From 2c8ebe3687301b2a94bb551d6d1a6570e2b407d2 Mon Sep 17 00:00:00 2001 From: Lingfeng Yang Date: Mon, 13 Jun 2016 09:24:07 -0700 Subject: [PATCH] ANDROID: goldfish: Add goldfish sync driver This is kernel driver for controlling the Goldfish sync device on the host. It is used to maintain ordering in critical OpenGL state changes while using GPU emulation. The guest open()'s the Goldfish sync device to create a context for possibly maintaining sync timeline and fences. There is a 1:1 correspondence between such sync contexts and OpenGL contexts in the guest that need synchronization (which in turn, is anything involving swapping buffers, SurfaceFlinger, or Hardware Composer). The ioctl QUEUE_WORK takes a handle to a sync object and attempts to tell the host GPU to wait on the sync object and deal with signaling it. It possibly outputs a fence FD on which the Android systems that use them (GLConsumer, SurfaceFlinger, anything employing EGL_ANDROID_native_fence_sync) can use to wait. Design decisions and work log: - New approach is to have the guest issue ioctls that trigger host wait, and then host increments timeline. - We need the host's sync object handle and sync thread handle as the necessary information for that. - ioctl() from guest can work simultaneously with the interrupt handling for commands from host. - optimization: don't write back on timeline inc - Change spin lock design to be much more lightweight; do not call sw_sync functions or loop too long anywhere. - Send read/write commands in batches to minimize guest/host transitions. - robustness: BUG if we will overrun the cmd buffer. - robustness: return fd -1 if we cannot get an unused fd. - correctness: remove global mutex - cleanup pass done, incl. but not limited to: - removal of clear_upto and - switching to devm_*** This is part of a sequential, multi-CL change: external/qemu: https://android-review.googlesource.com/239442 <- host-side device's host interface https://android-review.googlesource.com/221593 https://android-review.googlesource.com/248563 https://android-review.googlesource.com/248564 https://android-review.googlesource.com/223032 external/qemu-android: https://android-review.googlesource.com/238790 <- host-side device implementation kernel/goldfish: https://android-review.googlesource.com/232631 <- needed https://android-review.googlesource.com/238399 <- this CL Also squash following bug fixes from android-goldfish-3.18 branch. b44d486 goldfish_sync: provide a signal to detect reboot ad1f597 goldfish_sync: fix stalls by avoiding early kfree() de208e8 [goldfish-sync] Fix possible race between kernel and user space Change-Id: I22f8a0e824717a7e751b1b0e1b461455501502b6 --- arch/x86/configs/i386_ranchu_defconfig | 1 + arch/x86/configs/x86_64_ranchu_defconfig | 1 + drivers/staging/goldfish/Kconfig | 6 + drivers/staging/goldfish/Makefile | 5 + drivers/staging/goldfish/goldfish_sync.c | 987 +++++++++++++++++++++++ 5 files changed, 1000 insertions(+) create mode 100644 drivers/staging/goldfish/goldfish_sync.c diff --git a/arch/x86/configs/i386_ranchu_defconfig b/arch/x86/configs/i386_ranchu_defconfig index b0e4e0ed4b11..0206eb8cfb61 100644 --- a/arch/x86/configs/i386_ranchu_defconfig +++ b/arch/x86/configs/i386_ranchu_defconfig @@ -363,6 +363,7 @@ CONFIG_SYNC=y CONFIG_SW_SYNC=y CONFIG_ION=y CONFIG_GOLDFISH_AUDIO=y +CONFIG_GOLDFISH_SYNC=y CONFIG_SND_HDA_INTEL=y CONFIG_GOLDFISH=y CONFIG_GOLDFISH_PIPE=y diff --git a/arch/x86/configs/x86_64_ranchu_defconfig b/arch/x86/configs/x86_64_ranchu_defconfig index 8dae21ed3ede..dd389774bacb 100644 --- a/arch/x86/configs/x86_64_ranchu_defconfig +++ b/arch/x86/configs/x86_64_ranchu_defconfig @@ -360,6 +360,7 @@ CONFIG_SYNC=y CONFIG_SW_SYNC=y CONFIG_ION=y CONFIG_GOLDFISH_AUDIO=y +CONFIG_GOLDFISH_SYNC=y CONFIG_SND_HDA_INTEL=y CONFIG_GOLDFISH=y CONFIG_GOLDFISH_PIPE=y diff --git a/drivers/staging/goldfish/Kconfig b/drivers/staging/goldfish/Kconfig index 4e094602437c..c579141a7bed 100644 --- a/drivers/staging/goldfish/Kconfig +++ b/drivers/staging/goldfish/Kconfig @@ -4,6 +4,12 @@ config GOLDFISH_AUDIO ---help--- Emulated audio channel for the Goldfish Android Virtual Device +config GOLDFISH_SYNC + tristate "Goldfish AVD Sync Driver" + depends on GOLDFISH + ---help--- + Emulated sync fences for the Goldfish Android Virtual Device + config MTD_GOLDFISH_NAND tristate "Goldfish NAND device" depends on GOLDFISH diff --git a/drivers/staging/goldfish/Makefile b/drivers/staging/goldfish/Makefile index dec34ad58162..0cf525588210 100644 --- a/drivers/staging/goldfish/Makefile +++ b/drivers/staging/goldfish/Makefile @@ -4,3 +4,8 @@ obj-$(CONFIG_GOLDFISH_AUDIO) += goldfish_audio.o obj-$(CONFIG_MTD_GOLDFISH_NAND) += goldfish_nand.o + +# and sync + +ccflags-y := -Idrivers/staging/android +obj-$(CONFIG_GOLDFISH_SYNC) += goldfish_sync.o diff --git a/drivers/staging/goldfish/goldfish_sync.c b/drivers/staging/goldfish/goldfish_sync.c new file mode 100644 index 000000000000..ba8def29901e --- /dev/null +++ b/drivers/staging/goldfish/goldfish_sync.c @@ -0,0 +1,987 @@ +/* + * Copyright (C) 2016 Google, Inc. + * + * This software is licensed under the terms of the GNU General Public + * License version 2, as published by the Free Software Foundation, and + * may be copied, distributed, and modified under those terms. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + */ + +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include + +#include +#include +#include + +#include +#include + +#include "sw_sync.h" +#include "sync.h" + +#define ERR(...) printk(KERN_ERR __VA_ARGS__); + +#define INFO(...) printk(KERN_INFO __VA_ARGS__); + +#define DPRINT(...) pr_debug(__VA_ARGS__); + +#define DTRACE() DPRINT("%s: enter", __func__) + +/* The Goldfish sync driver is designed to provide a interface + * between the underlying host's sync device and the kernel's + * sw_sync. + * The purpose of the device/driver is to enable lightweight + * creation and signaling of timelines and fences + * in order to synchronize the guest with host-side graphics events. + * + * Each time the interrupt trips, the driver + * may perform a sw_sync operation. + */ + +/* The operations are: */ + +/* Ready signal - used to mark when irq should lower */ +#define CMD_SYNC_READY 0 + +/* Create a new timeline. writes timeline handle */ +#define CMD_CREATE_SYNC_TIMELINE 1 + +/* Create a fence object. reads timeline handle and time argument. + * Writes fence fd to the SYNC_REG_HANDLE register. */ +#define CMD_CREATE_SYNC_FENCE 2 + +/* Increments timeline. reads timeline handle and time argument */ +#define CMD_SYNC_TIMELINE_INC 3 + +/* Destroys a timeline. reads timeline handle */ +#define CMD_DESTROY_SYNC_TIMELINE 4 + +/* Starts a wait on the host with + * the given glsync object and sync thread handle. */ +#define CMD_TRIGGER_HOST_WAIT 5 + +/* The register layout is: */ + +#define SYNC_REG_BATCH_COMMAND 0x00 /* host->guest batch commands */ +#define SYNC_REG_BATCH_GUESTCOMMAND 0x04 /* guest->host batch commands */ +#define SYNC_REG_BATCH_COMMAND_ADDR 0x08 /* communicate physical address of host->guest batch commands */ +#define SYNC_REG_BATCH_COMMAND_ADDR_HIGH 0x0c /* 64-bit part */ +#define SYNC_REG_BATCH_GUESTCOMMAND_ADDR 0x10 /* communicate physical address of guest->host commands */ +#define SYNC_REG_BATCH_GUESTCOMMAND_ADDR_HIGH 0x14 /* 64-bit part */ +#define SYNC_REG_INIT 0x18 /* signals that the device has been probed */ + +/* There is an ioctl associated with goldfish sync driver. + * Make it conflict with ioctls that are not likely to be used + * in the emulator. + * + * '@' 00-0F linux/radeonfb.h conflict! + * '@' 00-0F drivers/video/aty/aty128fb.c conflict! + */ +#define GOLDFISH_SYNC_IOC_MAGIC '@' + +#define GOLDFISH_SYNC_IOC_QUEUE_WORK _IOWR(GOLDFISH_SYNC_IOC_MAGIC, 0, struct goldfish_sync_ioctl_info) + +/* The above definitions (command codes, register layout, ioctl definitions) + * need to be in sync with the following files: + * + * Host-side (emulator): + * external/qemu/android/emulation/goldfish_sync.h + * external/qemu-android/hw/misc/goldfish_sync.c + * + * Guest-side (system image): + * device/generic/goldfish-opengl/system/egl/goldfish_sync.h + * device/generic/goldfish/ueventd.ranchu.rc + * platform/build/target/board/generic/sepolicy/file_contexts + */ +struct goldfish_sync_hostcmd { + /* sorted for alignment */ + uint64_t handle; + uint64_t hostcmd_handle; + uint32_t cmd; + uint32_t time_arg; +}; + +struct goldfish_sync_guestcmd { + uint64_t host_command; /* uint64_t for alignment */ + uint64_t glsync_handle; + uint64_t thread_handle; + uint64_t guest_timeline_handle; +}; + +#define GOLDFISH_SYNC_MAX_CMDS 64 + +struct goldfish_sync_state { + char __iomem *reg_base; + int irq; + + /* Spinlock protects |to_do| / |to_do_end|. */ + spinlock_t lock; + /* |mutex_lock| protects all concurrent access + * to timelines for both kernel and user space. */ + struct mutex mutex_lock; + + /* Buffer holding commands issued from host. */ + struct goldfish_sync_hostcmd to_do[GOLDFISH_SYNC_MAX_CMDS]; + uint32_t to_do_end; + + /* Addresses for the reading or writing + * of individual commands. The host can directly write + * to |batch_hostcmd| (and then this driver immediately + * copies contents to |to_do|). This driver either replies + * through |batch_hostcmd| or simply issues a + * guest->host command through |batch_guestcmd|. + */ + struct goldfish_sync_hostcmd *batch_hostcmd; + struct goldfish_sync_guestcmd *batch_guestcmd; + + /* Used to give this struct itself to a work queue + * function for executing actual sync commands. */ + struct work_struct work_item; +}; + +static struct goldfish_sync_state global_sync_state[1]; + +struct goldfish_sync_timeline_obj { + struct sw_sync_timeline *sw_sync_tl; + uint32_t current_time; + /* We need to be careful about when we deallocate + * this |goldfish_sync_timeline_obj| struct. + * In order to ensure proper cleanup, we need to + * consider the triggered host-side wait that may + * still be in flight when the guest close()'s a + * goldfish_sync device's sync context fd (and + * destroys the |sw_sync_tl| field above). + * The host-side wait may raise IRQ + * and tell the kernel to increment the timeline _after_ + * the |sw_sync_tl| has already been set to null. + * + * From observations on OpenGL apps and CTS tests, this + * happens at some very low probability upon context + * destruction or process close, but it does happen + * and it needs to be handled properly. Otherwise, + * if we clean up the surrounding |goldfish_sync_timeline_obj| + * too early, any |handle| field of any host->guest command + * might not even point to a null |sw_sync_tl| field, + * but to garbage memory or even a reclaimed |sw_sync_tl|. + * If we do not count such "pending waits" and kfree the object + * immediately upon |goldfish_sync_timeline_destroy|, + * we might get mysterous RCU stalls after running a long + * time because the garbage memory that is being read + * happens to be interpretable as a |spinlock_t| struct + * that is currently in the locked state. + * + * To track when to free the |goldfish_sync_timeline_obj| + * itself, we maintain a kref. + * The kref essentially counts the timeline itself plus + * the number of waits in flight. kref_init/kref_put + * are issued on + * |goldfish_sync_timeline_create|/|goldfish_sync_timeline_destroy| + * and kref_get/kref_put are issued on + * |goldfish_sync_fence_create|/|goldfish_sync_timeline_inc|. + * + * The timeline is destroyed after reference count + * reaches zero, which would happen after + * |goldfish_sync_timeline_destroy| and all pending + * |goldfish_sync_timeline_inc|'s are fulfilled. + * + * NOTE (1): We assume that |fence_create| and + * |timeline_inc| calls are 1:1, otherwise the kref scheme + * will not work. This is a valid assumption as long + * as the host-side virtual device implementation + * does not insert any timeline increments + * that we did not trigger from here. + * + * NOTE (2): The use of kref by itself requires no locks, + * but this does not mean everything works without locks. + * Related timeline operations do require a lock of some sort, + * or at least are not proven to work without it. + * In particualr, we assume that all the operations + * done on the |kref| field above are done in contexts where + * |global_sync_state->mutex_lock| is held. Do not + * remove that lock until everything is proven to work + * without it!!! */ + struct kref kref; +}; + +/* We will call |delete_timeline_obj| when the last reference count + * of the kref is decremented. This deletes the sw_sync + * timeline object along with the wrapper itself. */ +static void delete_timeline_obj(struct kref* kref) { + struct goldfish_sync_timeline_obj* obj = + container_of(kref, struct goldfish_sync_timeline_obj, kref); + + sync_timeline_destroy(&obj->sw_sync_tl->obj); + obj->sw_sync_tl = NULL; + kfree(obj); +} + +static uint64_t gensym_ctr; +static void gensym(char *dst) +{ + sprintf(dst, "goldfish_sync:gensym:%llu", gensym_ctr); + gensym_ctr++; +} + +/* |goldfish_sync_timeline_create| assumes that |global_sync_state->mutex_lock| + * is held. */ +static struct goldfish_sync_timeline_obj* +goldfish_sync_timeline_create(void) +{ + + char timeline_name[256]; + struct sw_sync_timeline *res_sync_tl = NULL; + struct goldfish_sync_timeline_obj *res; + + DTRACE(); + + gensym(timeline_name); + + res_sync_tl = sw_sync_timeline_create(timeline_name); + if (!res_sync_tl) { + ERR("Failed to create sw_sync timeline."); + return NULL; + } + + res = kzalloc(sizeof(struct goldfish_sync_timeline_obj), GFP_KERNEL); + res->sw_sync_tl = res_sync_tl; + res->current_time = 0; + kref_init(&res->kref); + + DPRINT("new timeline_obj=0x%p", res); + return res; +} + +/* |goldfish_sync_fence_create| assumes that |global_sync_state->mutex_lock| + * is held. */ +static int +goldfish_sync_fence_create(struct goldfish_sync_timeline_obj *obj, + uint32_t val) +{ + + int fd; + char fence_name[256]; + struct sync_pt *syncpt = NULL; + struct sync_fence *sync_obj = NULL; + struct sw_sync_timeline *tl; + + DTRACE(); + + if (!obj) return -1; + + tl = obj->sw_sync_tl; + + syncpt = sw_sync_pt_create(tl, val); + if (!syncpt) { + ERR("could not create sync point! " + "sync_timeline=0x%p val=%d", + tl, val); + return -1; + } + + fd = get_unused_fd_flags(O_CLOEXEC); + if (fd < 0) { + ERR("could not get unused fd for sync fence. " + "errno=%d", fd); + goto err_cleanup_pt; + } + + gensym(fence_name); + + sync_obj = sync_fence_create(fence_name, syncpt); + if (!sync_obj) { + ERR("could not create sync fence! " + "sync_timeline=0x%p val=%d sync_pt=0x%p", + tl, val, syncpt); + goto err_cleanup_fd_pt; + } + + DPRINT("installing sync fence into fd %d sync_obj=0x%p", fd, sync_obj); + sync_fence_install(sync_obj, fd); + kref_get(&obj->kref); + + return fd; + +err_cleanup_fd_pt: + put_unused_fd(fd); +err_cleanup_pt: + sync_pt_free(syncpt); + return -1; +} + +/* |goldfish_sync_timeline_inc| assumes that |global_sync_state->mutex_lock| + * is held. */ +static void +goldfish_sync_timeline_inc(struct goldfish_sync_timeline_obj *obj, uint32_t inc) +{ + DTRACE(); + /* Just give up if someone else nuked the timeline. + * Whoever it was won't care that it doesn't get signaled. */ + if (!obj) return; + + DPRINT("timeline_obj=0x%p", obj); + sw_sync_timeline_inc(obj->sw_sync_tl, inc); + DPRINT("incremented timeline. increment max_time"); + obj->current_time += inc; + + /* Here, we will end up deleting the timeline object if it + * turns out that this call was a pending increment after + * |goldfish_sync_timeline_destroy| was called. */ + kref_put(&obj->kref, delete_timeline_obj); + DPRINT("done"); +} + +/* |goldfish_sync_timeline_destroy| assumes + * that |global_sync_state->mutex_lock| is held. */ +static void +goldfish_sync_timeline_destroy(struct goldfish_sync_timeline_obj *obj) +{ + DTRACE(); + /* See description of |goldfish_sync_timeline_obj| for why we + * should not immediately destroy |obj| */ + kref_put(&obj->kref, delete_timeline_obj); +} + +static inline void +goldfish_sync_cmd_queue(struct goldfish_sync_state *sync_state, + uint32_t cmd, + uint64_t handle, + uint32_t time_arg, + uint64_t hostcmd_handle) +{ + struct goldfish_sync_hostcmd *to_add; + + DTRACE(); + + BUG_ON(sync_state->to_do_end == GOLDFISH_SYNC_MAX_CMDS); + + to_add = &sync_state->to_do[sync_state->to_do_end]; + + to_add->cmd = cmd; + to_add->handle = handle; + to_add->time_arg = time_arg; + to_add->hostcmd_handle = hostcmd_handle; + + sync_state->to_do_end += 1; +} + +static inline void +goldfish_sync_hostcmd_reply(struct goldfish_sync_state *sync_state, + uint32_t cmd, + uint64_t handle, + uint32_t time_arg, + uint64_t hostcmd_handle) +{ + unsigned long irq_flags; + struct goldfish_sync_hostcmd *batch_hostcmd = + sync_state->batch_hostcmd; + + DTRACE(); + + spin_lock_irqsave(&sync_state->lock, irq_flags); + + batch_hostcmd->cmd = cmd; + batch_hostcmd->handle = handle; + batch_hostcmd->time_arg = time_arg; + batch_hostcmd->hostcmd_handle = hostcmd_handle; + writel(0, sync_state->reg_base + SYNC_REG_BATCH_COMMAND); + + spin_unlock_irqrestore(&sync_state->lock, irq_flags); +} + +static inline void +goldfish_sync_send_guestcmd(struct goldfish_sync_state *sync_state, + uint32_t cmd, + uint64_t glsync_handle, + uint64_t thread_handle, + uint64_t timeline_handle) +{ + unsigned long irq_flags; + struct goldfish_sync_guestcmd *batch_guestcmd = + sync_state->batch_guestcmd; + + DTRACE(); + + spin_lock_irqsave(&sync_state->lock, irq_flags); + + batch_guestcmd->host_command = (uint64_t)cmd; + batch_guestcmd->glsync_handle = (uint64_t)glsync_handle; + batch_guestcmd->thread_handle = (uint64_t)thread_handle; + batch_guestcmd->guest_timeline_handle = (uint64_t)timeline_handle; + writel(0, sync_state->reg_base + SYNC_REG_BATCH_GUESTCOMMAND); + + spin_unlock_irqrestore(&sync_state->lock, irq_flags); +} + +/* |goldfish_sync_interrupt| handles IRQ raises from the virtual device. + * In the context of OpenGL, this interrupt will fire whenever we need + * to signal a fence fd in the guest, with the command + * |CMD_SYNC_TIMELINE_INC|. + * However, because this function will be called in an interrupt context, + * it is necessary to do the actual work of signaling off of interrupt context. + * The shared work queue is used for this purpose. At the end when + * all pending commands are intercepted by the interrupt handler, + * we call |schedule_work|, which will later run the actual + * desired sync command in |goldfish_sync_work_item_fn|. + */ +static irqreturn_t goldfish_sync_interrupt(int irq, void *dev_id) +{ + + struct goldfish_sync_state *sync_state = dev_id; + + uint32_t nextcmd; + uint32_t command_r; + uint64_t handle_rw; + uint32_t time_r; + uint64_t hostcmd_handle_rw; + + int count = 0; + + DTRACE(); + + sync_state = dev_id; + + spin_lock(&sync_state->lock); + + for (;;) { + + readl(sync_state->reg_base + SYNC_REG_BATCH_COMMAND); + nextcmd = sync_state->batch_hostcmd->cmd; + + if (nextcmd == 0) + break; + + command_r = nextcmd; + handle_rw = sync_state->batch_hostcmd->handle; + time_r = sync_state->batch_hostcmd->time_arg; + hostcmd_handle_rw = sync_state->batch_hostcmd->hostcmd_handle; + + goldfish_sync_cmd_queue( + sync_state, + command_r, + handle_rw, + time_r, + hostcmd_handle_rw); + + count++; + } + + spin_unlock(&sync_state->lock); + + schedule_work(&sync_state->work_item); + + return (count == 0) ? IRQ_NONE : IRQ_HANDLED; +} + +/* |goldfish_sync_work_item_fn| does the actual work of servicing + * host->guest sync commands. This function is triggered whenever + * the IRQ for the goldfish sync device is raised. Once it starts + * running, it grabs the contents of the buffer containing the + * commands it needs to execute (there may be multiple, because + * our IRQ is active high and not edge triggered), and then + * runs all of them one after the other. + */ +static void goldfish_sync_work_item_fn(struct work_struct *input) +{ + + struct goldfish_sync_state *sync_state; + int sync_fence_fd; + + struct goldfish_sync_timeline_obj *timeline; + uint64_t timeline_ptr; + + uint64_t hostcmd_handle; + + uint32_t cmd; + uint64_t handle; + uint32_t time_arg; + + struct goldfish_sync_hostcmd *todo; + uint32_t todo_end; + + unsigned long irq_flags; + + struct goldfish_sync_hostcmd to_run[GOLDFISH_SYNC_MAX_CMDS]; + uint32_t i = 0; + + sync_state = container_of(input, struct goldfish_sync_state, work_item); + + mutex_lock(&sync_state->mutex_lock); + + spin_lock_irqsave(&sync_state->lock, irq_flags); { + + todo_end = sync_state->to_do_end; + + DPRINT("num sync todos: %u", sync_state->to_do_end); + + for (i = 0; i < todo_end; i++) + to_run[i] = sync_state->to_do[i]; + + /* We expect that commands will come in at a slow enough rate + * so that incoming items will not be more than + * GOLDFISH_SYNC_MAX_CMDS. + * + * This is because the way the sync device is used, + * it's only for managing buffer data transfers per frame, + * with a sequential dependency between putting things in + * to_do and taking them out. Once a set of commands is + * queued up in to_do, the user of the device waits for + * them to be processed before queuing additional commands, + * which limits the rate at which commands come in + * to the rate at which we take them out here. + * + * We also don't expect more than MAX_CMDS to be issued + * at once; there is a correspondence between + * which buffers need swapping to the (display / buffer queue) + * to particular commands, and we don't expect there to be + * enough display or buffer queues in operation at once + * to overrun GOLDFISH_SYNC_MAX_CMDS. + */ + sync_state->to_do_end = 0; + + } spin_unlock_irqrestore(&sync_state->lock, irq_flags); + + for (i = 0; i < todo_end; i++) { + DPRINT("todo index: %u", i); + + todo = &to_run[i]; + + cmd = todo->cmd; + + handle = (uint64_t)todo->handle; + time_arg = todo->time_arg; + hostcmd_handle = (uint64_t)todo->hostcmd_handle; + + DTRACE(); + + timeline = (struct goldfish_sync_timeline_obj *)(uintptr_t)handle; + + switch (cmd) { + case CMD_SYNC_READY: + break; + case CMD_CREATE_SYNC_TIMELINE: + DPRINT("exec CMD_CREATE_SYNC_TIMELINE: " + "handle=0x%llx time_arg=%d", + handle, time_arg); + timeline = goldfish_sync_timeline_create(); + timeline_ptr = (uintptr_t)timeline; + goldfish_sync_hostcmd_reply(sync_state, CMD_CREATE_SYNC_TIMELINE, + timeline_ptr, + 0, + hostcmd_handle); + DPRINT("sync timeline created: %p", timeline); + break; + case CMD_CREATE_SYNC_FENCE: + DPRINT("exec CMD_CREATE_SYNC_FENCE: " + "handle=0x%llx time_arg=%d", + handle, time_arg); + sync_fence_fd = goldfish_sync_fence_create(timeline, time_arg); + goldfish_sync_hostcmd_reply(sync_state, CMD_CREATE_SYNC_FENCE, + sync_fence_fd, + 0, + hostcmd_handle); + break; + case CMD_SYNC_TIMELINE_INC: + DPRINT("exec CMD_SYNC_TIMELINE_INC: " + "handle=0x%llx time_arg=%d", + handle, time_arg); + goldfish_sync_timeline_inc(timeline, time_arg); + break; + case CMD_DESTROY_SYNC_TIMELINE: + DPRINT("exec CMD_DESTROY_SYNC_TIMELINE: " + "handle=0x%llx time_arg=%d", + handle, time_arg); + goldfish_sync_timeline_destroy(timeline); + break; + } + DPRINT("Done executing sync command"); + } + mutex_unlock(&sync_state->mutex_lock); +} + +/* Guest-side interface: file operations */ + +/* Goldfish sync context and ioctl info. + * + * When a sync context is created by open()-ing the goldfish sync device, we + * create a sync context (|goldfish_sync_context|). + * + * Currently, the only data required to track is the sync timeline itself + * along with the current time, which are all packed up in the + * |goldfish_sync_timeline_obj| field. We use a |goldfish_sync_context| + * as the filp->private_data. + * + * Next, when a sync context user requests that work be queued and a fence + * fd provided, we use the |goldfish_sync_ioctl_info| struct, which holds + * information about which host handles to touch for this particular + * queue-work operation. We need to know about the host-side sync thread + * and the particular host-side GLsync object. We also possibly write out + * a file descriptor. + */ +struct goldfish_sync_context { + struct goldfish_sync_timeline_obj *timeline; +}; + +struct goldfish_sync_ioctl_info { + uint64_t host_glsync_handle_in; + uint64_t host_syncthread_handle_in; + int fence_fd_out; +}; + +static int goldfish_sync_open(struct inode *inode, struct file *file) +{ + + struct goldfish_sync_context *sync_context; + + DTRACE(); + + mutex_lock(&global_sync_state->mutex_lock); + + sync_context = kzalloc(sizeof(struct goldfish_sync_context), GFP_KERNEL); + + if (sync_context == NULL) { + ERR("Creation of goldfish sync context failed!"); + mutex_unlock(&global_sync_state->mutex_lock); + return -ENOMEM; + } + + sync_context->timeline = NULL; + + file->private_data = sync_context; + + DPRINT("successfully create a sync context @0x%p", sync_context); + + mutex_unlock(&global_sync_state->mutex_lock); + + return 0; +} + +static int goldfish_sync_release(struct inode *inode, struct file *file) +{ + + struct goldfish_sync_context *sync_context; + + DTRACE(); + + mutex_lock(&global_sync_state->mutex_lock); + + sync_context = file->private_data; + + if (sync_context->timeline) + goldfish_sync_timeline_destroy(sync_context->timeline); + + sync_context->timeline = NULL; + + kfree(sync_context); + + mutex_unlock(&global_sync_state->mutex_lock); + + return 0; +} + +/* |goldfish_sync_ioctl| is the guest-facing interface of goldfish sync + * and is used in conjunction with eglCreateSyncKHR to queue up the + * actual work of waiting for the EGL sync command to complete, + * possibly returning a fence fd to the guest. + */ +static long goldfish_sync_ioctl(struct file *file, + unsigned int cmd, + unsigned long arg) +{ + struct goldfish_sync_context *sync_context_data; + struct goldfish_sync_timeline_obj *timeline; + int fd_out; + struct goldfish_sync_ioctl_info ioctl_data; + + DTRACE(); + + sync_context_data = file->private_data; + fd_out = -1; + + switch (cmd) { + case GOLDFISH_SYNC_IOC_QUEUE_WORK: + + DPRINT("exec GOLDFISH_SYNC_IOC_QUEUE_WORK"); + + mutex_lock(&global_sync_state->mutex_lock); + + if (copy_from_user(&ioctl_data, + (void __user *)arg, + sizeof(ioctl_data))) { + ERR("Failed to copy memory for ioctl_data from user."); + mutex_unlock(&global_sync_state->mutex_lock); + return -EFAULT; + } + + if (ioctl_data.host_syncthread_handle_in == 0) { + DPRINT("Error: zero host syncthread handle!!!"); + mutex_unlock(&global_sync_state->mutex_lock); + return -EFAULT; + } + + if (!sync_context_data->timeline) { + DPRINT("no timeline yet, create one."); + sync_context_data->timeline = goldfish_sync_timeline_create(); + DPRINT("timeline: 0x%p", &sync_context_data->timeline); + } + + timeline = sync_context_data->timeline; + fd_out = goldfish_sync_fence_create(timeline, + timeline->current_time + 1); + DPRINT("Created fence with fd %d and current time %u (timeline: 0x%p)", + fd_out, + sync_context_data->timeline->current_time + 1, + sync_context_data->timeline); + + ioctl_data.fence_fd_out = fd_out; + + if (copy_to_user((void __user *)arg, + &ioctl_data, + sizeof(ioctl_data))) { + DPRINT("Error, could not copy to user!!!"); + + sys_close(fd_out); + /* We won't be doing an increment, kref_put immediately. */ + kref_put(&timeline->kref, delete_timeline_obj); + mutex_unlock(&global_sync_state->mutex_lock); + return -EFAULT; + } + + /* We are now about to trigger a host-side wait; + * accumulate on |pending_waits|. */ + goldfish_sync_send_guestcmd(global_sync_state, + CMD_TRIGGER_HOST_WAIT, + ioctl_data.host_glsync_handle_in, + ioctl_data.host_syncthread_handle_in, + (uint64_t)(uintptr_t)(sync_context_data->timeline)); + + mutex_unlock(&global_sync_state->mutex_lock); + return 0; + default: + return -ENOTTY; + } +} + +static const struct file_operations goldfish_sync_fops = { + .owner = THIS_MODULE, + .open = goldfish_sync_open, + .release = goldfish_sync_release, + .unlocked_ioctl = goldfish_sync_ioctl, + .compat_ioctl = goldfish_sync_ioctl, +}; + +static struct miscdevice goldfish_sync_device = { + .name = "goldfish_sync", + .fops = &goldfish_sync_fops, +}; + + +static bool setup_verify_batch_cmd_addr(struct goldfish_sync_state *sync_state, + void *batch_addr, + uint32_t addr_offset, + uint32_t addr_offset_high) +{ + uint64_t batch_addr_phys; + uint32_t batch_addr_phys_test_lo; + uint32_t batch_addr_phys_test_hi; + + if (!batch_addr) { + ERR("Could not use batch command address!"); + return false; + } + + batch_addr_phys = virt_to_phys(batch_addr); + writel((uint32_t)(batch_addr_phys), + sync_state->reg_base + addr_offset); + writel((uint32_t)(batch_addr_phys >> 32), + sync_state->reg_base + addr_offset_high); + + batch_addr_phys_test_lo = + readl(sync_state->reg_base + addr_offset); + batch_addr_phys_test_hi = + readl(sync_state->reg_base + addr_offset_high); + + if (virt_to_phys(batch_addr) != + (((uint64_t)batch_addr_phys_test_hi << 32) | + batch_addr_phys_test_lo)) { + ERR("Invalid batch command address!"); + return false; + } + + return true; +} + +int goldfish_sync_probe(struct platform_device *pdev) +{ + struct resource *ioresource; + struct goldfish_sync_state *sync_state = global_sync_state; + int status; + + DTRACE(); + + sync_state->to_do_end = 0; + + spin_lock_init(&sync_state->lock); + mutex_init(&sync_state->mutex_lock); + + platform_set_drvdata(pdev, sync_state); + + ioresource = platform_get_resource(pdev, IORESOURCE_MEM, 0); + if (ioresource == NULL) { + ERR("platform_get_resource failed"); + return -ENODEV; + } + + sync_state->reg_base = devm_ioremap(&pdev->dev, ioresource->start, PAGE_SIZE); + if (sync_state->reg_base == NULL) { + ERR("Could not ioremap"); + return -ENOMEM; + } + + sync_state->irq = platform_get_irq(pdev, 0); + if (sync_state->irq < 0) { + ERR("Could not platform_get_irq"); + return -ENODEV; + } + + status = devm_request_irq(&pdev->dev, + sync_state->irq, + goldfish_sync_interrupt, + IRQF_SHARED, + pdev->name, + sync_state); + if (status) { + ERR("request_irq failed"); + return -ENODEV; + } + + INIT_WORK(&sync_state->work_item, + goldfish_sync_work_item_fn); + + misc_register(&goldfish_sync_device); + + /* Obtain addresses for batch send/recv of commands. */ + { + struct goldfish_sync_hostcmd *batch_addr_hostcmd; + struct goldfish_sync_guestcmd *batch_addr_guestcmd; + + batch_addr_hostcmd = devm_kzalloc(&pdev->dev, sizeof(struct goldfish_sync_hostcmd), + GFP_KERNEL); + batch_addr_guestcmd = devm_kzalloc(&pdev->dev, sizeof(struct goldfish_sync_guestcmd), + GFP_KERNEL); + + if (!setup_verify_batch_cmd_addr(sync_state, + batch_addr_hostcmd, + SYNC_REG_BATCH_COMMAND_ADDR, + SYNC_REG_BATCH_COMMAND_ADDR_HIGH)) { + ERR("goldfish_sync: Could not setup batch command address"); + return -ENODEV; + } + + if (!setup_verify_batch_cmd_addr(sync_state, + batch_addr_guestcmd, + SYNC_REG_BATCH_GUESTCOMMAND_ADDR, + SYNC_REG_BATCH_GUESTCOMMAND_ADDR_HIGH)) { + ERR("goldfish_sync: Could not setup batch guest command address"); + return -ENODEV; + } + + sync_state->batch_hostcmd = batch_addr_hostcmd; + sync_state->batch_guestcmd = batch_addr_guestcmd; + } + + INFO("goldfish_sync: Initialized goldfish sync device"); + + writel(0, sync_state->reg_base + SYNC_REG_INIT); + + return 0; +} + +static int goldfish_sync_remove(struct platform_device *pdev) +{ + struct goldfish_sync_state *sync_state = global_sync_state; + + DTRACE(); + + misc_deregister(&goldfish_sync_device); + memset(sync_state, 0, sizeof(struct goldfish_sync_state)); + return 0; +} + +static const struct of_device_id goldfish_sync_of_match[] = { + { .compatible = "google,goldfish-sync", }, + {}, +}; +MODULE_DEVICE_TABLE(of, goldfish_sync_of_match); + +static const struct acpi_device_id goldfish_sync_acpi_match[] = { + { "GFSH0006", 0 }, + { }, +}; + +MODULE_DEVICE_TABLE(acpi, goldfish_sync_acpi_match); + +static struct platform_driver goldfish_sync = { + .probe = goldfish_sync_probe, + .remove = goldfish_sync_remove, + .driver = { + .name = "goldfish_sync", + .of_match_table = goldfish_sync_of_match, + .acpi_match_table = ACPI_PTR(goldfish_sync_acpi_match), + } +}; + +module_platform_driver(goldfish_sync); + +MODULE_AUTHOR("Google, Inc."); +MODULE_DESCRIPTION("Android QEMU Sync Driver"); +MODULE_LICENSE("GPL"); +MODULE_VERSION("1.0"); + +/* This function is only to run a basic test of sync framework. + * It creates a timeline and fence object whose signal point is at 1. + * The timeline is incremented, and we use the sync framework's + * sync_fence_wait on that fence object. If everything works out, + * we should not hang in the wait and return immediately. + * There is no way to explicitly run this test yet, but it + * can be used by inserting it at the end of goldfish_sync_probe. + */ +void test_kernel_sync(void) +{ + struct goldfish_sync_timeline_obj *test_timeline; + int test_fence_fd; + + DTRACE(); + + DPRINT("test sw_sync"); + + test_timeline = goldfish_sync_timeline_create(); + DPRINT("sw_sync_timeline_create -> 0x%p", test_timeline); + + test_fence_fd = goldfish_sync_fence_create(test_timeline, 1); + DPRINT("sync_fence_create -> %d", test_fence_fd); + + DPRINT("incrementing test timeline"); + goldfish_sync_timeline_inc(test_timeline, 1); + + DPRINT("test waiting (should NOT hang)"); + sync_fence_wait( + sync_fence_fdget(test_fence_fd), -1); + + DPRINT("test waiting (afterward)"); +} -- 2.20.1