November 2024 - Linux-stable-mirror

Performance Degradation After Upgrading to Kernel 6.8

by Chenbo Lu

Hello, I am experiencing a significant performance degradation after upgrading my kernel from version 6.6 to 6.8 and would appreciate any insights or suggestions. I am running a high-load simulation system that spawns more than 1000 threads and the overall CPU usage is 30%+ . Most of the threads are using real-time scheduling (SCHED_RR), and the threads of a model are using SCHED_DEADLINE. After upgrading the kernel, I noticed that the execution time of my model has increased from 4.5ms to 6ms. What I Have Done So Far: 1. I found this [bug report](https://bugzilla.kernel.org/show_bug.cgi?id=219366#c7) and reverted the commit efa7df3e3bb5da8e6abbe37727417f32a37fba47 mentioned in the post. Unfortunately, this did not resolve the issue. 2. I performed a git bisect and found that after these two commits related to scheduling (RT and deadline) were merged, the problem happened. They are 612f769edd06a6e42f7cd72425488e68ddaeef0a, 5fe7765997b139e2d922b58359dea181efe618f9 After reverting these two commits, the model execution time improved to around 5 ms. 3. I revert two more commits, and the execution time is back to 4.7ms: 63ba8422f876e32ee564ea95da9a7313b13ff0a1, efa7df3e3bb5da8e6abbe37727417f32a37fba47 My questions are: 1.Has anyone else experienced similar performance degradation after upgrading to kernel 6.8? 2.Can anyone explain why these two commits are causing the problem? I am not very familiar with the kernel code and would appreciate any insights. 3.Are there any additional settings or configurations I need to apply when using kernel 6.8 to avoid this issue? My CPU is AMD Ryzen Threadripper 3970X, and my desktop uses Ubuntu 24.04. Thanks again for any insights or suggestions. Please let me know if any other information is needed. Best, Chenbo -- This email and any relevant attachments may include confidential and/or proprietary information. Any distribution or use by anyone other than the intended recipient(s) or other than for the intended purpose(s) is prohibited and may be unlawful. If you are not the intended recipient of this message, please notify the sender by replying to this message and then delete it from your system.

9 months, 2 weeks

4
5
0 0

[PATCH v2] pps: Remove embedded cdev to fix a use-after-free

by Calvin Owens

On a board running ntpd and gpsd, I'm seeing a consistent use-after-free in sys_exit() from gpsd when rebooting: pps pps1: removed ------------[ cut here ]------------ kobject: '(null)' (00000000db4bec24): is not initialized, yet kobject_put() is being called. WARNING: CPU: 2 PID: 440 at lib/kobject.c:734 kobject_put+0x120/0x150 CPU: 2 UID: 299 PID: 440 Comm: gpsd Not tainted 6.11.0-rc6-00308-gb31c44928842 #1 Hardware name: Raspberry Pi 4 Model B Rev 1.1 (DT) pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--) pc : kobject_put+0x120/0x150 lr : kobject_put+0x120/0x150 sp : ffffffc0803d3ae0 x29: ffffffc0803d3ae0 x28: ffffff8042dc9738 x27: 0000000000000001 x26: 0000000000000000 x25: ffffff8042dc9040 x24: ffffff8042dc9440 x23: ffffff80402a4620 x22: ffffff8042ef4bd0 x21: ffffff80405cb600 x20: 000000000008001b x19: ffffff8040b3b6e0 x18: 0000000000000000 x17: 0000000000000000 x16: 0000000000000000 x15: 696e6920746f6e20 x14: 7369203a29343263 x13: 205d303434542020 x12: 0000000000000000 x11: 0000000000000000 x10: 0000000000000000 x9 : 0000000000000000 x8 : 0000000000000000 x7 : 0000000000000000 x6 : 0000000000000000 x5 : 0000000000000000 x4 : 0000000000000000 x3 : 0000000000000000 x2 : 0000000000000000 x1 : 0000000000000000 x0 : 0000000000000000 Call trace: kobject_put+0x120/0x150 cdev_put+0x20/0x3c __fput+0x2c4/0x2d8 ____fput+0x1c/0x38 task_work_run+0x70/0xfc do_exit+0x2a0/0x924 do_group_exit+0x34/0x90 get_signal+0x7fc/0x8c0 do_signal+0x128/0x13b4 do_notify_resume+0xdc/0x160 el0_svc+0xd4/0xf8 el0t_64_sync_handler+0x140/0x14c el0t_64_sync+0x190/0x194 ---[ end trace 0000000000000000 ]--- ...followed by more symptoms of corruption, with similar stacks: refcount_t: underflow; use-after-free. kernel BUG at lib/list_debug.c:62! Kernel panic - not syncing: Oops - BUG: Fatal exception This happens because pps_device_destruct() frees the pps_device with the embedded cdev immediately after calling cdev_del(), but, as the comment above cdev_del() notes, fops for previously opened cdevs are still callable even after cdev_del() returns. I think this bug has always been there: I can't explain why it suddenly started happening every time I reboot this particular board. In commit d953e0e837e6 ("pps: Fix a use-after free bug when unregistering a source."), George Spelvin suggested removing the embedded cdev. That seems like the simplest way to fix this, so I've implemented his suggestion, with pps_idr becoming the source of truth for which minor corresponds to which device. But now that pps_idr defines userspace visibility instead of cdev_add(), we need to be sure the pps->dev kobject refcount can't reach zero while userspace can still find it again. So, the idr_remove() call moves to pps_unregister_cdev(), and pps_idr now holds a reference to the pps->dev kobject. pps_core: source serial1 got cdev (251:1) <...> pps pps1: removed pps_core: unregistering pps1 pps_core: deallocating pps1 Fixes: d953e0e837e6 ("pps: Fix a use-after free bug when unregistering a source.") Cc: stable(a)vger.kernel.org Signed-off-by: Calvin Owens <calvin(a)wbinvd.org> --- Changes in v2: - Don't move pr_debug() from pps_device_destruct() to pps_unregister_cdev() - Actually add stable(a)vger.kernel.org to CC --- drivers/pps/pps.c | 83 ++++++++++++++++++++------------------ include/linux/pps_kernel.h | 1 - 2 files changed, 44 insertions(+), 40 deletions(-) diff --git a/drivers/pps/pps.c b/drivers/pps/pps.c index 5d19baae6a38..6980ab17f314 100644 --- a/drivers/pps/pps.c +++ b/drivers/pps/pps.c @@ -25,7 +25,7 @@ * Local variables */ -static dev_t pps_devt; +static int pps_major; static struct class *pps_class; static DEFINE_MUTEX(pps_idr_lock); @@ -296,19 +296,35 @@ static long pps_cdev_compat_ioctl(struct file *file, #define pps_cdev_compat_ioctl NULL #endif +static struct pps_device *pps_idr_get(unsigned long id) +{ + struct pps_device *pps; + + mutex_lock(&pps_idr_lock); + pps = idr_find(&pps_idr, id); + if (pps) + kobject_get(&pps->dev->kobj); + + mutex_unlock(&pps_idr_lock); + return pps; +} + static int pps_cdev_open(struct inode *inode, struct file *file) { - struct pps_device *pps = container_of(inode->i_cdev, - struct pps_device, cdev); + struct pps_device *pps = pps_idr_get(iminor(inode)); + + if (!pps) + return -ENODEV; + file->private_data = pps; - kobject_get(&pps->dev->kobj); return 0; } static int pps_cdev_release(struct inode *inode, struct file *file) { - struct pps_device *pps = container_of(inode->i_cdev, - struct pps_device, cdev); + struct pps_device *pps = file->private_data; + + WARN_ON(pps->id != iminor(inode)); kobject_put(&pps->dev->kobj); return 0; } @@ -332,14 +348,7 @@ static void pps_device_destruct(struct device *dev) { struct pps_device *pps = dev_get_drvdata(dev); - cdev_del(&pps->cdev); - - /* Now we can release the ID for re-use */ pr_debug("deallocating pps%d\n", pps->id); - mutex_lock(&pps_idr_lock); - idr_remove(&pps_idr, pps->id); - mutex_unlock(&pps_idr_lock); - kfree(dev); kfree(pps); } @@ -364,39 +373,26 @@ int pps_register_cdev(struct pps_device *pps) goto out_unlock; } pps->id = err; - mutex_unlock(&pps_idr_lock); - devt = MKDEV(MAJOR(pps_devt), pps->id); - - cdev_init(&pps->cdev, &pps_cdev_fops); - pps->cdev.owner = pps->info.owner; - - err = cdev_add(&pps->cdev, devt, 1); - if (err) { - pr_err("%s: failed to add char device %d:%d\n", - pps->info.name, MAJOR(pps_devt), pps->id); - goto free_idr; - } + devt = MKDEV(pps_major, pps->id); pps->dev = device_create(pps_class, pps->info.dev, devt, pps, "pps%d", pps->id); if (IS_ERR(pps->dev)) { err = PTR_ERR(pps->dev); - goto del_cdev; + goto free_idr; } /* Override the release function with our own */ pps->dev->release = pps_device_destruct; - pr_debug("source %s got cdev (%d:%d)\n", pps->info.name, - MAJOR(pps_devt), pps->id); + pr_debug("source %s got cdev (%d:%d)\n", pps->info.name, pps_major, + pps->id); + kobject_get(&pps->dev->kobj); + mutex_unlock(&pps_idr_lock); return 0; -del_cdev: - cdev_del(&pps->cdev); - free_idr: - mutex_lock(&pps_idr_lock); idr_remove(&pps_idr, pps->id); out_unlock: mutex_unlock(&pps_idr_lock); @@ -408,6 +404,12 @@ void pps_unregister_cdev(struct pps_device *pps) pr_debug("unregistering pps%d\n", pps->id); pps->lookup_cookie = NULL; device_destroy(pps_class, pps->dev->devt); + + /* Now we can release the ID for re-use */ + mutex_lock(&pps_idr_lock); + idr_remove(&pps_idr, pps->id); + kobject_put(&pps->dev->kobj); + mutex_unlock(&pps_idr_lock); } /* @@ -427,6 +429,11 @@ void pps_unregister_cdev(struct pps_device *pps) * so that it will not be used again, even if the pps device cannot * be removed from the idr due to pending references holding the minor * number in use. + * + * Since pps_idr holds a reference to the kobject, the returned + * pps_device is guaranteed to be valid until pps_unregister_cdev() is + * called on it. But after calling pps_unregister_cdev(), it may be + * freed at any time. */ struct pps_device *pps_lookup_dev(void const *cookie) { @@ -449,13 +456,11 @@ EXPORT_SYMBOL(pps_lookup_dev); static void __exit pps_exit(void) { class_destroy(pps_class); - unregister_chrdev_region(pps_devt, PPS_MAX_SOURCES); + __unregister_chrdev(pps_major, 0, PPS_MAX_SOURCES, "pps"); } static int __init pps_init(void) { - int err; - pps_class = class_create("pps"); if (IS_ERR(pps_class)) { pr_err("failed to allocate class\n"); @@ -463,8 +468,9 @@ static int __init pps_init(void) } pps_class->dev_groups = pps_groups; - err = alloc_chrdev_region(&pps_devt, 0, PPS_MAX_SOURCES, "pps"); - if (err < 0) { + pps_major = __register_chrdev(0, 0, PPS_MAX_SOURCES, "pps", + &pps_cdev_fops); + if (pps_major < 0) { pr_err("failed to allocate char device region\n"); goto remove_class; } @@ -477,8 +483,7 @@ static int __init pps_init(void) remove_class: class_destroy(pps_class); - - return err; + return pps_major; } subsys_initcall(pps_init); diff --git a/include/linux/pps_kernel.h b/include/linux/pps_kernel.h index 78c8ac4951b5..8ee312788118 100644 --- a/include/linux/pps_kernel.h +++ b/include/linux/pps_kernel.h @@ -56,7 +56,6 @@ struct pps_device { unsigned int id; /* PPS source unique ID */ void const *lookup_cookie; /* For pps_lookup_dev() only */ - struct cdev cdev; struct device *dev; struct fasync_struct *async_queue; /* fasync method */ spinlock_t lock; -- 2.45.2

9 months, 2 weeks

5
10
0 0

[PATCH 4.19.y 1/1] inet: inet_defrag: prevent sk release while still in use

by Saeed Mirzamohammadi

From: Florian Westphal <fw(a)strlen.de> commit 18685451fc4e546fc0e718580d32df3c0e5c8272 upstream. ip_local_out() and other functions can pass skb->sk as function argument. If the skb is a fragment and reassembly happens before such function call returns, the sk must not be released. This affects skb fragments reassembled via netfilter or similar modules, e.g. openvswitch or ct_act.c, when run as part of tx pipeline. Eric Dumazet made an initial analysis of this bug. Quoting Eric: Calling ip_defrag() in output path is also implying skb_orphan(), which is buggy because output path relies on sk not disappearing. A relevant old patch about the issue was : 8282f27449bf ("inet: frag: Always orphan skbs inside ip_defrag()") [..] net/ipv4/ip_output.c depends on skb->sk being set, and probably to an inet socket, not an arbitrary one. If we orphan the packet in ipvlan, then downstream things like FQ packet scheduler will not work properly. We need to change ip_defrag() to only use skb_orphan() when really needed, ie whenever frag_list is going to be used. Eric suggested to stash sk in fragment queue and made an initial patch. However there is a problem with this: If skb is refragmented again right after, ip_do_fragment() will copy head->sk to the new fragments, and sets up destructor to sock_wfree. IOW, we have no choice but to fix up sk_wmem accouting to reflect the fully reassembled skb, else wmem will underflow. This change moves the orphan down into the core, to last possible moment. As ip_defrag_offset is aliased with sk_buff->sk member, we must move the offset into the FRAG_CB, else skb->sk gets clobbered. This allows to delay the orphaning long enough to learn if the skb has to be queued or if the skb is completing the reasm queue. In the former case, things work as before, skb is orphaned. This is safe because skb gets queued/stolen and won't continue past reasm engine. In the latter case, we will steal the skb->sk reference, reattach it to the head skb, and fix up wmem accouting when inet_frag inflates truesize. Fixes: 7026b1ddb6b8 ("netfilter: Pass socket pointer down through okfn().") Diagnosed-by: Eric Dumazet <edumazet(a)google.com> Reported-by: xingwei lee <xrivendell7(a)gmail.com> Reported-by: yue sun <samsun1006219(a)gmail.com> Reported-by: syzbot+e5167d7144a62715044c(a)syzkaller.appspotmail.com Signed-off-by: Florian Westphal <fw(a)strlen.de> Reviewed-by: Eric Dumazet <edumazet(a)google.com> Link: https://lore.kernel.org/r/20240326101845.30836-1-fw@strlen.de Signed-off-by: Paolo Abeni <pabeni(a)redhat.com> Signed-off-by: Saeed Mirzamohammadi <saeed.mirzamohammadi(a)oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> (cherry picked from commit 1b6de5e6575b56502665c65cf93b0ae6aa0f51ab) Cc: <stable(a)vger.kernel.org> # 4.19+ Signed-off-by: Saeed Mirzamohammadi <saeed.mirzamohammadi(a)oracle.com> --- include/linux/skbuff.h | 5 +- net/core/sock_destructor.h | 12 +++++ net/ipv4/inet_fragment.c | 70 ++++++++++++++++++++----- net/ipv4/ip_fragment.c | 2 +- net/ipv6/netfilter/nf_conntrack_reasm.c | 2 +- 5 files changed, 72 insertions(+), 19 deletions(-) create mode 100644 net/core/sock_destructor.h diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h index f97734f34746..f5f76a04fdac 100644 --- a/include/linux/skbuff.h +++ b/include/linux/skbuff.h @@ -682,10 +682,7 @@ struct sk_buff { struct list_head list; }; - union { - struct sock *sk; - int ip_defrag_offset; - }; + struct sock *sk; union { ktime_t tstamp; diff --git a/net/core/sock_destructor.h b/net/core/sock_destructor.h new file mode 100644 index 000000000000..2f396e6bfba5 --- /dev/null +++ b/net/core/sock_destructor.h @@ -0,0 +1,12 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +#ifndef _NET_CORE_SOCK_DESTRUCTOR_H +#define _NET_CORE_SOCK_DESTRUCTOR_H +#include <net/tcp.h> + +static inline bool is_skb_wmem(const struct sk_buff *skb) +{ + return skb->destructor == sock_wfree || + skb->destructor == __sock_wfree || + (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree); +} +#endif diff --git a/net/ipv4/inet_fragment.c b/net/ipv4/inet_fragment.c index 9f69411251d0..9144c3cf984c 100644 --- a/net/ipv4/inet_fragment.c +++ b/net/ipv4/inet_fragment.c @@ -28,6 +28,8 @@ #include <net/ip.h> #include <net/ipv6.h> +#include "../core/sock_destructor.h" + /* Use skb->cb to track consecutive/adjacent fragments coming at * the end of the queue. Nodes in the rb-tree queue will * contain "runs" of one or more adjacent fragments. @@ -43,6 +45,7 @@ struct ipfrag_skb_cb { }; struct sk_buff *next_frag; int frag_run_len; + int ip_defrag_offset; }; #define FRAG_CB(skb) ((struct ipfrag_skb_cb *)((skb)->cb)) @@ -319,12 +322,12 @@ int inet_frag_queue_insert(struct inet_frag_queue *q, struct sk_buff *skb, */ if (!last) fragrun_create(q, skb); /* First fragment. */ - else if (last->ip_defrag_offset + last->len < end) { + else if (FRAG_CB(last)->ip_defrag_offset + last->len < end) { /* This is the common case: skb goes to the end. */ /* Detect and discard overlaps. */ - if (offset < last->ip_defrag_offset + last->len) + if (offset < FRAG_CB(last)->ip_defrag_offset + last->len) return IPFRAG_OVERLAP; - if (offset == last->ip_defrag_offset + last->len) + if (offset == FRAG_CB(last)->ip_defrag_offset + last->len) fragrun_append_to_last(q, skb); else fragrun_create(q, skb); @@ -341,13 +344,13 @@ int inet_frag_queue_insert(struct inet_frag_queue *q, struct sk_buff *skb, parent = *rbn; curr = rb_to_skb(parent); - curr_run_end = curr->ip_defrag_offset + + curr_run_end = FRAG_CB(curr)->ip_defrag_offset + FRAG_CB(curr)->frag_run_len; - if (end <= curr->ip_defrag_offset) + if (end <= FRAG_CB(curr)->ip_defrag_offset) rbn = &parent->rb_left; else if (offset >= curr_run_end) rbn = &parent->rb_right; - else if (offset >= curr->ip_defrag_offset && + else if (offset >= FRAG_CB(curr)->ip_defrag_offset && end <= curr_run_end) return IPFRAG_DUP; else @@ -361,7 +364,7 @@ int inet_frag_queue_insert(struct inet_frag_queue *q, struct sk_buff *skb, rb_insert_color(&skb->rbnode, &q->rb_fragments); } - skb->ip_defrag_offset = offset; + FRAG_CB(skb)->ip_defrag_offset = offset; return IPFRAG_OK; } @@ -371,13 +374,28 @@ void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb, struct sk_buff *parent) { struct sk_buff *fp, *head = skb_rb_first(&q->rb_fragments); - struct sk_buff **nextp; + void (*destructor)(struct sk_buff *); + unsigned int orig_truesize = 0; + struct sk_buff **nextp = NULL; + struct sock *sk = skb->sk; int delta; + if (sk && is_skb_wmem(skb)) { + /* TX: skb->sk might have been passed as argument to + * dst->output and must remain valid until tx completes. + * + * Move sk to reassembled skb and fix up wmem accounting. + */ + orig_truesize = skb->truesize; + destructor = skb->destructor; + } + if (head != skb) { fp = skb_clone(skb, GFP_ATOMIC); - if (!fp) - return NULL; + if (!fp) { + head = skb; + goto out_restore_sk; + } FRAG_CB(fp)->next_frag = FRAG_CB(skb)->next_frag; if (RB_EMPTY_NODE(&skb->rbnode)) FRAG_CB(parent)->next_frag = fp; @@ -386,6 +404,12 @@ void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb, &q->rb_fragments); if (q->fragments_tail == skb) q->fragments_tail = fp; + + if (orig_truesize) { + /* prevent skb_morph from releasing sk */ + skb->sk = NULL; + skb->destructor = NULL; + } skb_morph(skb, head); FRAG_CB(skb)->next_frag = FRAG_CB(head)->next_frag; rb_replace_node(&head->rbnode, &skb->rbnode, @@ -393,13 +417,13 @@ void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb, consume_skb(head); head = skb; } - WARN_ON(head->ip_defrag_offset != 0); + WARN_ON(FRAG_CB(head)->ip_defrag_offset != 0); delta = -head->truesize; /* Head of list must not be cloned. */ if (skb_unclone(head, GFP_ATOMIC)) - return NULL; + goto out_restore_sk; delta += head->truesize; if (delta) @@ -415,7 +439,7 @@ void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb, clone = alloc_skb(0, GFP_ATOMIC); if (!clone) - return NULL; + goto out_restore_sk; skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; skb_frag_list_init(head); for (i = 0; i < skb_shinfo(head)->nr_frags; i++) @@ -432,6 +456,21 @@ void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb, nextp = &skb_shinfo(head)->frag_list; } +out_restore_sk: + if (orig_truesize) { + int ts_delta = head->truesize - orig_truesize; + + /* if this reassembled skb is fragmented later, + * fraglist skbs will get skb->sk assigned from head->sk, + * and each frag skb will be released via sock_wfree. + * + * Update sk_wmem_alloc. + */ + head->sk = sk; + head->destructor = destructor; + refcount_add(ts_delta, &sk->sk_wmem_alloc); + } + return nextp; } EXPORT_SYMBOL(inet_frag_reasm_prepare); @@ -439,6 +478,8 @@ EXPORT_SYMBOL(inet_frag_reasm_prepare); void inet_frag_reasm_finish(struct inet_frag_queue *q, struct sk_buff *head, void *reasm_data) { + struct sock *sk = is_skb_wmem(head) ? head->sk : NULL; + const unsigned int head_truesize = head->truesize; struct sk_buff **nextp = (struct sk_buff **)reasm_data; struct rb_node *rbn; struct sk_buff *fp; @@ -484,6 +525,9 @@ void inet_frag_reasm_finish(struct inet_frag_queue *q, struct sk_buff *head, skb_mark_not_on_list(head); head->prev = NULL; head->tstamp = q->stamp; + + if (sk) + refcount_add(sum_truesize - head_truesize, &sk->sk_wmem_alloc); } EXPORT_SYMBOL(inet_frag_reasm_finish); diff --git a/net/ipv4/ip_fragment.c b/net/ipv4/ip_fragment.c index 5a1d39e32196..84544e5df7fc 100644 --- a/net/ipv4/ip_fragment.c +++ b/net/ipv4/ip_fragment.c @@ -380,6 +380,7 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb) } skb_dst_drop(skb); + skb_orphan(skb); return -EINPROGRESS; insert_error: @@ -477,7 +478,6 @@ int ip_defrag(struct net *net, struct sk_buff *skb, u32 user) struct ipq *qp; __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS); - skb_orphan(skb); /* Lookup (or create) queue header */ qp = ip_find(net, ip_hdr(skb), user, vif); diff --git a/net/ipv6/netfilter/nf_conntrack_reasm.c b/net/ipv6/netfilter/nf_conntrack_reasm.c index 35d5a76867d0..8aab62c330ef 100644 --- a/net/ipv6/netfilter/nf_conntrack_reasm.c +++ b/net/ipv6/netfilter/nf_conntrack_reasm.c @@ -307,6 +307,7 @@ static int nf_ct_frag6_queue(struct frag_queue *fq, struct sk_buff *skb, } skb_dst_drop(skb); + skb_orphan(skb); return -EINPROGRESS; insert_error: @@ -473,7 +474,6 @@ int nf_ct_frag6_gather(struct net *net, struct sk_buff *skb, u32 user) hdr = ipv6_hdr(skb); fhdr = (struct frag_hdr *)skb_transport_header(skb); - skb_orphan(skb); fq = fq_find(net, fhdr->identification, user, hdr, skb->dev ? skb->dev->ifindex : 0); if (fq == NULL) { -- 2.46.0

9 months, 2 weeks

2
2
0 0

[PATCH v5 1/1] dmaengine: dw: Select only supported masters for ACPI devices

by Andy Shevchenko

The recently submitted fix-commit revealed a problem in the iDMA 32-bit platform code. Even though the controller supported only a single master the dw_dma_acpi_filter() method hard-coded two master interfaces with IDs 0 and 1. As a result the sanity check implemented in the commit b336268dde75 ("dmaengine: dw: Add peripheral bus width verification") got incorrect interface data width and thus prevented the client drivers from configuring the DMA-channel with the EINVAL error returned. E.g., the next error was printed for the PXA2xx SPI controller driver trying to configure the requested channels: > [ 164.525604] pxa2xx_spi_pci 0000:00:07.1: DMA slave config failed > [ 164.536105] pxa2xx_spi_pci 0000:00:07.1: failed to get DMA TX descriptor > [ 164.543213] spidev spi-SPT0001:00: SPI transfer failed: -16 The problem would have been spotted much earlier if the iDMA 32-bit controller supported more than one master interfaces. But since it supports just a single master and the iDMA 32-bit specific code just ignores the master IDs in the CTLLO preparation method, the issue has been gone unnoticed so far. Fix the problem by specifying the default master ID for both memory and peripheral devices in the driver data. Thus the issue noticed for the iDMA 32-bit controllers will be eliminated and the ACPI-probed DW DMA controllers will be configured with the correct master ID by default. Cc: stable(a)vger.kernel.org Fixes: b336268dde75 ("dmaengine: dw: Add peripheral bus width verification") Fixes: 199244d69458 ("dmaengine: dw: add support of iDMA 32-bit hardware") Reported-by: Ferry Toth <fntoth(a)gmail.com> Closes: https://lore.kernel.org/dmaengine/ZuXbCKUs1iOqFu51@black.fi.intel.com/ Reported-by: Andy Shevchenko <andriy.shevchenko(a)linux.intel.com> Closes: https://lore.kernel.org/dmaengine/ZuXgI-VcHpMgbZ91@black.fi.intel.com/ Tested-by: Ferry Toth <fntoth(a)gmail.com> Signed-off-by: Andy Shevchenko <andriy.shevchenko(a)linux.intel.com> --- v5: rebranded to follow the compliances Vinod, please apply this for v6.12-rcX as we have a problem to fix. drivers/dma/dw/acpi.c | 6 ++++-- drivers/dma/dw/internal.h | 8 ++++++++ drivers/dma/dw/pci.c | 4 ++-- 3 files changed, 14 insertions(+), 4 deletions(-) diff --git a/drivers/dma/dw/acpi.c b/drivers/dma/dw/acpi.c index c510c109d2c3..b6452fffa657 100644 --- a/drivers/dma/dw/acpi.c +++ b/drivers/dma/dw/acpi.c @@ -8,13 +8,15 @@ static bool dw_dma_acpi_filter(struct dma_chan *chan, void *param) { + struct dw_dma *dw = to_dw_dma(chan->device); + struct dw_dma_chip_pdata *data = dev_get_drvdata(dw->dma.dev); struct acpi_dma_spec *dma_spec = param; struct dw_dma_slave slave = { .dma_dev = dma_spec->dev, .src_id = dma_spec->slave_id, .dst_id = dma_spec->slave_id, - .m_master = 0, - .p_master = 1, + .m_master = data->m_master, + .p_master = data->p_master, }; return dw_dma_filter(chan, &slave); diff --git a/drivers/dma/dw/internal.h b/drivers/dma/dw/internal.h index 563ce73488db..f1bd06a20cd6 100644 --- a/drivers/dma/dw/internal.h +++ b/drivers/dma/dw/internal.h @@ -51,11 +51,15 @@ struct dw_dma_chip_pdata { int (*probe)(struct dw_dma_chip *chip); int (*remove)(struct dw_dma_chip *chip); struct dw_dma_chip *chip; + u8 m_master; + u8 p_master; }; static __maybe_unused const struct dw_dma_chip_pdata dw_dma_chip_pdata = { .probe = dw_dma_probe, .remove = dw_dma_remove, + .m_master = 0, + .p_master = 1, }; static const struct dw_dma_platform_data idma32_pdata = { @@ -72,6 +76,8 @@ static __maybe_unused const struct dw_dma_chip_pdata idma32_chip_pdata = { .pdata = &idma32_pdata, .probe = idma32_dma_probe, .remove = idma32_dma_remove, + .m_master = 0, + .p_master = 0, }; static const struct dw_dma_platform_data xbar_pdata = { @@ -88,6 +94,8 @@ static __maybe_unused const struct dw_dma_chip_pdata xbar_chip_pdata = { .pdata = &xbar_pdata, .probe = idma32_dma_probe, .remove = idma32_dma_remove, + .m_master = 0, + .p_master = 0, }; #endif /* _DMA_DW_INTERNAL_H */ diff --git a/drivers/dma/dw/pci.c b/drivers/dma/dw/pci.c index ad2d4d012cf7..e8a0eb81726a 100644 --- a/drivers/dma/dw/pci.c +++ b/drivers/dma/dw/pci.c @@ -56,10 +56,10 @@ static int dw_pci_probe(struct pci_dev *pdev, const struct pci_device_id *pid) if (ret) return ret; - dw_dma_acpi_controller_register(chip->dw); - pci_set_drvdata(pdev, data); + dw_dma_acpi_controller_register(chip->dw); + return 0; } -- 2.43.0.rc1.1336.g36b5255a03ac

9 months, 2 weeks

2
1
0 0

[PATCH 6.11 00/31] xe/i915 fixes for 6.11

by Lucas De Marchi

Due to some issues with hibernation on Lunar Lake (integrated), it was decided to re-use the migration logic from Battle Mage (discrete). However in 6.11 there were several patches missing to allow that to work. A few patches were picked automatically for 6.11.10, but they are not sufficient. Bring the additional patches and some tests to make sure the backports work: this correspond to 20 of the patches here. Others were additional fixes or dependencies. This was tested on top of 6.11.10. Akshata Jahagirdar (5): drm/xe/migrate: Handle clear ccs logic for xe2 dgfx drm/xe/migrate: Add helper function to program identity map drm/xe/migrate: Add kunit to test clear functionality drm/xe/xe2: Introduce identity map for compressed pat for vram drm/xe/xe_migrate: Handle migration logic for xe2+ dgfx Aradhya Bhatia (1): drm/xe/xe2lpg: Extend Wa_15016589081 for xe2lpg Chaitanya Kumar Borah (1): drm/i915: Do not explicilty enable FEC in DP_TP_CTL for UHBR rates Daniele Ceraolo Spurio (1): drm/xe/uc: Use managed bo for HuC and GSC objects Gustavo Sousa (2): drm/xe/xe2: Extend performance tuning to media GT drm/xe/xe2: Add performance tuning for L3 cache flushing He Lugang (1): drm/xe: use devm_add_action_or_reset() helper Imre Deak (5): drm/xe: Handle polling only for system s/r in xe_display_pm_suspend/resume() drm/i915/dp: Assume panel power is off if runtime suspended drm/i915/dp: Disable unnecessary HPD polling for eDP drm/xe/display: Separate the d3cold and non-d3cold runtime PM handling drm/xe/display: Add missing HPD interrupt enabling during non-d3cold RPM resume Maarten Lankhorst (2): drm/xe: Remove runtime argument from display s/r functions drm/xe: Fix missing conversion to xe_display_pm_runtime_resume Matthew Auld (3): drm/xe/client: use mem_type from the current resource drm/xe/queue: move xa_alloc to prevent UAF drm/xe/bmg: improve cache flushing behaviour Matthew Brost (1): drm/xe: Do not run GPU page fault handler on a closed VM Michal Wajdeczko (4): drm/xe/kunit: Kill xe_cur_kunit() drm/xe/kunit: Simplify xe_bo live tests code layout drm/xe/kunit: Simplify xe_dma_buf live tests code layout drm/xe/kunit: Simplify xe_migrate live tests code layout Rodrigo Vivi (1): drm/{i915, xe}: Avoid direct inspection of dpt_vma from outside dpt Suraj Kandpal (2): drm/xe/display: Do not suspend resume dp mst during runtime drm/xe/display: Do not do intel_fbdev_set_suspend during runtime Thomas Hellström (1): drm/xe: Use separate rpm lockdep map for non-d3cold-capable devices Vinod Govindapillai (1): drm/xe/display: handle HPD polling in display runtime suspend/resume drivers/gpu/drm/i915/display/intel_dp.c | 16 +- drivers/gpu/drm/i915/display/intel_dpt.c | 4 + drivers/gpu/drm/i915/display/intel_dpt.h | 3 + .../drm/i915/display/skl_universal_plane.c | 3 +- drivers/gpu/drm/i915/intel_runtime_pm.h | 8 +- .../xe/compat-i915-headers/intel_runtime_pm.h | 8 + drivers/gpu/drm/xe/display/xe_display.c | 78 ++++- drivers/gpu/drm/xe/display/xe_display.h | 12 +- drivers/gpu/drm/xe/display/xe_fb_pin.c | 9 +- drivers/gpu/drm/xe/regs/xe_gt_regs.h | 12 +- drivers/gpu/drm/xe/tests/Makefile | 3 - drivers/gpu/drm/xe/tests/xe_bo.c | 24 +- drivers/gpu/drm/xe/tests/xe_bo_test.c | 21 -- drivers/gpu/drm/xe/tests/xe_bo_test.h | 14 - drivers/gpu/drm/xe/tests/xe_dma_buf.c | 20 +- drivers/gpu/drm/xe/tests/xe_dma_buf_test.c | 20 -- drivers/gpu/drm/xe/tests/xe_dma_buf_test.h | 13 - drivers/gpu/drm/xe/tests/xe_live_test_mod.c | 9 + drivers/gpu/drm/xe/tests/xe_migrate.c | 299 +++++++++++++++++- drivers/gpu/drm/xe/tests/xe_migrate_test.c | 20 -- drivers/gpu/drm/xe/tests/xe_migrate_test.h | 13 - drivers/gpu/drm/xe/tests/xe_mocs.c | 8 +- drivers/gpu/drm/xe/tests/xe_pci_test.c | 4 +- drivers/gpu/drm/xe/tests/xe_test.h | 8 +- drivers/gpu/drm/xe/xe_drm_client.c | 7 +- drivers/gpu/drm/xe/xe_exec_queue.c | 4 +- drivers/gpu/drm/xe/xe_gsc.c | 12 +- drivers/gpu/drm/xe/xe_gsc_proxy.c | 36 +-- drivers/gpu/drm/xe/xe_gt.c | 1 - drivers/gpu/drm/xe/xe_gt_freq.c | 4 +- drivers/gpu/drm/xe/xe_gt_pagefault.c | 6 + drivers/gpu/drm/xe/xe_gt_sysfs.c | 2 +- drivers/gpu/drm/xe/xe_huc.c | 19 +- drivers/gpu/drm/xe/xe_migrate.c | 185 +++++++---- drivers/gpu/drm/xe/xe_module.c | 9 + drivers/gpu/drm/xe/xe_pm.c | 100 ++++-- drivers/gpu/drm/xe/xe_pm.h | 1 + drivers/gpu/drm/xe/xe_tuning.c | 28 ++ drivers/gpu/drm/xe/xe_wa.c | 4 + 39 files changed, 735 insertions(+), 312 deletions(-) delete mode 100644 drivers/gpu/drm/xe/tests/xe_bo_test.c delete mode 100644 drivers/gpu/drm/xe/tests/xe_bo_test.h delete mode 100644 drivers/gpu/drm/xe/tests/xe_dma_buf_test.c delete mode 100644 drivers/gpu/drm/xe/tests/xe_dma_buf_test.h delete mode 100644 drivers/gpu/drm/xe/tests/xe_migrate_test.c delete mode 100644 drivers/gpu/drm/xe/tests/xe_migrate_test.h -- 2.47.0

9 months, 2 weeks

4
44
0 0

[PATCH v2] kbuild: Move -Wenum-enum-conversion to W=2

by Nathan Chancellor

-Wenum-enum-conversion was strengthened in clang-19 to warn for C, which caused the kernel to move it to W=1 in commit 75b5ab134bb5 ("kbuild: Move -Wenum-{compare-conditional,enum-conversion} into W=1") because there were numerous instances that would break builds with -Werror. Unfortunately, this is not a full solution, as more and more developers, subsystems, and distributors are building with W=1 as well, so they continue to see the numerous instances of this warning. Since the move to W=1, there have not been many new instances that have appeared through various build reports and the ones that have appeared seem to be following similar existing patterns, suggesting that most instances of this warning will not be real issues. The only alternatives for silencing this warning are adding casts (which is generally seen as an ugly practice) or refactoring the enums to macro defines or a unified enum (which may be undesirable because of type safety in other parts of the code). Move the warning to W=2, where warnings that occur frequently but may be relevant should reside. Cc: stable(a)vger.kernel.org Fixes: 75b5ab134bb5 ("kbuild: Move -Wenum-{compare-conditional,enum-conversion} into W=1") Link: https://lore.kernel.org/ZwRA9SOcOjjLJcpi@google.com/ Signed-off-by: Nathan Chancellor <nathan(a)kernel.org> --- Changes in v2: - Move -Wenum-enum-conversion to W=2, instead of disabling it outright (Arnd) - Leave -Wenum-compare-conditional in W=1, as there are not that many instances, so it can be turned on fully at some point (Arnd) - Link to v1: https://lore.kernel.org/r/20241016-disable-two-clang-enum-warnings-v1-1-ae8… --- scripts/Makefile.extrawarn | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/scripts/Makefile.extrawarn b/scripts/Makefile.extrawarn index 1d13cecc7cc7808610e635ddc03476cf92b3a8c1..04faf15ed316a9c291dc952b6cc40fb6c8c330cf 100644 --- a/scripts/Makefile.extrawarn +++ b/scripts/Makefile.extrawarn @@ -130,7 +130,6 @@ KBUILD_CFLAGS += $(call cc-disable-warning, pointer-to-enum-cast) KBUILD_CFLAGS += -Wno-tautological-constant-out-of-range-compare KBUILD_CFLAGS += $(call cc-disable-warning, unaligned-access) KBUILD_CFLAGS += -Wno-enum-compare-conditional -KBUILD_CFLAGS += -Wno-enum-enum-conversion endif endif @@ -154,6 +153,10 @@ KBUILD_CFLAGS += -Wno-missing-field-initializers KBUILD_CFLAGS += -Wno-type-limits KBUILD_CFLAGS += -Wno-shift-negative-value +ifdef CONFIG_CC_IS_CLANG +KBUILD_CFLAGS += -Wno-enum-enum-conversion +endif + ifdef CONFIG_CC_IS_GCC KBUILD_CFLAGS += -Wno-maybe-uninitialized endif --- base-commit: 8e929cb546ee42c9a61d24fae60605e9e3192354 change-id: 20241016-disable-two-clang-enum-warnings-e7994d44f948 Best regards, -- Nathan Chancellor <nathan(a)kernel.org>

9 months, 2 weeks

2
2
0 0

[PATCH] arm64: dts: qcom: sa8775p: Fix the size of 'addr_space' regions

by Manivannan Sadhasivam

For both the controller instances, size of the 'addr_space' region should be 0x1fe00000 as per the hardware memory layout. Otherwise, endpoint drivers cannot request even reasonable BAR size of 1MB. Cc: stable(a)vger.kernel.org # 6.11 Fixes: c5f5de8434ec ("arm64: dts: qcom: sa8775p: Add ep pcie1 controller node") Fixes: 1924f5518224 ("arm64: dts: qcom: sa8775p: Add ep pcie0 controller node") Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam(a)linaro.org> --- arch/arm64/boot/dts/qcom/sa8775p.dtsi | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/arch/arm64/boot/dts/qcom/sa8775p.dtsi b/arch/arm64/boot/dts/qcom/sa8775p.dtsi index e8dbc8d820a6..320a94dcac5c 100644 --- a/arch/arm64/boot/dts/qcom/sa8775p.dtsi +++ b/arch/arm64/boot/dts/qcom/sa8775p.dtsi @@ -5587,7 +5587,7 @@ pcie0_ep: pcie-ep@1c00000 { <0x0 0x40000000 0x0 0xf20>, <0x0 0x40000f20 0x0 0xa8>, <0x0 0x40001000 0x0 0x4000>, - <0x0 0x40200000 0x0 0x100000>, + <0x0 0x40200000 0x0 0x1fe00000>, <0x0 0x01c03000 0x0 0x1000>, <0x0 0x40005000 0x0 0x2000>; reg-names = "parf", "dbi", "elbi", "atu", "addr_space", @@ -5744,7 +5744,7 @@ pcie1_ep: pcie-ep@1c10000 { <0x0 0x60000000 0x0 0xf20>, <0x0 0x60000f20 0x0 0xa8>, <0x0 0x60001000 0x0 0x4000>, - <0x0 0x60200000 0x0 0x100000>, + <0x0 0x60200000 0x0 0x1fe00000>, <0x0 0x01c13000 0x0 0x1000>, <0x0 0x60005000 0x0 0x2000>; reg-names = "parf", "dbi", "elbi", "atu", "addr_space", -- 2.25.1

9 months, 2 weeks

3
2
0 0

[PATCH] mmc: core: Further prevent card detect during shutdown

by Ulf Hansson

Disabling card detect from the host's ->shutdown_pre() callback turned out to not be the complete solution. More precisely, beyond the point when the mmc_bus->shutdown() has been called, to gracefully power off the card, we need to prevent card detect. Otherwise the mmc_rescan work may poll for the card with a CMD13, to see if it's still alive, which then will fail and hang as the card has already been powered off. To fix this problem, let's disable mmc_rescan prior to power off the card during shutdown. Reported-by: Anthony Pighin <anthony.pighin(a)nokia.com> Fixes: 66c915d09b94 ("mmc: core: Disable card detect during shutdown") Cc: stable(a)vger.kernel.org Signed-off-by: Ulf Hansson <ulf.hansson(a)linaro.org> --- drivers/mmc/core/bus.c | 2 ++ drivers/mmc/core/core.c | 3 +++ 2 files changed, 5 insertions(+) diff --git a/drivers/mmc/core/bus.c b/drivers/mmc/core/bus.c index 9283b28bc69f..1cf64e0952fb 100644 --- a/drivers/mmc/core/bus.c +++ b/drivers/mmc/core/bus.c @@ -149,6 +149,8 @@ static void mmc_bus_shutdown(struct device *dev) if (dev->driver && drv->shutdown) drv->shutdown(card); + __mmc_stop_host(host); + if (host->bus_ops->shutdown) { ret = host->bus_ops->shutdown(host); if (ret) diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c index a499f3c59de5..d996d39c0d6f 100644 --- a/drivers/mmc/core/core.c +++ b/drivers/mmc/core/core.c @@ -2335,6 +2335,9 @@ void mmc_start_host(struct mmc_host *host) void __mmc_stop_host(struct mmc_host *host) { + if (host->rescan_disable) + return; + if (host->slot.cd_irq >= 0) { mmc_gpio_set_cd_wake(host, false); disable_irq(host->slot.cd_irq); -- 2.43.0

9 months, 2 weeks

2
3
0 0

[PATCH 5.4 0/1] Address CVE-2024-49974

by cel＠kernel.org

From: Chuck Lever <chuck.lever(a)oracle.com> Backport the upstream patch that disables background NFSv4.2 COPY operations. Unlike later LTS kernels, the patches that limit the number of background COPY operations do not apply at all to v5.4. Because there is no support for server-to-server COPY in v5.4, disabling background COPY operations should not be noticeable. Chuck Lever (1): NFSD: Force all NFSv4.2 COPY requests to be synchronous fs/nfsd/nfs4proc.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) -- 2.47.0

9 months, 2 weeks

4
5
0 0

[PATCH v2 0/4] Venus driver fixes to avoid possible OOB accesses

by Vikash Garodia

v1 -> v2: - init_codec to always update with latest payload from firmware (Dmitry/Bryan) - Rewrite the logic of packet parsing to consider payload size for different packet type (Bryan) - Consider reading sfr data till available space (Dmitry) - Add reviewed-by tags v1: https://lore.kernel.org/all/20241105-venus_oob-v1-0-8d4feedfe2bb@quicinc.co… This series primarily adds check at relevant places in venus driver where there are possible OOB accesses due to unexpected payload from venus firmware. The patches describes the specific OOB possibility. Please review and share your feedback. Validated on sc7180(v4) and rb5(v6). Stan, please help to extend the test on db410c(v1). Signed-off-by: Vikash Garodia <quic_vgarodia(a)quicinc.com> --- Vikash Garodia (4): media: venus: hfi_parser: add check to avoid out of bound access media: venus: hfi_parser: avoid OOB access beyond payload word count media: venus: hfi: add check to handle incorrect queue size media: venus: hfi: add a check to handle OOB in sfr region drivers/media/platform/qcom/venus/hfi_parser.c | 58 +++++++++++++++++++++----- drivers/media/platform/qcom/venus/hfi_venus.c | 15 ++++++- 2 files changed, 60 insertions(+), 13 deletions(-) --- base-commit: c7ccf3683ac9746b263b0502255f5ce47f64fe0a change-id: 20241115-venus_oob_2-21708239176a Best regards, -- Vikash Garodia <quic_vgarodia(a)quicinc.com>

9 months, 2 weeks

2
8
0 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

Linux-stable-mirror November 2024