From 861b11a968b1f51f45832486e62bfe23fc29fc19 Mon Sep 17 00:00:00 2001 From: Deepak Nibade Date: Thu, 23 Nov 2017 01:03:24 -0800 Subject: gpu: nvgpu: move snapshot_client memory handling to linux We right now store dmabuf fd and dma_buf pointer for gk20a_cs_snapshot_client But since dma_buf and all related APIs are linux specific, we need to remove them from common code and move them to linux specific code Add new linux specific structure gk20a_cs_snapshot_client_linux which includes struct gk20a_cs_snapshot_client and linux specific dma_buf pointer In gk20a_attach_cycle_stats_snapshot(), we first handle all dma_buf related operations and then call gr_gk20a_css_attach() Move gk20a_channel_free_cycle_stats_snapshot() to ioctl_channel.c In gk20a_channel_free_cycle_stats_snapshot(), we call gr_gk20a_css_detach() and then free up dma_buf in linux specific code We also need to call gk20a_channel_free_cycle_stats_snapshot() while closing the channel, so call it from linux specific nvgpu_channel_close_linux() Jira NVGPU-397 Jira NVGPU-415 Change-Id: Ida27240541f6adf31f28d7d7ee4f51651c6d3de2 Signed-off-by: Deepak Nibade Reviewed-on: https://git-master.nvidia.com/r/1603908 GVS: Gerrit_Virtual_Submit Reviewed-by: Terje Bergstrom Reviewed-by: mobile promotions Tested-by: mobile promotions --- drivers/gpu/nvgpu/common/linux/channel.c | 5 ++ drivers/gpu/nvgpu/common/linux/ioctl_channel.c | 95 +++++++++++++++++++++++--- drivers/gpu/nvgpu/common/linux/ioctl_channel.h | 11 +++ 3 files changed, 103 insertions(+), 8 deletions(-) (limited to 'drivers/gpu/nvgpu/common') diff --git a/drivers/gpu/nvgpu/common/linux/channel.c b/drivers/gpu/nvgpu/common/linux/channel.c index 1ae2d444..0ed596ac 100644 --- a/drivers/gpu/nvgpu/common/linux/channel.c +++ b/drivers/gpu/nvgpu/common/linux/channel.c @@ -29,6 +29,7 @@ #include "gk20a/gk20a.h" #include "channel.h" +#include "ioctl_channel.h" #include "os_linux.h" #include @@ -242,6 +243,10 @@ static void nvgpu_channel_open_linux(struct channel_gk20a *ch) static void nvgpu_channel_close_linux(struct channel_gk20a *ch) { nvgpu_channel_work_completion_clear(ch); + +#if defined(CONFIG_GK20A_CYCLE_STATS) + gk20a_channel_free_cycle_stats_snapshot(ch); +#endif } static int nvgpu_channel_alloc_linux(struct gk20a *g, struct channel_gk20a *ch) diff --git a/drivers/gpu/nvgpu/common/linux/ioctl_channel.c b/drivers/gpu/nvgpu/common/linux/ioctl_channel.c index 67bec31b..13355605 100644 --- a/drivers/gpu/nvgpu/common/linux/ioctl_channel.c +++ b/drivers/gpu/nvgpu/common/linux/ioctl_channel.c @@ -42,6 +42,11 @@ #include "os_linux.h" #include "ctxsw_trace.h" +/* the minimal size of client buffer */ +#define CSS_MIN_CLIENT_SNAPSHOT_SIZE \ + (sizeof(struct gk20a_cs_snapshot_fifo) + \ + sizeof(struct gk20a_cs_snapshot_fifo_entry) * 256) + static const char *gr_gk20a_graphics_preempt_mode_name(u32 graphics_preempt_mode) { switch (graphics_preempt_mode) { @@ -157,18 +162,92 @@ static int gk20a_attach_cycle_stats_snapshot(struct channel_gk20a *ch, u32 perfmon_id_count, u32 *perfmon_id_start) { - int ret; + int ret = 0; + struct gk20a *g = ch->g; + struct gk20a_cs_snapshot_client_linux *client_linux; + struct gk20a_cs_snapshot_client *client; nvgpu_mutex_acquire(&ch->cs_client_mutex); if (ch->cs_client) { - ret = -EEXIST; - } else { - ret = gr_gk20a_css_attach(ch, - dmabuf_fd, - perfmon_id_count, - perfmon_id_start, - &ch->cs_client); + nvgpu_mutex_release(&ch->cs_client_mutex); + return -EEXIST; + } + + client_linux = nvgpu_kzalloc(g, sizeof(*client_linux)); + if (!client_linux) { + ret = -ENOMEM; + goto err; + } + + client_linux->dmabuf_fd = dmabuf_fd; + client_linux->dma_handler = dma_buf_get(client_linux->dmabuf_fd); + if (IS_ERR(client_linux->dma_handler)) { + ret = PTR_ERR(client_linux->dma_handler); + client_linux->dma_handler = NULL; + goto err_free; + } + + client = &client_linux->cs_client; + client->snapshot_size = client_linux->dma_handler->size; + if (client->snapshot_size < CSS_MIN_CLIENT_SNAPSHOT_SIZE) { + ret = -ENOMEM; + goto err_put; + } + + client->snapshot = (struct gk20a_cs_snapshot_fifo *) + dma_buf_vmap(client_linux->dma_handler); + if (!client->snapshot) { + ret = -ENOMEM; + goto err_put; + } + + ch->cs_client = client; + + ret = gr_gk20a_css_attach(ch, + perfmon_id_count, + perfmon_id_start, + ch->cs_client); + + nvgpu_mutex_release(&ch->cs_client_mutex); + + return ret; + +err_put: + dma_buf_put(client_linux->dma_handler); +err_free: + nvgpu_kfree(g, client_linux); +err: + nvgpu_mutex_release(&ch->cs_client_mutex); + return ret; +} + +int gk20a_channel_free_cycle_stats_snapshot(struct channel_gk20a *ch) +{ + int ret; + struct gk20a_cs_snapshot_client_linux *client_linux; + + nvgpu_mutex_acquire(&ch->cs_client_mutex); + if (!ch->cs_client) { + nvgpu_mutex_release(&ch->cs_client_mutex); + return 0; } + + client_linux = container_of(ch->cs_client, + struct gk20a_cs_snapshot_client_linux, + cs_client); + + ret = gr_gk20a_css_detach(ch, ch->cs_client); + + if (client_linux->dma_handler) { + if (ch->cs_client->snapshot) + dma_buf_vunmap(client_linux->dma_handler, + ch->cs_client->snapshot); + dma_buf_put(client_linux->dma_handler); + } + + ch->cs_client = NULL; + nvgpu_kfree(ch->g, client_linux); + nvgpu_mutex_release(&ch->cs_client_mutex); return ret; diff --git a/drivers/gpu/nvgpu/common/linux/ioctl_channel.h b/drivers/gpu/nvgpu/common/linux/ioctl_channel.h index 235d84ef..3ea8d765 100644 --- a/drivers/gpu/nvgpu/common/linux/ioctl_channel.h +++ b/drivers/gpu/nvgpu/common/linux/ioctl_channel.h @@ -15,11 +15,20 @@ #include +#include "gk20a/css_gr_gk20a.h" + struct inode; struct file; struct gk20a; struct nvgpu_channel_open_args; +struct gk20a_cs_snapshot_client_linux { + struct gk20a_cs_snapshot_client cs_client; + + u32 dmabuf_fd; + struct dma_buf *dma_handler; +}; + int gk20a_channel_open(struct inode *inode, struct file *filp); int gk20a_channel_release(struct inode *inode, struct file *filp); long gk20a_channel_ioctl(struct file *filp, @@ -27,6 +36,8 @@ long gk20a_channel_ioctl(struct file *filp, int gk20a_channel_open_ioctl(struct gk20a *g, struct nvgpu_channel_open_args *args); +int gk20a_channel_free_cycle_stats_snapshot(struct channel_gk20a *ch); + extern const struct file_operations gk20a_event_id_ops; extern const struct file_operations gk20a_channel_ops; -- cgit v1.2.2