gpu:nvgpu:gm20b: disable irqs when hs pmu executes

bug 200040021

polling halt irq to check for hs bin completion
keep irqs disabled to avoid executing irq handler

Change-Id: Ic245d89580444dcbf1cf5ec34bfe0f8b0c5bbc0f
Signed-off-by: Vijayakumar <vsubbu@nvidia.com>
Reviewed-on: http://git-master/r/554659
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>

4 files changed, 45 insertions, 7 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
index 64b126f6..7a62f05c 100644
--- a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.c
@@ -1048,7 +1048,7 @@ static int pmu_idle(struct pmu_gk20a *pmu)
         return 0;
 }
 
-static void pmu_enable_irq(struct pmu_gk20a *pmu, bool enable)
+void pmu_enable_irq(struct pmu_gk20a *pmu, bool enable)
 {
         struct gk20a *g = gk20a_from_pmu(pmu);
 
diff --git a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
index 048520f9..bc5e474a 100644
--- a/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/pmu_gk20a.h
@@ -1145,5 +1145,5 @@ int gk20a_pmu_ap_send_command(struct gk20a *g,
                 union pmu_ap_cmd *p_ap_cmd, bool b_block);
 int gk20a_aelpg_init(struct gk20a *g);
 int gk20a_aelpg_init_and_enable(struct gk20a *g, u8 ctrl_id);
-
+void pmu_enable_irq(struct pmu_gk20a *pmu, bool enable);
 #endif /*__PMU_GK20A_H__*/
diff --git a/drivers/gpu/nvgpu/gm20b/acr_gm20b.c b/drivers/gpu/nvgpu/gm20b/acr_gm20b.c
index b717272c..acfcf41b 100644
--- a/drivers/gpu/nvgpu/gm20b/acr_gm20b.c
+++ b/drivers/gpu/nvgpu/gm20b/acr_gm20b.c
@@ -63,6 +63,11 @@ get_ucode_details pmu_acr_supp_ucode_list[] = {
 /*Once is LS mode, cpuctl_alias is only accessible*/
 void start_gm20b_pmu(struct gk20a *g)
 {
+        /*disable irqs for hs falcon booting as we will poll for halt*/
+        mutex_lock(&g->pmu.isr_mutex);
+        pmu_enable_irq(&g->pmu, true);
+        g->pmu.isr_enabled = true;
+        mutex_unlock(&g->pmu.isr_mutex);
         gk20a_writel(g, pwr_falcon_cpuctl_alias_r(),
                 pwr_falcon_cpuctl_startcpu_f(1));
 }
@@ -1116,6 +1121,11 @@ int gm20b_init_pmu_setup_hw1(struct gk20a *g, struct flcn_bl_dmem_desc *desc,
         pmu_copy_to_dmem(pmu, g->acr.pmu_args,
                         (u8 *)(g->ops.pmu_ver.get_pmu_cmdline_args_ptr(pmu)),
                         g->ops.pmu_ver.get_pmu_cmdline_args_size(pmu), 0);
+        /*disable irqs for hs falcon booting as we will poll for halt*/
+        mutex_lock(&pmu->isr_mutex);
+        pmu_enable_irq(pmu, false);
+        pmu->isr_enabled = false;
+        mutex_unlock(&pmu->isr_mutex);
         err = bl_bootstrap(pmu, desc, bl_sz);
         if (err)
                 return err;
@@ -1216,8 +1226,9 @@ int pmu_exec_gen_bl(struct gk20a *g, void *desc, u8 b_wait_for_halt)
          * to PMU halt
          */
 
-        gk20a_writel(g, pwr_falcon_irqsclr_r(),
-                gk20a_readl(g, pwr_falcon_irqsclr_r()) & (~(0x10)));
+        if (clear_halt_interrupt_status(g, GPU_TIMEOUT_DEFAULT))
+                goto err_unmap_bl;
+
         gm20b_dbg_pmu("err reg :%x\n", readl(mc +
                 MC_ERR_GENERALIZED_CARVEOUT_STATUS_0));
         gm20b_dbg_pmu("phys sec reg %x\n", gk20a_readl(g,
@@ -1228,10 +1239,11 @@ int pmu_exec_gen_bl(struct gk20a *g, void *desc, u8 b_wait_for_halt)
         /* Poll for HALT */
         if (b_wait_for_halt) {
                 err = pmu_wait_for_halt(g, GPU_TIMEOUT_DEFAULT);
-                if (err == 0)
+                if (err == 0) {
                         /* Clear the HALT interrupt */
-                        gk20a_writel(g, pwr_falcon_irqsclr_r(),
-                        gk20a_readl(g, pwr_falcon_irqsclr_r()) & (~(0x10)));
+                  if (clear_halt_interrupt_status(g, GPU_TIMEOUT_DEFAULT))
+                        goto err_unmap_bl;
+                }
                 else
                         goto err_unmap_bl;
         }
@@ -1281,3 +1293,28 @@ int pmu_wait_for_halt(struct gk20a *g, unsigned int timeout)
                 return -EBUSY;
         return 0;
 }
+
+/*!
+*       Wait for PMU halt interrupt status to be cleared
+*       @param[in]      g               GPU object pointer
+*       @param[in]      timeout_us      Timeout in Us for PMU to halt
+*       @return '0' if PMU halt irq status is clear
+*/
+int clear_halt_interrupt_status(struct gk20a *g, unsigned int timeout)
+{
+        u32 data = 0;
+        while (timeout != 0) {
+                gk20a_writel(g, pwr_falcon_irqsclr_r(),
+                             gk20a_readl(g, pwr_falcon_irqsclr_r()) | (0x10));
+                data = gk20a_readl(g, (pwr_falcon_irqstat_r()));
+                if ((data & pwr_falcon_irqstat_halt_true_f()) !=
+                        pwr_falcon_irqstat_halt_true_f())
+                        /*halt irq is clear*/
+                        break;
+                timeout--;
+                udelay(1);
+        }
+        if (timeout == 0)
+                return -EBUSY;
+        return 0;
+}
diff --git a/drivers/gpu/nvgpu/gm20b/acr_gm20b.h b/drivers/gpu/nvgpu/gm20b/acr_gm20b.h
index d1c42e46..5dddc0b2 100644
--- a/drivers/gpu/nvgpu/gm20b/acr_gm20b.h
+++ b/drivers/gpu/nvgpu/gm20b/acr_gm20b.h
@@ -383,4 +383,5 @@ int gm20b_bootstrap_hs_flcn(struct gk20a *g);
 int gm20b_pmu_setup_sw(struct gk20a *g);
 int pmu_exec_gen_bl(struct gk20a *g, void *desc, u8 b_wait_for_halt);
 int pmu_wait_for_halt(struct gk20a *g, unsigned int timeout_us);
+int clear_halt_interrupt_status(struct gk20a *g, unsigned int timeout);
 #endif /*__ACR_GM20B_H_*/