- Linux-stable-mirror - lists.linaro.org

[Linux-stable-mirror] Patch "net/9p: Switch to wait_event_killable()" has been added to the 4.9-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled net/9p: Switch to wait_event_killable() to the 4.9-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: net-9p-switch-to-wait_event_killable.patch and it can be found in the queue-4.9 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From 9523feac272ccad2ad8186ba4fcc89103754de52 Mon Sep 17 00:00:00 2001 From: Tuomas Tynkkynen <tuomas(a)tuxera.com> Date: Wed, 6 Sep 2017 17:59:08 +0300 Subject: net/9p: Switch to wait_event_killable() From: Tuomas Tynkkynen <tuomas(a)tuxera.com> commit 9523feac272ccad2ad8186ba4fcc89103754de52 upstream. Because userspace gets Very Unhappy when calls like stat() and execve() return -EINTR on 9p filesystem mounts. For instance, when bash is looking in PATH for things to execute and some SIGCHLD interrupts stat(), bash can throw a spurious 'command not found' since it doesn't retry the stat(). In practice, hitting the problem is rare and needs a really slow/bogged down 9p server. Signed-off-by: Tuomas Tynkkynen <tuomas(a)tuxera.com> Signed-off-by: Al Viro <viro(a)zeniv.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- net/9p/client.c | 3 +-- net/9p/trans_virtio.c | 13 ++++++------- 2 files changed, 7 insertions(+), 9 deletions(-) --- a/net/9p/client.c +++ b/net/9p/client.c @@ -749,8 +749,7 @@ p9_client_rpc(struct p9_client *c, int8_ } again: /* Wait for the response */ - err = wait_event_interruptible(*req->wq, - req->status >= REQ_STATUS_RCVD); + err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD); /* * Make sure our req is coherent with regard to updates in other --- a/net/9p/trans_virtio.c +++ b/net/9p/trans_virtio.c @@ -286,8 +286,8 @@ req_retry: if (err == -ENOSPC) { chan->ring_bufs_avail = 0; spin_unlock_irqrestore(&chan->lock, flags); - err = wait_event_interruptible(*chan->vc_wq, - chan->ring_bufs_avail); + err = wait_event_killable(*chan->vc_wq, + chan->ring_bufs_avail); if (err == -ERESTARTSYS) return err; @@ -327,7 +327,7 @@ static int p9_get_mapped_pages(struct vi * Other zc request to finish here */ if (atomic_read(&vp_pinned) >= chan->p9_max_pages) { - err = wait_event_interruptible(vp_wq, + err = wait_event_killable(vp_wq, (atomic_read(&vp_pinned) < chan->p9_max_pages)); if (err == -ERESTARTSYS) return err; @@ -471,8 +471,8 @@ req_retry_pinned: if (err == -ENOSPC) { chan->ring_bufs_avail = 0; spin_unlock_irqrestore(&chan->lock, flags); - err = wait_event_interruptible(*chan->vc_wq, - chan->ring_bufs_avail); + err = wait_event_killable(*chan->vc_wq, + chan->ring_bufs_avail); if (err == -ERESTARTSYS) goto err_out; @@ -489,8 +489,7 @@ req_retry_pinned: virtqueue_kick(chan->vq); spin_unlock_irqrestore(&chan->lock, flags); p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); - err = wait_event_interruptible(*req->wq, - req->status >= REQ_STATUS_RCVD); + err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD); /* * Non kernel buffers are pinned, unpin them */ Patches currently in stable-queue which might be from tuomas(a)tuxera.com are queue-4.9/net-9p-switch-to-wait_event_killable.patch queue-4.9/fs-9p-compare-qid.path-in-v9fs_test_inode.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] Patch "fscrypt: lock mutex before checking for bounce page pool" has been added to the 4.9-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled fscrypt: lock mutex before checking for bounce page pool to the 4.9-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: fscrypt-lock-mutex-before-checking-for-bounce-page-pool.patch and it can be found in the queue-4.9 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From a0b3bc855374c50b5ea85273553485af48caf2f7 Mon Sep 17 00:00:00 2001 From: Eric Biggers <ebiggers(a)google.com> Date: Sun, 29 Oct 2017 06:30:19 -0400 Subject: fscrypt: lock mutex before checking for bounce page pool From: Eric Biggers <ebiggers(a)google.com> commit a0b3bc855374c50b5ea85273553485af48caf2f7 upstream. fscrypt_initialize(), which allocates the global bounce page pool when an encrypted file is first accessed, uses "double-checked locking" to try to avoid locking fscrypt_init_mutex. However, it doesn't use any memory barriers, so it's theoretically possible for a thread to observe a bounce page pool which has not been fully initialized. This is a classic bug with "double-checked locking". While "only a theoretical issue" in the latest kernel, in pre-4.8 kernels the pointer that was checked was not even the last to be initialized, so it was easily possible for a crash (NULL pointer dereference) to happen. This was changed only incidentally by the large refactor to use fs/crypto/. Solve both problems in a trivial way that can easily be backported: just always take the mutex. It's theoretically less efficient, but it shouldn't be noticeable in practice as the mutex is only acquired very briefly once per encrypted file. Later I'd like to make this use a helper macro like DO_ONCE(). However, DO_ONCE() runs in atomic context, so we'd need to add a new macro that allows blocking. Signed-off-by: Eric Biggers <ebiggers(a)google.com> Signed-off-by: Theodore Ts'o <tytso(a)mit.edu> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- fs/crypto/crypto.c | 3 --- 1 file changed, 3 deletions(-) --- a/fs/crypto/crypto.c +++ b/fs/crypto/crypto.c @@ -484,9 +484,6 @@ int fscrypt_initialize(void) { int i, res = -ENOMEM; - if (fscrypt_bounce_page_pool) - return 0; - mutex_lock(&fscrypt_init_mutex); if (fscrypt_bounce_page_pool) goto already_initialized; Patches currently in stable-queue which might be from ebiggers(a)google.com are queue-4.9/lib-mpi-call-cond_resched-from-mpi_powm-loop.patch queue-4.9/fscrypt-lock-mutex-before-checking-for-bounce-page-pool.patch queue-4.9/dm-bufio-fix-integer-overflow-when-limiting-maximum-cache-size.patch queue-4.9/libceph-don-t-warn-if-user-tries-to-add-invalid-key.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] Patch "PM / OPP: Add missing of_node_put(np)" has been added to the 4.4-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled PM / OPP: Add missing of_node_put(np) to the 4.4-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: pm-opp-add-missing-of_node_put-np.patch and it can be found in the queue-4.4 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From 7978db344719dab1e56d05e6fc04aaaddcde0a5e Mon Sep 17 00:00:00 2001 From: Tobias Jordan <Tobias.Jordan(a)elektrobit.com> Date: Wed, 4 Oct 2017 11:35:03 +0530 Subject: PM / OPP: Add missing of_node_put(np) From: Tobias Jordan <Tobias.Jordan(a)elektrobit.com> commit 7978db344719dab1e56d05e6fc04aaaddcde0a5e upstream. The for_each_available_child_of_node() loop in _of_add_opp_table_v2() doesn't drop the reference to "np" on errors. Fix that. Fixes: 274659029c9d (PM / OPP: Add support to parse "operating-points-v2" bindings) Signed-off-by: Tobias Jordan <Tobias.Jordan(a)elektrobit.com> [ VK: Improved commit log. ] Signed-off-by: Viresh Kumar <viresh.kumar(a)linaro.org> Reviewed-by: Stephen Boyd <sboyd(a)codeaurora.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki(a)intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- drivers/base/power/opp/core.c | 1 + 1 file changed, 1 insertion(+) --- a/drivers/base/power/opp/core.c +++ b/drivers/base/power/opp/core.c @@ -1205,6 +1205,7 @@ static int _of_add_opp_table_v2(struct d if (ret) { dev_err(dev, "%s: Failed to add OPP, %d\n", __func__, ret); + of_node_put(np); goto free_table; } } Patches currently in stable-queue which might be from Tobias.Jordan(a)elektrobit.com are queue-4.4/pm-opp-add-missing-of_node_put-np.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] Patch "net/9p: Switch to wait_event_killable()" has been added to the 4.4-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled net/9p: Switch to wait_event_killable() to the 4.4-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: net-9p-switch-to-wait_event_killable.patch and it can be found in the queue-4.4 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From 9523feac272ccad2ad8186ba4fcc89103754de52 Mon Sep 17 00:00:00 2001 From: Tuomas Tynkkynen <tuomas(a)tuxera.com> Date: Wed, 6 Sep 2017 17:59:08 +0300 Subject: net/9p: Switch to wait_event_killable() From: Tuomas Tynkkynen <tuomas(a)tuxera.com> commit 9523feac272ccad2ad8186ba4fcc89103754de52 upstream. Because userspace gets Very Unhappy when calls like stat() and execve() return -EINTR on 9p filesystem mounts. For instance, when bash is looking in PATH for things to execute and some SIGCHLD interrupts stat(), bash can throw a spurious 'command not found' since it doesn't retry the stat(). In practice, hitting the problem is rare and needs a really slow/bogged down 9p server. Signed-off-by: Tuomas Tynkkynen <tuomas(a)tuxera.com> Signed-off-by: Al Viro <viro(a)zeniv.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- net/9p/client.c | 3 +-- net/9p/trans_virtio.c | 13 ++++++------- 2 files changed, 7 insertions(+), 9 deletions(-) --- a/net/9p/client.c +++ b/net/9p/client.c @@ -749,8 +749,7 @@ p9_client_rpc(struct p9_client *c, int8_ } again: /* Wait for the response */ - err = wait_event_interruptible(*req->wq, - req->status >= REQ_STATUS_RCVD); + err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD); /* * Make sure our req is coherent with regard to updates in other --- a/net/9p/trans_virtio.c +++ b/net/9p/trans_virtio.c @@ -290,8 +290,8 @@ req_retry: if (err == -ENOSPC) { chan->ring_bufs_avail = 0; spin_unlock_irqrestore(&chan->lock, flags); - err = wait_event_interruptible(*chan->vc_wq, - chan->ring_bufs_avail); + err = wait_event_killable(*chan->vc_wq, + chan->ring_bufs_avail); if (err == -ERESTARTSYS) return err; @@ -331,7 +331,7 @@ static int p9_get_mapped_pages(struct vi * Other zc request to finish here */ if (atomic_read(&vp_pinned) >= chan->p9_max_pages) { - err = wait_event_interruptible(vp_wq, + err = wait_event_killable(vp_wq, (atomic_read(&vp_pinned) < chan->p9_max_pages)); if (err == -ERESTARTSYS) return err; @@ -475,8 +475,8 @@ req_retry_pinned: if (err == -ENOSPC) { chan->ring_bufs_avail = 0; spin_unlock_irqrestore(&chan->lock, flags); - err = wait_event_interruptible(*chan->vc_wq, - chan->ring_bufs_avail); + err = wait_event_killable(*chan->vc_wq, + chan->ring_bufs_avail); if (err == -ERESTARTSYS) goto err_out; @@ -493,8 +493,7 @@ req_retry_pinned: virtqueue_kick(chan->vq); spin_unlock_irqrestore(&chan->lock, flags); p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); - err = wait_event_interruptible(*req->wq, - req->status >= REQ_STATUS_RCVD); + err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD); /* * Non kernel buffers are pinned, unpin them */ Patches currently in stable-queue which might be from tuomas(a)tuxera.com are queue-4.4/net-9p-switch-to-wait_event_killable.patch queue-4.4/fs-9p-compare-qid.path-in-v9fs_test_inode.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] Patch "fscrypt: lock mutex before checking for bounce page pool" has been added to the 4.4-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled fscrypt: lock mutex before checking for bounce page pool to the 4.4-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: fscrypt-lock-mutex-before-checking-for-bounce-page-pool.patch and it can be found in the queue-4.4 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From a0b3bc855374c50b5ea85273553485af48caf2f7 Mon Sep 17 00:00:00 2001 From: Eric Biggers <ebiggers(a)google.com> Date: Sun, 29 Oct 2017 06:30:19 -0400 Subject: fscrypt: lock mutex before checking for bounce page pool From: Eric Biggers <ebiggers(a)google.com> commit a0b3bc855374c50b5ea85273553485af48caf2f7 upstream. fscrypt_initialize(), which allocates the global bounce page pool when an encrypted file is first accessed, uses "double-checked locking" to try to avoid locking fscrypt_init_mutex. However, it doesn't use any memory barriers, so it's theoretically possible for a thread to observe a bounce page pool which has not been fully initialized. This is a classic bug with "double-checked locking". While "only a theoretical issue" in the latest kernel, in pre-4.8 kernels the pointer that was checked was not even the last to be initialized, so it was easily possible for a crash (NULL pointer dereference) to happen. This was changed only incidentally by the large refactor to use fs/crypto/. Solve both problems in a trivial way that can easily be backported: just always take the mutex. It's theoretically less efficient, but it shouldn't be noticeable in practice as the mutex is only acquired very briefly once per encrypted file. Later I'd like to make this use a helper macro like DO_ONCE(). However, DO_ONCE() runs in atomic context, so we'd need to add a new macro that allows blocking. Signed-off-by: Eric Biggers <ebiggers(a)google.com> Signed-off-by: Theodore Ts'o <tytso(a)mit.edu> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- fs/ext4/crypto_key.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) --- a/fs/ext4/crypto_key.c +++ b/fs/ext4/crypto_key.c @@ -129,11 +129,9 @@ int ext4_get_encryption_info(struct inod if (ei->i_crypt_info) return 0; - if (!ext4_read_workqueue) { - res = ext4_init_crypto(); - if (res) - return res; - } + res = ext4_init_crypto(); + if (res) + return res; res = ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION, EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, Patches currently in stable-queue which might be from ebiggers(a)google.com are queue-4.4/lib-mpi-call-cond_resched-from-mpi_powm-loop.patch queue-4.4/fscrypt-lock-mutex-before-checking-for-bounce-page-pool.patch queue-4.4/dm-bufio-fix-integer-overflow-when-limiting-maximum-cache-size.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] Patch "ARM64: dts: meson-gxl: Add alternate ARM Trusted Firmware reserved memory zone" has been added to the 4.14-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled ARM64: dts: meson-gxl: Add alternate ARM Trusted Firmware reserved memory zone to the 4.14-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: arm64-dts-meson-gxl-add-alternate-arm-trusted-firmware-reserved-memory-zone.patch and it can be found in the queue-4.14 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From 4ee8e51b9edfe7845a094690a365c844e5a35b4b Mon Sep 17 00:00:00 2001 From: Neil Armstrong <narmstrong(a)baylibre.com> Date: Wed, 11 Oct 2017 17:23:12 +0200 Subject: ARM64: dts: meson-gxl: Add alternate ARM Trusted Firmware reserved memory zone From: Neil Armstrong <narmstrong(a)baylibre.com> commit 4ee8e51b9edfe7845a094690a365c844e5a35b4b upstream. This year, Amlogic updated the ARM Trusted Firmware reserved memory mapping for Meson GXL SoCs and products sold since May 2017 uses this alternate reserved memory mapping. But products had been sold using the previous mapping. This issue has been explained in [1] and a dynamic solution is yet to be found to avoid loosing another 3Mbytes of reservable memory. In the meantime, this patch adds this alternate memory zone only for the GXL and GXM SoCs since GXBB based new products stopped earlier. [1] http://lists.infradead.org/pipermail/linux-amlogic/2017-October/004860.html Fixes: bba8e3f42736 ("ARM64: dts: meson-gx: Add firmware reserved memory zones") Reported-by: Jerome Brunet <jbrunet(a)baylibre.com> Signed-off-by: Neil Armstrong <narmstrong(a)baylibre.com> Signed-off-by: Kevin Hilman <khilman(a)baylibre.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- arch/arm64/boot/dts/amlogic/meson-gxl.dtsi | 8 ++++++++ 1 file changed, 8 insertions(+) --- a/arch/arm64/boot/dts/amlogic/meson-gxl.dtsi +++ b/arch/arm64/boot/dts/amlogic/meson-gxl.dtsi @@ -49,6 +49,14 @@ / { compatible = "amlogic,meson-gxl"; + + reserved-memory { + /* Alternate 3 MiB reserved for ARM Trusted Firmware (BL31) */ + secmon_reserved_alt: secmon@05000000 { + reg = <0x0 0x05000000 0x0 0x300000>; + no-map; + }; + }; }; &ethmac { Patches currently in stable-queue which might be from narmstrong(a)baylibre.com are queue-4.14/arm64-dts-meson-gxl-add-alternate-arm-trusted-firmware-reserved-memory-zone.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] [PATCH v4.9] PM / OPP: Add missing of_node_put(np)

by Viresh Kumar

From: Tobias Jordan <Tobias.Jordan(a)elektrobit.com> commit 7978db344719dab1e56d05e6fc04aaaddcde0a5e upstream. The for_each_available_child_of_node() loop in _of_add_opp_table_v2() doesn't drop the reference to "np" on errors. Fix that. Fixes: 274659029c9d (PM / OPP: Add support to parse "operating-points-v2" bindings) Cc: 4.9 <stable(a)vger.kernel.org> # 4.9 Signed-off-by: Tobias Jordan <Tobias.Jordan(a)elektrobit.com> [ VK: Improved commit log. ] Signed-off-by: Viresh Kumar <viresh.kumar(a)linaro.org> Reviewed-by: Stephen Boyd <sboyd(a)codeaurora.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki(a)intel.com> --- drivers/base/power/opp/of.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/base/power/opp/of.c b/drivers/base/power/opp/of.c index 5552211e6fcd..af208884bbed 100644 --- a/drivers/base/power/opp/of.c +++ b/drivers/base/power/opp/of.c @@ -348,6 +348,7 @@ static int _of_add_opp_table_v2(struct device *dev, struct device_node *opp_np) if (ret) { dev_err(dev, "%s: Failed to add OPP, %d\n", __func__, ret); + of_node_put(np); goto free_table; } } -- 2.15.0.194.g9af6a3dea062

7 years, 9 months

2
1
0 0

[Linux-stable-mirror] Patch "mtd: nand: Fix writing mtdoops to nand flash." has been added to the 3.18-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled mtd: nand: Fix writing mtdoops to nand flash. to the 3.18-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: mtd-nand-fix-writing-mtdoops-to-nand-flash.patch and it can be found in the queue-3.18 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From 30863e38ebeb500a31cecee8096fb5002677dd9b Mon Sep 17 00:00:00 2001 From: Brent Taylor <motobud(a)gmail.com> Date: Mon, 30 Oct 2017 22:32:45 -0500 Subject: mtd: nand: Fix writing mtdoops to nand flash. From: Brent Taylor <motobud(a)gmail.com> commit 30863e38ebeb500a31cecee8096fb5002677dd9b upstream. When mtdoops calls mtd_panic_write(), it eventually calls panic_nand_write() in nand_base.c. In order to properly wait for the nand chip to be ready in panic_nand_wait(), the chip must first be selected. When using the atmel nand flash controller, a panic would occur due to a NULL pointer exception. Fixes: 2af7c6539931 ("mtd: Add panic_write for NAND flashes") Signed-off-by: Brent Taylor <motobud(a)gmail.com> Signed-off-by: Boris Brezillon <boris.brezillon(a)free-electrons.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- drivers/mtd/nand/nand_base.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) --- a/drivers/mtd/nand/nand_base.c +++ b/drivers/mtd/nand/nand_base.c @@ -2501,15 +2501,18 @@ static int panic_nand_write(struct mtd_i size_t *retlen, const uint8_t *buf) { struct nand_chip *chip = mtd->priv; + int chipnr = (int)(to >> chip->chip_shift); struct mtd_oob_ops ops; int ret; - /* Wait for the device to get ready */ - panic_nand_wait(mtd, chip, 400); - /* Grab the device */ panic_nand_get_device(chip, mtd, FL_WRITING); + chip->select_chip(mtd, chipnr); + + /* Wait for the device to get ready */ + panic_nand_wait(mtd, chip, 400); + ops.len = len; ops.datbuf = (uint8_t *)buf; ops.oobbuf = NULL; Patches currently in stable-queue which might be from motobud(a)gmail.com are queue-3.18/mtd-nand-fix-writing-mtdoops-to-nand-flash.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] Patch "net/9p: Switch to wait_event_killable()" has been added to the 3.18-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled net/9p: Switch to wait_event_killable() to the 3.18-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: net-9p-switch-to-wait_event_killable.patch and it can be found in the queue-3.18 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From 9523feac272ccad2ad8186ba4fcc89103754de52 Mon Sep 17 00:00:00 2001 From: Tuomas Tynkkynen <tuomas(a)tuxera.com> Date: Wed, 6 Sep 2017 17:59:08 +0300 Subject: net/9p: Switch to wait_event_killable() From: Tuomas Tynkkynen <tuomas(a)tuxera.com> commit 9523feac272ccad2ad8186ba4fcc89103754de52 upstream. Because userspace gets Very Unhappy when calls like stat() and execve() return -EINTR on 9p filesystem mounts. For instance, when bash is looking in PATH for things to execute and some SIGCHLD interrupts stat(), bash can throw a spurious 'command not found' since it doesn't retry the stat(). In practice, hitting the problem is rare and needs a really slow/bogged down 9p server. Signed-off-by: Tuomas Tynkkynen <tuomas(a)tuxera.com> Signed-off-by: Al Viro <viro(a)zeniv.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- net/9p/client.c | 3 +-- net/9p/trans_virtio.c | 13 ++++++------- 2 files changed, 7 insertions(+), 9 deletions(-) --- a/net/9p/client.c +++ b/net/9p/client.c @@ -753,8 +753,7 @@ p9_client_rpc(struct p9_client *c, int8_ } again: /* Wait for the response */ - err = wait_event_interruptible(*req->wq, - req->status >= REQ_STATUS_RCVD); + err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD); /* * Make sure our req is coherent with regard to updates in other --- a/net/9p/trans_virtio.c +++ b/net/9p/trans_virtio.c @@ -292,8 +292,8 @@ req_retry: if (err == -ENOSPC) { chan->ring_bufs_avail = 0; spin_unlock_irqrestore(&chan->lock, flags); - err = wait_event_interruptible(*chan->vc_wq, - chan->ring_bufs_avail); + err = wait_event_killable(*chan->vc_wq, + chan->ring_bufs_avail); if (err == -ERESTARTSYS) return err; @@ -324,7 +324,7 @@ static int p9_get_mapped_pages(struct vi * Other zc request to finish here */ if (atomic_read(&vp_pinned) >= chan->p9_max_pages) { - err = wait_event_interruptible(vp_wq, + err = wait_event_killable(vp_wq, (atomic_read(&vp_pinned) < chan->p9_max_pages)); if (err == -ERESTARTSYS) return err; @@ -454,8 +454,8 @@ req_retry_pinned: if (err == -ENOSPC) { chan->ring_bufs_avail = 0; spin_unlock_irqrestore(&chan->lock, flags); - err = wait_event_interruptible(*chan->vc_wq, - chan->ring_bufs_avail); + err = wait_event_killable(*chan->vc_wq, + chan->ring_bufs_avail); if (err == -ERESTARTSYS) goto err_out; @@ -472,8 +472,7 @@ req_retry_pinned: virtqueue_kick(chan->vq); spin_unlock_irqrestore(&chan->lock, flags); p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); - err = wait_event_interruptible(*req->wq, - req->status >= REQ_STATUS_RCVD); + err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD); /* * Non kernel buffers are pinned, unpin them */ Patches currently in stable-queue which might be from tuomas(a)tuxera.com are queue-3.18/net-9p-switch-to-wait_event_killable.patch queue-3.18/fs-9p-compare-qid.path-in-v9fs_test_inode.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] [PATCH] mtd: nand: Fix writing mtdoops to nand flash

by motobud＠gmail.com

From: Brent Taylor <motobud(a)gmail.com> commit 30863e38ebeb500a31cecee8096fb5002677dd9b upstream. When mtdoops calls mtd_panic_write(), it eventually calls panic_nand_write() in nand_base.c. In order to properly wait for the nand chip to be ready in panic_nand_wait(), the chip must first be selected. When using the atmel nand flash controller, a panic would occur due to a NULL pointer exception. Fixes: 2af7c6539931 ("mtd: Add panic_write for NAND flashes") Cc: <stable(a)vger.kernel.org> # 4.4.x Cc: <stable(a)vger.kernel.org> # 3.18.x Signed-off-by: Brent Taylor <motobud(a)gmail.com> --- drivers/mtd/nand/nand_base.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/drivers/mtd/nand/nand_base.c b/drivers/mtd/nand/nand_base.c index 5b5c62712814..f0b95212c6b7 100644 --- a/drivers/mtd/nand/nand_base.c +++ b/drivers/mtd/nand/nand_base.c @@ -2501,15 +2501,18 @@ static int panic_nand_write(struct mtd_info *mtd, loff_t to, size_t len, size_t *retlen, const uint8_t *buf) { struct nand_chip *chip = mtd->priv; + int chipnr = (int)(to >> chip->chip_shift); struct mtd_oob_ops ops; int ret; - /* Wait for the device to get ready */ - panic_nand_wait(mtd, chip, 400); - /* Grab the device */ panic_nand_get_device(chip, mtd, FL_WRITING); + chip->select_chip(mtd, chipnr); + + /* Wait for the device to get ready */ + panic_nand_wait(mtd, chip, 400); + ops.len = len; ops.datbuf = (uint8_t *)buf; ops.oobbuf = NULL; -- 2.15.0

7 years, 9 months

2
1
0 0

[Linux-stable-mirror] [PATCH stable backport] net/9p: Switch to wait_event_killable()

by Tuomas Tynkkynen

commit 9523feac272ccad2ad8186ba4fcc89103754de52 upstream, with changes to net/9p/trans_xen.c dropped as that driver was only added in 4.12. Because userspace gets Very Unhappy when calls like stat() and execve() return -EINTR on 9p filesystem mounts. For instance, when bash is looking in PATH for things to execute and some SIGCHLD interrupts stat(), bash can throw a spurious 'command not found' since it doesn't retry the stat(). In practice, hitting the problem is rare and needs a really slow/bogged down 9p server. Cc: stable(a)vger.kernel.org Signed-off-by: Tuomas Tynkkynen <tuomas(a)tuxera.com> Signed-off-by: Al Viro <viro(a)zeniv.linux.org.uk> --- This backport applies to stable trees for 4.9 and below. --- net/9p/client.c | 3 +-- net/9p/trans_virtio.c | 13 ++++++------- 2 files changed, 7 insertions(+), 9 deletions(-) diff --git a/net/9p/client.c b/net/9p/client.c index cf129fec7329..1fd60190177e 100644 --- a/net/9p/client.c +++ b/net/9p/client.c @@ -749,8 +749,7 @@ p9_client_rpc(struct p9_client *c, int8_t type, const char *fmt, ...) } again: /* Wait for the response */ - err = wait_event_interruptible(*req->wq, - req->status >= REQ_STATUS_RCVD); + err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD); /* * Make sure our req is coherent with regard to updates in other diff --git a/net/9p/trans_virtio.c b/net/9p/trans_virtio.c index f24b25c25106..f3a4efcf1456 100644 --- a/net/9p/trans_virtio.c +++ b/net/9p/trans_virtio.c @@ -286,8 +286,8 @@ p9_virtio_request(struct p9_client *client, struct p9_req_t *req) if (err == -ENOSPC) { chan->ring_bufs_avail = 0; spin_unlock_irqrestore(&chan->lock, flags); - err = wait_event_interruptible(*chan->vc_wq, - chan->ring_bufs_avail); + err = wait_event_killable(*chan->vc_wq, + chan->ring_bufs_avail); if (err == -ERESTARTSYS) return err; @@ -327,7 +327,7 @@ static int p9_get_mapped_pages(struct virtio_chan *chan, * Other zc request to finish here */ if (atomic_read(&vp_pinned) >= chan->p9_max_pages) { - err = wait_event_interruptible(vp_wq, + err = wait_event_killable(vp_wq, (atomic_read(&vp_pinned) < chan->p9_max_pages)); if (err == -ERESTARTSYS) return err; @@ -471,8 +471,8 @@ p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req, if (err == -ENOSPC) { chan->ring_bufs_avail = 0; spin_unlock_irqrestore(&chan->lock, flags); - err = wait_event_interruptible(*chan->vc_wq, - chan->ring_bufs_avail); + err = wait_event_killable(*chan->vc_wq, + chan->ring_bufs_avail); if (err == -ERESTARTSYS) goto err_out; @@ -489,8 +489,7 @@ p9_virtio_zc_request(struct p9_client *client, struct p9_req_t *req, virtqueue_kick(chan->vq); spin_unlock_irqrestore(&chan->lock, flags); p9_debug(P9_DEBUG_TRANS, "virtio request kicked\n"); - err = wait_event_interruptible(*req->wq, - req->status >= REQ_STATUS_RCVD); + err = wait_event_killable(*req->wq, req->status >= REQ_STATUS_RCVD); /* * Non kernel buffers are pinned, unpin them */ -- 2.15.0

7 years, 9 months

2
1
0 0

Re: [Linux-stable-mirror] [PATCH] ARM64: dts: meson-gxl: Add alternate ARM Trusted Firmware reserved memory zone

by Kevin Hilman

Greg, Neil Armstrong <narmstrong(a)baylibre.com> writes: > This year, Amlogic updated the ARM Trusted Firmware reserved memory mapping > for Meson GXL SoCs and products sold since May 2017 uses this alternate > reserved memory mapping. > But products had been sold using the previous mapping. > > This issue has been explained in [1] and a dynamic solution is yet to be > found to avoid loosing another 3Mbytes of reservable memory. > > In the meantime, this patch adds this alternate memory zone only for > the GXL and GXM SoCs since GXBB based new products stopped earlier. > > [1] http://lists.infradead.org/pipermail/linux-amlogic/2017-October/004860.html > > Fixes: bba8e3f42736 ("ARM64: dts: meson-gx: Add firmware reserved memory zones") > Reported-by: Jerome Brunet <jbrunet(a)baylibre.com> > Signed-off-by: Neil Armstrong <narmstrong(a)baylibre.com> I missed the Fixes tag on this one, but it's now commit 4ee8e51b9edf upstream. Please apply to stable v4.10+ Thanks, Kevin

7 years, 9 months

2
1
0 0

[Linux-stable-mirror] [PATCH 4.4] fscrypt: lock mutex before checking for bounce page pool

by Eric Biggers

commit a0b3bc855374c50b5ea85273553485af48caf2f7 upstream. [Please apply to 4.4-stable.] fscrypt_initialize(), which allocates the global bounce page pool when an encrypted file is first accessed, uses "double-checked locking" to try to avoid locking fscrypt_init_mutex. However, it doesn't use any memory barriers, so it's theoretically possible for a thread to observe a bounce page pool which has not been fully initialized. This is a classic bug with "double-checked locking". While "only a theoretical issue" in the latest kernel, in pre-4.8 kernels the pointer that was checked was not even the last to be initialized, so it was easily possible for a crash (NULL pointer dereference) to happen. This was changed only incidentally by the large refactor to use fs/crypto/. Solve both problems in a trivial way that can easily be backported: just always take the mutex. It's theoretically less efficient, but it shouldn't be noticeable in practice as the mutex is only acquired very briefly once per encrypted file. Later I'd like to make this use a helper macro like DO_ONCE(). However, DO_ONCE() runs in atomic context, so we'd need to add a new macro that allows blocking. Cc: stable(a)vger.kernel.org # v4.1+ Signed-off-by: Eric Biggers <ebiggers(a)google.com> Signed-off-by: Theodore Ts'o <tytso(a)mit.edu> --- fs/ext4/crypto_key.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) diff --git a/fs/ext4/crypto_key.c b/fs/ext4/crypto_key.c index 9a1bc638abce..9308fe4b66e6 100644 --- a/fs/ext4/crypto_key.c +++ b/fs/ext4/crypto_key.c @@ -129,11 +129,9 @@ int ext4_get_encryption_info(struct inode *inode) if (ei->i_crypt_info) return 0; - if (!ext4_read_workqueue) { - res = ext4_init_crypto(); - if (res) - return res; - } + res = ext4_init_crypto(); + if (res) + return res; res = ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION, EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, -- 2.15.0.417.g466bffb3ac-goog

7 years, 9 months

2
1
0 0

[Linux-stable-mirror] [PATCH 4.9] fscrypt: lock mutex before checking for bounce page pool

by Eric Biggers

commit a0b3bc855374c50b5ea85273553485af48caf2f7 upstream. [Please apply to 4.9-stable.] fscrypt_initialize(), which allocates the global bounce page pool when an encrypted file is first accessed, uses "double-checked locking" to try to avoid locking fscrypt_init_mutex. However, it doesn't use any memory barriers, so it's theoretically possible for a thread to observe a bounce page pool which has not been fully initialized. This is a classic bug with "double-checked locking". While "only a theoretical issue" in the latest kernel, in pre-4.8 kernels the pointer that was checked was not even the last to be initialized, so it was easily possible for a crash (NULL pointer dereference) to happen. This was changed only incidentally by the large refactor to use fs/crypto/. Solve both problems in a trivial way that can easily be backported: just always take the mutex. It's theoretically less efficient, but it shouldn't be noticeable in practice as the mutex is only acquired very briefly once per encrypted file. Later I'd like to make this use a helper macro like DO_ONCE(). However, DO_ONCE() runs in atomic context, so we'd need to add a new macro that allows blocking. Cc: stable(a)vger.kernel.org # v4.1+ Signed-off-by: Eric Biggers <ebiggers(a)google.com> Signed-off-by: Theodore Ts'o <tytso(a)mit.edu> --- fs/crypto/crypto.c | 3 --- 1 file changed, 3 deletions(-) diff --git a/fs/crypto/crypto.c b/fs/crypto/crypto.c index 61cfccea77bc..73de1446c8d4 100644 --- a/fs/crypto/crypto.c +++ b/fs/crypto/crypto.c @@ -484,9 +484,6 @@ int fscrypt_initialize(void) { int i, res = -ENOMEM; - if (fscrypt_bounce_page_pool) - return 0; - mutex_lock(&fscrypt_init_mutex); if (fscrypt_bounce_page_pool) goto already_initialized; -- 2.15.0.417.g466bffb3ac-goog

7 years, 9 months

2
1
0 0

Re: [Linux-stable-mirror] [PATCH] Ensure that the SCSI error handler gets woken up

by Pavel Tikhomirov

1-st, Stuart - thanks for adding me to CC, 2-nd Bart - no idea why you didn't? =) > If scsi_eh_scmd_add() is called concurrently with scsi_host_queue_ready() > while shost->host_blocked > 0 then it can happen that neither function > wakes up the SCSI error handler. Fix this by making every function that > decreases the host_busy counter to wake up the error handler if necessary. Bart, you've added a comment to my initial patch() about performance, let me quote it here: > An important achievement of the scsi-mq code was removal of all > spin_lock_irq(shost->host_lock) statements from the hot path. The above > changes will have a significant negative performance impact, especially if > multiple LUNs associated with the same SCSI host are involved. Can the > reported race be fixed without slowing down the hot path significantly? I > think that both adding spin lock or smp_mb() calls in the hot path will > have a significant negative performance impact. These was a tricky question so I had no immediate answer. Here is the one: a) We need to check if scsi_eh_wakeup needs to wake up error handler thread in every place where we change host_busy. Else we have a chance that these change will break the wake up check in other existing places and will lead to deadlock. b) If we have several variables and change them (one different variable in in different thread) and after that we want to check the joint state of these variables, we sould surely have some kind of memory barriers to have a consistent state at some point. c) We already have spinlocks in scsi_schedule_eh, scsi_eh_scmd_add and scsi_device_unbusy, so it seems the good thing to reuse them for new checks too. I don't see another way to fix these problem. Your patch puts spinlocks under check which should itself be under spinlock, and breaks the initial fix (see Stuart's comment that the problem still reproduces). And you does _not_ answer your own question. > > Reported-by: Pavel Tikhomirov <ptikhomirov@xxxxxxxxxxxxx> > Fixes: commit 746650160866 ("scsi: convert host_busy to atomic_t") > Signed-off-by: Bart Van Assche <bart.vanassche@xxxxxxx> As your patch is based on my initial patch (https://patchwork.kernel.org/patch/9938919/), when all problems will be resolved with it, can you please add here: Signed-off-by: Pavel Tikhomirov <ptikhomirov(a)virtuozzo.com> > Cc: Konstantin Khorenko <khorenko@xxxxxxxxxxxxx> > Cc: Stuart Hayes <stuart.w.hayes@xxxxxxxxx> > Cc: Christoph Hellwig <hch@xxxxxx> > Cc: Hannes Reinecke <hare@xxxxxxxx> > Cc: Johannes Thumshirn <jthumshirn@xxxxxxx> > Cc: <stable@xxxxxxxxxxxxxxx> > --- > drivers/scsi/scsi_error.c | 3 ++- > drivers/scsi/scsi_lib.c | 22 ++++++++++++++-------- > 2 files changed, 16 insertions(+), 9 deletions(-) > > diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c > index 5e89049e9b4e..f7f014c755d7 100644 > --- a/drivers/scsi/scsi_error.c > +++ b/drivers/scsi/scsi_error.c > @@ -61,9 +61,10 @@ static int scsi_eh_try_stu(struct scsi_cmnd *scmd); > static int scsi_try_to_abort_cmd(struct scsi_host_template *, > struct scsi_cmnd *); > > -/* called with shost->host_lock held */ > void scsi_eh_wakeup(struct Scsi_Host *shost) > { > + lockdep_assert_held(shost->host_lock); > + > if (atomic_read(&shost->host_busy) == shost->host_failed) { > trace_scsi_eh_wakeup(shost); > wake_up_process(shost->ehandler); I also think these hunk is just an additional precaution and should be in separate patch. > diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c > index 1e05e1885ac8..abd37d77af2d 100644 > --- a/drivers/scsi/scsi_lib.c > +++ b/drivers/scsi/scsi_lib.c > @@ -318,22 +318,28 @@ static void scsi_init_cmd_errh(struct scsi_cmnd *cmd) > cmd->cmd_len = scsi_command_size(cmd->cmnd); > } > > -void scsi_device_unbusy(struct scsi_device *sdev) > +static void scsi_dec_host_busy(struct Scsi_Host *shost) > { > - struct Scsi_Host *shost = sdev->host; > - struct scsi_target *starget = scsi_target(sdev); > unsigned long flags; > > atomic_dec(&shost->host_busy); > - if (starget->can_queue > 0) > - atomic_dec(&starget->target_busy); > - > if (unlikely(scsi_host_in_recovery(shost) && > (shost->host_failed || shost->host_eh_scheduled))) { As I've wrote above you do wrong locking here in scsi_dec_host_busy. Note that the above check reads host_failed and can load host_failed before host_busy is decremented due to reordering. > spin_lock_irqsave(shost->host_lock, flags); > scsi_eh_wakeup(shost); > spin_unlock_irqrestore(shost->host_lock, flags); > } > +} > + > +void scsi_device_unbusy(struct scsi_device *sdev) > +{ > + struct Scsi_Host *shost = sdev->host; > + struct scsi_target *starget = scsi_target(sdev); > + > + scsi_dec_host_busy(shost); > + > + if (starget->can_queue > 0) > + atomic_dec(&starget->target_busy); > > atomic_dec(&sdev->device_busy); > } > @@ -1532,7 +1538,7 @@ static inline int scsi_host_queue_ready(struct request_queue *q, > list_add_tail(&sdev->starved_entry, &shost->starved_list); > spin_unlock_irq(shost->host_lock); > out_dec: > - atomic_dec(&shost->host_busy); > + scsi_dec_host_busy(shost); > return 0; > } > > @@ -2020,7 +2026,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, > return BLK_STS_OK; > > out_dec_host_busy: > - atomic_dec(&shost->host_busy); > + scsi_dec_host_busy(shost); > out_dec_target_busy: > if (scsi_target(sdev)->can_queue > 0) > atomic_dec(&scsi_target(sdev)->target_busy); > -- > 2.15.0 -- Best regards, Tikhomirov Pavel Software Developer, Virtuozzo.

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] Patch "sched/rt: Simplify the IPI based RT balancing logic" has been added to the 4.9-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled sched/rt: Simplify the IPI based RT balancing logic to the 4.9-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: sched-rt-simplify-the-ipi-based-rt-balancing-logic.patch and it can be found in the queue-4.9 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From 4bdced5c9a2922521e325896a7bbbf0132c94e56 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" <rostedt(a)goodmis.org> Date: Fri, 6 Oct 2017 14:05:04 -0400 Subject: sched/rt: Simplify the IPI based RT balancing logic From: Steven Rostedt (Red Hat) <rostedt(a)goodmis.org> commit 4bdced5c9a2922521e325896a7bbbf0132c94e56 upstream. When a CPU lowers its priority (schedules out a high priority task for a lower priority one), a check is made to see if any other CPU has overloaded RT tasks (more than one). It checks the rto_mask to determine this and if so it will request to pull one of those tasks to itself if the non running RT task is of higher priority than the new priority of the next task to run on the current CPU. When we deal with large number of CPUs, the original pull logic suffered from large lock contention on a single CPU run queue, which caused a huge latency across all CPUs. This was caused by only having one CPU having overloaded RT tasks and a bunch of other CPUs lowering their priority. To solve this issue, commit: b6366f048e0c ("sched/rt: Use IPI to trigger RT task push migration instead of pulling") changed the way to request a pull. Instead of grabbing the lock of the overloaded CPU's runqueue, it simply sent an IPI to that CPU to do the work. Although the IPI logic worked very well in removing the large latency build up, it still could suffer from a large number of IPIs being sent to a single CPU. On a 80 CPU box, I measured over 200us of processing IPIs. Worse yet, when I tested this on a 120 CPU box, with a stress test that had lots of RT tasks scheduling on all CPUs, it actually triggered the hard lockup detector! One CPU had so many IPIs sent to it, and due to the restart mechanism that is triggered when the source run queue has a priority status change, the CPU spent minutes! processing the IPIs. Thinking about this further, I realized there's no reason for each run queue to send its own IPI. As all CPUs with overloaded tasks must be scanned regardless if there's one or many CPUs lowering their priority, because there's no current way to find the CPU with the highest priority task that can schedule to one of these CPUs, there really only needs to be one IPI being sent around at a time. This greatly simplifies the code! The new approach is to have each root domain have its own irq work, as the rto_mask is per root domain. The root domain has the following fields attached to it: rto_push_work - the irq work to process each CPU set in rto_mask rto_lock - the lock to protect some of the other rto fields rto_loop_start - an atomic that keeps contention down on rto_lock the first CPU scheduling in a lower priority task is the one to kick off the process. rto_loop_next - an atomic that gets incremented for each CPU that schedules in a lower priority task. rto_loop - a variable protected by rto_lock that is used to compare against rto_loop_next rto_cpu - The cpu to send the next IPI to, also protected by the rto_lock. When a CPU schedules in a lower priority task and wants to make sure overloaded CPUs know about it. It increments the rto_loop_next. Then it atomically sets rto_loop_start with a cmpxchg. If the old value is not "0", then it is done, as another CPU is kicking off the IPI loop. If the old value is "0", then it will take the rto_lock to synchronize with a possible IPI being sent around to the overloaded CPUs. If rto_cpu is greater than or equal to nr_cpu_ids, then there's either no IPI being sent around, or one is about to finish. Then rto_cpu is set to the first CPU in rto_mask and an IPI is sent to that CPU. If there's no CPUs set in rto_mask, then there's nothing to be done. When the CPU receives the IPI, it will first try to push any RT tasks that is queued on the CPU but can't run because a higher priority RT task is currently running on that CPU. Then it takes the rto_lock and looks for the next CPU in the rto_mask. If it finds one, it simply sends an IPI to that CPU and the process continues. If there's no more CPUs in the rto_mask, then rto_loop is compared with rto_loop_next. If they match, everything is done and the process is over. If they do not match, then a CPU scheduled in a lower priority task as the IPI was being passed around, and the process needs to start again. The first CPU in rto_mask is sent the IPI. This change removes this duplication of work in the IPI logic, and greatly lowers the latency caused by the IPIs. This removed the lockup happening on the 120 CPU machine. It also simplifies the code tremendously. What else could anyone ask for? Thanks to Peter Zijlstra for simplifying the rto_loop_start atomic logic and supplying me with the rto_start_trylock() and rto_start_unlock() helper functions. Signed-off-by: Steven Rostedt (VMware) <rostedt(a)goodmis.org> Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Cc: Clark Williams <williams(a)redhat.com> Cc: Daniel Bristot de Oliveira <bristot(a)redhat.com> Cc: John Kacur <jkacur(a)redhat.com> Cc: Linus Torvalds <torvalds(a)linux-foundation.org> Cc: Mike Galbraith <efault(a)gmx.de> Cc: Peter Zijlstra <peterz(a)infradead.org> Cc: Scott Wood <swood(a)redhat.com> Cc: Thomas Gleixner <tglx(a)linutronix.de> Link: http://lkml.kernel.org/r/20170424114732.1aac6dc4@gandalf.local.home Signed-off-by: Ingo Molnar <mingo(a)kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- kernel/sched/core.c | 6 + kernel/sched/rt.c | 235 ++++++++++++++++++++++++--------------------------- kernel/sched/sched.h | 24 +++-- 3 files changed, 138 insertions(+), 127 deletions(-) --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -5877,6 +5877,12 @@ static int init_rootdomain(struct root_d if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) goto free_dlo_mask; +#ifdef HAVE_RT_PUSH_IPI + rd->rto_cpu = -1; + raw_spin_lock_init(&rd->rto_lock); + init_irq_work(&rd->rto_push_work, rto_push_irq_work_func); +#endif + init_dl_bw(&rd->dl_bw); if (cpudl_init(&rd->cpudl) != 0) goto free_dlo_mask; --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -72,10 +72,6 @@ static void start_rt_bandwidth(struct rt raw_spin_unlock(&rt_b->rt_runtime_lock); } -#if defined(CONFIG_SMP) && defined(HAVE_RT_PUSH_IPI) -static void push_irq_work_func(struct irq_work *work); -#endif - void init_rt_rq(struct rt_rq *rt_rq) { struct rt_prio_array *array; @@ -95,13 +91,6 @@ void init_rt_rq(struct rt_rq *rt_rq) rt_rq->rt_nr_migratory = 0; rt_rq->overloaded = 0; plist_head_init(&rt_rq->pushable_tasks); - -#ifdef HAVE_RT_PUSH_IPI - rt_rq->push_flags = 0; - rt_rq->push_cpu = nr_cpu_ids; - raw_spin_lock_init(&rt_rq->push_lock); - init_irq_work(&rt_rq->push_work, push_irq_work_func); -#endif #endif /* CONFIG_SMP */ /* We start is dequeued state, because no RT tasks are queued */ rt_rq->rt_queued = 0; @@ -1864,160 +1853,166 @@ static void push_rt_tasks(struct rq *rq) } #ifdef HAVE_RT_PUSH_IPI + /* - * The search for the next cpu always starts at rq->cpu and ends - * when we reach rq->cpu again. It will never return rq->cpu. - * This returns the next cpu to check, or nr_cpu_ids if the loop - * is complete. + * When a high priority task schedules out from a CPU and a lower priority + * task is scheduled in, a check is made to see if there's any RT tasks + * on other CPUs that are waiting to run because a higher priority RT task + * is currently running on its CPU. In this case, the CPU with multiple RT + * tasks queued on it (overloaded) needs to be notified that a CPU has opened + * up that may be able to run one of its non-running queued RT tasks. + * + * All CPUs with overloaded RT tasks need to be notified as there is currently + * no way to know which of these CPUs have the highest priority task waiting + * to run. Instead of trying to take a spinlock on each of these CPUs, + * which has shown to cause large latency when done on machines with many + * CPUs, sending an IPI to the CPUs to have them push off the overloaded + * RT tasks waiting to run. + * + * Just sending an IPI to each of the CPUs is also an issue, as on large + * count CPU machines, this can cause an IPI storm on a CPU, especially + * if its the only CPU with multiple RT tasks queued, and a large number + * of CPUs scheduling a lower priority task at the same time. + * + * Each root domain has its own irq work function that can iterate over + * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT + * tassk must be checked if there's one or many CPUs that are lowering + * their priority, there's a single irq work iterator that will try to + * push off RT tasks that are waiting to run. + * + * When a CPU schedules a lower priority task, it will kick off the + * irq work iterator that will jump to each CPU with overloaded RT tasks. + * As it only takes the first CPU that schedules a lower priority task + * to start the process, the rto_start variable is incremented and if + * the atomic result is one, then that CPU will try to take the rto_lock. + * This prevents high contention on the lock as the process handles all + * CPUs scheduling lower priority tasks. + * + * All CPUs that are scheduling a lower priority task will increment the + * rt_loop_next variable. This will make sure that the irq work iterator + * checks all RT overloaded CPUs whenever a CPU schedules a new lower + * priority task, even if the iterator is in the middle of a scan. Incrementing + * the rt_loop_next will cause the iterator to perform another scan. * - * rq->rt.push_cpu holds the last cpu returned by this function, - * or if this is the first instance, it must hold rq->cpu. */ static int rto_next_cpu(struct rq *rq) { - int prev_cpu = rq->rt.push_cpu; + struct root_domain *rd = rq->rd; + int next; int cpu; - cpu = cpumask_next(prev_cpu, rq->rd->rto_mask); - /* - * If the previous cpu is less than the rq's CPU, then it already - * passed the end of the mask, and has started from the beginning. - * We end if the next CPU is greater or equal to rq's CPU. + * When starting the IPI RT pushing, the rto_cpu is set to -1, + * rt_next_cpu() will simply return the first CPU found in + * the rto_mask. + * + * If rto_next_cpu() is called with rto_cpu is a valid cpu, it + * will return the next CPU found in the rto_mask. + * + * If there are no more CPUs left in the rto_mask, then a check is made + * against rto_loop and rto_loop_next. rto_loop is only updated with + * the rto_lock held, but any CPU may increment the rto_loop_next + * without any locking. */ - if (prev_cpu < rq->cpu) { - if (cpu >= rq->cpu) - return nr_cpu_ids; + for (;;) { - } else if (cpu >= nr_cpu_ids) { - /* - * We passed the end of the mask, start at the beginning. - * If the result is greater or equal to the rq's CPU, then - * the loop is finished. - */ - cpu = cpumask_first(rq->rd->rto_mask); - if (cpu >= rq->cpu) - return nr_cpu_ids; - } - rq->rt.push_cpu = cpu; + /* When rto_cpu is -1 this acts like cpumask_first() */ + cpu = cpumask_next(rd->rto_cpu, rd->rto_mask); - /* Return cpu to let the caller know if the loop is finished or not */ - return cpu; -} + rd->rto_cpu = cpu; -static int find_next_push_cpu(struct rq *rq) -{ - struct rq *next_rq; - int cpu; + if (cpu < nr_cpu_ids) + return cpu; - while (1) { - cpu = rto_next_cpu(rq); - if (cpu >= nr_cpu_ids) - break; - next_rq = cpu_rq(cpu); + rd->rto_cpu = -1; + + /* + * ACQUIRE ensures we see the @rto_mask changes + * made prior to the @next value observed. + * + * Matches WMB in rt_set_overload(). + */ + next = atomic_read_acquire(&rd->rto_loop_next); - /* Make sure the next rq can push to this rq */ - if (next_rq->rt.highest_prio.next < rq->rt.highest_prio.curr) + if (rd->rto_loop == next) break; + + rd->rto_loop = next; } - return cpu; + return -1; } -#define RT_PUSH_IPI_EXECUTING 1 -#define RT_PUSH_IPI_RESTART 2 +static inline bool rto_start_trylock(atomic_t *v) +{ + return !atomic_cmpxchg_acquire(v, 0, 1); +} -static void tell_cpu_to_push(struct rq *rq) +static inline void rto_start_unlock(atomic_t *v) { - int cpu; + atomic_set_release(v, 0); +} - if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { - raw_spin_lock(&rq->rt.push_lock); - /* Make sure it's still executing */ - if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { - /* - * Tell the IPI to restart the loop as things have - * changed since it started. - */ - rq->rt.push_flags |= RT_PUSH_IPI_RESTART; - raw_spin_unlock(&rq->rt.push_lock); - return; - } - raw_spin_unlock(&rq->rt.push_lock); - } +static void tell_cpu_to_push(struct rq *rq) +{ + int cpu = -1; - /* When here, there's no IPI going around */ + /* Keep the loop going if the IPI is currently active */ + atomic_inc(&rq->rd->rto_loop_next); - rq->rt.push_cpu = rq->cpu; - cpu = find_next_push_cpu(rq); - if (cpu >= nr_cpu_ids) + /* Only one CPU can initiate a loop at a time */ + if (!rto_start_trylock(&rq->rd->rto_loop_start)) return; - rq->rt.push_flags = RT_PUSH_IPI_EXECUTING; + raw_spin_lock(&rq->rd->rto_lock); + + /* + * The rto_cpu is updated under the lock, if it has a valid cpu + * then the IPI is still running and will continue due to the + * update to loop_next, and nothing needs to be done here. + * Otherwise it is finishing up and an ipi needs to be sent. + */ + if (rq->rd->rto_cpu < 0) + cpu = rto_next_cpu(rq); + + raw_spin_unlock(&rq->rd->rto_lock); - irq_work_queue_on(&rq->rt.push_work, cpu); + rto_start_unlock(&rq->rd->rto_loop_start); + + if (cpu >= 0) + irq_work_queue_on(&rq->rd->rto_push_work, cpu); } /* Called from hardirq context */ -static void try_to_push_tasks(void *arg) +void rto_push_irq_work_func(struct irq_work *work) { - struct rt_rq *rt_rq = arg; - struct rq *rq, *src_rq; - int this_cpu; + struct rq *rq; int cpu; - this_cpu = rt_rq->push_cpu; + rq = this_rq(); - /* Paranoid check */ - BUG_ON(this_cpu != smp_processor_id()); - - rq = cpu_rq(this_cpu); - src_rq = rq_of_rt_rq(rt_rq); - -again: + /* + * We do not need to grab the lock to check for has_pushable_tasks. + * When it gets updated, a check is made if a push is possible. + */ if (has_pushable_tasks(rq)) { raw_spin_lock(&rq->lock); - push_rt_task(rq); + push_rt_tasks(rq); raw_spin_unlock(&rq->lock); } - /* Pass the IPI to the next rt overloaded queue */ - raw_spin_lock(&rt_rq->push_lock); - /* - * If the source queue changed since the IPI went out, - * we need to restart the search from that CPU again. - */ - if (rt_rq->push_flags & RT_PUSH_IPI_RESTART) { - rt_rq->push_flags &= ~RT_PUSH_IPI_RESTART; - rt_rq->push_cpu = src_rq->cpu; - } + raw_spin_lock(&rq->rd->rto_lock); - cpu = find_next_push_cpu(src_rq); + /* Pass the IPI to the next rt overloaded queue */ + cpu = rto_next_cpu(rq); - if (cpu >= nr_cpu_ids) - rt_rq->push_flags &= ~RT_PUSH_IPI_EXECUTING; - raw_spin_unlock(&rt_rq->push_lock); + raw_spin_unlock(&rq->rd->rto_lock); - if (cpu >= nr_cpu_ids) + if (cpu < 0) return; - /* - * It is possible that a restart caused this CPU to be - * chosen again. Don't bother with an IPI, just see if we - * have more to push. - */ - if (unlikely(cpu == rq->cpu)) - goto again; - /* Try the next RT overloaded CPU */ - irq_work_queue_on(&rt_rq->push_work, cpu); -} - -static void push_irq_work_func(struct irq_work *work) -{ - struct rt_rq *rt_rq = container_of(work, struct rt_rq, push_work); - - try_to_push_tasks(rt_rq); + irq_work_queue_on(&rq->rd->rto_push_work, cpu); } #endif /* HAVE_RT_PUSH_IPI */ --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -463,7 +463,7 @@ static inline int rt_bandwidth_enabled(v } /* RT IPI pull logic requires IRQ_WORK */ -#ifdef CONFIG_IRQ_WORK +#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) # define HAVE_RT_PUSH_IPI #endif @@ -485,12 +485,6 @@ struct rt_rq { unsigned long rt_nr_total; int overloaded; struct plist_head pushable_tasks; -#ifdef HAVE_RT_PUSH_IPI - int push_flags; - int push_cpu; - struct irq_work push_work; - raw_spinlock_t push_lock; -#endif #endif /* CONFIG_SMP */ int rt_queued; @@ -572,6 +566,19 @@ struct root_domain { struct dl_bw dl_bw; struct cpudl cpudl; +#ifdef HAVE_RT_PUSH_IPI + /* + * For IPI pull requests, loop across the rto_mask. + */ + struct irq_work rto_push_work; + raw_spinlock_t rto_lock; + /* These are only updated and read within rto_lock */ + int rto_loop; + int rto_cpu; + /* These atomics are updated outside of a lock */ + atomic_t rto_loop_next; + atomic_t rto_loop_start; +#endif /* * The "RT overload" flag: it gets set if a CPU has more than * one runnable RT task. @@ -584,6 +591,9 @@ struct root_domain { extern struct root_domain def_root_domain; +#ifdef HAVE_RT_PUSH_IPI +extern void rto_push_irq_work_func(struct irq_work *work); +#endif #endif /* CONFIG_SMP */ /* Patches currently in stable-queue which might be from rostedt(a)goodmis.org are queue-4.9/sched-make-resched_cpu-unconditional.patch queue-4.9/sched-rt-simplify-the-ipi-based-rt-balancing-logic.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] Patch "sched/rt: Simplify the IPI based RT balancing logic" has been added to the 4.4-stable tree

by gregkh＠linuxfoundation.org

This is a note to let you know that I've just added the patch titled sched/rt: Simplify the IPI based RT balancing logic to the 4.4-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=sum… The filename of the patch is: sched-rt-simplify-the-ipi-based-rt-balancing-logic.patch and it can be found in the queue-4.4 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable(a)vger.kernel.org> know about it. >From 4bdced5c9a2922521e325896a7bbbf0132c94e56 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" <rostedt(a)goodmis.org> Date: Fri, 6 Oct 2017 14:05:04 -0400 Subject: sched/rt: Simplify the IPI based RT balancing logic From: Steven Rostedt (Red Hat) <rostedt(a)goodmis.org> commit 4bdced5c9a2922521e325896a7bbbf0132c94e56 upstream. When a CPU lowers its priority (schedules out a high priority task for a lower priority one), a check is made to see if any other CPU has overloaded RT tasks (more than one). It checks the rto_mask to determine this and if so it will request to pull one of those tasks to itself if the non running RT task is of higher priority than the new priority of the next task to run on the current CPU. When we deal with large number of CPUs, the original pull logic suffered from large lock contention on a single CPU run queue, which caused a huge latency across all CPUs. This was caused by only having one CPU having overloaded RT tasks and a bunch of other CPUs lowering their priority. To solve this issue, commit: b6366f048e0c ("sched/rt: Use IPI to trigger RT task push migration instead of pulling") changed the way to request a pull. Instead of grabbing the lock of the overloaded CPU's runqueue, it simply sent an IPI to that CPU to do the work. Although the IPI logic worked very well in removing the large latency build up, it still could suffer from a large number of IPIs being sent to a single CPU. On a 80 CPU box, I measured over 200us of processing IPIs. Worse yet, when I tested this on a 120 CPU box, with a stress test that had lots of RT tasks scheduling on all CPUs, it actually triggered the hard lockup detector! One CPU had so many IPIs sent to it, and due to the restart mechanism that is triggered when the source run queue has a priority status change, the CPU spent minutes! processing the IPIs. Thinking about this further, I realized there's no reason for each run queue to send its own IPI. As all CPUs with overloaded tasks must be scanned regardless if there's one or many CPUs lowering their priority, because there's no current way to find the CPU with the highest priority task that can schedule to one of these CPUs, there really only needs to be one IPI being sent around at a time. This greatly simplifies the code! The new approach is to have each root domain have its own irq work, as the rto_mask is per root domain. The root domain has the following fields attached to it: rto_push_work - the irq work to process each CPU set in rto_mask rto_lock - the lock to protect some of the other rto fields rto_loop_start - an atomic that keeps contention down on rto_lock the first CPU scheduling in a lower priority task is the one to kick off the process. rto_loop_next - an atomic that gets incremented for each CPU that schedules in a lower priority task. rto_loop - a variable protected by rto_lock that is used to compare against rto_loop_next rto_cpu - The cpu to send the next IPI to, also protected by the rto_lock. When a CPU schedules in a lower priority task and wants to make sure overloaded CPUs know about it. It increments the rto_loop_next. Then it atomically sets rto_loop_start with a cmpxchg. If the old value is not "0", then it is done, as another CPU is kicking off the IPI loop. If the old value is "0", then it will take the rto_lock to synchronize with a possible IPI being sent around to the overloaded CPUs. If rto_cpu is greater than or equal to nr_cpu_ids, then there's either no IPI being sent around, or one is about to finish. Then rto_cpu is set to the first CPU in rto_mask and an IPI is sent to that CPU. If there's no CPUs set in rto_mask, then there's nothing to be done. When the CPU receives the IPI, it will first try to push any RT tasks that is queued on the CPU but can't run because a higher priority RT task is currently running on that CPU. Then it takes the rto_lock and looks for the next CPU in the rto_mask. If it finds one, it simply sends an IPI to that CPU and the process continues. If there's no more CPUs in the rto_mask, then rto_loop is compared with rto_loop_next. If they match, everything is done and the process is over. If they do not match, then a CPU scheduled in a lower priority task as the IPI was being passed around, and the process needs to start again. The first CPU in rto_mask is sent the IPI. This change removes this duplication of work in the IPI logic, and greatly lowers the latency caused by the IPIs. This removed the lockup happening on the 120 CPU machine. It also simplifies the code tremendously. What else could anyone ask for? Thanks to Peter Zijlstra for simplifying the rto_loop_start atomic logic and supplying me with the rto_start_trylock() and rto_start_unlock() helper functions. Signed-off-by: Steven Rostedt (VMware) <rostedt(a)goodmis.org> Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Cc: Clark Williams <williams(a)redhat.com> Cc: Daniel Bristot de Oliveira <bristot(a)redhat.com> Cc: John Kacur <jkacur(a)redhat.com> Cc: Linus Torvalds <torvalds(a)linux-foundation.org> Cc: Mike Galbraith <efault(a)gmx.de> Cc: Peter Zijlstra <peterz(a)infradead.org> Cc: Scott Wood <swood(a)redhat.com> Cc: Thomas Gleixner <tglx(a)linutronix.de> Link: http://lkml.kernel.org/r/20170424114732.1aac6dc4@gandalf.local.home Signed-off-by: Ingo Molnar <mingo(a)kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> --- kernel/sched/core.c | 6 + kernel/sched/rt.c | 235 ++++++++++++++++++++++++--------------------------- kernel/sched/sched.h | 24 +++-- 3 files changed, 138 insertions(+), 127 deletions(-) --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -5907,6 +5907,12 @@ static int init_rootdomain(struct root_d if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) goto free_dlo_mask; +#ifdef HAVE_RT_PUSH_IPI + rd->rto_cpu = -1; + raw_spin_lock_init(&rd->rto_lock); + init_irq_work(&rd->rto_push_work, rto_push_irq_work_func); +#endif + init_dl_bw(&rd->dl_bw); if (cpudl_init(&rd->cpudl) != 0) goto free_dlo_mask; --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -64,10 +64,6 @@ static void start_rt_bandwidth(struct rt raw_spin_unlock(&rt_b->rt_runtime_lock); } -#if defined(CONFIG_SMP) && defined(HAVE_RT_PUSH_IPI) -static void push_irq_work_func(struct irq_work *work); -#endif - void init_rt_rq(struct rt_rq *rt_rq) { struct rt_prio_array *array; @@ -87,13 +83,6 @@ void init_rt_rq(struct rt_rq *rt_rq) rt_rq->rt_nr_migratory = 0; rt_rq->overloaded = 0; plist_head_init(&rt_rq->pushable_tasks); - -#ifdef HAVE_RT_PUSH_IPI - rt_rq->push_flags = 0; - rt_rq->push_cpu = nr_cpu_ids; - raw_spin_lock_init(&rt_rq->push_lock); - init_irq_work(&rt_rq->push_work, push_irq_work_func); -#endif #endif /* CONFIG_SMP */ /* We start is dequeued state, because no RT tasks are queued */ rt_rq->rt_queued = 0; @@ -1802,160 +1791,166 @@ static void push_rt_tasks(struct rq *rq) } #ifdef HAVE_RT_PUSH_IPI + /* - * The search for the next cpu always starts at rq->cpu and ends - * when we reach rq->cpu again. It will never return rq->cpu. - * This returns the next cpu to check, or nr_cpu_ids if the loop - * is complete. + * When a high priority task schedules out from a CPU and a lower priority + * task is scheduled in, a check is made to see if there's any RT tasks + * on other CPUs that are waiting to run because a higher priority RT task + * is currently running on its CPU. In this case, the CPU with multiple RT + * tasks queued on it (overloaded) needs to be notified that a CPU has opened + * up that may be able to run one of its non-running queued RT tasks. + * + * All CPUs with overloaded RT tasks need to be notified as there is currently + * no way to know which of these CPUs have the highest priority task waiting + * to run. Instead of trying to take a spinlock on each of these CPUs, + * which has shown to cause large latency when done on machines with many + * CPUs, sending an IPI to the CPUs to have them push off the overloaded + * RT tasks waiting to run. + * + * Just sending an IPI to each of the CPUs is also an issue, as on large + * count CPU machines, this can cause an IPI storm on a CPU, especially + * if its the only CPU with multiple RT tasks queued, and a large number + * of CPUs scheduling a lower priority task at the same time. + * + * Each root domain has its own irq work function that can iterate over + * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT + * tassk must be checked if there's one or many CPUs that are lowering + * their priority, there's a single irq work iterator that will try to + * push off RT tasks that are waiting to run. + * + * When a CPU schedules a lower priority task, it will kick off the + * irq work iterator that will jump to each CPU with overloaded RT tasks. + * As it only takes the first CPU that schedules a lower priority task + * to start the process, the rto_start variable is incremented and if + * the atomic result is one, then that CPU will try to take the rto_lock. + * This prevents high contention on the lock as the process handles all + * CPUs scheduling lower priority tasks. + * + * All CPUs that are scheduling a lower priority task will increment the + * rt_loop_next variable. This will make sure that the irq work iterator + * checks all RT overloaded CPUs whenever a CPU schedules a new lower + * priority task, even if the iterator is in the middle of a scan. Incrementing + * the rt_loop_next will cause the iterator to perform another scan. * - * rq->rt.push_cpu holds the last cpu returned by this function, - * or if this is the first instance, it must hold rq->cpu. */ static int rto_next_cpu(struct rq *rq) { - int prev_cpu = rq->rt.push_cpu; + struct root_domain *rd = rq->rd; + int next; int cpu; - cpu = cpumask_next(prev_cpu, rq->rd->rto_mask); - /* - * If the previous cpu is less than the rq's CPU, then it already - * passed the end of the mask, and has started from the beginning. - * We end if the next CPU is greater or equal to rq's CPU. + * When starting the IPI RT pushing, the rto_cpu is set to -1, + * rt_next_cpu() will simply return the first CPU found in + * the rto_mask. + * + * If rto_next_cpu() is called with rto_cpu is a valid cpu, it + * will return the next CPU found in the rto_mask. + * + * If there are no more CPUs left in the rto_mask, then a check is made + * against rto_loop and rto_loop_next. rto_loop is only updated with + * the rto_lock held, but any CPU may increment the rto_loop_next + * without any locking. */ - if (prev_cpu < rq->cpu) { - if (cpu >= rq->cpu) - return nr_cpu_ids; + for (;;) { - } else if (cpu >= nr_cpu_ids) { - /* - * We passed the end of the mask, start at the beginning. - * If the result is greater or equal to the rq's CPU, then - * the loop is finished. - */ - cpu = cpumask_first(rq->rd->rto_mask); - if (cpu >= rq->cpu) - return nr_cpu_ids; - } - rq->rt.push_cpu = cpu; + /* When rto_cpu is -1 this acts like cpumask_first() */ + cpu = cpumask_next(rd->rto_cpu, rd->rto_mask); - /* Return cpu to let the caller know if the loop is finished or not */ - return cpu; -} + rd->rto_cpu = cpu; -static int find_next_push_cpu(struct rq *rq) -{ - struct rq *next_rq; - int cpu; + if (cpu < nr_cpu_ids) + return cpu; - while (1) { - cpu = rto_next_cpu(rq); - if (cpu >= nr_cpu_ids) - break; - next_rq = cpu_rq(cpu); + rd->rto_cpu = -1; + + /* + * ACQUIRE ensures we see the @rto_mask changes + * made prior to the @next value observed. + * + * Matches WMB in rt_set_overload(). + */ + next = atomic_read_acquire(&rd->rto_loop_next); - /* Make sure the next rq can push to this rq */ - if (next_rq->rt.highest_prio.next < rq->rt.highest_prio.curr) + if (rd->rto_loop == next) break; + + rd->rto_loop = next; } - return cpu; + return -1; } -#define RT_PUSH_IPI_EXECUTING 1 -#define RT_PUSH_IPI_RESTART 2 +static inline bool rto_start_trylock(atomic_t *v) +{ + return !atomic_cmpxchg_acquire(v, 0, 1); +} -static void tell_cpu_to_push(struct rq *rq) +static inline void rto_start_unlock(atomic_t *v) { - int cpu; + atomic_set_release(v, 0); +} - if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { - raw_spin_lock(&rq->rt.push_lock); - /* Make sure it's still executing */ - if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { - /* - * Tell the IPI to restart the loop as things have - * changed since it started. - */ - rq->rt.push_flags |= RT_PUSH_IPI_RESTART; - raw_spin_unlock(&rq->rt.push_lock); - return; - } - raw_spin_unlock(&rq->rt.push_lock); - } +static void tell_cpu_to_push(struct rq *rq) +{ + int cpu = -1; - /* When here, there's no IPI going around */ + /* Keep the loop going if the IPI is currently active */ + atomic_inc(&rq->rd->rto_loop_next); - rq->rt.push_cpu = rq->cpu; - cpu = find_next_push_cpu(rq); - if (cpu >= nr_cpu_ids) + /* Only one CPU can initiate a loop at a time */ + if (!rto_start_trylock(&rq->rd->rto_loop_start)) return; - rq->rt.push_flags = RT_PUSH_IPI_EXECUTING; + raw_spin_lock(&rq->rd->rto_lock); + + /* + * The rto_cpu is updated under the lock, if it has a valid cpu + * then the IPI is still running and will continue due to the + * update to loop_next, and nothing needs to be done here. + * Otherwise it is finishing up and an ipi needs to be sent. + */ + if (rq->rd->rto_cpu < 0) + cpu = rto_next_cpu(rq); + + raw_spin_unlock(&rq->rd->rto_lock); - irq_work_queue_on(&rq->rt.push_work, cpu); + rto_start_unlock(&rq->rd->rto_loop_start); + + if (cpu >= 0) + irq_work_queue_on(&rq->rd->rto_push_work, cpu); } /* Called from hardirq context */ -static void try_to_push_tasks(void *arg) +void rto_push_irq_work_func(struct irq_work *work) { - struct rt_rq *rt_rq = arg; - struct rq *rq, *src_rq; - int this_cpu; + struct rq *rq; int cpu; - this_cpu = rt_rq->push_cpu; + rq = this_rq(); - /* Paranoid check */ - BUG_ON(this_cpu != smp_processor_id()); - - rq = cpu_rq(this_cpu); - src_rq = rq_of_rt_rq(rt_rq); - -again: + /* + * We do not need to grab the lock to check for has_pushable_tasks. + * When it gets updated, a check is made if a push is possible. + */ if (has_pushable_tasks(rq)) { raw_spin_lock(&rq->lock); - push_rt_task(rq); + push_rt_tasks(rq); raw_spin_unlock(&rq->lock); } - /* Pass the IPI to the next rt overloaded queue */ - raw_spin_lock(&rt_rq->push_lock); - /* - * If the source queue changed since the IPI went out, - * we need to restart the search from that CPU again. - */ - if (rt_rq->push_flags & RT_PUSH_IPI_RESTART) { - rt_rq->push_flags &= ~RT_PUSH_IPI_RESTART; - rt_rq->push_cpu = src_rq->cpu; - } + raw_spin_lock(&rq->rd->rto_lock); - cpu = find_next_push_cpu(src_rq); + /* Pass the IPI to the next rt overloaded queue */ + cpu = rto_next_cpu(rq); - if (cpu >= nr_cpu_ids) - rt_rq->push_flags &= ~RT_PUSH_IPI_EXECUTING; - raw_spin_unlock(&rt_rq->push_lock); + raw_spin_unlock(&rq->rd->rto_lock); - if (cpu >= nr_cpu_ids) + if (cpu < 0) return; - /* - * It is possible that a restart caused this CPU to be - * chosen again. Don't bother with an IPI, just see if we - * have more to push. - */ - if (unlikely(cpu == rq->cpu)) - goto again; - /* Try the next RT overloaded CPU */ - irq_work_queue_on(&rt_rq->push_work, cpu); -} - -static void push_irq_work_func(struct irq_work *work) -{ - struct rt_rq *rt_rq = container_of(work, struct rt_rq, push_work); - - try_to_push_tasks(rt_rq); + irq_work_queue_on(&rq->rd->rto_push_work, cpu); } #endif /* HAVE_RT_PUSH_IPI */ --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -429,7 +429,7 @@ static inline int rt_bandwidth_enabled(v } /* RT IPI pull logic requires IRQ_WORK */ -#ifdef CONFIG_IRQ_WORK +#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) # define HAVE_RT_PUSH_IPI #endif @@ -450,12 +450,6 @@ struct rt_rq { unsigned long rt_nr_total; int overloaded; struct plist_head pushable_tasks; -#ifdef HAVE_RT_PUSH_IPI - int push_flags; - int push_cpu; - struct irq_work push_work; - raw_spinlock_t push_lock; -#endif #endif /* CONFIG_SMP */ int rt_queued; @@ -537,6 +531,19 @@ struct root_domain { struct dl_bw dl_bw; struct cpudl cpudl; +#ifdef HAVE_RT_PUSH_IPI + /* + * For IPI pull requests, loop across the rto_mask. + */ + struct irq_work rto_push_work; + raw_spinlock_t rto_lock; + /* These are only updated and read within rto_lock */ + int rto_loop; + int rto_cpu; + /* These atomics are updated outside of a lock */ + atomic_t rto_loop_next; + atomic_t rto_loop_start; +#endif /* * The "RT overload" flag: it gets set if a CPU has more than * one runnable RT task. @@ -547,6 +554,9 @@ struct root_domain { extern struct root_domain def_root_domain; +#ifdef HAVE_RT_PUSH_IPI +extern void rto_push_irq_work_func(struct irq_work *work); +#endif #endif /* CONFIG_SMP */ /* Patches currently in stable-queue which might be from rostedt(a)goodmis.org are queue-4.4/sched-make-resched_cpu-unconditional.patch queue-4.4/sched-rt-simplify-the-ipi-based-rt-balancing-logic.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] FAILED: patch "[PATCH] sched/rt: Simplify the IPI based RT balancing logic" failed to apply to 4.9-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 4.9-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. thanks, greg k-h ------------------ original commit in Linus's tree ------------------ >From 4bdced5c9a2922521e325896a7bbbf0132c94e56 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" <rostedt(a)goodmis.org> Date: Fri, 6 Oct 2017 14:05:04 -0400 Subject: [PATCH] sched/rt: Simplify the IPI based RT balancing logic When a CPU lowers its priority (schedules out a high priority task for a lower priority one), a check is made to see if any other CPU has overloaded RT tasks (more than one). It checks the rto_mask to determine this and if so it will request to pull one of those tasks to itself if the non running RT task is of higher priority than the new priority of the next task to run on the current CPU. When we deal with large number of CPUs, the original pull logic suffered from large lock contention on a single CPU run queue, which caused a huge latency across all CPUs. This was caused by only having one CPU having overloaded RT tasks and a bunch of other CPUs lowering their priority. To solve this issue, commit: b6366f048e0c ("sched/rt: Use IPI to trigger RT task push migration instead of pulling") changed the way to request a pull. Instead of grabbing the lock of the overloaded CPU's runqueue, it simply sent an IPI to that CPU to do the work. Although the IPI logic worked very well in removing the large latency build up, it still could suffer from a large number of IPIs being sent to a single CPU. On a 80 CPU box, I measured over 200us of processing IPIs. Worse yet, when I tested this on a 120 CPU box, with a stress test that had lots of RT tasks scheduling on all CPUs, it actually triggered the hard lockup detector! One CPU had so many IPIs sent to it, and due to the restart mechanism that is triggered when the source run queue has a priority status change, the CPU spent minutes! processing the IPIs. Thinking about this further, I realized there's no reason for each run queue to send its own IPI. As all CPUs with overloaded tasks must be scanned regardless if there's one or many CPUs lowering their priority, because there's no current way to find the CPU with the highest priority task that can schedule to one of these CPUs, there really only needs to be one IPI being sent around at a time. This greatly simplifies the code! The new approach is to have each root domain have its own irq work, as the rto_mask is per root domain. The root domain has the following fields attached to it: rto_push_work - the irq work to process each CPU set in rto_mask rto_lock - the lock to protect some of the other rto fields rto_loop_start - an atomic that keeps contention down on rto_lock the first CPU scheduling in a lower priority task is the one to kick off the process. rto_loop_next - an atomic that gets incremented for each CPU that schedules in a lower priority task. rto_loop - a variable protected by rto_lock that is used to compare against rto_loop_next rto_cpu - The cpu to send the next IPI to, also protected by the rto_lock. When a CPU schedules in a lower priority task and wants to make sure overloaded CPUs know about it. It increments the rto_loop_next. Then it atomically sets rto_loop_start with a cmpxchg. If the old value is not "0", then it is done, as another CPU is kicking off the IPI loop. If the old value is "0", then it will take the rto_lock to synchronize with a possible IPI being sent around to the overloaded CPUs. If rto_cpu is greater than or equal to nr_cpu_ids, then there's either no IPI being sent around, or one is about to finish. Then rto_cpu is set to the first CPU in rto_mask and an IPI is sent to that CPU. If there's no CPUs set in rto_mask, then there's nothing to be done. When the CPU receives the IPI, it will first try to push any RT tasks that is queued on the CPU but can't run because a higher priority RT task is currently running on that CPU. Then it takes the rto_lock and looks for the next CPU in the rto_mask. If it finds one, it simply sends an IPI to that CPU and the process continues. If there's no more CPUs in the rto_mask, then rto_loop is compared with rto_loop_next. If they match, everything is done and the process is over. If they do not match, then a CPU scheduled in a lower priority task as the IPI was being passed around, and the process needs to start again. The first CPU in rto_mask is sent the IPI. This change removes this duplication of work in the IPI logic, and greatly lowers the latency caused by the IPIs. This removed the lockup happening on the 120 CPU machine. It also simplifies the code tremendously. What else could anyone ask for? Thanks to Peter Zijlstra for simplifying the rto_loop_start atomic logic and supplying me with the rto_start_trylock() and rto_start_unlock() helper functions. Signed-off-by: Steven Rostedt (VMware) <rostedt(a)goodmis.org> Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Cc: Clark Williams <williams(a)redhat.com> Cc: Daniel Bristot de Oliveira <bristot(a)redhat.com> Cc: John Kacur <jkacur(a)redhat.com> Cc: Linus Torvalds <torvalds(a)linux-foundation.org> Cc: Mike Galbraith <efault(a)gmx.de> Cc: Peter Zijlstra <peterz(a)infradead.org> Cc: Scott Wood <swood(a)redhat.com> Cc: Thomas Gleixner <tglx(a)linutronix.de> Link: http://lkml.kernel.org/r/20170424114732.1aac6dc4@gandalf.local.home Signed-off-by: Ingo Molnar <mingo(a)kernel.org> diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c index 0af5ca9e3e3f..fda27991699a 100644 --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -73,10 +73,6 @@ static void start_rt_bandwidth(struct rt_bandwidth *rt_b) raw_spin_unlock(&rt_b->rt_runtime_lock); } -#if defined(CONFIG_SMP) && defined(HAVE_RT_PUSH_IPI) -static void push_irq_work_func(struct irq_work *work); -#endif - void init_rt_rq(struct rt_rq *rt_rq) { struct rt_prio_array *array; @@ -96,13 +92,6 @@ void init_rt_rq(struct rt_rq *rt_rq) rt_rq->rt_nr_migratory = 0; rt_rq->overloaded = 0; plist_head_init(&rt_rq->pushable_tasks); - -#ifdef HAVE_RT_PUSH_IPI - rt_rq->push_flags = 0; - rt_rq->push_cpu = nr_cpu_ids; - raw_spin_lock_init(&rt_rq->push_lock); - init_irq_work(&rt_rq->push_work, push_irq_work_func); -#endif #endif /* CONFIG_SMP */ /* We start is dequeued state, because no RT tasks are queued */ rt_rq->rt_queued = 0; @@ -1875,241 +1864,166 @@ static void push_rt_tasks(struct rq *rq) } #ifdef HAVE_RT_PUSH_IPI + /* - * The search for the next cpu always starts at rq->cpu and ends - * when we reach rq->cpu again. It will never return rq->cpu. - * This returns the next cpu to check, or nr_cpu_ids if the loop - * is complete. + * When a high priority task schedules out from a CPU and a lower priority + * task is scheduled in, a check is made to see if there's any RT tasks + * on other CPUs that are waiting to run because a higher priority RT task + * is currently running on its CPU. In this case, the CPU with multiple RT + * tasks queued on it (overloaded) needs to be notified that a CPU has opened + * up that may be able to run one of its non-running queued RT tasks. + * + * All CPUs with overloaded RT tasks need to be notified as there is currently + * no way to know which of these CPUs have the highest priority task waiting + * to run. Instead of trying to take a spinlock on each of these CPUs, + * which has shown to cause large latency when done on machines with many + * CPUs, sending an IPI to the CPUs to have them push off the overloaded + * RT tasks waiting to run. + * + * Just sending an IPI to each of the CPUs is also an issue, as on large + * count CPU machines, this can cause an IPI storm on a CPU, especially + * if its the only CPU with multiple RT tasks queued, and a large number + * of CPUs scheduling a lower priority task at the same time. + * + * Each root domain has its own irq work function that can iterate over + * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT + * tassk must be checked if there's one or many CPUs that are lowering + * their priority, there's a single irq work iterator that will try to + * push off RT tasks that are waiting to run. + * + * When a CPU schedules a lower priority task, it will kick off the + * irq work iterator that will jump to each CPU with overloaded RT tasks. + * As it only takes the first CPU that schedules a lower priority task + * to start the process, the rto_start variable is incremented and if + * the atomic result is one, then that CPU will try to take the rto_lock. + * This prevents high contention on the lock as the process handles all + * CPUs scheduling lower priority tasks. + * + * All CPUs that are scheduling a lower priority task will increment the + * rt_loop_next variable. This will make sure that the irq work iterator + * checks all RT overloaded CPUs whenever a CPU schedules a new lower + * priority task, even if the iterator is in the middle of a scan. Incrementing + * the rt_loop_next will cause the iterator to perform another scan. * - * rq->rt.push_cpu holds the last cpu returned by this function, - * or if this is the first instance, it must hold rq->cpu. */ static int rto_next_cpu(struct rq *rq) { - int prev_cpu = rq->rt.push_cpu; + struct root_domain *rd = rq->rd; + int next; int cpu; - cpu = cpumask_next(prev_cpu, rq->rd->rto_mask); - /* - * If the previous cpu is less than the rq's CPU, then it already - * passed the end of the mask, and has started from the beginning. - * We end if the next CPU is greater or equal to rq's CPU. + * When starting the IPI RT pushing, the rto_cpu is set to -1, + * rt_next_cpu() will simply return the first CPU found in + * the rto_mask. + * + * If rto_next_cpu() is called with rto_cpu is a valid cpu, it + * will return the next CPU found in the rto_mask. + * + * If there are no more CPUs left in the rto_mask, then a check is made + * against rto_loop and rto_loop_next. rto_loop is only updated with + * the rto_lock held, but any CPU may increment the rto_loop_next + * without any locking. */ - if (prev_cpu < rq->cpu) { - if (cpu >= rq->cpu) - return nr_cpu_ids; + for (;;) { - } else if (cpu >= nr_cpu_ids) { - /* - * We passed the end of the mask, start at the beginning. - * If the result is greater or equal to the rq's CPU, then - * the loop is finished. - */ - cpu = cpumask_first(rq->rd->rto_mask); - if (cpu >= rq->cpu) - return nr_cpu_ids; - } - rq->rt.push_cpu = cpu; + /* When rto_cpu is -1 this acts like cpumask_first() */ + cpu = cpumask_next(rd->rto_cpu, rd->rto_mask); - /* Return cpu to let the caller know if the loop is finished or not */ - return cpu; -} + rd->rto_cpu = cpu; -static int find_next_push_cpu(struct rq *rq) -{ - struct rq *next_rq; - int cpu; + if (cpu < nr_cpu_ids) + return cpu; - while (1) { - cpu = rto_next_cpu(rq); - if (cpu >= nr_cpu_ids) - break; - next_rq = cpu_rq(cpu); + rd->rto_cpu = -1; + + /* + * ACQUIRE ensures we see the @rto_mask changes + * made prior to the @next value observed. + * + * Matches WMB in rt_set_overload(). + */ + next = atomic_read_acquire(&rd->rto_loop_next); - /* Make sure the next rq can push to this rq */ - if (next_rq->rt.highest_prio.next < rq->rt.highest_prio.curr) + if (rd->rto_loop == next) break; + + rd->rto_loop = next; } - return cpu; + return -1; } -#define RT_PUSH_IPI_EXECUTING 1 -#define RT_PUSH_IPI_RESTART 2 +static inline bool rto_start_trylock(atomic_t *v) +{ + return !atomic_cmpxchg_acquire(v, 0, 1); +} -/* - * When a high priority task schedules out from a CPU and a lower priority - * task is scheduled in, a check is made to see if there's any RT tasks - * on other CPUs that are waiting to run because a higher priority RT task - * is currently running on its CPU. In this case, the CPU with multiple RT - * tasks queued on it (overloaded) needs to be notified that a CPU has opened - * up that may be able to run one of its non-running queued RT tasks. - * - * On large CPU boxes, there's the case that several CPUs could schedule - * a lower priority task at the same time, in which case it will look for - * any overloaded CPUs that it could pull a task from. To do this, the runqueue - * lock must be taken from that overloaded CPU. Having 10s of CPUs all fighting - * for a single overloaded CPU's runqueue lock can produce a large latency. - * (This has actually been observed on large boxes running cyclictest). - * Instead of taking the runqueue lock of the overloaded CPU, each of the - * CPUs that scheduled a lower priority task simply sends an IPI to the - * overloaded CPU. An IPI is much cheaper than taking an runqueue lock with - * lots of contention. The overloaded CPU will look to push its non-running - * RT task off, and if it does, it can then ignore the other IPIs coming - * in, and just pass those IPIs off to any other overloaded CPU. - * - * When a CPU schedules a lower priority task, it only sends an IPI to - * the "next" CPU that has overloaded RT tasks. This prevents IPI storms, - * as having 10 CPUs scheduling lower priority tasks and 10 CPUs with - * RT overloaded tasks, would cause 100 IPIs to go out at once. - * - * The overloaded RT CPU, when receiving an IPI, will try to push off its - * overloaded RT tasks and then send an IPI to the next CPU that has - * overloaded RT tasks. This stops when all CPUs with overloaded RT tasks - * have completed. Just because a CPU may have pushed off its own overloaded - * RT task does not mean it should stop sending the IPI around to other - * overloaded CPUs. There may be another RT task waiting to run on one of - * those CPUs that are of higher priority than the one that was just - * pushed. - * - * An optimization that could possibly be made is to make a CPU array similar - * to the cpupri array mask of all running RT tasks, but for the overloaded - * case, then the IPI could be sent to only the CPU with the highest priority - * RT task waiting, and that CPU could send off further IPIs to the CPU with - * the next highest waiting task. Since the overloaded case is much less likely - * to happen, the complexity of this implementation may not be worth it. - * Instead, just send an IPI around to all overloaded CPUs. - * - * The rq->rt.push_flags holds the status of the IPI that is going around. - * A run queue can only send out a single IPI at a time. The possible flags - * for rq->rt.push_flags are: - * - * (None or zero): No IPI is going around for the current rq - * RT_PUSH_IPI_EXECUTING: An IPI for the rq is being passed around - * RT_PUSH_IPI_RESTART: The priority of the running task for the rq - * has changed, and the IPI should restart - * circulating the overloaded CPUs again. - * - * rq->rt.push_cpu contains the CPU that is being sent the IPI. It is updated - * before sending to the next CPU. - * - * Instead of having all CPUs that schedule a lower priority task send - * an IPI to the same "first" CPU in the RT overload mask, they send it - * to the next overloaded CPU after their own CPU. This helps distribute - * the work when there's more than one overloaded CPU and multiple CPUs - * scheduling in lower priority tasks. - * - * When a rq schedules a lower priority task than what was currently - * running, the next CPU with overloaded RT tasks is examined first. - * That is, if CPU 1 and 5 are overloaded, and CPU 3 schedules a lower - * priority task, it will send an IPI first to CPU 5, then CPU 5 will - * send to CPU 1 if it is still overloaded. CPU 1 will clear the - * rq->rt.push_flags if RT_PUSH_IPI_RESTART is not set. - * - * The first CPU to notice IPI_RESTART is set, will clear that flag and then - * send an IPI to the next overloaded CPU after the rq->cpu and not the next - * CPU after push_cpu. That is, if CPU 1, 4 and 5 are overloaded when CPU 3 - * schedules a lower priority task, and the IPI_RESTART gets set while the - * handling is being done on CPU 5, it will clear the flag and send it back to - * CPU 4 instead of CPU 1. - * - * Note, the above logic can be disabled by turning off the sched_feature - * RT_PUSH_IPI. Then the rq lock of the overloaded CPU will simply be - * taken by the CPU requesting a pull and the waiting RT task will be pulled - * by that CPU. This may be fine for machines with few CPUs. - */ -static void tell_cpu_to_push(struct rq *rq) +static inline void rto_start_unlock(atomic_t *v) { - int cpu; + atomic_set_release(v, 0); +} - if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { - raw_spin_lock(&rq->rt.push_lock); - /* Make sure it's still executing */ - if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { - /* - * Tell the IPI to restart the loop as things have - * changed since it started. - */ - rq->rt.push_flags |= RT_PUSH_IPI_RESTART; - raw_spin_unlock(&rq->rt.push_lock); - return; - } - raw_spin_unlock(&rq->rt.push_lock); - } +static void tell_cpu_to_push(struct rq *rq) +{ + int cpu = -1; - /* When here, there's no IPI going around */ + /* Keep the loop going if the IPI is currently active */ + atomic_inc(&rq->rd->rto_loop_next); - rq->rt.push_cpu = rq->cpu; - cpu = find_next_push_cpu(rq); - if (cpu >= nr_cpu_ids) + /* Only one CPU can initiate a loop at a time */ + if (!rto_start_trylock(&rq->rd->rto_loop_start)) return; - rq->rt.push_flags = RT_PUSH_IPI_EXECUTING; + raw_spin_lock(&rq->rd->rto_lock); + + /* + * The rto_cpu is updated under the lock, if it has a valid cpu + * then the IPI is still running and will continue due to the + * update to loop_next, and nothing needs to be done here. + * Otherwise it is finishing up and an ipi needs to be sent. + */ + if (rq->rd->rto_cpu < 0) + cpu = rto_next_cpu(rq); - irq_work_queue_on(&rq->rt.push_work, cpu); + raw_spin_unlock(&rq->rd->rto_lock); + + rto_start_unlock(&rq->rd->rto_loop_start); + + if (cpu >= 0) + irq_work_queue_on(&rq->rd->rto_push_work, cpu); } /* Called from hardirq context */ -static void try_to_push_tasks(void *arg) +void rto_push_irq_work_func(struct irq_work *work) { - struct rt_rq *rt_rq = arg; - struct rq *rq, *src_rq; - int this_cpu; + struct rq *rq; int cpu; - this_cpu = rt_rq->push_cpu; + rq = this_rq(); - /* Paranoid check */ - BUG_ON(this_cpu != smp_processor_id()); - - rq = cpu_rq(this_cpu); - src_rq = rq_of_rt_rq(rt_rq); - -again: + /* + * We do not need to grab the lock to check for has_pushable_tasks. + * When it gets updated, a check is made if a push is possible. + */ if (has_pushable_tasks(rq)) { raw_spin_lock(&rq->lock); - push_rt_task(rq); + push_rt_tasks(rq); raw_spin_unlock(&rq->lock); } - /* Pass the IPI to the next rt overloaded queue */ - raw_spin_lock(&rt_rq->push_lock); - /* - * If the source queue changed since the IPI went out, - * we need to restart the search from that CPU again. - */ - if (rt_rq->push_flags & RT_PUSH_IPI_RESTART) { - rt_rq->push_flags &= ~RT_PUSH_IPI_RESTART; - rt_rq->push_cpu = src_rq->cpu; - } + raw_spin_lock(&rq->rd->rto_lock); - cpu = find_next_push_cpu(src_rq); + /* Pass the IPI to the next rt overloaded queue */ + cpu = rto_next_cpu(rq); - if (cpu >= nr_cpu_ids) - rt_rq->push_flags &= ~RT_PUSH_IPI_EXECUTING; - raw_spin_unlock(&rt_rq->push_lock); + raw_spin_unlock(&rq->rd->rto_lock); - if (cpu >= nr_cpu_ids) + if (cpu < 0) return; - /* - * It is possible that a restart caused this CPU to be - * chosen again. Don't bother with an IPI, just see if we - * have more to push. - */ - if (unlikely(cpu == rq->cpu)) - goto again; - /* Try the next RT overloaded CPU */ - irq_work_queue_on(&rt_rq->push_work, cpu); -} - -static void push_irq_work_func(struct irq_work *work) -{ - struct rt_rq *rt_rq = container_of(work, struct rt_rq, push_work); - - try_to_push_tasks(rt_rq); + irq_work_queue_on(&rq->rd->rto_push_work, cpu); } #endif /* HAVE_RT_PUSH_IPI */ diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index a81c9782e98c..8aa24b41f652 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -505,7 +505,7 @@ static inline int rt_bandwidth_enabled(void) } /* RT IPI pull logic requires IRQ_WORK */ -#ifdef CONFIG_IRQ_WORK +#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) # define HAVE_RT_PUSH_IPI #endif @@ -527,12 +527,6 @@ struct rt_rq { unsigned long rt_nr_total; int overloaded; struct plist_head pushable_tasks; -#ifdef HAVE_RT_PUSH_IPI - int push_flags; - int push_cpu; - struct irq_work push_work; - raw_spinlock_t push_lock; -#endif #endif /* CONFIG_SMP */ int rt_queued; @@ -641,6 +635,19 @@ struct root_domain { struct dl_bw dl_bw; struct cpudl cpudl; +#ifdef HAVE_RT_PUSH_IPI + /* + * For IPI pull requests, loop across the rto_mask. + */ + struct irq_work rto_push_work; + raw_spinlock_t rto_lock; + /* These are only updated and read within rto_lock */ + int rto_loop; + int rto_cpu; + /* These atomics are updated outside of a lock */ + atomic_t rto_loop_next; + atomic_t rto_loop_start; +#endif /* * The "RT overload" flag: it gets set if a CPU has more than * one runnable RT task. @@ -658,6 +665,9 @@ extern void init_defrootdomain(void); extern int sched_init_domains(const struct cpumask *cpu_map); extern void rq_attach_root(struct rq *rq, struct root_domain *rd); +#ifdef HAVE_RT_PUSH_IPI +extern void rto_push_irq_work_func(struct irq_work *work); +#endif #endif /* CONFIG_SMP */ /* diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index f51d123f9fe1..e50450c2fed8 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -268,6 +268,12 @@ static int init_rootdomain(struct root_domain *rd) if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) goto free_dlo_mask; +#ifdef HAVE_RT_PUSH_IPI + rd->rto_cpu = -1; + raw_spin_lock_init(&rd->rto_lock); + init_irq_work(&rd->rto_push_work, rto_push_irq_work_func); +#endif + init_dl_bw(&rd->dl_bw); if (cpudl_init(&rd->cpudl) != 0) goto free_rto_mask;

7 years, 9 months

3
3
0 0

[Linux-stable-mirror] [PATCH v2] drm/i915/gvt: Correct ADDR_4K/2M/1G_MASK definition

by Xiong Zhang

For ADDR_4K_MASK, bit[45..12] should be 1, all other bits should be 0. The current definition wrongly set bit[46] as 1 also. This path fixes this. v2: Add commit message, fixes and cc stable.(Zhenyu) Fixes: 2707e4446688("drm/i915/gvt: vGPU graphics memory virtualization") Signed-off-by: Xiong Zhang <xiong.y.zhang(a)intel.com> Cc: stable(a)vger.kernel.org --- drivers/gpu/drm/i915/gvt/gtt.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/i915/gvt/gtt.c b/drivers/gpu/drm/i915/gvt/gtt.c index 3d6008b..71a0f2b 100644 --- a/drivers/gpu/drm/i915/gvt/gtt.c +++ b/drivers/gpu/drm/i915/gvt/gtt.c @@ -331,9 +331,9 @@ static inline int gtt_set_entry64(void *pt, #define GTT_HAW 46 -#define ADDR_1G_MASK (((1UL << (GTT_HAW - 30 + 1)) - 1) << 30) -#define ADDR_2M_MASK (((1UL << (GTT_HAW - 21 + 1)) - 1) << 21) -#define ADDR_4K_MASK (((1UL << (GTT_HAW - 12 + 1)) - 1) << 12) +#define ADDR_1G_MASK (((1UL << (GTT_HAW - 30)) - 1) << 30) +#define ADDR_2M_MASK (((1UL << (GTT_HAW - 21)) - 1) << 21) +#define ADDR_4K_MASK (((1UL << (GTT_HAW - 12)) - 1) << 12) static unsigned long gen8_gtt_get_pfn(struct intel_gvt_gtt_entry *e) { -- 2.7.4

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] + mmmadvise-bugfix-of-madvise-systemcall-infinite-loop-under-special-circumstances.patch added to -mm tree

by akpm＠linux-foundation.org

The patch titled Subject: mm/madvise.c: fix madvise() infinite loop under special circumstances has been added to the -mm tree. Its filename is mmmadvise-bugfix-of-madvise-systemcall-infinite-loop-under-special-circumstances.patch This patch should soon appear at http://ozlabs.org/~akpm/mmots/broken-out/mmmadvise-bugfix-of-madvise-system… and later at http://ozlabs.org/~akpm/mmotm/broken-out/mmmadvise-bugfix-of-madvise-system… Before you just go and hit "reply", please: a) Consider who else should be cc'ed b) Prefer to cc a suitable mailing list as well c) Ideally: find the original patch on the mailing list and do a reply-to-all to that, adding suitable additional cc's *** Remember to use Documentation/SubmitChecklist when testing your code *** The -mm tree is included into linux-next and is updated there every 3-4 working days ------------------------------------------------------ From: chenjie <chenjie6(a)huawei.com> Subject: mm/madvise.c: fix madvise() infinite loop under special circumstances MADVISE_WILLNEED has always been a noop for DAX (formerly XIP) mappings. Unfortunatelly madvise_willneed() doesn't communicate this information properly to the generic madvise syscall implementation. The calling convention is quite subtle there. madvise_vma() is supposed to either return an error or update &prev otherwise the main loop will never advance to the next vma and it will keep looping for ever without a way to get out of the kernel. It seems this has been broken since introduction. Nobody has noticed because nobody seems to be using MADVISE_WILLNEED on these DAX mappings. [mhocko(a)suse.com: rewrite changelog] Link: http://lkml.kernel.org/r/20171127115318.911-1-guoxuenan@huawei.com Fixes: fe77ba6f4f97 ("[PATCH] xip: madvice/fadvice: execute in place") Signed-off-by: chenjie <chenjie6(a)huawei.com> Signed-off-by: guoxuenan <guoxuenan(a)huawei.com> Acked-by: Michal Hocko <mhocko(a)suse.com> Cc: Minchan Kim <minchan(a)kernel.org> Cc: zhangyi (F) <yi.zhang(a)huawei.com> Cc: Miao Xie <miaoxie(a)huawei.com> Cc: Mike Rapoport <rppt(a)linux.vnet.ibm.com> Cc: Shaohua Li <shli(a)fb.com> Cc: Andrea Arcangeli <aarcange(a)redhat.com> Cc: Mel Gorman <mgorman(a)techsingularity.net> Cc: Kirill A. Shutemov <kirill.shutemov(a)linux.intel.com> Cc: David Rientjes <rientjes(a)google.com> Cc: Anshuman Khandual <khandual(a)linux.vnet.ibm.com> Cc: Rik van Riel <riel(a)redhat.com> Cc: Carsten Otte <cotte(a)de.ibm.com> Cc: Dan Williams <dan.j.williams(a)intel.com> Cc: <stable(a)vger.kernel.org> Signed-off-by: Andrew Morton <akpm(a)linux-foundation.org> --- mm/madvise.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff -puN mm/madvise.c~mmmadvise-bugfix-of-madvise-systemcall-infinite-loop-under-special-circumstances mm/madvise.c --- a/mm/madvise.c~mmmadvise-bugfix-of-madvise-systemcall-infinite-loop-under-special-circumstances +++ a/mm/madvise.c @@ -276,15 +276,14 @@ static long madvise_willneed(struct vm_a { struct file *file = vma->vm_file; + *prev = vma; #ifdef CONFIG_SWAP if (!file) { - *prev = vma; force_swapin_readahead(vma, start, end); return 0; } if (shmem_mapping(file->f_mapping)) { - *prev = vma; force_shm_swapin_readahead(vma, start, end, file->f_mapping); return 0; @@ -299,7 +298,6 @@ static long madvise_willneed(struct vm_a return 0; } - *prev = vma; start = ((start - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; if (end > vma->vm_end) end = vma->vm_end; _ Patches currently in -mm which might be from chenjie6(a)huawei.com are mmmadvise-bugfix-of-madvise-systemcall-infinite-loop-under-special-circumstances.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] + exec-avoid-rlimit_stack-races-with-prlimit.patch added to -mm tree

by akpm＠linux-foundation.org

The patch titled Subject: exec: avoid RLIMIT_STACK races with prlimit() has been added to the -mm tree. Its filename is exec-avoid-rlimit_stack-races-with-prlimit.patch This patch should soon appear at http://ozlabs.org/~akpm/mmots/broken-out/exec-avoid-rlimit_stack-races-with… and later at http://ozlabs.org/~akpm/mmotm/broken-out/exec-avoid-rlimit_stack-races-with… Before you just go and hit "reply", please: a) Consider who else should be cc'ed b) Prefer to cc a suitable mailing list as well c) Ideally: find the original patch on the mailing list and do a reply-to-all to that, adding suitable additional cc's *** Remember to use Documentation/SubmitChecklist when testing your code *** The -mm tree is included into linux-next and is updated there every 3-4 working days ------------------------------------------------------ From: Kees Cook <keescook(a)chromium.org> Subject: exec: avoid RLIMIT_STACK races with prlimit() While the defense-in-depth RLIMIT_STACK limit on setuid processes was protected against races from other threads calling setrlimit(), I missed protecting it against races from external processes calling prlimit(). This adds locking around the change and makes sure that rlim_max is set too. Link: http://lkml.kernel.org/r/20171127193457.GA11348@beast Fixes: 64701dee4178e ("exec: Use sane stack rlimit under secureexec") Signed-off-by: Kees Cook <keescook(a)chromium.org> Reported-by: Ben Hutchings <ben.hutchings(a)codethink.co.uk> Reported-by: Brad Spengler <spender(a)grsecurity.net> Cc: James Morris <james.l.morris(a)oracle.com> Cc: Serge Hallyn <serge(a)hallyn.com> Cc: Andy Lutomirski <luto(a)kernel.org> Cc: Oleg Nesterov <oleg(a)redhat.com> Cc: Jiri Slaby <jslaby(a)suse.cz> Cc: <stable(a)vger.kernel.org> Signed-off-by: Andrew Morton <akpm(a)linux-foundation.org> --- fs/exec.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff -puN fs/exec.c~exec-avoid-rlimit_stack-races-with-prlimit fs/exec.c --- a/fs/exec.c~exec-avoid-rlimit_stack-races-with-prlimit +++ a/fs/exec.c @@ -1340,10 +1340,15 @@ void setup_new_exec(struct linux_binprm * avoid bad behavior from the prior rlimits. This has to * happen before arch_pick_mmap_layout(), which examines * RLIMIT_STACK, but after the point of no return to avoid - * needing to clean up the change on failure. + * races from other threads changing the limits. This also + * must be protected from races with prlimit() calls. */ + task_lock(current->group_leader); if (current->signal->rlim[RLIMIT_STACK].rlim_cur > _STK_LIM) current->signal->rlim[RLIMIT_STACK].rlim_cur = _STK_LIM; + if (current->signal->rlim[RLIMIT_STACK].rlim_max > _STK_LIM) + current->signal->rlim[RLIMIT_STACK].rlim_max = _STK_LIM; + task_unlock(current->group_leader); } arch_pick_mmap_layout(current->mm); _ Patches currently in -mm which might be from keescook(a)chromium.org are exec-avoid-rlimit_stack-races-with-prlimit.patch makefile-move-stack-protector-compiler-breakage-test-earlier.patch makefile-move-stack-protector-availability-out-of-kconfig.patch makefile-introduce-config_cc_stackprotector_auto.patch

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] [PATCH v2] drm/i915/gvt: Correct ADDR_4K/2M/1G_MASK definition

by Xiong Zhang

For ADDR_4K_MASK, bit[45..12] should be 1, all other bits should be 0. The current definition wrongly set bit[46] as 1 also. This path fixes this. v2: Add commit message, fixes and cc stable.(Zhenyu) Fixes: 2707e4446688("drm/i915/gvt: vGPU graphics memory virtualization") Signed-off-by: Xiong Zhang <xiong.y.zhang(a)intel.com> --- drivers/gpu/drm/i915/gvt/gtt.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/i915/gvt/gtt.c b/drivers/gpu/drm/i915/gvt/gtt.c index 3d6008b..71a0f2b 100644 --- a/drivers/gpu/drm/i915/gvt/gtt.c +++ b/drivers/gpu/drm/i915/gvt/gtt.c @@ -331,9 +331,9 @@ static inline int gtt_set_entry64(void *pt, #define GTT_HAW 46 -#define ADDR_1G_MASK (((1UL << (GTT_HAW - 30 + 1)) - 1) << 30) -#define ADDR_2M_MASK (((1UL << (GTT_HAW - 21 + 1)) - 1) << 21) -#define ADDR_4K_MASK (((1UL << (GTT_HAW - 12 + 1)) - 1) << 12) +#define ADDR_1G_MASK (((1UL << (GTT_HAW - 30)) - 1) << 30) +#define ADDR_2M_MASK (((1UL << (GTT_HAW - 21)) - 1) << 21) +#define ADDR_4K_MASK (((1UL << (GTT_HAW - 12)) - 1) << 12) static unsigned long gen8_gtt_get_pfn(struct intel_gvt_gtt_entry *e) { -- 2.7.4

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] [PATCH 4.1] fscrypt: lock mutex before checking for bounce page pool

by Eric Biggers

commit a0b3bc855374c50b5ea85273553485af48caf2f7 upstream. [Please apply to 4.1-stable.] fscrypt_initialize(), which allocates the global bounce page pool when an encrypted file is first accessed, uses "double-checked locking" to try to avoid locking fscrypt_init_mutex. However, it doesn't use any memory barriers, so it's theoretically possible for a thread to observe a bounce page pool which has not been fully initialized. This is a classic bug with "double-checked locking". While "only a theoretical issue" in the latest kernel, in pre-4.8 kernels the pointer that was checked was not even the last to be initialized, so it was easily possible for a crash (NULL pointer dereference) to happen. This was changed only incidentally by the large refactor to use fs/crypto/. Solve both problems in a trivial way that can easily be backported: just always take the mutex. It's theoretically less efficient, but it shouldn't be noticeable in practice as the mutex is only acquired very briefly once per encrypted file. Later I'd like to make this use a helper macro like DO_ONCE(). However, DO_ONCE() runs in atomic context, so we'd need to add a new macro that allows blocking. Cc: stable(a)vger.kernel.org # v4.1+ Signed-off-by: Eric Biggers <ebiggers(a)google.com> Signed-off-by: Theodore Ts'o <tytso(a)mit.edu> --- fs/ext4/crypto.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/fs/ext4/crypto.c b/fs/ext4/crypto.c index 8ff15273ab0c..fdda7a5a1180 100644 --- a/fs/ext4/crypto.c +++ b/fs/ext4/crypto.c @@ -120,8 +120,7 @@ struct ext4_crypto_ctx *ext4_get_crypto_ctx(struct inode *inode) unsigned long flags; struct ext4_encryption_key *key = &EXT4_I(inode)->i_encryption_key; - if (!ext4_read_workqueue) - ext4_init_crypto(); + ext4_init_crypto(); /* * We first try getting the ctx from a free list because in -- 2.15.0.417.g466bffb3ac-goog

7 years, 9 months

1
0
0 0

[Linux-stable-mirror] [PATCH] eeprom: at24: fix the magic value for at24mac402

by Bartosz Golaszewski

There's an ilog2() expansion in AT24_DEVICE_MAGIC() which rounds down the actual size of EUI-48 byte array in at24mac402 eeproms to 4 from 6, making it impossible to read it all. Fix it by creating the magic value manually with the correct size. This patch contains a temporary fix that is suitable for stable branches. Eventually we'll probably remove the call to ilog2() while converting the magic values to actual structs. This fix will cause a warning message to be emitted by probe() for at24mac402 chips but theys should be initialized correctly anyway. Cc: stable(a)vger.kernel.org Fixes: 0b813658c115 ("eeprom: at24: add support for at24mac series") Signed-off-by: Bartosz Golaszewski <brgl(a)bgdev.pl> --- drivers/misc/eeprom/at24.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-) diff --git a/drivers/misc/eeprom/at24.c b/drivers/misc/eeprom/at24.c index e0b4b36ef010..72adf34627f5 100644 --- a/drivers/misc/eeprom/at24.c +++ b/drivers/misc/eeprom/at24.c @@ -141,8 +141,17 @@ static const struct i2c_device_id at24_ids[] = { { "24c02", AT24_DEVICE_MAGIC(2048 / 8, 0) }, { "24cs02", AT24_DEVICE_MAGIC(16, AT24_FLAG_SERIAL | AT24_FLAG_READONLY) }, - { "24mac402", AT24_DEVICE_MAGIC(48 / 8, - AT24_FLAG_MAC | AT24_FLAG_READONLY) }, + /* + * REVISIT: the size of the EUI-48 byte array is 6 in at24mac402, while + * the call to ilog2() in AT24_DEVICE_MAGIC() rounds it down to 4. We + * might want to eventually remove it altogether or maybe even convert + * the magic values to real structs. + * + * For now just create the magic value manually with the right size. + */ + { "24mac402", ((1 << AT24_SIZE_FLAGS | + (AT24_FLAG_MAC | AT24_FLAG_READONLY)) + << AT24_SIZE_BYTELEN | (48 / 8)) }, { "24mac602", AT24_DEVICE_MAGIC(64 / 8, AT24_FLAG_MAC | AT24_FLAG_READONLY) }, /* spd is a 24c02 in memory DIMMs */ -- 2.15.0

7 years, 9 months

2
2
0 0

[Linux-stable-mirror] FAILED: patch "[PATCH] sched/rt: Simplify the IPI based RT balancing logic" failed to apply to 3.18-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 3.18-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. thanks, greg k-h ------------------ original commit in Linus's tree ------------------ >From 4bdced5c9a2922521e325896a7bbbf0132c94e56 Mon Sep 17 00:00:00 2001 From: "Steven Rostedt (Red Hat)" <rostedt(a)goodmis.org> Date: Fri, 6 Oct 2017 14:05:04 -0400 Subject: [PATCH] sched/rt: Simplify the IPI based RT balancing logic When a CPU lowers its priority (schedules out a high priority task for a lower priority one), a check is made to see if any other CPU has overloaded RT tasks (more than one). It checks the rto_mask to determine this and if so it will request to pull one of those tasks to itself if the non running RT task is of higher priority than the new priority of the next task to run on the current CPU. When we deal with large number of CPUs, the original pull logic suffered from large lock contention on a single CPU run queue, which caused a huge latency across all CPUs. This was caused by only having one CPU having overloaded RT tasks and a bunch of other CPUs lowering their priority. To solve this issue, commit: b6366f048e0c ("sched/rt: Use IPI to trigger RT task push migration instead of pulling") changed the way to request a pull. Instead of grabbing the lock of the overloaded CPU's runqueue, it simply sent an IPI to that CPU to do the work. Although the IPI logic worked very well in removing the large latency build up, it still could suffer from a large number of IPIs being sent to a single CPU. On a 80 CPU box, I measured over 200us of processing IPIs. Worse yet, when I tested this on a 120 CPU box, with a stress test that had lots of RT tasks scheduling on all CPUs, it actually triggered the hard lockup detector! One CPU had so many IPIs sent to it, and due to the restart mechanism that is triggered when the source run queue has a priority status change, the CPU spent minutes! processing the IPIs. Thinking about this further, I realized there's no reason for each run queue to send its own IPI. As all CPUs with overloaded tasks must be scanned regardless if there's one or many CPUs lowering their priority, because there's no current way to find the CPU with the highest priority task that can schedule to one of these CPUs, there really only needs to be one IPI being sent around at a time. This greatly simplifies the code! The new approach is to have each root domain have its own irq work, as the rto_mask is per root domain. The root domain has the following fields attached to it: rto_push_work - the irq work to process each CPU set in rto_mask rto_lock - the lock to protect some of the other rto fields rto_loop_start - an atomic that keeps contention down on rto_lock the first CPU scheduling in a lower priority task is the one to kick off the process. rto_loop_next - an atomic that gets incremented for each CPU that schedules in a lower priority task. rto_loop - a variable protected by rto_lock that is used to compare against rto_loop_next rto_cpu - The cpu to send the next IPI to, also protected by the rto_lock. When a CPU schedules in a lower priority task and wants to make sure overloaded CPUs know about it. It increments the rto_loop_next. Then it atomically sets rto_loop_start with a cmpxchg. If the old value is not "0", then it is done, as another CPU is kicking off the IPI loop. If the old value is "0", then it will take the rto_lock to synchronize with a possible IPI being sent around to the overloaded CPUs. If rto_cpu is greater than or equal to nr_cpu_ids, then there's either no IPI being sent around, or one is about to finish. Then rto_cpu is set to the first CPU in rto_mask and an IPI is sent to that CPU. If there's no CPUs set in rto_mask, then there's nothing to be done. When the CPU receives the IPI, it will first try to push any RT tasks that is queued on the CPU but can't run because a higher priority RT task is currently running on that CPU. Then it takes the rto_lock and looks for the next CPU in the rto_mask. If it finds one, it simply sends an IPI to that CPU and the process continues. If there's no more CPUs in the rto_mask, then rto_loop is compared with rto_loop_next. If they match, everything is done and the process is over. If they do not match, then a CPU scheduled in a lower priority task as the IPI was being passed around, and the process needs to start again. The first CPU in rto_mask is sent the IPI. This change removes this duplication of work in the IPI logic, and greatly lowers the latency caused by the IPIs. This removed the lockup happening on the 120 CPU machine. It also simplifies the code tremendously. What else could anyone ask for? Thanks to Peter Zijlstra for simplifying the rto_loop_start atomic logic and supplying me with the rto_start_trylock() and rto_start_unlock() helper functions. Signed-off-by: Steven Rostedt (VMware) <rostedt(a)goodmis.org> Signed-off-by: Peter Zijlstra (Intel) <peterz(a)infradead.org> Cc: Clark Williams <williams(a)redhat.com> Cc: Daniel Bristot de Oliveira <bristot(a)redhat.com> Cc: John Kacur <jkacur(a)redhat.com> Cc: Linus Torvalds <torvalds(a)linux-foundation.org> Cc: Mike Galbraith <efault(a)gmx.de> Cc: Peter Zijlstra <peterz(a)infradead.org> Cc: Scott Wood <swood(a)redhat.com> Cc: Thomas Gleixner <tglx(a)linutronix.de> Link: http://lkml.kernel.org/r/20170424114732.1aac6dc4@gandalf.local.home Signed-off-by: Ingo Molnar <mingo(a)kernel.org> diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c index 0af5ca9e3e3f..fda27991699a 100644 --- a/kernel/sched/rt.c +++ b/kernel/sched/rt.c @@ -73,10 +73,6 @@ static void start_rt_bandwidth(struct rt_bandwidth *rt_b) raw_spin_unlock(&rt_b->rt_runtime_lock); } -#if defined(CONFIG_SMP) && defined(HAVE_RT_PUSH_IPI) -static void push_irq_work_func(struct irq_work *work); -#endif - void init_rt_rq(struct rt_rq *rt_rq) { struct rt_prio_array *array; @@ -96,13 +92,6 @@ void init_rt_rq(struct rt_rq *rt_rq) rt_rq->rt_nr_migratory = 0; rt_rq->overloaded = 0; plist_head_init(&rt_rq->pushable_tasks); - -#ifdef HAVE_RT_PUSH_IPI - rt_rq->push_flags = 0; - rt_rq->push_cpu = nr_cpu_ids; - raw_spin_lock_init(&rt_rq->push_lock); - init_irq_work(&rt_rq->push_work, push_irq_work_func); -#endif #endif /* CONFIG_SMP */ /* We start is dequeued state, because no RT tasks are queued */ rt_rq->rt_queued = 0; @@ -1875,241 +1864,166 @@ static void push_rt_tasks(struct rq *rq) } #ifdef HAVE_RT_PUSH_IPI + /* - * The search for the next cpu always starts at rq->cpu and ends - * when we reach rq->cpu again. It will never return rq->cpu. - * This returns the next cpu to check, or nr_cpu_ids if the loop - * is complete. + * When a high priority task schedules out from a CPU and a lower priority + * task is scheduled in, a check is made to see if there's any RT tasks + * on other CPUs that are waiting to run because a higher priority RT task + * is currently running on its CPU. In this case, the CPU with multiple RT + * tasks queued on it (overloaded) needs to be notified that a CPU has opened + * up that may be able to run one of its non-running queued RT tasks. + * + * All CPUs with overloaded RT tasks need to be notified as there is currently + * no way to know which of these CPUs have the highest priority task waiting + * to run. Instead of trying to take a spinlock on each of these CPUs, + * which has shown to cause large latency when done on machines with many + * CPUs, sending an IPI to the CPUs to have them push off the overloaded + * RT tasks waiting to run. + * + * Just sending an IPI to each of the CPUs is also an issue, as on large + * count CPU machines, this can cause an IPI storm on a CPU, especially + * if its the only CPU with multiple RT tasks queued, and a large number + * of CPUs scheduling a lower priority task at the same time. + * + * Each root domain has its own irq work function that can iterate over + * all CPUs with RT overloaded tasks. Since all CPUs with overloaded RT + * tassk must be checked if there's one or many CPUs that are lowering + * their priority, there's a single irq work iterator that will try to + * push off RT tasks that are waiting to run. + * + * When a CPU schedules a lower priority task, it will kick off the + * irq work iterator that will jump to each CPU with overloaded RT tasks. + * As it only takes the first CPU that schedules a lower priority task + * to start the process, the rto_start variable is incremented and if + * the atomic result is one, then that CPU will try to take the rto_lock. + * This prevents high contention on the lock as the process handles all + * CPUs scheduling lower priority tasks. + * + * All CPUs that are scheduling a lower priority task will increment the + * rt_loop_next variable. This will make sure that the irq work iterator + * checks all RT overloaded CPUs whenever a CPU schedules a new lower + * priority task, even if the iterator is in the middle of a scan. Incrementing + * the rt_loop_next will cause the iterator to perform another scan. * - * rq->rt.push_cpu holds the last cpu returned by this function, - * or if this is the first instance, it must hold rq->cpu. */ static int rto_next_cpu(struct rq *rq) { - int prev_cpu = rq->rt.push_cpu; + struct root_domain *rd = rq->rd; + int next; int cpu; - cpu = cpumask_next(prev_cpu, rq->rd->rto_mask); - /* - * If the previous cpu is less than the rq's CPU, then it already - * passed the end of the mask, and has started from the beginning. - * We end if the next CPU is greater or equal to rq's CPU. + * When starting the IPI RT pushing, the rto_cpu is set to -1, + * rt_next_cpu() will simply return the first CPU found in + * the rto_mask. + * + * If rto_next_cpu() is called with rto_cpu is a valid cpu, it + * will return the next CPU found in the rto_mask. + * + * If there are no more CPUs left in the rto_mask, then a check is made + * against rto_loop and rto_loop_next. rto_loop is only updated with + * the rto_lock held, but any CPU may increment the rto_loop_next + * without any locking. */ - if (prev_cpu < rq->cpu) { - if (cpu >= rq->cpu) - return nr_cpu_ids; + for (;;) { - } else if (cpu >= nr_cpu_ids) { - /* - * We passed the end of the mask, start at the beginning. - * If the result is greater or equal to the rq's CPU, then - * the loop is finished. - */ - cpu = cpumask_first(rq->rd->rto_mask); - if (cpu >= rq->cpu) - return nr_cpu_ids; - } - rq->rt.push_cpu = cpu; + /* When rto_cpu is -1 this acts like cpumask_first() */ + cpu = cpumask_next(rd->rto_cpu, rd->rto_mask); - /* Return cpu to let the caller know if the loop is finished or not */ - return cpu; -} + rd->rto_cpu = cpu; -static int find_next_push_cpu(struct rq *rq) -{ - struct rq *next_rq; - int cpu; + if (cpu < nr_cpu_ids) + return cpu; - while (1) { - cpu = rto_next_cpu(rq); - if (cpu >= nr_cpu_ids) - break; - next_rq = cpu_rq(cpu); + rd->rto_cpu = -1; + + /* + * ACQUIRE ensures we see the @rto_mask changes + * made prior to the @next value observed. + * + * Matches WMB in rt_set_overload(). + */ + next = atomic_read_acquire(&rd->rto_loop_next); - /* Make sure the next rq can push to this rq */ - if (next_rq->rt.highest_prio.next < rq->rt.highest_prio.curr) + if (rd->rto_loop == next) break; + + rd->rto_loop = next; } - return cpu; + return -1; } -#define RT_PUSH_IPI_EXECUTING 1 -#define RT_PUSH_IPI_RESTART 2 +static inline bool rto_start_trylock(atomic_t *v) +{ + return !atomic_cmpxchg_acquire(v, 0, 1); +} -/* - * When a high priority task schedules out from a CPU and a lower priority - * task is scheduled in, a check is made to see if there's any RT tasks - * on other CPUs that are waiting to run because a higher priority RT task - * is currently running on its CPU. In this case, the CPU with multiple RT - * tasks queued on it (overloaded) needs to be notified that a CPU has opened - * up that may be able to run one of its non-running queued RT tasks. - * - * On large CPU boxes, there's the case that several CPUs could schedule - * a lower priority task at the same time, in which case it will look for - * any overloaded CPUs that it could pull a task from. To do this, the runqueue - * lock must be taken from that overloaded CPU. Having 10s of CPUs all fighting - * for a single overloaded CPU's runqueue lock can produce a large latency. - * (This has actually been observed on large boxes running cyclictest). - * Instead of taking the runqueue lock of the overloaded CPU, each of the - * CPUs that scheduled a lower priority task simply sends an IPI to the - * overloaded CPU. An IPI is much cheaper than taking an runqueue lock with - * lots of contention. The overloaded CPU will look to push its non-running - * RT task off, and if it does, it can then ignore the other IPIs coming - * in, and just pass those IPIs off to any other overloaded CPU. - * - * When a CPU schedules a lower priority task, it only sends an IPI to - * the "next" CPU that has overloaded RT tasks. This prevents IPI storms, - * as having 10 CPUs scheduling lower priority tasks and 10 CPUs with - * RT overloaded tasks, would cause 100 IPIs to go out at once. - * - * The overloaded RT CPU, when receiving an IPI, will try to push off its - * overloaded RT tasks and then send an IPI to the next CPU that has - * overloaded RT tasks. This stops when all CPUs with overloaded RT tasks - * have completed. Just because a CPU may have pushed off its own overloaded - * RT task does not mean it should stop sending the IPI around to other - * overloaded CPUs. There may be another RT task waiting to run on one of - * those CPUs that are of higher priority than the one that was just - * pushed. - * - * An optimization that could possibly be made is to make a CPU array similar - * to the cpupri array mask of all running RT tasks, but for the overloaded - * case, then the IPI could be sent to only the CPU with the highest priority - * RT task waiting, and that CPU could send off further IPIs to the CPU with - * the next highest waiting task. Since the overloaded case is much less likely - * to happen, the complexity of this implementation may not be worth it. - * Instead, just send an IPI around to all overloaded CPUs. - * - * The rq->rt.push_flags holds the status of the IPI that is going around. - * A run queue can only send out a single IPI at a time. The possible flags - * for rq->rt.push_flags are: - * - * (None or zero): No IPI is going around for the current rq - * RT_PUSH_IPI_EXECUTING: An IPI for the rq is being passed around - * RT_PUSH_IPI_RESTART: The priority of the running task for the rq - * has changed, and the IPI should restart - * circulating the overloaded CPUs again. - * - * rq->rt.push_cpu contains the CPU that is being sent the IPI. It is updated - * before sending to the next CPU. - * - * Instead of having all CPUs that schedule a lower priority task send - * an IPI to the same "first" CPU in the RT overload mask, they send it - * to the next overloaded CPU after their own CPU. This helps distribute - * the work when there's more than one overloaded CPU and multiple CPUs - * scheduling in lower priority tasks. - * - * When a rq schedules a lower priority task than what was currently - * running, the next CPU with overloaded RT tasks is examined first. - * That is, if CPU 1 and 5 are overloaded, and CPU 3 schedules a lower - * priority task, it will send an IPI first to CPU 5, then CPU 5 will - * send to CPU 1 if it is still overloaded. CPU 1 will clear the - * rq->rt.push_flags if RT_PUSH_IPI_RESTART is not set. - * - * The first CPU to notice IPI_RESTART is set, will clear that flag and then - * send an IPI to the next overloaded CPU after the rq->cpu and not the next - * CPU after push_cpu. That is, if CPU 1, 4 and 5 are overloaded when CPU 3 - * schedules a lower priority task, and the IPI_RESTART gets set while the - * handling is being done on CPU 5, it will clear the flag and send it back to - * CPU 4 instead of CPU 1. - * - * Note, the above logic can be disabled by turning off the sched_feature - * RT_PUSH_IPI. Then the rq lock of the overloaded CPU will simply be - * taken by the CPU requesting a pull and the waiting RT task will be pulled - * by that CPU. This may be fine for machines with few CPUs. - */ -static void tell_cpu_to_push(struct rq *rq) +static inline void rto_start_unlock(atomic_t *v) { - int cpu; + atomic_set_release(v, 0); +} - if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { - raw_spin_lock(&rq->rt.push_lock); - /* Make sure it's still executing */ - if (rq->rt.push_flags & RT_PUSH_IPI_EXECUTING) { - /* - * Tell the IPI to restart the loop as things have - * changed since it started. - */ - rq->rt.push_flags |= RT_PUSH_IPI_RESTART; - raw_spin_unlock(&rq->rt.push_lock); - return; - } - raw_spin_unlock(&rq->rt.push_lock); - } +static void tell_cpu_to_push(struct rq *rq) +{ + int cpu = -1; - /* When here, there's no IPI going around */ + /* Keep the loop going if the IPI is currently active */ + atomic_inc(&rq->rd->rto_loop_next); - rq->rt.push_cpu = rq->cpu; - cpu = find_next_push_cpu(rq); - if (cpu >= nr_cpu_ids) + /* Only one CPU can initiate a loop at a time */ + if (!rto_start_trylock(&rq->rd->rto_loop_start)) return; - rq->rt.push_flags = RT_PUSH_IPI_EXECUTING; + raw_spin_lock(&rq->rd->rto_lock); + + /* + * The rto_cpu is updated under the lock, if it has a valid cpu + * then the IPI is still running and will continue due to the + * update to loop_next, and nothing needs to be done here. + * Otherwise it is finishing up and an ipi needs to be sent. + */ + if (rq->rd->rto_cpu < 0) + cpu = rto_next_cpu(rq); - irq_work_queue_on(&rq->rt.push_work, cpu); + raw_spin_unlock(&rq->rd->rto_lock); + + rto_start_unlock(&rq->rd->rto_loop_start); + + if (cpu >= 0) + irq_work_queue_on(&rq->rd->rto_push_work, cpu); } /* Called from hardirq context */ -static void try_to_push_tasks(void *arg) +void rto_push_irq_work_func(struct irq_work *work) { - struct rt_rq *rt_rq = arg; - struct rq *rq, *src_rq; - int this_cpu; + struct rq *rq; int cpu; - this_cpu = rt_rq->push_cpu; + rq = this_rq(); - /* Paranoid check */ - BUG_ON(this_cpu != smp_processor_id()); - - rq = cpu_rq(this_cpu); - src_rq = rq_of_rt_rq(rt_rq); - -again: + /* + * We do not need to grab the lock to check for has_pushable_tasks. + * When it gets updated, a check is made if a push is possible. + */ if (has_pushable_tasks(rq)) { raw_spin_lock(&rq->lock); - push_rt_task(rq); + push_rt_tasks(rq); raw_spin_unlock(&rq->lock); } - /* Pass the IPI to the next rt overloaded queue */ - raw_spin_lock(&rt_rq->push_lock); - /* - * If the source queue changed since the IPI went out, - * we need to restart the search from that CPU again. - */ - if (rt_rq->push_flags & RT_PUSH_IPI_RESTART) { - rt_rq->push_flags &= ~RT_PUSH_IPI_RESTART; - rt_rq->push_cpu = src_rq->cpu; - } + raw_spin_lock(&rq->rd->rto_lock); - cpu = find_next_push_cpu(src_rq); + /* Pass the IPI to the next rt overloaded queue */ + cpu = rto_next_cpu(rq); - if (cpu >= nr_cpu_ids) - rt_rq->push_flags &= ~RT_PUSH_IPI_EXECUTING; - raw_spin_unlock(&rt_rq->push_lock); + raw_spin_unlock(&rq->rd->rto_lock); - if (cpu >= nr_cpu_ids) + if (cpu < 0) return; - /* - * It is possible that a restart caused this CPU to be - * chosen again. Don't bother with an IPI, just see if we - * have more to push. - */ - if (unlikely(cpu == rq->cpu)) - goto again; - /* Try the next RT overloaded CPU */ - irq_work_queue_on(&rt_rq->push_work, cpu); -} - -static void push_irq_work_func(struct irq_work *work) -{ - struct rt_rq *rt_rq = container_of(work, struct rt_rq, push_work); - - try_to_push_tasks(rt_rq); + irq_work_queue_on(&rq->rd->rto_push_work, cpu); } #endif /* HAVE_RT_PUSH_IPI */ diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index a81c9782e98c..8aa24b41f652 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -505,7 +505,7 @@ static inline int rt_bandwidth_enabled(void) } /* RT IPI pull logic requires IRQ_WORK */ -#ifdef CONFIG_IRQ_WORK +#if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) # define HAVE_RT_PUSH_IPI #endif @@ -527,12 +527,6 @@ struct rt_rq { unsigned long rt_nr_total; int overloaded; struct plist_head pushable_tasks; -#ifdef HAVE_RT_PUSH_IPI - int push_flags; - int push_cpu; - struct irq_work push_work; - raw_spinlock_t push_lock; -#endif #endif /* CONFIG_SMP */ int rt_queued; @@ -641,6 +635,19 @@ struct root_domain { struct dl_bw dl_bw; struct cpudl cpudl; +#ifdef HAVE_RT_PUSH_IPI + /* + * For IPI pull requests, loop across the rto_mask. + */ + struct irq_work rto_push_work; + raw_spinlock_t rto_lock; + /* These are only updated and read within rto_lock */ + int rto_loop; + int rto_cpu; + /* These atomics are updated outside of a lock */ + atomic_t rto_loop_next; + atomic_t rto_loop_start; +#endif /* * The "RT overload" flag: it gets set if a CPU has more than * one runnable RT task. @@ -658,6 +665,9 @@ extern void init_defrootdomain(void); extern int sched_init_domains(const struct cpumask *cpu_map); extern void rq_attach_root(struct rq *rq, struct root_domain *rd); +#ifdef HAVE_RT_PUSH_IPI +extern void rto_push_irq_work_func(struct irq_work *work); +#endif #endif /* CONFIG_SMP */ /* diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c index f51d123f9fe1..e50450c2fed8 100644 --- a/kernel/sched/topology.c +++ b/kernel/sched/topology.c @@ -268,6 +268,12 @@ static int init_rootdomain(struct root_domain *rd) if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) goto free_dlo_mask; +#ifdef HAVE_RT_PUSH_IPI + rd->rto_cpu = -1; + raw_spin_lock_init(&rd->rto_lock); + init_irq_work(&rd->rto_push_work, rto_push_irq_work_func); +#endif + init_dl_bw(&rd->dl_bw); if (cpudl_init(&rd->cpudl) != 0) goto free_rto_mask;

7 years, 9 months

2
1
0 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

Linux-stable-mirror