Re: [Devel] [PATCH] scsi/eh: fix hang adding ehandler wakeups after decrementing host_busy
Are there any issues with this patch (https://patchwork.kernel.org/patch/9938919/) that Pavel Tikhomirov submitted back in September? I am willing to help if there's anything I can do to help get it accepted. The failing case I'm working on involves lots of servers with disk read/write activity with periodic INQUIRY commands that aren't supported by the target device, so the error handler gets invoked periodically, and some of the systems will hang after a day or two. Thanks Stuart --- This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
[Devel] [PATCH v5 2/2] nfs: protect callback execution against per-net callback thread shutdown
From: Stanislav Kinsburskiy Here is the race: CPU #0 CPU#1 cleanup_mnt nfs41_callback_svc (get xprt from the list) nfs_callback_down ... ... ... svc_close_net ... ... ... svc_xprt_free ... svc_bc_sock_freebc_svc_process kfree(xprt) svc_process_common rqstp->rq_xprt->xpt_ops (use after free) The problem is that per-net SUNRPC transports shutdown is done regardless current callback execution. This is a race leading to transport use-after-free in callback handler. This patch fixes it in stright-forward way. I.e. it protects callback execution with the same mutex used for per-net data creation and destruction. Hopefully, it won't slow down NFS client significantly. https://jira.sw.ru/browse/PSBM-75751 v5: destroy all per-net backchannel requests before transports on in nfs_callback_down_net v4: use another mutex to protect callback execution agains per-net transports shutdown. This guarantees, that transports won't be destroyed by shutdown callback while execution is in progress and vice versa. v3: Fix mutex deadlock, when shutdown callback waits for thread to exit (with mutex taken), while thread wait for the mutex to take. The idea is to simply check if thread has to exit, if mutex lock has failed. This is a busy loop, but it shouldn't happend often and for long. Signed-off-by: Stanislav Kinsburskiy --- fs/nfs/callback.c | 17 + 1 file changed, 17 insertions(+) diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c index 0beb275..e18d774 100644 --- a/fs/nfs/callback.c +++ b/fs/nfs/callback.c @@ -99,6 +99,8 @@ nfs4_callback_up(struct svc_serv *serv) } #if defined(CONFIG_NFS_V4_1) +static DEFINE_MUTEX(nfs41_callback_mutex); + /* * The callback service for NFSv4.1 callbacks */ @@ -117,6 +119,12 @@ nfs41_callback_svc(void *vrqstp) if (try_to_freeze()) continue; + mutex_lock(&nfs41_callback_mutex); + if (kthread_should_stop()) { + mutex_unlock(&nfs41_callback_mutex); + return 0; + } + prepare_to_wait(&serv->sv_cb_waitq, &wq, TASK_INTERRUPTIBLE); spin_lock_bh(&serv->sv_cb_lock); if (!list_empty(&serv->sv_cb_list)) { @@ -129,8 +137,10 @@ nfs41_callback_svc(void *vrqstp) error = bc_svc_process(serv, req, rqstp); dprintk("bc_svc_process() returned w/ error code= %d\n", error); + mutex_unlock(&nfs41_callback_mutex); } else { spin_unlock_bh(&serv->sv_cb_lock); + mutex_unlock(&nfs41_callback_mutex); schedule(); finish_wait(&serv->sv_cb_waitq, &wq); } @@ -242,6 +252,7 @@ static void nfs_callback_down_net(u32 minorversion, struct svc_serv *serv, struc return; dprintk("NFS: destroy per-net callback data; net=%p\n", net); + bc_svc_flush_queue_net(serv, net); svc_shutdown_net(serv, net); } @@ -377,7 +388,13 @@ void nfs_callback_down(int minorversion, struct net *net) struct nfs_callback_data *cb_info = &nfs_callback_info[minorversion]; mutex_lock(&nfs_callback_mutex); +#if defined(CONFIG_NFS_V4_1) + mutex_lock(&nfs41_callback_mutex); + nfs_callback_down_net(minorversion, cb_info->serv, net); + mutex_unlock(&nfs41_callback_mutex); +#else nfs_callback_down_net(minorversion, cb_info->serv, net); +#endif cb_info->users--; if (cb_info->users == 0 && cb_info->task != NULL) { kthread_stop(cb_info->task); ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
[Devel] [PATCH v5 0/2] nfs: fix race between callback shutdown and execution
The idea is to use mutex for protecting callback execution agains per-net callback shutdown and destroying all the net-related backchannel requests before transports destruction. --- Stanislav Kinsburskiy (2): sunrpc: bc_svc_flush_queue_net() helper introduced nfs: protect callback execution against per-net callback thread shutdown fs/nfs/callback.c | 17 + include/linux/sunrpc/svc.h |2 ++ net/sunrpc/svc.c | 15 +++ 3 files changed, 34 insertions(+) -- ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
[Devel] [PATCH v5 1/2] sunrpc: bc_svc_flush_queue_net() helper introduced
From: Stanislav Kinsburskiy This helper can be used to remove backchannel requests from callback queue on per-net basis. Signed-off-by: Stanislav Kinsburskiy --- include/linux/sunrpc/svc.h |2 ++ net/sunrpc/svc.c | 15 +++ 2 files changed, 17 insertions(+) diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h index 2b30868..fe70ff0 100644 --- a/include/linux/sunrpc/svc.h +++ b/include/linux/sunrpc/svc.h @@ -484,6 +484,8 @@ void svc_reserve(struct svc_rqst *rqstp, int space); struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu); char *svc_print_addr(struct svc_rqst *, char *, size_t); +void bc_svc_flush_queue_net(struct svc_serv *serv, struct net *net); + #defineRPC_MAX_ADDRBUFLEN (63U) /* diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c index de8cded..2ca4ff7 100644 --- a/net/sunrpc/svc.c +++ b/net/sunrpc/svc.c @@ -1338,6 +1338,21 @@ svc_process(struct svc_rqst *rqstp) EXPORT_SYMBOL_GPL(svc_process); #if defined(CONFIG_SUNRPC_BACKCHANNEL) +void bc_svc_flush_queue_net(struct svc_serv *serv, struct net *net) +{ + struct rpc_rqst *req, *tmp; + + spin_lock_bh(&serv->sv_cb_lock); + list_for_each_entry_safe(req, tmp, &serv->sv_cb_list, rq_bc_list) { + if (req->rq_xprt->xprt_net == net) { + list_del(&req->rq_bc_list); + xprt_free_bc_request(req); + } + } + spin_unlock_bh(&serv->sv_cb_lock); +} +EXPORT_SYMBOL_GPL(bc_svc_flush_queue_net); + /* * Process a backchannel RPC request that arrived over an existing * outbound connection ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
[Devel] [PATCH RHEL7 COMMIT] tswap, tcache: Increase shrinkers seeks
The commit is pushed to "branch-rh7-3.10.0-693.1.1.vz7.37.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git after rh7-3.10.0-693.1.1.vz7.37.24 --> commit d6cdd02bdcf3f264f18b1d187cb419ecc5607644 Author: Kirill Tkhai Date: Wed Nov 8 18:28:06 2017 +0300 tswap, tcache: Increase shrinkers seeks Commit e008b95a28ef95dd4bb08f69c89d26fc5fa7411a "ms/mm: use sc->priority for slab shrink targets" exposed the fact we shrinks too many tcache pages. Shrinkers of {in,}active pages shrink up to 32 pages, while tcache and tswap shrinks 128 pages. This became a reason of tcache active test fail. This patch makes numbers of shrinked pages of tcache and tswap in consistent state with pages shrinkers, and restores the test-expected behaviour. https://jira.sw.ru/browse/PSBM-72584 Signed-off-by: Kirill Tkhai Acked-by: Andrey Ryabinin --- mm/tcache.c | 2 +- mm/tswap.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/mm/tcache.c b/mm/tcache.c index d1a2c53..8b893e6 100644 --- a/mm/tcache.c +++ b/mm/tcache.c @@ -1202,7 +1202,7 @@ static unsigned long tcache_shrink_scan(struct shrinker *shrink, struct shrinker tcache_shrinker = { .count_objects = tcache_shrink_count, .scan_objects = tcache_shrink_scan, - .seeks = 1, + .seeks = 4, .batch = TCACHE_SCAN_BATCH, .flags = SHRINKER_NUMA_AWARE, }; diff --git a/mm/tswap.c b/mm/tswap.c index 38a389f..b7a990e 100644 --- a/mm/tswap.c +++ b/mm/tswap.c @@ -271,7 +271,7 @@ static unsigned long tswap_shrink_scan(struct shrinker *shrink, static struct shrinker tswap_shrinker = { .count_objects = tswap_shrink_count, .scan_objects = tswap_shrink_scan, - .seeks = 1, + .seeks = 4, .flags = SHRINKER_NUMA_AWARE, }; ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
Re: [Devel] [PATCH v4] nfs: protect callback execution against per-net callback thread shutdown
On 08.11.2017 17:07, Stanislav Kinsburskiy wrote: > From: Stanislav Kinsburskiy > > Here is the race: > > CPU #0CPU#1 > > cleanup_mnt nfs41_callback_svc (get xprt from the list) > nfs_callback_down ... > ... ... > svc_close_net ... > ... ... > svc_xprt_free ... > svc_bc_sock_free bc_svc_process > kfree(xprt) svc_process_common > rqstp->rq_xprt->xpt_ops (use after free) > > The problem is that per-net SUNRPC transports shutdown is done regardless > current callback execution. This is a race leading to transport use-after-free > in callback handler. > This patch fixes it in stright-forward way. I.e. it protects callback > execution with the same mutex used for per-net data creation and destruction. > Hopefully, it won't slow down NFS client significantly. > > https://jira.sw.ru/browse/PSBM-75751 > > v4: use another mutex to protect callback execution agains per-net transports > shutdown. > This guarantees, that transports won't be destroyed by shutdown callback while > execution is in progress and vice versa. > > v3: Fix mutex deadlock, when shutdown callback waits for thread to exit (with > mutex taken), while thread wait for the mutex to take. > The idea is to simply check if thread has to exit, if mutex lock has failed. > This is a busy loop, but it shouldn't happend often and for long. > > Signed-off-by: Stanislav Kinsburskiy > --- > fs/nfs/callback.c | 16 > 1 file changed, 16 insertions(+) > > diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c > index 0beb275..4d7979b 100644 > --- a/fs/nfs/callback.c > +++ b/fs/nfs/callback.c > @@ -99,6 +99,8 @@ nfs4_callback_up(struct svc_serv *serv) > } > > #if defined(CONFIG_NFS_V4_1) > +static DEFINE_MUTEX(nfs41_callback_mutex); > + > /* > * The callback service for NFSv4.1 callbacks > */ > @@ -117,6 +119,12 @@ nfs41_callback_svc(void *vrqstp) > if (try_to_freeze()) > continue; > > + mutex_lock(&nfs41_callback_mutex); > + if (kthread_should_stop()) { > + mutex_unlock(&nfs41_callback_mutex); > + return 0; > + } > + > prepare_to_wait(&serv->sv_cb_waitq, &wq, TASK_INTERRUPTIBLE); > spin_lock_bh(&serv->sv_cb_lock); > if (!list_empty(&serv->sv_cb_list)) { > @@ -129,8 +137,10 @@ nfs41_callback_svc(void *vrqstp) > error = bc_svc_process(serv, req, rqstp); > dprintk("bc_svc_process() returned w/ error code= %d\n", > error); > + mutex_unlock(&nfs41_callback_mutex); > } else { > spin_unlock_bh(&serv->sv_cb_lock); > + mutex_unlock(&nfs41_callback_mutex); > schedule(); > finish_wait(&serv->sv_cb_waitq, &wq); > } > @@ -377,7 +387,13 @@ void nfs_callback_down(int minorversion, struct net *net) > struct nfs_callback_data *cb_info = &nfs_callback_info[minorversion]; > > mutex_lock(&nfs_callback_mutex); > +#if defined(CONFIG_NFS_V4_1) > + mutex_lock(&nfs41_callback_mutex); > + nfs_callback_down_net(minorversion, cb_info->serv, net); > + mutex_unlock(&nfs41_callback_mutex); It seems we also should flush queue here. > +#else > nfs_callback_down_net(minorversion, cb_info->serv, net); > +#endif > cb_info->users--; > if (cb_info->users == 0 && cb_info->task != NULL) { > kthread_stop(cb_info->task); > ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
Re: [Devel] [PATCH] nfs: protect callback execution against per-net callback thread shutdown
Please, ignore 08.11.2017 15:07, Stanislav Kinsburskiy пишет: > From: Stanislav Kinsburskiy > > Here is the race: > > CPU #0CPU#1 > > cleanup_mnt nfs41_callback_svc (get xprt from the list) > nfs_callback_down ... > ... ... > svc_close_net ... > ... ... > svc_xprt_free ... > svc_bc_sock_free bc_svc_process > kfree(xprt) svc_process_common > rqstp->rq_xprt->xpt_ops (use after free) > > The problem is that per-net SUNRPC transports shutdown is done regardless > current callback execution. This is a race leading to transport use-after-free > in callback handler. > This patch fixes it in stright-forward way. I.e. it protects callback > execution with the same mutex used for per-net data creation and destruction. > Hopefully, it won't slow down NFS client significantly. > > https://jira.sw.ru/browse/PSBM-75751 > > v4: use another mutex to protect callback execution agains per-net transports > shutdown. > This guarantees, that transports won't be destroyed by shutdown callback while > execution is in progress and vice versa. > > v3: Fix mutex deadlock, when shutdown callback waits for thread to exit (with > mutex taken), while thread wait for the mutex to take. > The idea is to simply check if thread has to exit, if mutex lock has failed. > This is a busy loop, but it shouldn't happend often and for long. > > Signed-off-by: Stanislav Kinsburskiy > --- > fs/nfs/callback.c | 16 > 1 file changed, 16 insertions(+) > > diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c > index 0beb275..4d7979b 100644 > --- a/fs/nfs/callback.c > +++ b/fs/nfs/callback.c > @@ -99,6 +99,8 @@ nfs4_callback_up(struct svc_serv *serv) > } > > #if defined(CONFIG_NFS_V4_1) > +static DEFINE_MUTEX(nfs41_callback_mutex); > + > /* > * The callback service for NFSv4.1 callbacks > */ > @@ -117,6 +119,12 @@ nfs41_callback_svc(void *vrqstp) > if (try_to_freeze()) > continue; > > + mutex_lock(&nfs41_callback_mutex); > + if (kthread_should_stop()) { > + mutex_unlock(&nfs41_callback_mutex); > + return 0; > + } > + > prepare_to_wait(&serv->sv_cb_waitq, &wq, TASK_INTERRUPTIBLE); > spin_lock_bh(&serv->sv_cb_lock); > if (!list_empty(&serv->sv_cb_list)) { > @@ -129,8 +137,10 @@ nfs41_callback_svc(void *vrqstp) > error = bc_svc_process(serv, req, rqstp); > dprintk("bc_svc_process() returned w/ error code= %d\n", > error); > + mutex_unlock(&nfs41_callback_mutex); > } else { > spin_unlock_bh(&serv->sv_cb_lock); > + mutex_unlock(&nfs41_callback_mutex); > schedule(); > finish_wait(&serv->sv_cb_waitq, &wq); > } > @@ -377,7 +387,13 @@ void nfs_callback_down(int minorversion, struct net *net) > struct nfs_callback_data *cb_info = &nfs_callback_info[minorversion]; > > mutex_lock(&nfs_callback_mutex); > +#if defined(CONFIG_NFS_V4_1) > + mutex_lock(&nfs41_callback_mutex); > + nfs_callback_down_net(minorversion, cb_info->serv, net); > + mutex_unlock(&nfs41_callback_mutex); > +#else > nfs_callback_down_net(minorversion, cb_info->serv, net); > +#endif > cb_info->users--; > if (cb_info->users == 0 && cb_info->task != NULL) { > kthread_stop(cb_info->task); > > ___ > Devel mailing list > Devel@openvz.org > https://lists.openvz.org/mailman/listinfo/devel > ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
[Devel] [PATCH] nfs: protect callback execution against per-net callback thread shutdown
From: Stanislav Kinsburskiy Here is the race: CPU #0 CPU#1 cleanup_mnt nfs41_callback_svc (get xprt from the list) nfs_callback_down ... ... ... svc_close_net ... ... ... svc_xprt_free ... svc_bc_sock_freebc_svc_process kfree(xprt) svc_process_common rqstp->rq_xprt->xpt_ops (use after free) The problem is that per-net SUNRPC transports shutdown is done regardless current callback execution. This is a race leading to transport use-after-free in callback handler. This patch fixes it in stright-forward way. I.e. it protects callback execution with the same mutex used for per-net data creation and destruction. Hopefully, it won't slow down NFS client significantly. https://jira.sw.ru/browse/PSBM-75751 v4: use another mutex to protect callback execution agains per-net transports shutdown. This guarantees, that transports won't be destroyed by shutdown callback while execution is in progress and vice versa. v3: Fix mutex deadlock, when shutdown callback waits for thread to exit (with mutex taken), while thread wait for the mutex to take. The idea is to simply check if thread has to exit, if mutex lock has failed. This is a busy loop, but it shouldn't happend often and for long. Signed-off-by: Stanislav Kinsburskiy --- fs/nfs/callback.c | 16 1 file changed, 16 insertions(+) diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c index 0beb275..4d7979b 100644 --- a/fs/nfs/callback.c +++ b/fs/nfs/callback.c @@ -99,6 +99,8 @@ nfs4_callback_up(struct svc_serv *serv) } #if defined(CONFIG_NFS_V4_1) +static DEFINE_MUTEX(nfs41_callback_mutex); + /* * The callback service for NFSv4.1 callbacks */ @@ -117,6 +119,12 @@ nfs41_callback_svc(void *vrqstp) if (try_to_freeze()) continue; + mutex_lock(&nfs41_callback_mutex); + if (kthread_should_stop()) { + mutex_unlock(&nfs41_callback_mutex); + return 0; + } + prepare_to_wait(&serv->sv_cb_waitq, &wq, TASK_INTERRUPTIBLE); spin_lock_bh(&serv->sv_cb_lock); if (!list_empty(&serv->sv_cb_list)) { @@ -129,8 +137,10 @@ nfs41_callback_svc(void *vrqstp) error = bc_svc_process(serv, req, rqstp); dprintk("bc_svc_process() returned w/ error code= %d\n", error); + mutex_unlock(&nfs41_callback_mutex); } else { spin_unlock_bh(&serv->sv_cb_lock); + mutex_unlock(&nfs41_callback_mutex); schedule(); finish_wait(&serv->sv_cb_waitq, &wq); } @@ -377,7 +387,13 @@ void nfs_callback_down(int minorversion, struct net *net) struct nfs_callback_data *cb_info = &nfs_callback_info[minorversion]; mutex_lock(&nfs_callback_mutex); +#if defined(CONFIG_NFS_V4_1) + mutex_lock(&nfs41_callback_mutex); + nfs_callback_down_net(minorversion, cb_info->serv, net); + mutex_unlock(&nfs41_callback_mutex); +#else nfs_callback_down_net(minorversion, cb_info->serv, net); +#endif cb_info->users--; if (cb_info->users == 0 && cb_info->task != NULL) { kthread_stop(cb_info->task); ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
[Devel] [PATCH v4] nfs: protect callback execution against per-net callback thread shutdown
From: Stanislav Kinsburskiy Here is the race: CPU #0 CPU#1 cleanup_mnt nfs41_callback_svc (get xprt from the list) nfs_callback_down ... ... ... svc_close_net ... ... ... svc_xprt_free ... svc_bc_sock_freebc_svc_process kfree(xprt) svc_process_common rqstp->rq_xprt->xpt_ops (use after free) The problem is that per-net SUNRPC transports shutdown is done regardless current callback execution. This is a race leading to transport use-after-free in callback handler. This patch fixes it in stright-forward way. I.e. it protects callback execution with the same mutex used for per-net data creation and destruction. Hopefully, it won't slow down NFS client significantly. https://jira.sw.ru/browse/PSBM-75751 v4: use another mutex to protect callback execution agains per-net transports shutdown. This guarantees, that transports won't be destroyed by shutdown callback while execution is in progress and vice versa. v3: Fix mutex deadlock, when shutdown callback waits for thread to exit (with mutex taken), while thread wait for the mutex to take. The idea is to simply check if thread has to exit, if mutex lock has failed. This is a busy loop, but it shouldn't happend often and for long. Signed-off-by: Stanislav Kinsburskiy --- fs/nfs/callback.c | 16 1 file changed, 16 insertions(+) diff --git a/fs/nfs/callback.c b/fs/nfs/callback.c index 0beb275..4d7979b 100644 --- a/fs/nfs/callback.c +++ b/fs/nfs/callback.c @@ -99,6 +99,8 @@ nfs4_callback_up(struct svc_serv *serv) } #if defined(CONFIG_NFS_V4_1) +static DEFINE_MUTEX(nfs41_callback_mutex); + /* * The callback service for NFSv4.1 callbacks */ @@ -117,6 +119,12 @@ nfs41_callback_svc(void *vrqstp) if (try_to_freeze()) continue; + mutex_lock(&nfs41_callback_mutex); + if (kthread_should_stop()) { + mutex_unlock(&nfs41_callback_mutex); + return 0; + } + prepare_to_wait(&serv->sv_cb_waitq, &wq, TASK_INTERRUPTIBLE); spin_lock_bh(&serv->sv_cb_lock); if (!list_empty(&serv->sv_cb_list)) { @@ -129,8 +137,10 @@ nfs41_callback_svc(void *vrqstp) error = bc_svc_process(serv, req, rqstp); dprintk("bc_svc_process() returned w/ error code= %d\n", error); + mutex_unlock(&nfs41_callback_mutex); } else { spin_unlock_bh(&serv->sv_cb_lock); + mutex_unlock(&nfs41_callback_mutex); schedule(); finish_wait(&serv->sv_cb_waitq, &wq); } @@ -377,7 +387,13 @@ void nfs_callback_down(int minorversion, struct net *net) struct nfs_callback_data *cb_info = &nfs_callback_info[minorversion]; mutex_lock(&nfs_callback_mutex); +#if defined(CONFIG_NFS_V4_1) + mutex_lock(&nfs41_callback_mutex); + nfs_callback_down_net(minorversion, cb_info->serv, net); + mutex_unlock(&nfs41_callback_mutex); +#else nfs_callback_down_net(minorversion, cb_info->serv, net); +#endif cb_info->users--; if (cb_info->users == 0 && cb_info->task != NULL) { kthread_stop(cb_info->task); ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
Re: [Devel] [PATCH] tswap, tcache: Increase shrinkers seeks
On 11/08/2017 01:21 PM, Kirill Tkhai wrote: > Commit e008b95a28ef95dd4bb08f69c89d26fc5fa7411a > "ms/mm: use sc->priority for slab shrink targets" > exposed the fact we shrinks too many tcache pages. > > Shrinkers of {in,}active pages shrink up to 32 > pages, while tcache and tswap shrinks 128 pages. > This became a reason of tcache active test fail. > > This patch makes numbers of shrinked pages of tcache > and tswap in consistent state with pages shrinkers, > and restores the test-expected behaviour. > > https://jira.sw.ru/browse/PSBM-72584 > > Signed-off-by: Kirill Tkhai Acked-by: Andrey Ryabinin ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
[Devel] [PATCH] tswap, tcache: Increase shrinkers seeks
Commit e008b95a28ef95dd4bb08f69c89d26fc5fa7411a "ms/mm: use sc->priority for slab shrink targets" exposed the fact we shrinks too many tcache pages. Shrinkers of {in,}active pages shrink up to 32 pages, while tcache and tswap shrinks 128 pages. This became a reason of tcache active test fail. This patch makes numbers of shrinked pages of tcache and tswap in consistent state with pages shrinkers, and restores the test-expected behaviour. https://jira.sw.ru/browse/PSBM-72584 Signed-off-by: Kirill Tkhai --- mm/tcache.c |2 +- mm/tswap.c |2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/mm/tcache.c b/mm/tcache.c index d1a2c53e11a..8b893e6b520 100644 --- a/mm/tcache.c +++ b/mm/tcache.c @@ -1202,7 +1202,7 @@ static unsigned long tcache_shrink_scan(struct shrinker *shrink, struct shrinker tcache_shrinker = { .count_objects = tcache_shrink_count, .scan_objects = tcache_shrink_scan, - .seeks = 1, + .seeks = 4, .batch = TCACHE_SCAN_BATCH, .flags = SHRINKER_NUMA_AWARE, }; diff --git a/mm/tswap.c b/mm/tswap.c index 38a389fb0ca..b7a990e8cd8 100644 --- a/mm/tswap.c +++ b/mm/tswap.c @@ -271,7 +271,7 @@ static unsigned long tswap_shrink_scan(struct shrinker *shrink, static struct shrinker tswap_shrinker = { .count_objects = tswap_shrink_count, .scan_objects = tswap_shrink_scan, - .seeks = 1, + .seeks = 4, .flags = SHRINKER_NUMA_AWARE, }; ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel
[Devel] [PATCH] mm: Change formula of calculation of default min_free_kbytes
Parameter min_free_kbytes acts on per zone watermarks. It is used to calculate the zones free memory value, below which the direct reclaim starts and becomes throttled (the called task sleeps). This patch makes default min_free_kbytes to be 2% of available physical memory, but not more than 4GB. And this is more, than previous formula gave (it was a sqrt). Why do we need that. We bumped in the situation, when intense disc write inside a CT on a node, having very few free memory, may lead to the state, when almost all tasks are spining in direct reclaim. The tasks can't do effective reclaim as generated dirty pages are written and released by ploop threads, and thus the tasks in practically are just busy looping. Ploop threads can't produce the effective reclaim, as processors are occupied by the busylooping tasks and also they need free pages to do that. So, the system is looping and becomes very slow and unresponsible. https://jira.sw.ru/browse/PSBM-69296 Signed-off-by: Kirill Tkhai --- mm/page_alloc.c | 27 +++ 1 file changed, 3 insertions(+), 24 deletions(-) diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 137d1d86ddf..2108034bd80 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -6399,27 +6399,6 @@ void setup_per_zone_wmarks(void) /* * Initialise min_free_kbytes. - * - * For small machines we want it small (128k min). For large machines - * we want it large (64MB max). But it is not linear, because network - * bandwidth does not increase linearly with machine size. We use - * - * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: - * min_free_kbytes = sqrt(lowmem_kbytes * 16) - * - * which yields - * - * 16MB: 512k - * 32MB: 724k - * 64MB: 1024k - * 128MB: 1448k - * 256MB: 2048k - * 512MB: 2896k - * 1024MB: 4096k - * 2048MB: 5792k - * 4096MB: 8192k - * 8192MB: 11584k - * 16384MB:16384k */ int __meminit init_per_zone_wmark_min(void) { @@ -6427,14 +6406,14 @@ int __meminit init_per_zone_wmark_min(void) int new_min_free_kbytes; lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); - new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16); + new_min_free_kbytes = lowmem_kbytes * 2 / 100; /* 2% */ if (new_min_free_kbytes > user_min_free_kbytes) { min_free_kbytes = new_min_free_kbytes; if (min_free_kbytes < 128) min_free_kbytes = 128; - if (min_free_kbytes > 65536) - min_free_kbytes = 65536; + if (min_free_kbytes > 4194304) + min_free_kbytes = 4194304; } else { pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n", new_min_free_kbytes, user_min_free_kbytes); ___ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel