wbsd.c: after spin_lock_bh uses spin_unlock instead spin_unlock_bh
KERNEL_VERSION: 2.6.31 DESCRIBE:
Driver ./drivers/mmc/host/wbsd.c calls spin_lock_bh and then spin_unlock instead of spin_unlock_bh:
753 static void wbsd_request(struct mmc_host *mmc, struct mmc_request *mrq)
...
761 spin_lock_bh(&host->lock);
...
844 done:
845 wbsd_request_end(host, mrq);
846
847 spin_unlock_bh(&host->lock);
848 }
But in wsdb_request calls spin_unlock/spin_lock instead of spin_unlock_bh/spin_lock_bh;
206 static void wbsd_request_end(struct wbsd_host *host, struct mmc_request
...
230 spin_unlock(&host->lock); 231 mmc_request_done(host->mmc, mrq); 232 spin_lock(&host->lock);
233 }
Found by : Linux Driver Verification
Quoted 18 lines · Alexander Strakh
KERNEL_VERSION: 2.6.31 DESCRIBE:Driver
./drivers/mmc/host/wbsd.ccallsspin_lock_bhand thenspin_unlockinstead ofspin_unlock_bh:753 static void
wbsd_request(struct mmc_host *mmc, struct mmc_request *mrq)
....
761spin_lock_bh(&host->lock);
....
844 done:
845wbsd_request_end(host, mrq);
846
847spin_unlock_bh(&host->lock);
848 }But in
wsdb_requestcalls spin_unlock/spin_lock instead of spin_unlock_bh/spin_lock_bh;
Quoted 6 lines
206 static void wbsd_request_end(struct
wbsd_host*host, structmmc_request
....230 spin_unlock(&host->lock); 231 mmc_request_done(host->mmc, mrq); 232 spin_lock(&host->lock);233 }
Doesn't look too odd, it simply leaves BH disabled, but unlocks the spinlock.
Quoted 18 lines · Alexander Strakh
KERNEL_VERSION: 2.6.31 DESCRIBE:Driver
./drivers/mmc/host/wbsd.ccallsspin_lock_bhand thenspin_unlockinstead ofspin_unlock_bh:753 static void
wbsd_request(struct mmc_host *mmc, struct mmc_request *mrq)
...
761spin_lock_bh(&host->lock);
...
844 done:
845wbsd_request_end(host, mrq);
846
847spin_unlock_bh(&host->lock);
848 }But in
wsdb_requestcalls spin_unlock/spin_lock instead of spin_unlock_bh/spin_lock_bh;
Quoted 6 lines
206 static void wbsd_request_end(struct
wbsd_host*host, structmmc_request
...230 spin_unlock(&host->lock); 231 mmc_request_done(host->mmc, mrq); 232 spin_lock(&host->lock);233 }
And where exactly you see a bug?
wbsd_request_end() can be called in the following ways:
- through wbsd_finish_data() from tasklet (therefore no need to disable tasklets there)
- from wbsd_request() directly. The sequence here is
spin_lock_bh(&host->lock); ... wbsd_request_end(host, mrq); spin_unlock(&host->lock); .... spin_lock(&host->lock); spin_unlock_bh(&host->lock);
So the whole thing runs with tasklets disabled, no matter what you do with the host->lock, and they are only enabled at the very end of wbsd_request()
So, what exact problem scenario do you see here?
Signature · 2 lines
Jiri Kosina SUSE Labs, Novell Inc.