->qpd); @@ -827,7 +830,8 @@ static void destroy_kernel_queue_cpsch(struct device_queue_manager *dqm, pr_debug("kfd: In %s\n", __func__); mutex_lock(&dqm->lock); - destroy_queues_cpsch(dqm, false); + /* here we actually preempt the DIQ */ + destroy_queues_cpsch(dqm, true, false); list_del(&kq->list); dqm->queue_count--; qpd->is_debug = false; @@ -883,8 +887,7 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q, return -ENOMEM; } - init_sdma_vm(dqm, q, qpd); - + dqm->ops_asic_specific.init_sdma_vm(dqm, q, qpd); retval = mqd->init_mqd(mqd, &q->mqd, &q->mqd_mem_obj, &q->gart_mqd_addr, &q->properties); if (retval != 0) @@ -912,7 +915,7 @@ out: return retval; } -static int amdkfd_fence_wait_timeout(unsigned int *fence_addr, +int amdkfd_fence_wait_timeout(unsigned int *fence_addr, unsigned int fence_value, unsigned long timeout) { @@ -934,13 +937,16 @@ static int destroy_sdma_queues(struct device_queue_manager *dqm, unsigned int sdma_engine) { return pm_send_unmap_queue(&dqm->packets, KFD_QUEUE_TYPE_SDMA, - KFD_PREEMPT_TYPE_FILTER_ALL_QUEUES, 0, false, + KFD_PREEMPT_TYPE_FILTER_DYNAMIC_QUEUES, 0, false, sdma_engine); } -static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock) +static int destroy_queues_cpsch(struct device_queue_manager *dqm, + bool preempt_static_queues, bool lock) { int retval; + enum kfd_preempt_type_filter preempt_type; + struct kfd_process_device *pdd; BUG_ON(!dqm); @@ -959,8 +965,12 @@ static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock) destroy_sdma_queues(dqm, 1); } + preempt_type = preempt_static_queues ? + KFD_PREEMPT_TYPE_FILTER_ALL_QUEUES : + KFD_PREEMPT_TYPE_FILTER_DYNAMIC_QUEUES; + retval = pm_send_unmap_queue(&dqm->packets, KFD_QUEUE_TYPE_COMPUTE, - KFD_PREEMPT_TYPE_FILTER_ALL_QUEUES, 0, false, 0); + preempt_type, 0, false, 0); if (retval != 0) goto out; @@ -968,8 +978,14 @@ static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock) pm_send_query_status(&dqm->packets, dqm->fence_gpu_addr, KFD_FENCE_COMPLETED); /* should be timed out */ - amdkfd_fence_wait_timeout(dqm->fence_addr, KFD_FENCE_COMPLETED, + retval = amdkfd_fence_wait_timeout(dqm->fence_addr, KFD_FENCE_COMPLETED, QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS); + if (retval != 0) { + pdd = kfd_get_process_device_data(dqm->dev, + kfd_get_process(current)); + pdd->reset_wavefronts = true; + goto out; + } pm_release_ib(&dqm->packets); dqm->active_runlist = false; @@ -988,7 +1004,7 @@ static int execute_queues_cpsch(struct device_queue_manager *dqm, bool lock) if (lock) mutex_lock(&dqm->lock); - retval = destroy_queues_cpsch(dqm, false); + retval = destroy_queues_cpsch(dqm, false, false); if (retval != 0) { pr_err("kfd: the cp might be in an unrecoverable state due to an unsuccessful queues preemption"); goto out; @@ -1023,13 +1039,27 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm, { int retval; struct mqd_manager *mqd; + bool preempt_all_queues; BUG_ON(!dqm || !qpd || !q); + preempt_all_queues = false; + retval = 0; /* remove queue from list to prevent rescheduling after preemption */ mutex_lock(&dqm->lock); + + if (qpd->is_debug) { + /* + * error, currently we do not allow to destroy a queue + * of a currently debugged process + */ + retval = -EBUSY; + goto failed_try_destroy_debugged_queue; + + } + mqd = dqm->ops.get_mqd_manager(dqm, get_mqd_type_from_queue_type(q->properties.type)); if (!mqd) { @@ -1061,6 +1091,8 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm, return 0; failed: +failed_try_destroy_debugged_queue: + mutex_unlock(&dqm->lock); return retval; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h index 488f51d19427..ec4036a09f3e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h @@ -88,9 +88,11 @@ struct device_queue_manager_ops { struct queue *q, struct qcm_process_device *qpd, int *allocate_vmid); + int (*destroy_queue)(struct device_queue_manager *dqm, struct qcm_process_device *qpd, struct queue *q); + int (*update_queue)(struct device_queue_manager *dqm, struct queue *q); @@ -100,8 +102,10 @@ struct device_queue_manager_ops { int (*register_process)(struct device_queue_manager *dqm, struct qcm_process_device *qpd); + int (*unregister_process)(struct device_queue_manager *dqm, struct qcm_process_device *qpd); + int (*initialize)(struct device_queue_manager *dqm); int (*start)(struct device_queue_manager *dqm); int (*stop)(struct device_queue_manager *dqm); @@ -109,15 +113,32 @@ struct device_queue_manager_ops { int (*create_kernel_queue)(struct device_queue_manager *dqm, struct kernel_queue *kq, struct qcm_process_device *qpd); + void (*destroy_kernel_queue)(struct device_queue_manager *dqm, struct kernel_queue *kq, struct qcm_process_device *qpd); + + bool (*set_cache_memory_policy)(struct device_queue_manager *dqm, + struct qcm_process_device *qpd, + enum cache_policy default_policy, + enum cache_policy alternate_policy, + void __user *alternate_aperture_base, + uint64_t alternate_aperture_size); +}; + +struct device_queue_manager_asic_ops { + int (*register_process)(struct device_queue_manager *dqm, + struct qcm_process_device *qpd); + int (*initialize)(struct device_queue_manager *dqm); bool (*set_cache_memory_policy)(struct device_queue_manager *dqm, struct qcm_process_device *qpd, enum cache_policy default_policy, enum cache_policy alternate_policy, void __user *alternate_aperture_base, uint64_t alternate_aperture_size); + void (*init_sdma_vm)(struct device_queue_manager *dqm, + struct queue *q, + struct qcm_process_device *qpd); }; /** @@ -134,7 +155,7 @@ struct device_queue_manager_ops { struct device_queue_manager { struct device_queue_manager_ops ops; - struct device_queue_manager_ops ops_asic_specific; + struct device_queue_manager_asic_ops ops_asic_specific; struct mqd_manager *mqds[KFD_MQD_TYPE_MAX]; struct packet_manager packets; @@ -157,8 +178,8 @@ struct device_queue_manager { bool active_runlist; }; -void device_queue_manager_init_cik(struct device_queue_manager_ops *ops); -void device_queue_manager_init_vi(struct device_queue_manager_ops *ops); +void device_queue_manager_init_cik(struct device_queue_manager_asic_ops *ops); +void device_queue_manager_init_vi(struct device_queue_manager_asic_ops *ops); void program_sh_mem_settings(struct device_queue_manager *dqm, struct qcm_process_device *qpd); int init_pipelines(struct device_queue_manager *dqm, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c index 5469efe0523e..9ce8a20a7aff 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c @@ -33,12 +33,15 @@ static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm, static int register_process_cik(struct device_queue_manager *dqm, struct qcm_process_device *qpd); static int initialize_cpsch_cik(struct device_queue_manager *dqm); +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd); -void device_queue_manager_init_cik(struct device_queue_manager_ops *ops) +void device_queue_manager_init_cik(struct device_queue_manager_asic_ops *ops) { ops->set_cache_memory_policy = set_cache_memory_policy_cik; ops->register_process = register_process_cik; ops->initialize = initialize_cpsch_cik; + ops->init_sdma_vm = init_sdma_vm; } static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble) @@ -129,6 +132,19 @@ static int register_process_cik(struct device_queue_manager *dqm, return 0; } +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd) +{ + uint32_t value = SDMA_ATC; + + if (q->process->is_32bit_user_mode) + value |= SDMA_VA_PTR32 | get_sh_mem_bases_32(qpd_to_pdd(qpd)); + else + value |= SDMA_VA_SHARED_BASE(get_sh_mem_bases_nybble_64( + qpd_to_pdd(qpd))); + q->properties.sdma_vm_addr = value; +} + static int initialize_cpsch_cik(struct device_queue_manager *dqm) { return init_pipelines(dqm, get_pipes_num(dqm), get_first_pipe(dqm)); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c index 20553dcd257d..4c15212a3899 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c @@ -32,14 +32,17 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm, static int register_process_vi(struct device_queue_manager *dqm, struct qcm_process_device *qpd); static int initialize_cpsch_vi(struct device_queue_manager *dqm); +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd); -void device_queue_manager_init_vi(struct device_queue_manager_ops *ops) +void device_queue_manager_init_vi(struct device_queue_manager_asic_ops *ops) { pr_warn("amdkfd: VI DQM is not currently supported\n"); ops->set_cache_memory_policy = set_cache_memory_policy_vi; ops->register_process = register_process_vi; ops->initialize = initialize_cpsch_vi; + ops->init_sdma_vm = init_sdma_vm; } static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm, @@ -58,6 +61,11 @@ static int register_process_vi(struct device_queue_manager *dqm, return -1; } +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd) +{ +} + static int initialize_cpsch_vi(struct device_queue_manager *dqm) { return 0; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c b/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c index 17e56dcc8540..e621eba63126 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c @@ -142,14 +142,13 @@ int kfd_doorbell_mmap(struct kfd_process *process, struct vm_area_struct *vma) vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); - pr_debug("kfd: mapping doorbell page in kfd_doorbell_mmap\n" - " target user address == 0x%08llX\n" - " physical address == 0x%08llX\n" - " vm_flags == 0x%04lX\n" - " size == 0x%04lX\n", - (unsigned long long) vma->vm_start, address, vma->vm_flags, - doorbell_process_allocation()); - + pr_debug("mapping doorbell page:\n"); + pr_debug(" target user address == 0x%08llX\n", + (unsigned long long) vma->vm_start); + pr_debug(" physical address == 0x%08llX\n", address); + pr_debug(" vm_flags == 0x%04lX\n", vma->vm_flags); + pr_debug(" size == 0x%04lX\n", + doorbell_process_allocation()); return io_remap_pfn_range(vma, vma->vm_start, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c new file mode 100644 index 000000000000..b6e28dcaea1d --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c @@ -0,0 +1,969 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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 <linux/mm_types.h> +#include <linux/slab.h> +#include <linux/types.h> +#include <linux/sched.h> +#include <linux/uaccess.h> +#include <linux/mm.h> +#include <linux/mman.h> +#include <linux/memory.h> +#include "kfd_priv.h" +#include "kfd_events.h" +#include <linux/device.h> + +/* + * A task can only be on a single wait_queue at a time, but we need to support + * waiting on multiple events (any/all). + * Instead of each event simply having a wait_queue with sleeping tasks, it + * has a singly-linked list of tasks. + * A thread that wants to sleep creates an array of these, one for each event + * and adds one to each event's waiter chain. + */ +struct kfd_event_waiter { + struct list_head waiters; + struct task_struct *sleeping_task; + + /* Transitions to true when the event this belongs to is signaled. */ + bool activated; + + /* Event */ + struct kfd_event *event; + uint32_t input_index; +}; + +/* + * Over-complicated pooled allocator for event notification slots. + * + * Each signal event needs a 64-bit signal slot where the signaler will write + * a 1 before sending an interrupt.l (This is needed because some interrupts + * do not contain enough spare data bits to identify an event.) + * We get whole pages from vmalloc and map them to the process VA. + * Individual signal events are then allocated a slot in a page. + */ + +struct signal_page { + struct list_head event_pages; /* kfd_process.signal_event_pages */ + uint64_t *kernel_address; + uint64_t __user *user_address; + uint32_t page_index; /* Index into the mmap aperture. */ + unsigned int free_slots; + unsigned long used_slot_bitmap[0]; +}; + +#define SLOTS_PER_PAGE KFD_SIGNAL_EVENT_LIMIT +#define SLOT_BITMAP_SIZE BITS_TO_LONGS(SLOTS_PER_PAGE) +#define BITS_PER_PAGE (ilog2(SLOTS_PER_PAGE)+1) +#define SIGNAL_PAGE_SIZE (sizeof(struct signal_page) + \ + SLOT_BITMAP_SIZE * sizeof(long)) + +/* + * For signal events, the event ID is used as the interrupt user data. + * For SQ s_sendmsg interrupts, this is limited to 8 bits. + */ + +#define INTERRUPT_DATA_BITS 8 +#define SIGNAL_EVENT_ID_SLOT_SHIFT 0 + +static uint64_t *page_slots(struct signal_page *page) +{ + return page->kernel_address; +} + +static bool allocate_free_slot(struct kfd_process *process, + struct signal_page **out_page, + unsigned int *out_slot_index) +{ + struct signal_page *page; + + list_for_each_entry(page, &process->signal_event_pages, event_pages) { + if (page->free_slots > 0) { + unsigned int slot = + find_first_zero_bit(page->used_slot_bitmap, + SLOTS_PER_PAGE); + + __set_bit(slot, page->used_slot_bitmap); + page->free_slots--; + + page_slots(page)[slot] = UNSIGNALED_EVENT_SLOT; + + *out_page = page; + *out_slot_index = slot; + + pr_debug("allocated event signal slot in page %p, slot %d\n", + page, slot); + + return true; + } + } + + pr_debug("No free event signal slots were found for process %p\n", + process); + + return false; +} + +#define list_tail_entry(head, type, member) \ + list_entry((head)->prev, type, member) + +static bool allocate_signal_page(struct file *devkfd, struct kfd_process *p) +{ + void *backing_store; + struct signal_page *page; + + page = kzalloc(SIGNAL_PAGE_SIZE, GFP_KERNEL); + if (!page) + goto fail_alloc_signal_page; + + page->free_slots = SLOTS_PER_PAGE; + + backing_store = (void *) __get_free_pages(GFP_KERNEL | __GFP_ZERO, + get_order(KFD_SIGNAL_EVENT_LIMIT * 8)); + if (!backing_store) + goto fail_alloc_signal_store; + + /* prevent user-mode info leaks */ + memset(backing_store, (uint8_t) UNSIGNALED_EVENT_SLOT, + KFD_SIGNAL_EVENT_LIMIT * 8); + + page->kernel_address = backing_store; + + if (list_empty(&p->signal_event_pages)) + page->page_index = 0; + else + page->page_index = list_tail_entry(&p->signal_event_pages, + struct signal_page, + event_pages)->page_index + 1; + + pr_debug("allocated new event signal page at %p, for process %p\n", + page, p); + pr_debug("page index is %d\n", page->page_index); + + list_add(&page->event_pages, &p->signal_event_pages); + + return true; + +fail_alloc_signal_store: + kfree(page); +fail_alloc_signal_page: + return false; +} + +static bool allocate_event_notification_slot(struct file *devkfd, + struct kfd_process *p, + struct signal_page **page, + unsigned int *signal_slot_index) +{ + bool ret; + + ret = allocate_free_slot(p, page, signal_slot_index); + if (ret == false) { + ret = allocate_signal_page(devkfd, p); + if (ret == true) + ret = allocate_free_slot(p, page, signal_slot_index); + } + + return ret; +} + +/* Assumes that the process's event_mutex is locked. */ +static void release_event_notification_slot(struct signal_page *page, + size_t slot_index) +{ + __clear_bit(slot_index, page->used_slot_bitmap); + page->free_slots++; + + /* We don't free signal pages, they are retained by the process + * and reused until it exits. */ +} + +static struct signal_page *lookup_signal_page_by_index(struct kfd_process *p, + unsigned int page_index) +{ + struct signal_page *page; + + /* + * This is safe because we don't delete signal pages until the + * process exits. + */ + list_for_each_entry(page, &p->signal_event_pages, event_pages) + if (page->page_index == page_index) + return page; + + return NULL; +} + +/* + * Assumes that p->event_mutex is held and of course that p is not going + * away (current or locked). + */ +static struct kfd_event *lookup_event_by_id(struct kfd_process *p, uint32_t id) +{ + struct kfd_event *ev; + + hash_for_each_possible(p->events, ev, events, id) + if (ev->event_id == id) + return ev; + + return NULL; +} + +static u32 make_signal_event_id(struct signal_page *page, + unsigned int signal_slot_index) +{ + return page->page_index | + (signal_slot_index << SIGNAL_EVENT_ID_SLOT_SHIFT); +} + +/* + * Produce a kfd event id for a nonsignal event. + * These are arbitrary numbers, so we do a sequential search through + * the hash table for an unused number. + */ +static u32 make_nonsignal_event_id(struct kfd_process *p) +{ + u32 id; + + for (id = p->next_nonsignal_event_id; + id < KFD_LAST_NONSIGNAL_EVENT_ID && + lookup_event_by_id(p, id) != NULL; + id++) + ; + + if (id < KFD_LAST_NONSIGNAL_EVENT_ID) { + + /* + * What if id == LAST_NONSIGNAL_EVENT_ID - 1? + * Then next_nonsignal_event_id = LAST_NONSIGNAL_EVENT_ID so + * the first loop fails immediately and we proceed with the + * wraparound loop below. + */ + p->next_nonsignal_event_id = id + 1; + + return id; + } + + for (id = KFD_FIRST_NONSIGNAL_EVENT_ID; + id < KFD_LAST_NONSIGNAL_EVENT_ID && + lookup_event_by_id(p, id) != NULL; + id++) + ; + + + if (id < KFD_LAST_NONSIGNAL_EVENT_ID) { + p->next_nonsignal_event_id = id + 1; + return id; + } + + p->next_nonsignal_event_id = KFD_FIRST_NONSIGNAL_EVENT_ID; + return 0; +} + +static struct kfd_event *lookup_event_by_page_slot(struct kfd_process *p, + struct signal_page *page, + unsigned int signal_slot) +{ + return lookup_event_by_id(p, make_signal_event_id(page, signal_slot)); +} + +static int create_signal_event(struct file *devkfd, + struct kfd_process *p, + struct kfd_event *ev) +{ + if (p->signal_event_count == KFD_SIGNAL_EVENT_LIMIT) { + pr_warn("amdkfd: Signal event wasn't created because limit was reached\n"); + return -ENOMEM; + } + + if (!allocate_event_notification_slot(devkfd, p, &ev->signal_page, + &ev->signal_slot_index)) { + pr_warn("amdkfd: Signal event wasn't created because out of kernel memory\n"); + return -ENOMEM; + } + + p->signal_event_count++; + + ev->user_signal_address = + &ev->signal_page->user_address[ev->signal_slot_index]; + + ev->event_id = make_signal_event_id(ev->signal_page, + ev->signal_slot_index); + + pr_debug("signal event number %zu created with id %d, address %p\n", + p->signal_event_count, ev->event_id, + ev->user_signal_address); + + pr_debug("signal event number %zu created with id %d, address %p\n", + p->signal_event_count, ev->event_id, + ev->user_signal_address); + + return 0; +} + +/* + * No non-signal events are supported yet. + * We create them as events that never signal. + * Set event calls from user-mode are failed. + */ +static int create_other_event(struct kfd_process *p, struct kfd_event *ev) +{ + ev->event_id = make_nonsignal_event_id(p); + if (ev->event_id == 0) + return -ENOMEM; + + return 0; +} + +void kfd_event_init_process(struct kfd_process *p) +{ + mutex_init(&p->event_mutex); + hash_init(p->events); + INIT_LIST_HEAD(&p->signal_event_pages); + p->next_nonsignal_event_id = KFD_FIRST_NONSIGNAL_EVENT_ID; + p->signal_event_count = 0; +} + +static void destroy_event(struct kfd_process *p, struct kfd_event *ev) +{ + if (ev->signal_page != NULL) { + release_event_notification_slot(ev->signal_page, + ev->signal_slot_index); + p->signal_event_count--; + } + + /* + * Abandon the list of waiters. Individual waiting threads will + * clean up their own data. + */ + list_del(&ev->waiters); + + hash_del(&ev->events); + kfree(ev); +} + +static void destroy_events(struct kfd_process *p) +{ + struct kfd_event *ev; + struct hlist_node *tmp; + unsigned int hash_bkt; + + hash_for_each_safe(p->events, hash_bkt, tmp, ev, events) + destroy_event(p, ev); +} + +/* + * We assume that the process is being destroyed and there is no need to + * unmap the pages or keep bookkeeping data in order. + */ +static void shutdown_signal_pages(struct kfd_process *p) +{ + struct signal_page *page, *tmp; + + list_for_each_entry_safe(page, tmp, &p->signal_event_pages, + event_pages) { + free_pages((unsigned long)page->kernel_address, + get_order(KFD_SIGNAL_EVENT_LIMIT * 8)); + kfree(page); + } +} + +void kfd_event_free_process(struct kfd_process *p) +{ + destroy_events(p); + shutdown_signal_pages(p); +} + +static bool event_can_be_gpu_signaled(const struct kfd_event *ev) +{ + return ev->type == KFD_EVENT_TYPE_SIGNAL || + ev->type == KFD_EVENT_TYPE_DEBUG; +} + +static bool event_can_be_cpu_signaled(const struct kfd_event *ev) +{ + return ev->type == KFD_EVENT_TYPE_SIGNAL; +} + +int kfd_event_create(struct file *devkfd, struct kfd_process *p, + uint32_t event_type, bool auto_reset, uint32_t node_id, + uint32_t *event_id, uint32_t *event_trigger_data, + uint64_t *event_page_offset, uint32_t *event_slot_index) +{ + int ret = 0; + struct kfd_event *ev = kzalloc(sizeof(*ev), GFP_KERNEL); + + if (!ev) + return -ENOMEM; + + ev->type = event_type; + ev->auto_reset = auto_reset; + ev->signaled = false; + + INIT_LIST_HEAD(&ev->waiters); + + *event_page_offset = 0; + + mutex_lock(&p->event_mutex); + + switch (event_type) { + case KFD_EVENT_TYPE_SIGNAL: + case KFD_EVENT_TYPE_DEBUG: + ret = create_signal_event(devkfd, p, ev); + if (!ret) { + *event_page_offset = (ev->signal_page->page_index | + KFD_MMAP_EVENTS_MASK); + *event_page_offset <<= PAGE_SHIFT; + *event_slot_index = ev->signal_slot_index; + } + break; + default: + ret = create_other_event(p, ev); + break; + } + + if (!ret) { + hash_add(p->events, &ev->events, ev->event_id); + + *event_id = ev->event_id; + *event_trigger_data = ev->event_id; + } else { + kfree(ev); + } + + mutex_unlock(&p->event_mutex); + + return ret; +} + +/* Assumes that p is current. */ +int kfd_event_destroy(struct kfd_process *p, uint32_t event_id) +{ + struct kfd_event *ev; + int ret = 0; + + mutex_lock(&p->event_mutex); + + ev = lookup_event_by_id(p, event_id); + + if (ev) + destroy_event(p, ev); + else + ret = -EINVAL; + + mutex_unlock(&p->event_mutex); + return ret; +} + +static void set_event(struct kfd_event *ev) +{ + struct kfd_event_waiter *waiter; + struct kfd_event_waiter *next; + + /* Auto reset if the list is non-empty and we're waking someone. */ + ev->signaled = !ev->auto_reset || list_empty(&ev->waiters); + + list_for_each_entry_safe(waiter, next, &ev->waiters, waiters) { + waiter->activated = true; + + /* _init because free_waiters will call list_del */ + list_del_init(&waiter->waiters); + + wake_up_process(waiter->sleeping_task); + } +} + +/* Assumes that p is current. */ +int kfd_set_event(struct kfd_process *p, uint32_t event_id) +{ + int ret = 0; + struct kfd_event *ev; + + mutex_lock(&p->event_mutex); + + ev = lookup_event_by_id(p, event_id); + + if (ev && event_can_be_cpu_signaled(ev)) + set_event(ev); + else + ret = -EINVAL; + + mutex_unlock(&p->event_mutex); + return ret; +} + +static void reset_event(struct kfd_event *ev) +{ + ev->signaled = false; +} + +/* Assumes that p is current. */ +int kfd_reset_event(struct kfd_process *p, uint32_t event_id) +{ + int ret = 0; + struct kfd_event *ev; + + mutex_lock(&p->event_mutex); + + ev = lookup_event_by_id(p, event_id); + + if (ev && event_can_be_cpu_signaled(ev)) + reset_event(ev); + else + ret = -EINVAL; + + mutex_unlock(&p->event_mutex); + return ret; + +} + +static void acknowledge_signal(struct kfd_process *p, struct kfd_event *ev) +{ + page_slots(ev->signal_page)[ev->signal_slot_index] = + UNSIGNALED_EVENT_SLOT; +} + +static bool is_slot_signaled(struct signal_page *page, unsigned int index) +{ + return page_slots(page)[index] != UNSIGNALED_EVENT_SLOT; +} + +static void set_event_from_interrupt(struct kfd_process *p, + struct kfd_event *ev) +{ + if (ev && event_can_be_gpu_signaled(ev)) { + acknowledge_signal(p, ev); + set_event(ev); + } +} + +void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id, + uint32_t valid_id_bits) +{ + struct kfd_event *ev; + + /* + * Because we are called from arbitrary context (workqueue) as opposed + * to process context, kfd_process could attempt to exit while we are + * running so the lookup function returns a locked process. + */ + struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); + + if (!p) + return; /* Presumably process exited. */ + + mutex_lock(&p->event_mutex); + + if (valid_id_bits >= INTERRUPT_DATA_BITS) { + /* Partial ID is a full ID. */ + ev = lookup_event_by_id(p, partial_id); + set_event_from_interrupt(p, ev); + } else { + /* + * Partial ID is in fact partial. For now we completely + * ignore it, but we could use any bits we did receive to + * search faster. + */ + struct signal_page *page; + unsigned i; + + list_for_each_entry(page, &p->signal_event_pages, event_pages) + for (i = 0; i < SLOTS_PER_PAGE; i++) + if (is_slot_signaled(page, i)) { + ev = lookup_event_by_page_slot(p, + page, i); + set_event_from_interrupt(p, ev); + } + } + + mutex_unlock(&p->event_mutex); + mutex_unlock(&p->mutex); +} + +static struct kfd_event_waiter *alloc_event_waiters(uint32_t num_events) +{ + struct kfd_event_waiter *event_waiters; + uint32_t i; + + event_waiters = kmalloc_array(num_events, + sizeof(struct kfd_event_waiter), + GFP_KERNEL); + + for (i = 0; (event_waiters) && (i < num_events) ; i++) { + INIT_LIST_HEAD(&event_waiters[i].waiters); + event_waiters[i].sleeping_task = current; + event_waiters[i].activated = false; + } + + return event_waiters; +} + +static int init_event_waiter(struct kfd_process *p, + struct kfd_event_waiter *waiter, + uint32_t event_id, + uint32_t input_index) +{ + struct kfd_event *ev = lookup_event_by_id(p, event_id); + + if (!ev) + return -EINVAL; + + waiter->event = ev; + waiter->input_index = input_index; + waiter->activated = ev->signaled; + ev->signaled = ev->signaled && !ev->auto_reset; + + list_add(&waiter->waiters, &ev->waiters); + + return 0; +} + +static bool test_event_condition(bool all, uint32_t num_events, + struct kfd_event_waiter *event_waiters) +{ + uint32_t i; + uint32_t activated_count = 0; + + for (i = 0; i < num_events; i++) { + if (event_waiters[i].activated) { + if (!all) + return true; + + activated_count++; + } + } + + return activated_count == num_events; +} + +/* + * Copy event specific data, if defined. + * Currently only memory exception events have additional data to copy to user + */ +static bool copy_signaled_event_data(uint32_t num_events, + struct kfd_event_waiter *event_waiters, + struct kfd_event_data __user *data) +{ + struct kfd_hsa_memory_exception_data *src; + struct kfd_hsa_memory_exception_data __user *dst; + struct kfd_event_waiter *waiter; + struct kfd_event *event; + uint32_t i; + + for (i = 0; i < num_events; i++) { + waiter = &event_waiters[i]; + event = waiter->event; + if (waiter->activated && event->type == KFD_EVENT_TYPE_MEMORY) { + dst = &data[waiter->input_index].memory_exception_data; + src = &event->memory_exception_data; + if (copy_to_user(dst, src, + sizeof(struct kfd_hsa_memory_exception_data))) + return false; + } + } + + return true; + +} + + + +static long user_timeout_to_jiffies(uint32_t user_timeout_ms) +{ + if (user_timeout_ms == KFD_EVENT_TIMEOUT_IMMEDIATE) + return 0; + + if (user_timeout_ms == KFD_EVENT_TIMEOUT_INFINITE) + return MAX_SCHEDULE_TIMEOUT; + + /* + * msecs_to_jiffies interprets all values above 2^31-1 as infinite, + * but we consider them finite. + * This hack is wrong, but nobody is likely to notice. + */ + user_timeout_ms = min_t(uint32_t, user_timeout_ms, 0x7FFFFFFF); + + return msecs_to_jiffies(user_timeout_ms) + 1; +} + +static void free_waiters(uint32_t num_events, struct kfd_event_waiter *waiters) +{ + uint32_t i; + + for (i = 0; i < num_events; i++) + list_del(&waiters[i].waiters); + + kfree(waiters); +} + +int kfd_wait_on_events(struct kfd_process *p, + uint32_t num_events, void __user *data, + bool all, uint32_t user_timeout_ms, + enum kfd_event_wait_result *wait_result) +{ + struct kfd_event_data __user *events = + (struct kfd_event_data __user *) data; + uint32_t i; + int ret = 0; + struct kfd_event_waiter *event_waiters = NULL; + long timeout = user_timeout_to_jiffies(user_timeout_ms); + + mutex_lock(&p->event_mutex); + + event_waiters = alloc_event_waiters(num_events); + if (!event_waiters) { + ret = -ENOMEM; + goto fail; + } + + for (i = 0; i < num_events; i++) { + struct kfd_event_data event_data; + + if (copy_from_user(&event_data, &events[i], + sizeof(struct kfd_event_data))) + goto fail; + + ret = init_event_waiter(p, &event_waiters[i], + event_data.event_id, i); + if (ret) + goto fail; + } + + mutex_unlock(&p->event_mutex); + + while (true) { + if (fatal_signal_pending(current)) { + ret = -EINTR; + break; + } + + if (signal_pending(current)) { + /* + * This is wrong when a nonzero, non-infinite timeout + * is specified. We need to use + * ERESTARTSYS_RESTARTBLOCK, but struct restart_block + * contains a union with data for each user and it's + * in generic kernel code that I don't want to + * touch yet. + */ + ret = -ERESTARTSYS; + break; + } + + if (test_event_condition(all, num_events, event_waiters)) { + if (copy_signaled_event_data(num_events, + event_waiters, events)) + *wait_result = KFD_WAIT_COMPLETE; + else + *wait_result = KFD_WAIT_ERROR; + break; + } + + if (timeout <= 0) { + *wait_result = KFD_WAIT_TIMEOUT; + break; + } + + timeout = schedule_timeout_interruptible(timeout); + } + __set_current_state(TASK_RUNNING); + + mutex_lock(&p->event_mutex); + free_waiters(num_events, event_waiters); + mutex_unlock(&p->event_mutex); + + return ret; + +fail: + if (event_waiters) + free_waiters(num_events, event_waiters); + + mutex_unlock(&p->event_mutex); + + *wait_result = KFD_WAIT_ERROR; + + return ret; +} + +int kfd_event_mmap(struct kfd_process *p, struct vm_area_struct *vma) +{ + + unsigned int page_index; + unsigned long pfn; + struct signal_page *page; + + /* check required size is logical */ + if (get_order(KFD_SIGNAL_EVENT_LIMIT * 8) != + get_order(vma->vm_end - vma->vm_start)) { + pr_err("amdkfd: event page mmap requested illegal size\n"); + return -EINVAL; + } + + page_index = vma->vm_pgoff; + + page = lookup_signal_page_by_index(p, page_index); + if (!page) { + /* Probably KFD bug, but mmap is user-accessible. */ + pr_debug("signal page could not be found for page_index %u\n", + page_index); + return -EINVAL; + } + + pfn = __pa(page->kernel_address); + pfn >>= PAGE_SHIFT; + + vma->vm_flags |= VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_NORESERVE + | VM_DONTDUMP | VM_PFNMAP; + + pr_debug("mapping signal page\n"); + pr_debug(" start user address == 0x%08lx\n", vma->vm_start); + pr_debug(" end user address == 0x%08lx\n", vma->vm_end); + pr_debug(" pfn == 0x%016lX\n", pfn); + pr_debug(" vm_flags == 0x%08lX\n", vma->vm_flags); + pr_debug(" size == 0x%08lX\n", + vma->vm_end - vma->vm_start); + + page->user_address = (uint64_t __user *)vma->vm_start; + + /* mapping the page to user process */ + return remap_pfn_range(vma, vma->vm_start, pfn, + vma->vm_end - vma->vm_start, vma->vm_page_prot); +} + +/* + * Assumes that p->event_mutex is held and of course + * that p is not going away (current or locked). + */ +static void lookup_events_by_type_and_signal(struct kfd_process *p, + int type, void *event_data) +{ + struct kfd_hsa_memory_exception_data *ev_data; + struct kfd_event *ev; + int bkt; + bool send_signal = true; + + ev_data = (struct kfd_hsa_memory_exception_data *) event_data; + + hash_for_each(p->events, bkt, ev, events) + if (ev->type == type) { + send_signal = false; + dev_dbg(kfd_device, + "Event found: id %X type %d", + ev->event_id, ev->type); + set_event(ev); + if (ev->type == KFD_EVENT_TYPE_MEMORY && ev_data) + ev->memory_exception_data = *ev_data; + } + + /* Send SIGTERM no event of type "type" has been found*/ + if (send_signal) { + if (send_sigterm) { + dev_warn(kfd_device, + "Sending SIGTERM to HSA Process with PID %d ", + p->lead_thread->pid); + send_sig(SIGTERM, p->lead_thread, 0); + } else { + dev_err(kfd_device, + "HSA Process (PID %d) got unhandled exception", + p->lead_thread->pid); + } + } +} + +void kfd_signal_iommu_event(struct kfd_dev *dev, unsigned int pasid, + unsigned long address, bool is_write_requested, + bool is_execute_requested) +{ + struct kfd_hsa_memory_exception_data memory_exception_data; + struct vm_area_struct *vma; + + /* + * Because we are called from arbitrary context (workqueue) as opposed + * to process context, kfd_process could attempt to exit while we are + * running so the lookup function returns a locked process. + */ + struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); + + if (!p) + return; /* Presumably process exited. */ + + memset(&memory_exception_data, 0, sizeof(memory_exception_data)); + + down_read(&p->mm->mmap_sem); + vma = find_vma(p->mm, address); + + memory_exception_data.gpu_id = dev->id; + memory_exception_data.va = address; + /* Set failure reason */ + memory_exception_data.failure.NotPresent = 1; + memory_exception_data.failure.NoExecute = 0; + memory_exception_data.failure.ReadOnly = 0; + if (vma) { + if (vma->vm_start > address) { + memory_exception_data.failure.NotPresent = 1; + memory_exception_data.failure.NoExecute = 0; + memory_exception_data.failure.ReadOnly = 0; + } else { + memory_exception_data.failure.NotPresent = 0; + if (is_write_requested && !(vma->vm_flags & VM_WRITE)) + memory_exception_data.failure.ReadOnly = 1; + else + memory_exception_data.failure.ReadOnly = 0; + if (is_execute_requested && !(vma->vm_flags & VM_EXEC)) + memory_exception_data.failure.NoExecute = 1; + else + memory_exception_data.failure.NoExecute = 0; + } + } + + up_read(&p->mm->mmap_sem); + + mutex_lock(&p->event_mutex); + + /* Lookup events by type and signal them */ + lookup_events_by_type_and_signal(p, KFD_EVENT_TYPE_MEMORY, + &memory_exception_data); + + mutex_unlock(&p->event_mutex); + mutex_unlock(&p->mutex); +} + +void kfd_signal_hw_exception_event(unsigned int pasid) +{ + /* + * Because we are called from arbitrary context (workqueue) as opposed + * to process context, kfd_process could attempt to exit while we are + * running so the lookup function returns a locked process. + */ + struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); + + if (!p) + return; /* Presumably process exited. */ + + mutex_lock(&p->event_mutex); + + /* Lookup events by type and signal them */ + lookup_events_by_type_and_signal(p, KFD_EVENT_TYPE_HW_EXCEPTION, NULL); + + mutex_unlock(&p->event_mutex); + mutex_unlock(&p->mutex); +} diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.h b/drivers/gpu/drm/amd/amdkfd/kfd_events.h new file mode 100644 index 000000000000..28f6838b1f4c --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.h @@ -0,0 +1,84 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + */ + +#ifndef KFD_EVENTS_H_INCLUDED +#define KFD_EVENTS_H_INCLUDED + +#include <linux/kernel.h> +#include <linux/hashtable.h> +#include <linux/types.h> +#include <linux/list.h> +#include "kfd_priv.h" +#include <uapi/linux/kfd_ioctl.h> + +#define KFD_EVENT_ID_NONSIGNAL_MASK 0x80000000U +#define KFD_FIRST_NONSIGNAL_EVENT_ID KFD_EVENT_ID_NONSIGNAL_MASK +#define KFD_LAST_NONSIGNAL_EVENT_ID UINT_MAX + +/* + * Written into kfd_signal_slot_t to indicate that the event is not signaled. + * Since the event protocol may need to write the event ID into memory, this + * must not be a valid event ID. + * For the sake of easy memset-ing, this must be a byte pattern. + */ +#define UNSIGNALED_EVENT_SLOT ((uint64_t)-1) + +struct kfd_event_waiter; +struct signal_page; + +struct kfd_event { + /* All events in process, rooted at kfd_process.events. */ + struct hlist_node events; + + u32 event_id; + + bool signaled; + bool auto_reset; + + int type; + + struct list_head waiters; /* List of kfd_event_waiter by waiters. */ + + /* Only for signal events. */ + struct signal_page *signal_page; + unsigned int signal_slot_index; + uint64_t __user *user_signal_address; + + /* type specific data */ + union { + struct kfd_hsa_memory_exception_data memory_exception_data; + }; +}; + +#define KFD_EVENT_TIMEOUT_IMMEDIATE 0 +#define KFD_EVENT_TIMEOUT_INFINITE 0xFFFFFFFFu + +/* Matching HSA_EVENTTYPE */ +#define KFD_EVENT_TYPE_SIGNAL 0 +#define KFD_EVENT_TYPE_HW_EXCEPTION 3 +#define KFD_EVENT_TYPE_DEBUG 5 +#define KFD_EVENT_TYPE_MEMORY 8 + +extern void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id, + uint32_t valid_id_bits); + +#endif diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c new file mode 100644 index 000000000000..7f134aa9bfd3 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c @@ -0,0 +1,188 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + */ + +/* + * KFD Interrupts. + * + * AMD GPUs deliver interrupts by pushing an interrupt description onto the + * interrupt ring and then sending an interrupt. KGD receives the interrupt + * in ISR and sends us a pointer to each new entry on the interrupt ring. + * + * We generally can't process interrupt-signaled events from ISR, so we call + * out to each interrupt client module (currently only the scheduler) to ask if + * each interrupt is interesting. If they return true, then it requires further + * processing so we copy it to an internal interrupt ring and call each + * interrupt client again from a work-queue. + * + * There's no acknowledgment for the interrupts we use. The hardware simply + * queues a new interrupt each time without waiting. + * + * The fixed-size internal queue means that it's possible for us to lose + * interrupts because we have no back-pressure to the hardware. + */ + +#include <linux/slab.h> +#include <linux/device.h> +#include "kfd_priv.h" + +#define KFD_INTERRUPT_RING_SIZE 1024 + +static void interrupt_wq(struct work_struct *); + +int kfd_interrupt_init(struct kfd_dev *kfd) +{ + void *interrupt_ring = kmalloc_array(KFD_INTERRUPT_RING_SIZE, + kfd->device_info->ih_ring_entry_size, + GFP_KERNEL); + if (!interrupt_ring) + return -ENOMEM; + + kfd->interrupt_ring = interrupt_ring; + kfd->interrupt_ring_size = + KFD_INTERRUPT_RING_SIZE * kfd->device_info->ih_ring_entry_size; + atomic_set(&kfd->interrupt_ring_wptr, 0); + atomic_set(&kfd->interrupt_ring_rptr, 0); + + spin_lock_init(&kfd->interrupt_lock); + + INIT_WORK(&kfd->interrupt_work, interrupt_wq); + + kfd->interrupts_active = true; + + /* + * After this function returns, the interrupt will be enabled. This + * barrier ensures that the interrupt running on a different processor + * sees all the above writes. + */ + smp_wmb(); + + return 0; +} + +void kfd_interrupt_exit(struct kfd_dev *kfd) +{ + /* + * Stop the interrupt handler from writing to the ring and scheduling + * workqueue items. The spinlock ensures that any interrupt running + * after we have unlocked sees interrupts_active = false. + */ + unsigned long flags; + + spin_lock_irqsave(&kfd->interrupt_lock, flags); + kfd->interrupts_active = false; + spin_unlock_irqrestore(&kfd->interrupt_lock, flags); + + /* + * Flush_scheduled_work ensures that there are no outstanding + * work-queue items that will access interrupt_ring. New work items + * can't be created because we stopped interrupt handling above. + */ + flush_scheduled_work(); + + kfree(kfd->interrupt_ring); +} + +/* + * This assumes that it can't be called concurrently with itself + * but only with dequeue_ih_ring_entry. + */ +bool enqueue_ih_ring_entry(struct kfd_dev *kfd, const void *ih_ring_entry) +{ + unsigned int rptr = atomic_read(&kfd->interrupt_ring_rptr); + unsigned int wptr = atomic_read(&kfd->interrupt_ring_wptr); + + if ((rptr - wptr) % kfd->interrupt_ring_size == + kfd->device_info->ih_ring_entry_size) { + /* This is very bad, the system is likely to hang. */ + dev_err_ratelimited(kfd_chardev(), + "Interrupt ring overflow, dropping interrupt.\n"); + return false; + } + + memcpy(kfd->interrupt_ring + wptr, ih_ring_entry, + kfd->device_info->ih_ring_entry_size); + + wptr = (wptr + kfd->device_info->ih_ring_entry_size) % + kfd->interrupt_ring_size; + smp_wmb(); /* Ensure memcpy'd data is visible before wptr update. */ + atomic_set(&kfd->interrupt_ring_wptr, wptr); + + return true; +} + +/* + * This assumes that it can't be called concurrently with itself + * but only with enqueue_ih_ring_entry. + */ +static bool dequeue_ih_ring_entry(struct kfd_dev *kfd, void *ih_ring_entry) +{ + /* + * Assume that wait queues have an implicit barrier, i.e. anything that + * happened in the ISR before it queued work is visible. + */ + + unsigned int wptr = atomic_read(&kfd->interrupt_ring_wptr); + unsigned int rptr = atomic_read(&kfd->interrupt_ring_rptr); + + if (rptr == wptr) + return false; + + memcpy(ih_ring_entry, kfd->interrupt_ring + rptr, + kfd->device_info->ih_ring_entry_size); + + rptr = (rptr + kfd->device_info->ih_ring_entry_size) % + kfd->interrupt_ring_size; + + /* + * Ensure the rptr write update is not visible until + * memcpy has finished reading. + */ + smp_mb(); + atomic_set(&kfd->interrupt_ring_rptr, rptr); + + return true; +} + +static void interrupt_wq(struct work_struct *work) +{ + struct kfd_dev *dev = container_of(work, struct kfd_dev, + interrupt_work); + + uint32_t ih_ring_entry[DIV_ROUND_UP( + dev->device_info->ih_ring_entry_size, + sizeof(uint32_t))]; + + while (dequeue_ih_ring_entry(dev, ih_ring_entry)) + dev->device_info->event_interrupt_class->interrupt_wq(dev, + ih_ring_entry); +} + +bool interrupt_is_wanted(struct kfd_dev *dev, const uint32_t *ih_ring_entry) +{ + /* integer and bitwise OR so there is no boolean short-circuiting */ + unsigned wanted = 0; + + wanted |= dev->device_info->event_interrupt_class->interrupt_isr(dev, + ih_ring_entry); + + return wanted != 0; +} diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c index c7d298e62c96..8fa894100290 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c @@ -215,8 +215,9 @@ static int acquire_packet_buffer(struct kernel_queue *kq, queue_address = (unsigned int *)kq->pq_kernel_addr; queue_size_dwords = kq->queue->properties.queue_size / sizeof(uint32_t); - pr_debug("amdkfd: In func %s\nrptr: %d\nwptr: %d\nqueue_address 0x%p\n", - __func__, rptr, wptr, queue_address); + pr_debug("rptr: %d\n", rptr); + pr_debug("wptr: %d\n", wptr); + pr_debug("queue_address 0x%p\n", queue_address); available_size = (rptr - 1 - wptr + queue_size_dwords) % queue_size_dwords; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_module.c b/drivers/gpu/drm/amd/amdkfd/kfd_module.c index 4e0a68f13a77..ca8410e8683d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_module.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_module.c @@ -29,10 +29,10 @@ #define KFD_DRIVER_AUTHOR "AMD Inc. and others" #define KFD_DRIVER_DESC "Standalone HSA driver for AMD's GPUs" -#define KFD_DRIVER_DATE "20150122" +#define KFD_DRIVER_DATE "20150421" #define KFD_DRIVER_MAJOR 0 #define KFD_DRIVER_MINOR 7 -#define KFD_DRIVER_PATCHLEVEL 1 +#define KFD_DRIVER_PATCHLEVEL 2 static const struct kgd2kfd_calls kgd2kfd = { .exit = kgd2kfd_exit, @@ -54,6 +54,11 @@ module_param(max_num_of_queues_per_device, int, 0444); MODULE_PARM_DESC(max_num_of_queues_per_device, "Maximum number of supported queues per device (1 = Minimum, 4096 = default)"); +int send_sigterm; +module_param(send_sigterm, int, 0444); +MODULE_PARM_DESC(send_sigterm, + "Send sigterm to HSA process on unhandled exception (0 = disable, 1 = enable)"); + bool kgd2kfd_init(unsigned interface_version, const struct kgd2kfd_calls **g2f) { /* diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c index e2533d875f43..99b6d28a11c3 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c @@ -163,7 +163,7 @@ static int pm_create_map_process(struct packet_manager *pm, uint32_t *buffer, num_queues = 0; list_for_each_entry(cur, &qpd->queues_list, list) num_queues++; - packet->bitfields10.num_queues = num_queues; + packet->bitfields10.num_queues = (qpd->is_debug) ? 0 : num_queues; packet->sh_mem_config = qpd->sh_mem_config; packet->sh_mem_bases = qpd->sh_mem_bases; @@ -177,9 +177,10 @@ static int pm_create_map_process(struct packet_manager *pm, uint32_t *buffer, } static int pm_create_map_queue(struct packet_manager *pm, uint32_t *buffer, - struct queue *q) + struct queue *q, bool is_static) { struct pm4_map_queues *packet; + bool use_static = is_static; BUG_ON(!pm || !buffer || !q); @@ -209,6 +210,7 @@ static int pm_create_map_queue(struct packet_manager *pm, uint32_t *buffer, case KFD_QUEUE_TYPE_SDMA: packet->bitfields2.engine_sel = engine_sel__mes_map_queues__sdma0; + use_static = false; /* no static queues under SDMA */ break; default: BUG(); @@ -218,6 +220,9 @@ static int pm_create_map_queue(struct packet_manager *pm, uint32_t *buffer, packet->mes_map_queues_ordinals[0].bitfields3.doorbell_offset = q->properties.doorbell_off; + packet->mes_map_queues_ordinals[0].bitfields3.is_static = + (use_static == true) ? 1 : 0; + packet->mes_map_queues_ordinals[0].mqd_addr_lo = lower_32_bits(q->gart_mqd_addr); @@ -271,9 +276,11 @@ static int pm_create_runlist_ib(struct packet_manager *pm, pm_release_ib(pm); return -ENOMEM; } + retval = pm_create_map_process(pm, &rl_buffer[rl_wptr], qpd); if (retval != 0) return retval; + proccesses_mapped++; inc_wptr(&rl_wptr, sizeof(struct pm4_map_process), alloc_size_bytes); @@ -281,23 +288,36 @@ static int pm_create_runlist_ib(struct packet_manager *pm, list_for_each_entry(kq, &qpd->priv_queue_list, list) { if (kq->queue->properties.is_active != true) continue; + + pr_debug("kfd: static_queue, mapping kernel q %d, is debug status %d\n", + kq->queue->queue, qpd->is_debug); + retval = pm_create_map_queue(pm, &rl_buffer[rl_wptr], - kq->queue); + kq->queue, qpd->is_debug); if (retval != 0) return retval; - inc_wptr(&rl_wptr, sizeof(struct pm4_map_queues), - alloc_size_bytes); + + inc_wptr(&rl_wptr, + sizeof(struct pm4_map_queues), + alloc_size_bytes); } list_for_each_entry(q, &qpd->queues_list, list) { if (q->properties.is_active != true) continue; - retval = pm_create_map_queue(pm, - &rl_buffer[rl_wptr], q); + + pr_debug("kfd: static_queue, mapping user queue %d, is debug status %d\n", + q->queue, qpd->is_debug); + + retval = pm_create_map_queue(pm, &rl_buffer[rl_wptr], + q, qpd->is_debug); + if (retval != 0) return retval; - inc_wptr(&rl_wptr, sizeof(struct pm4_map_queues), - alloc_size_bytes); + + inc_wptr(&rl_wptr, + sizeof(struct pm4_map_queues), + alloc_size_bytes); } } @@ -488,7 +508,8 @@ int pm_send_unmap_queue(struct packet_manager *pm, enum kfd_queue_type type, packet = (struct pm4_unmap_queues *)buffer; memset(buffer, 0, sizeof(struct pm4_unmap_queues)); - + pr_debug("kfd: static_queue: unmapping queues: mode is %d , reset is %d , type is %d\n", + mode, reset, type); packet->header.u32all = build_pm4_header(IT_UNMAP_QUEUES, sizeof(struct pm4_unmap_queues)); switch (type) { @@ -529,6 +550,11 @@ int pm_send_unmap_queue(struct packet_manager *pm, enum kfd_queue_type type, packet->bitfields2.queue_sel = queue_sel__mes_unmap_queues__perform_request_on_all_active_queues; break; + case KFD_PREEMPT_TYPE_FILTER_DYNAMIC_QUEUES: + /* in this case, we do not preempt static queues */ + packet->bitfields2.queue_sel = + queue_sel__mes_unmap_queues__perform_request_on_dynamic_queues_only; + break; default: BUG(); break; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers.h b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers.h index 071ad5724bd2..5b393f3e34a9 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers.h @@ -237,7 +237,8 @@ struct pm4_map_queues { struct { union { struct { - uint32_t reserved5:2; + uint32_t is_static:1; + uint32_t reserved5:1; uint32_t doorbell_offset:21; uint32_t reserved6:3; uint32_t queue:6; @@ -328,7 +329,8 @@ enum unmap_queues_action_enum { enum unmap_queues_queue_sel_enum { queue_sel__mes_unmap_queues__perform_request_on_specified_queues = 0, queue_sel__mes_unmap_queues__perform_request_on_pasid_queues = 1, - queue_sel__mes_unmap_queues__perform_request_on_all_active_queues = 2 + queue_sel__mes_unmap_queues__perform_request_on_all_active_queues = 2, + queue_sel__mes_unmap_queues__perform_request_on_dynamic_queues_only = 3 }; enum unmap_queues_engine_sel_enum { diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_diq.h b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_diq.h new file mode 100644 index 000000000000..a0ff34878163 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_diq.h @@ -0,0 +1,290 @@ +/* + * Copyright 2014 Advanced Micro Devices, Inc. + * + * 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 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) 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. + * + */ + +#ifndef KFD_PM4_HEADERS_DIQ_H_ +#define KFD_PM4_HEADERS_DIQ_H_ + +/*--------------------_INDIRECT_BUFFER-------------------- */ + +#ifndef _PM4__INDIRECT_BUFFER_DEFINED +#define _PM4__INDIRECT_BUFFER_DEFINED +enum _INDIRECT_BUFFER_cache_policy_enum { + cache_policy___indirect_buffer__lru = 0, + cache_policy___indirect_buffer__stream = 1, + cache_policy___indirect_buffer__bypass = 2 +}; + +enum { + IT_INDIRECT_BUFFER_PASID = 0x5C +}; + +struct pm4__indirect_buffer_pasid { + union { + union PM4_MES_TYPE_3_HEADER header; /* header */ + unsigned int ordinal1; + }; + + union { + struct { + unsigned int reserved1:2; + unsigned int ib_base_lo:30; + } bitfields2; + unsigned int ordinal2; + }; + + union { + struct { + unsigned int ib_base_hi:16; + unsigned int reserved2:16; + } bitfields3; + unsigned int ordinal3; + }; + + union { + unsigned int control; + unsigned int ordinal4; + }; + + union { + struct { + unsigned int pasid:10; + unsigned int reserved4:22; + } bitfields5; + unsigned int ordinal5; + }; + +}; + +#endif + +/*--------------------_RELEASE_MEM-------------------- */ + +#ifndef _PM4__RELEASE_MEM_DEFINED +#define _PM4__RELEASE_MEM_DEFINED +enum _RELEASE_MEM_event_index_enum { + event_index___release_mem__end_of_pipe = 5, + event_index___release_mem__shader_done = 6 +}; + +enum _RELEASE_MEM_cache_policy_enum { + cache_policy___release_mem__lru = 0, + cache_policy___release_mem__stream = 1, + cache_policy___release_mem__bypass = 2 +}; + +enum _RELEASE_MEM_dst_sel_enum { + dst_sel___release_mem__memory_controller = 0, + dst_sel___release_mem__tc_l2 = 1, + dst_sel___release_mem__queue_write_pointer_register = 2, + dst_sel___release_mem__queue_write_pointer_poll_mask_bit = 3 +}; + +enum _RELEASE_MEM_int_sel_enum { + int_sel___release_mem__none = 0, + int_sel___release_mem__send_interrupt_only = 1, + int_sel___release_mem__send_interrupt_after_write_confirm = 2, + int_sel___release_mem__send_data_after_write_confirm = 3 +}; + +enum _RELEASE_MEM_data_sel_enum { + data_sel___release_mem__none = 0, + data_sel___release_mem__send_32_bit_low = 1, + data_sel___release_mem__send_64_bit_data = 2, + data_sel___release_mem__send_gpu_clock_counter = 3, + data_sel___release_mem__send_cp_perfcounter_hi_lo = 4, + data_sel___release_mem__store_gds_data_to_memory = 5 +}; + +struct pm4__release_mem { + union { + union PM4_MES_TYPE_3_HEADER header; /*header */ + unsigned int ordinal1; + }; + + union { + struct { + unsigned int event_type:6; + unsigned int reserved1:2; + enum _RELEASE_MEM_event_index_enum event_index:4; + unsigned int tcl1_vol_action_ena:1; + unsigned int tc_vol_action_ena:1; + unsigned int reserved2:1; + unsigned int tc_wb_action_ena:1; + unsigned int tcl1_action_ena:1; + unsigned int tc_action_ena:1; + unsigned int reserved3:6; + unsigned int atc:1; + enum _RELEASE_MEM_cache_policy_enum cache_policy:2; + unsigned int reserved4:5; + } bitfields2; + unsigned int ordinal2; + }; + + union { + struct { + unsigned int reserved5:16; + enum _RELEASE_MEM_dst_sel_enum dst_sel:2; + unsigned int reserved6:6; + enum _RELEASE_MEM_int_sel_enum int_sel:3; + unsigned int reserved7:2; + enum _RELEASE_MEM_data_sel_enum data_sel:3; + } bitfields3; + unsigned int ordinal3; + }; + + union { + struct { + unsigned int reserved8:2; + unsigned int address_lo_32b:30; + } bitfields4; + struct { + unsigned int reserved9:3; + unsigned int address_lo_64b:29; + } bitfields5; + unsigned int ordinal4; + }; + + unsigned int address_hi; + + unsigned int data_lo; + + unsigned int data_hi; + +}; +#endif + + +/*--------------------_SET_CONFIG_REG-------------------- */ + +#ifndef _PM4__SET_CONFIG_REG_DEFINED +#define _PM4__SET_CONFIG_REG_DEFINED + +struct pm4__set_config_reg { + union { + union PM4_MES_TYPE_3_HEADER header; /*header */ + unsigned int ordinal1; + }; + + union { + struct { + unsigned int reg_offset:16; + unsigned int reserved1:7; + unsigned int vmid_shift:5; + unsigned int insert_vmid:1; + unsigned int reserved2:3; + } bitfields2; + unsigned int ordinal2; + }; + + unsigned int reg_data[1]; /*1..N of these fields */ + +}; +#endif + +/*--------------------_WAIT_REG_MEM-------------------- */ + +#ifndef _PM4__WAIT_REG_MEM_DEFINED +#define _PM4__WAIT_REG_MEM_DEFINED +enum _WAIT_REG_MEM_function_enum { + function___wait_reg_mem__always_pass = 0, + function___wait_reg_mem__less_than_ref_value = 1, + function___wait_reg_mem__less_than_equal_to_the_ref_value = 2, + function___wait_reg_mem__equal_to_the_reference_value = 3, + function___wait_reg_mem__not_equal_reference_value = 4, + function___wait_reg_mem__greater_than_or_equal_reference_value = 5, + function___wait_reg_mem__greater_than_reference_value = 6, + function___wait_reg_mem__reserved = 7 +}; + +enum _WAIT_REG_MEM_mem_space_enum { + mem_space___wait_reg_mem__register_space = 0, + mem_space___wait_reg_mem__memory_space = 1 +}; + +enum _WAIT_REG_MEM_operation_enum { + operation___wait_reg_mem__wait_reg_mem = 0, + operation___wait_reg_mem__wr_wait_wr_reg = 1 +}; + +struct pm4__wait_reg_mem { + union { + union PM4_MES_TYPE_3_HEADER header; /*header */ + unsigned int ordinal1; + }; + + union { + struct { + enum _WAIT_REG_MEM_function_enum function:3; + unsigned int reserved1:1; + enum _WAIT_REG_MEM_mem_space_enum mem_space:2; + enum _WAIT_REG_MEM_operation_enum operation:2; + unsigned int reserved2:24; + } bitfields2; + unsigned int ordinal2; + }; + + union { + struct { + unsigned int reserved3:2; + unsigned int memory_poll_addr_lo:30; + } bitfields3; + struct { + unsigned int register_poll_addr:16; + unsigned int reserved4:16; + } bitfields4; + struct { + unsigned int register_write_addr:16; + unsigned int reserved5:16; + } bitfields5; + unsigned int ordinal3; + }; + + union { + struct { + unsigned int poll_address_hi:16; + unsigned int reserved6:16; + } bitfields6; + struct { + unsigned int register_write_addr:16; + unsigned int reserved7:16; + } bitfields7; + unsigned int ordinal4; + }; + + unsigned int reference; + + unsigned int mask; + + union { + struct { + unsigned int poll_interval:16; + unsigned int reserved8:16; + } bitfields8; + unsigned int ordinal7; + }; + +}; +#endif + + +#endif /* KFD_PM4_HEADERS_DIQ_H_ */ diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index f21fccebd75b..d0d5f4baf72d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -35,6 +35,9 @@ #define KFD_SYSFS_FILE_MODE 0444 +#define KFD_MMAP_DOORBELL_MASK 0x8000000000000 +#define KFD_MMAP_EVENTS_MASK 0x4000000000000 + /* * When working with cp scheduler we should assign the HIQ manually or via * the radeon driver to a fixed hqd slot, here are the fixed HIQ hqd slot @@ -71,6 +74,12 @@ extern int max_num_of_queues_per_device; /* Kernel module parameter to specify the scheduling policy */ extern int sched_policy; +/* + * Kernel module parameter to specify whether to send sigterm to HSA process on + * unhandled exception + */ +extern int send_sigterm; + /** * enum kfd_sched_policy * @@ -108,9 +117,18 @@ enum asic_family_type { CHIP_CARRIZO }; +struct kfd_event_interrupt_class { + bool (*interrupt_isr)(struct kfd_dev *dev, + const uint32_t *ih_ring_entry); + void (*interrupt_wq)(struct kfd_dev *dev, + const uint32_t *ih_ring_entry); +}; + struct kfd_device_info { unsigned int asic_family; + const struct kfd_event_interrupt_class *event_interrupt_class; unsigned int max_pasid_bits; + unsigned int max_no_of_hqd; size_t ih_ring_entry_size; uint8_t num_of_watch_points; uint16_t mqd_size_aligned; @@ -150,8 +168,8 @@ struct kfd_dev { const struct kfd2kgd_calls *kfd2kgd; struct mutex doorbell_mutex; - unsigned long doorbell_available_index[DIV_ROUND_UP( - KFD_MAX_NUM_OF_QUEUES_PER_PROCESS, BITS_PER_LONG)]; + DECLARE_BITMAP(doorbell_available_index, + KFD_MAX_NUM_OF_QUEUES_PER_PROCESS); void *gtt_mem; uint64_t gtt_start_gpu_addr; @@ -161,10 +179,26 @@ struct kfd_dev { unsigned int gtt_sa_chunk_size; unsigned int gtt_sa_num_of_chunks; + /* Interrupts */ + void *interrupt_ring; + size_t interrupt_ring_size; + atomic_t interrupt_ring_rptr; + atomic_t interrupt_ring_wptr; + struct work_struct interrupt_work; + spinlock_t interrupt_lock; + /* QCM Device instance */ struct device_queue_manager *dqm; bool init_complete; + /* + * Interrupts of interest to KFD are copied + * from the HW ring into a SW ring. + */ + bool interrupts_active; + + /* Debug manager */ + struct kfd_dbgmgr *dbgmgr; }; /* KGD2KFD callbacks */ @@ -201,6 +235,7 @@ struct device *kfd_chardev(void); enum kfd_preempt_type_filter { KFD_PREEMPT_TYPE_FILTER_SINGLE_QUEUE, KFD_PREEMPT_TYPE_FILTER_ALL_QUEUES, + KFD_PREEMPT_TYPE_FILTER_DYNAMIC_QUEUES, KFD_PREEMPT_TYPE_FILTER_BY_PASID }; @@ -428,6 +463,11 @@ struct kfd_process_device { /* Is this process/pasid bound to this device? (amd_iommu_bind_pasid) */ bool bound; + + /* This flag tells if we should reset all + * wavefronts on process termination + */ + bool reset_wavefronts; }; #define qpd_to_pdd(x) container_of(x, struct kfd_process_device, qpd) @@ -473,10 +513,17 @@ struct kfd_process { /* Size is queue_array_size, up to MAX_PROCESS_QUEUES. */ struct kfd_queue **queues; - unsigned long allocated_queue_bitmap[DIV_ROUND_UP(KFD_MAX_NUM_OF_QUEUES_PER_PROCESS, BITS_PER_LONG)]; - /*Is the user space process 32 bit?*/ bool is_32bit_user_mode; + + /* Event-related data */ + struct mutex event_mutex; + /* All events in process hashed by ID, linked on kfd_event.events. */ + DECLARE_HASHTABLE(events, 4); + struct list_head signal_event_pages; /* struct slot_page_header. + event_pages */ + u32 next_nonsignal_event_id; + size_t signal_event_count; }; /** @@ -501,6 +548,7 @@ void kfd_process_create_wq(void); void kfd_process_destroy_wq(void); struct kfd_process *kfd_create_process(const struct task_struct *); struct kfd_process *kfd_get_process(const struct task_struct *); +struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid); struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev, struct kfd_process *p); @@ -555,7 +603,11 @@ struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev); struct kfd_dev *kfd_topology_enum_kfd_devices(uint8_t idx); /* Interrupts */ +int kfd_interrupt_init(struct kfd_dev *dev); +void kfd_interrupt_exit(struct kfd_dev *dev); void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry); +bool enqueue_ih_ring_entry(struct kfd_dev *kfd, const void *ih_ring_entry); +bool interrupt_is_wanted(struct kfd_dev *dev, const uint32_t *ih_ring_entry); /* Power Management */ void kgd2kfd_suspend(struct kfd_dev *kfd); @@ -606,6 +658,12 @@ int pqm_create_queue(struct process_queue_manager *pqm, int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid); int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid, struct queue_properties *p); +struct kernel_queue *pqm_get_kernel_queue(struct process_queue_manager *pqm, + unsigned int qid); + +int amdkfd_fence_wait_timeout(unsigned int *fence_addr, + unsigned int fence_value, + unsigned long timeout); /* Packet Manager */ @@ -642,4 +700,37 @@ uint64_t kfd_get_number_elems(struct kfd_dev *kfd); phys_addr_t kfd_get_process_doorbells(struct kfd_dev *dev, struct kfd_process *process); +/* Events */ +extern const struct kfd_event_interrupt_class event_interrupt_class_cik; +extern const struct kfd_device_global_init_class device_global_init_class_cik; + +enum kfd_event_wait_result { + KFD_WAIT_COMPLETE, + KFD_WAIT_TIMEOUT, + KFD_WAIT_ERROR +}; + +void kfd_event_init_process(struct kfd_process *p); +void kfd_event_free_process(struct kfd_process *p); +int kfd_event_mmap(struct kfd_process *process, struct vm_area_struct *vma); +int kfd_wait_on_events(struct kfd_process *p, + uint32_t num_events, void __user *data, + bool all, uint32_t user_timeout_ms, + enum kfd_event_wait_result *wait_result); +void kfd_signal_event_interrupt(unsigned int pasid, uint32_t partial_id, + uint32_t valid_id_bits); +void kfd_signal_iommu_event(struct kfd_dev *dev, + unsigned int pasid, unsigned long address, + bool is_write_requested, bool is_execute_requested); +void kfd_signal_hw_exception_event(unsigned int pasid); +int kfd_set_event(struct kfd_process *p, uint32_t event_id); +int kfd_reset_event(struct kfd_process *p, uint32_t event_id); +int kfd_event_create(struct file *devkfd, struct kfd_process *p, + uint32_t event_type, bool auto_reset, uint32_t node_id, + uint32_t *event_id, uint32_t *event_trigger_data, + uint64_t *event_page_offset, uint32_t *event_slot_index); +int kfd_event_destroy(struct kfd_process *p, uint32_t event_id); + +int dbgdev_wave_reset_wavefronts(struct kfd_dev *dev, struct kfd_process *p); + #endif diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 945d6226dc51..8a1f999daa24 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -31,6 +31,7 @@ struct mm_struct; #include "kfd_priv.h" +#include "kfd_dbgmgr.h" /* * Initial size for the array of queues. @@ -172,12 +173,17 @@ static void kfd_process_wq_release(struct work_struct *work) pr_debug("Releasing pdd (topology id %d) for process (pasid %d) in workqueue\n", pdd->dev->id, p->pasid); + if (pdd->reset_wavefronts) + dbgdev_wave_reset_wavefronts(pdd->dev, p); + amd_iommu_unbind_pasid(pdd->dev->pdev, p->pasid); list_del(&pdd->per_device_list); kfree(pdd); } + kfd_event_free_process(p); + kfd_pasid_free(p->pasid); mutex_unlock(&p->mutex); @@ -203,8 +209,7 @@ static void kfd_process_destroy_delayed(struct rcu_head *rcu) mmdrop(p->mm); - work = (struct kfd_process_release_work *) - kmalloc(sizeof(struct kfd_process_release_work), GFP_ATOMIC); + work = kmalloc(sizeof(struct kfd_process_release_work), GFP_ATOMIC); if (work) { INIT_WORK((struct work_struct *) work, kfd_process_wq_release); @@ -217,6 +222,7 @@ static void kfd_process_notifier_release(struct mmu_notifier *mn, struct mm_struct *mm) { struct kfd_process *p; + struct kfd_process_device *pdd = NULL; /* * The kfd_process structure can not be free because the @@ -235,6 +241,15 @@ static void kfd_process_notifier_release(struct mmu_notifier *mn, /* In case our notifier is called before IOMMU notifier */ pqm_uninit(&p->pqm); + /* Iterate over all process device data structure and check + * if we should reset all wavefronts */ + list_for_each_entry(pdd, &p->per_device_data, per_device_list) + if (pdd->reset_wavefronts) { + pr_warn("amdkfd: Resetting all wave fronts\n"); + dbgdev_wave_reset_wavefronts(pdd->dev, p); + pdd->reset_wavefronts = false; + } + mutex_unlock(&p->mutex); /* @@ -289,6 +304,8 @@ static struct kfd_process *create_process(const struct task_struct *thread) INIT_LIST_HEAD(&process->per_device_data); + kfd_event_init_process(process); + err = pqm_init(&process->pqm, process); if (err != 0) goto err_process_pqm_init; @@ -339,6 +356,7 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev, INIT_LIST_HEAD(&pdd->qpd.queues_list); INIT_LIST_HEAD(&pdd->qpd.priv_queue_list); pdd->qpd.dqm = dev->dqm; + pdd->reset_wavefronts = false; list_add(&pdd->per_device_list, &p->per_device_data); } @@ -396,9 +414,16 @@ void kfd_unbind_process_from_device(struct kfd_dev *dev, unsigned int pasid) mutex_lock(&p->mutex); + if ((dev->dbgmgr) && (dev->dbgmgr->pasid == p->pasid)) + kfd_dbgmgr_destroy(dev->dbgmgr); + pqm_uninit(&p->pqm); pdd = kfd_get_process_device_data(dev, p); + if (pdd->reset_wavefronts) { + dbgdev_wave_reset_wavefronts(pdd->dev, p); + pdd->reset_wavefronts = false; + } /* * Just mark pdd as unbound, because we still need it to call @@ -431,3 +456,23 @@ bool kfd_has_process_device_data(struct kfd_process *p) { return !(list_empty(&p->per_device_data)); } + +/* This returns with process->mutex locked. */ +struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid) +{ + struct kfd_process *p; + unsigned int temp; + + int idx = srcu_read_lock(&kfd_processes_srcu); + + hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) { + if (p->pasid == pasid) { + mutex_lock(&p->mutex); + break; + } + } + + srcu_read_unlock(&kfd_processes_srcu, idx); + + return p; +} diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c index 530b82c4e78b..7b69070f7ecc 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c @@ -158,6 +158,8 @@ int pqm_create_queue(struct process_queue_manager *pqm, struct queue *q; struct process_queue_node *pqn; struct kernel_queue *kq; + int num_queues = 0; + struct queue *cur; BUG_ON(!pqm || !dev || !properties || !qid); @@ -172,6 +174,20 @@ int pqm_create_queue(struct process_queue_manager *pqm, return -1; } + /* + * for debug process, verify that it is within the static queues limit + * currently limit is set to half of the total avail HQD slots + * If we are just about to create DIQ, the is_debug flag is not set yet + * Hence we also check the type as well + */ + if ((pdd->qpd.is_debug) || + (type == KFD_QUEUE_TYPE_DIQ)) { + list_for_each_entry(cur, &pdd->qpd.queues_list, list) + num_queues++; + if (num_queues >= dev->device_info->max_no_of_hqd/2) + return (-ENOSPC); + } + retval = find_available_queue_slot(pqm, qid); if (retval != 0) return retval; @@ -341,7 +357,7 @@ int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid, return 0; } -static __attribute__((unused)) struct kernel_queue *pqm_get_kernel_queue( +struct kernel_queue *pqm_get_kernel_queue( struct process_queue_manager *pqm, unsigned int qid) { |