nv-tegra-mirror/linux-nvgpu
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nvgpu.git synced 2025-12-24 10:34:43 +03:00
Code Issues Packages Projects Releases Wiki Activity

gpu: nvgpu: check ctx valid bit at right place

Browse Source 
When contexts are unloaded from GR, the valid bit is reset
but the instance pointer information remains intact.
Check valid bit in *is_channel_ctx_resident* function as
valid bit might not be set when *get_channel_from_ctx function*
is called from gr_isr

Bug 200289491

Change-Id: I4da7f04794c7e7e80b511756dbd851205cd76fbc
Signed-off-by: Seema Khowala <seemaj@nvidia.com>
Reviewed-on: http://git-master/r/1505908
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This commit is contained in:
Seema Khowala
2017-06-20 12:25:05 -07:00
committed by mobile promotions
parent 8ca81687c7
commit 581e8b4ec8
1 changed files with 17 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
drivers/gpu/nvgpu/gk20a/gr_gk20a.c
View File

@@ -5906,11 +5906,10 @@ static struct channel_gk20a *gk20a_gr_get_channel_from_ctx(
struct channel_gk20a *ret = NULL;
/* when contexts are unloaded from GR, the valid bit is reset
* but the instance pointer information remains intact. So the
* valid bit must be checked to be absolutely certain that a
* valid context is currently resident. */
if (!gr_fecs_current_ctx_valid_v(curr_ctx))
return NULL;
* but the instance pointer information remains intact.
* This might be called from gr_isr where contexts might be
* unloaded. No need to check ctx_valid bit
*/
nvgpu_spinlock_acquire(&gr->ch_tlb_lock);
@@ -6475,10 +6474,12 @@ int gk20a_gr_isr(struct gk20a *g)
isr_data.class_num = gr_fe_object_table_nvclass_v(obj_table);
ch = gk20a_gr_get_channel_from_ctx(g, isr_data.curr_ctx, &tsgid);
if (ch)
if (ch) {
isr_data.chid = ch->hw_chid;
else
} else {
isr_data.chid = FIFO_INVAL_CHANNEL_ID;
nvgpu_err(g, "ch id is INVALID 0xffffffff");
}
gk20a_dbg(gpu_dbg_intr | gpu_dbg_gpu_dbg,
"channel %d: addr 0x%08x, "
@@ -8283,6 +8284,15 @@ bool gk20a_is_channel_ctx_resident(struct channel_gk20a *ch)
bool ret = false;
curr_gr_ctx = gk20a_readl(g, gr_fecs_current_ctx_r());
/* when contexts are unloaded from GR, the valid bit is reset
* but the instance pointer information remains intact. So the
* valid bit must be checked to be absolutely certain that a
* valid context is currently resident.
*/
if (!gr_fecs_current_ctx_valid_v(curr_gr_ctx))
return NULL;
curr_ch = gk20a_gr_get_channel_from_ctx(g, curr_gr_ctx,
&curr_gr_tsgid);