diff options
author | Debarshi Dutta <ddutta@nvidia.com> | 2018-08-08 08:06:01 -0400 |
---|---|---|
committer | mobile promotions <svcmobile_promotions@nvidia.com> | 2018-09-20 13:50:25 -0400 |
commit | 85c323c3e89d6e1b624b839c3325ae072952e545 (patch) | |
tree | 0822589079047c74fb278f512a51f80baff07b85 /drivers/gpu/nvgpu/os | |
parent | 519948a9c664020fd0b37118749faad2dfd73d97 (diff) |
gpu: nvgpu: restrict devfreq during active clk_arb set requests
Restrict access to devfreq when there are active set requests in the
clk_arbiter. We make the following changes in the patch.
1) Add a global counter in the struct gk20a named clk_arb_global_nr which
is used to track the number of active set requests in the system.
2) Anytime a set request is successfully made by the userspace,
clk_arb_global_nr is incremented by 1 and during the completion of
request(releasing the corresponding file handle), clk_arb_global_nr is
decremented by 1.
3) gk20a_scale_target(invoked by devfreq to set the new frequency based
on load) atomically checks clk_arb_global_nr. If the value = 0, the
code simply continue or else if its > 0, it quits thus making devfreq
requests mutually exclusive with the clk_arbiter.
Bug 2061372
Change-Id: I5d19de03e45520f4ff8fccb97b1f1589d04c8ab8
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1790002
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Diffstat (limited to 'drivers/gpu/nvgpu/os')
-rw-r--r-- | drivers/gpu/nvgpu/os/linux/debug_clk_gm20b.c | 3 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/os/linux/driver_common.c | 3 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/os/linux/ioctl_clk_arb.c | 1 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/os/linux/scale.c | 3 | ||||
-rw-r--r-- | drivers/gpu/nvgpu/os/posix/clk_arb.c | 5 |
5 files changed, 15 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/os/linux/debug_clk_gm20b.c b/drivers/gpu/nvgpu/os/linux/debug_clk_gm20b.c index 52eea83f..b8b95fd7 100644 --- a/drivers/gpu/nvgpu/os/linux/debug_clk_gm20b.c +++ b/drivers/gpu/nvgpu/os/linux/debug_clk_gm20b.c | |||
@@ -17,6 +17,7 @@ | |||
17 | #include <linux/seq_file.h> | 17 | #include <linux/seq_file.h> |
18 | 18 | ||
19 | #include <nvgpu/io.h> | 19 | #include <nvgpu/io.h> |
20 | #include <nvgpu/clk_arb.h> | ||
20 | 21 | ||
21 | #include "gm20b/clk_gm20b.h" | 22 | #include "gm20b/clk_gm20b.h" |
22 | #include "os_linux.h" | 23 | #include "os_linux.h" |
@@ -33,6 +34,8 @@ static int rate_get(void *data, u64 *val) | |||
33 | static int rate_set(void *data, u64 val) | 34 | static int rate_set(void *data, u64 val) |
34 | { | 35 | { |
35 | struct gk20a *g = (struct gk20a *)data; | 36 | struct gk20a *g = (struct gk20a *)data; |
37 | if (nvgpu_clk_arb_has_active_req(g)) | ||
38 | return 0; | ||
36 | return g->ops.clk.set_rate(g, CTRL_CLK_DOMAIN_GPCCLK, (u32)val); | 39 | return g->ops.clk.set_rate(g, CTRL_CLK_DOMAIN_GPCCLK, (u32)val); |
37 | } | 40 | } |
38 | DEFINE_SIMPLE_ATTRIBUTE(rate_fops, rate_get, rate_set, "%llu\n"); | 41 | DEFINE_SIMPLE_ATTRIBUTE(rate_fops, rate_get, rate_set, "%llu\n"); |
diff --git a/drivers/gpu/nvgpu/os/linux/driver_common.c b/drivers/gpu/nvgpu/os/linux/driver_common.c index ebd0b40b..0a25cd3f 100644 --- a/drivers/gpu/nvgpu/os/linux/driver_common.c +++ b/drivers/gpu/nvgpu/os/linux/driver_common.c | |||
@@ -66,6 +66,9 @@ static void nvgpu_init_vars(struct gk20a *g) | |||
66 | nvgpu_mutex_init(&g->tpc_pg_lock); | 66 | nvgpu_mutex_init(&g->tpc_pg_lock); |
67 | nvgpu_mutex_init(&g->clk_arb_enable_lock); | 67 | nvgpu_mutex_init(&g->clk_arb_enable_lock); |
68 | 68 | ||
69 | /* Init the clock req count to 0 */ | ||
70 | nvgpu_atomic_set(&g->clk_arb_global_nr, 0); | ||
71 | |||
69 | l->regs_saved = l->regs; | 72 | l->regs_saved = l->regs; |
70 | l->bar1_saved = l->bar1; | 73 | l->bar1_saved = l->bar1; |
71 | 74 | ||
diff --git a/drivers/gpu/nvgpu/os/linux/ioctl_clk_arb.c b/drivers/gpu/nvgpu/os/linux/ioctl_clk_arb.c index 2d9946b1..ff1cc00a 100644 --- a/drivers/gpu/nvgpu/os/linux/ioctl_clk_arb.c +++ b/drivers/gpu/nvgpu/os/linux/ioctl_clk_arb.c | |||
@@ -436,6 +436,7 @@ int nvgpu_clk_arb_commit_request_fd(struct gk20a *g, | |||
436 | clk_arb_dbg(g, "requested target = %u\n", | 436 | clk_arb_dbg(g, "requested target = %u\n", |
437 | (u32)dev->gpc2clk_target_mhz); | 437 | (u32)dev->gpc2clk_target_mhz); |
438 | 438 | ||
439 | nvgpu_atomic_inc(&g->clk_arb_global_nr); | ||
439 | nvgpu_ref_get(&dev->refcount); | 440 | nvgpu_ref_get(&dev->refcount); |
440 | nvgpu_spinlock_acquire(&session->session_lock); | 441 | nvgpu_spinlock_acquire(&session->session_lock); |
441 | nvgpu_list_add(&dev->node, &session->targets); | 442 | nvgpu_list_add(&dev->node, &session->targets); |
diff --git a/drivers/gpu/nvgpu/os/linux/scale.c b/drivers/gpu/nvgpu/os/linux/scale.c index ecc8207a..cbb9fdc5 100644 --- a/drivers/gpu/nvgpu/os/linux/scale.c +++ b/drivers/gpu/nvgpu/os/linux/scale.c | |||
@@ -25,6 +25,7 @@ | |||
25 | 25 | ||
26 | #include <nvgpu/kmem.h> | 26 | #include <nvgpu/kmem.h> |
27 | #include <nvgpu/log.h> | 27 | #include <nvgpu/log.h> |
28 | #include <nvgpu/clk_arb.h> | ||
28 | 29 | ||
29 | #include "gk20a/gk20a.h" | 30 | #include "gk20a/gk20a.h" |
30 | #include "platform_gk20a.h" | 31 | #include "platform_gk20a.h" |
@@ -154,6 +155,8 @@ static int gk20a_scale_target(struct device *dev, unsigned long *freq, | |||
154 | unsigned long rounded_rate; | 155 | unsigned long rounded_rate; |
155 | unsigned long min_freq = 0, max_freq = 0; | 156 | unsigned long min_freq = 0, max_freq = 0; |
156 | 157 | ||
158 | if (nvgpu_clk_arb_has_active_req(g)) | ||
159 | return 0; | ||
157 | /* | 160 | /* |
158 | * Calculate floor and cap frequency values | 161 | * Calculate floor and cap frequency values |
159 | * | 162 | * |
diff --git a/drivers/gpu/nvgpu/os/posix/clk_arb.c b/drivers/gpu/nvgpu/os/posix/clk_arb.c index 63ab0f13..fcba0a26 100644 --- a/drivers/gpu/nvgpu/os/posix/clk_arb.c +++ b/drivers/gpu/nvgpu/os/posix/clk_arb.c | |||
@@ -52,6 +52,11 @@ int nvgpu_clk_arb_worker_init(struct gk20a *g) | |||
52 | return -ENOSYS; | 52 | return -ENOSYS; |
53 | } | 53 | } |
54 | 54 | ||
55 | bool nvgpu_clk_arb_has_active_req(struct gk20a *g) | ||
56 | { | ||
57 | return false; | ||
58 | } | ||
59 | |||
55 | int nvgpu_clk_arb_get_arbiter_actual_mhz(struct gk20a *g, | 60 | int nvgpu_clk_arb_get_arbiter_actual_mhz(struct gk20a *g, |
56 | u32 api_domain, u16 *actual_mhz) | 61 | u32 api_domain, u16 *actual_mhz) |
57 | { | 62 | { |