linux-ti33x-psp: Check if scheduling is required after DMA

After processing one DMA completion, checking if rescheduling is
required instead of scheduling another completion request if the
FIFO is non empty.

This fixes high CPU load issue seen with g_mass_storage on a beaglebone
and could potentially fix similar issues seen with other gadget drivers

Signed-off-by: Joel A Fernandes <joelagnel@ti.com>
Signed-off-by: Koen Kooi <koen@dominion.thruhere.net>

2 files changed, 44 insertions, 1 deletions

diff --git a/recipes-kernel/linux/linux-ti33x-psp-3.1/0006-usb-musb-cppi41_dma-Check-if-scheduling-is-required-.patch b/recipes-kernel/linux/linux-ti33x-psp-3.1/0006-usb-musb-cppi41_dma-Check-if-scheduling-is-required-.patch
new file mode 100644
index 00000000..701c09d8
--- /dev/null
+++ b/recipes-kernel/linux/linux-ti33x-psp-3.1/0006-usb-musb-cppi41_dma-Check-if-scheduling-is-required-.patch
@@ -0,0 +1,42 @@
+From b564457d26931a00f8c5645981a00c1eb78e9a90 Mon Sep 17 00:00:00 2001
+From: Joel A Fernandes <joelagnel@ti.com>
+Date: Sat, 12 Nov 2011 23:42:26 -0600
+Subject: [PATCH] usb::musb::cppi41_dma: Check if scheduling is required after DMA
+
+Even after TX DMA, data can still be in the FIFO and we would keep
+rescheduling the worker thread in a polling fashion taking up a lot
+of CPU if the controller was slow to respond. We check if rescheduling
+is required and sleep if so.
+
+This fixes high CPU load issue seen with g_mass_storage on a BeagleBone
+and could potentially fix similar issues seen with other gadget drivers.
+
+Signed-off-by: Joel A Fernandes <joelagnel@ti.com>
+---
+ drivers/usb/musb/cppi41_dma.c |   10 ++++++++++
+ 1 files changed, 10 insertions(+), 0 deletions(-)
+
+diff --git a/drivers/usb/musb/cppi41_dma.c b/drivers/usb/musb/cppi41_dma.c
+index a24707c..e68c5d7 100644
+--- a/drivers/usb/musb/cppi41_dma.c
++++ b/drivers/usb/musb/cppi41_dma.c
+@@ -1319,6 +1319,16 @@ void txdma_completion_work(struct work_struct *data)
+        struct musb *musb = cppi->musb;
+        unsigned long flags;
+ 
++       /*
++        * txdma worker thread can call schedule_work on itself and cause
++        * itself to be scheduled immediately and because the data might still
++        * be in FIFO if it hasn't been pushed out after DMA, it is possible for
++        * the worker to consume lot of CPU when the controller is slow, so we
++        * reschedule if necessary.
++        */
++       if (need_resched())
++               cond_resched();
++
+        spin_lock_irqsave(&musb->lock, flags);
+        cppi41_check_fifo_empty(cppi);
+        spin_unlock_irqrestore(&musb->lock, flags);
+-- 
+1.7.4.1
+
diff --git a/recipes-kernel/linux/linux-ti33x-psp_3.1.bb b/recipes-kernel/linux/linux-ti33x-psp_3.1.bb
index c4cdfea9..37f59cad 100644
--- a/recipes-kernel/linux/linux-ti33x-psp_3.1.bb
+++ b/recipes-kernel/linux/linux-ti33x-psp_3.1.bb
@@ -11,7 +11,7 @@ MULTI_CONFIG_BASE_SUFFIX = ""
 
 BRANCH = "v3.1-staging"
 SRCREV = "1d84d8853fa30cf3db2571a5aec572accca4e29d"
-MACHINE_KERNEL_PR_append = "q+gitr${SRCREV}"
+MACHINE_KERNEL_PR_append = "r+gitr${SRCREV}"
 
 COMPATIBLE_MACHINE = "(ti33x)"
 
@@ -35,6 +35,7 @@ PATCHES_OVER_PSP = " \
         file://0003-arm-omap-mcspi-correct-memory-range-when-requesting-.patch \
         file://0004-arm-omap-mcspi-follow-proper-pm_runtime-enable-disab.patch \
         file://0005-arm-omap-mcspi-follow-proper-probe-remove-steps.patch \
+        file://0006-usb-musb-cppi41_dma-Check-if-scheduling-is-required-.patch \
         file://can/0001-can-d_can-Added-support-for-Bosch-D_CAN-controller.patch \
         file://can/0002-can-d_can-Added-platform-data-for-am33xx-device.patch \
         file://can/0003-can-d_can-DCAN-config-added-to-am335x_evm_defconfig.patch \