Hi Tomi,

On Fri, Aug 21, 2020 at 03:06:59PM +0300, Tomi Valkeinen wrote:

On 21/08/2020 10:45, Dinghao Liu wrote:

...

pm_runtime_get_sync() increments the runtime PM usage counter even when it returns an error code. However, users of dsi_runtime_get(), a direct wrapper of pm_runtime_get_sync(), assume that PM usage counter will not change on error. Thus a pairing decrement is needed on the error handling path to keep the counter balanced.

Fixes: 4fbafaf371be7 ("OMAP: DSS2: Use PM runtime & HWMOD support") Signed-off-by: Dinghao Liu dinghao.liu@zju.edu.cn

---

drivers/gpu/drm/omapdrm/dss/dsi.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/omapdrm/dss/dsi.c b/drivers/gpu/drm/omapdrm/dss/dsi.c index eeccf40bae41..973bfa14a104 100644 --- a/drivers/gpu/drm/omapdrm/dss/dsi.c +++ b/drivers/gpu/drm/omapdrm/dss/dsi.c @@ -1112,8 +1112,11 @@ static int dsi_runtime_get(struct dsi_data *dsi) DSSDBG("dsi_runtime_get\n");

r = pm_runtime_get_sync(dsi->dev);

• WARN_ON(r < 0);
• return r < 0 ? r : 0;

• if (WARN_ON(r < 0)) {

• pm_runtime_put_noidle(dsi->dev);
  

• return r;
  

• }
• return 0;

}

Thanks! Good catch. I think this is broken in all the other modules in omapdrm too (e.g. dispc.c, venc.c, etc).

Would you like to update the patch to cover the whole omapdrm?

Just for yoru information, there has been quite a few similar patches submitted all across the kernel. I believe this is an issue of the pm_runtime_get_sync() API, which really shouldn't require a put() when it fails. For drivers that really don't expect pm_runtime_get_sync() to fail (no I2C access to a regulator for instance, only SoC-internal operations) I've instead decided to ignore the error completely. I don't think poluting the whole kernel code base with this kind of "fixes" is a good idea.