[PATCH 3/5] iommu/rockchip: Fix the poll period and timeout of FORCE_RESET

[Arun Iyengar via B4 Relay]

From: Arun Iyengar <aiyengar1990 at gmail.com>

rk_iommu_force_reset() passes RK_MMU_FORCE_RESET_TIMEOUT_US (100 ms) as
the sleep between two reads of the reset status and RK_MMU_POLL_TIMEOUT_US
(1 ms) as the timeout, the other way round from what the names and every
other poll in the driver say. A reset that is not done at the first read
makes readx_poll_timeout() sleep for 25 to 100 ms, read once more and
give up, so the reset is both slower than needed when it succeeds and
not polled for the 100 ms the driver meant to allow when it is slow.

When the reset is done by the first read, as is normal (rk_iommu_resume()
of an RK3588 video decoder IOMMU takes 9 to 18 us), nothing changes; the
swap matters when a reset is slow or never completes, as on an RK3588
video decoder IOMMU whose master has not gone idle.

Poll every RK_MMU_POLL_PERIOD_US for up to RK_MMU_FORCE_RESET_TIMEOUT_US.

Fixes: 0416bf6479ac ("iommu/rockchip: Use iopoll helpers to wait for hardware")
Signed-off-by: Arun Iyengar <aiyengar1990 at gmail.com>
---
 drivers/iommu/rockchip-iommu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/iommu/rockchip-iommu.c b/drivers/iommu/rockchip-iommu.c
index c3c3502db27d..a81da8115085 100644
--- a/drivers/iommu/rockchip-iommu.c
+++ b/drivers/iommu/rockchip-iommu.c
@@ -535,8 +535,8 @@ static int rk_iommu_force_reset(struct rk_iommu *iommu)
 	rk_iommu_command(iommu, RK_MMU_CMD_FORCE_RESET);
 
 	ret = readx_poll_timeout(rk_iommu_is_reset_done, iommu, val,
-				 val, RK_MMU_FORCE_RESET_TIMEOUT_US,
-				 RK_MMU_POLL_TIMEOUT_US);
+				 val, RK_MMU_POLL_PERIOD_US,
+				 RK_MMU_FORCE_RESET_TIMEOUT_US);
 	if (ret) {
 		dev_err(iommu->dev, "FORCE_RESET command timed out\n");
 		return ret;

-- 
2.55.0