Skip to contentSkip to conversation

irqchip/versatile-fpga: Handle chained IRQs properly

PATCH New Sasha Levin 1 participant 9 messages Apr 10, 2020

Subsystems (4)
lore
Share & tools
lore
author maintainer
#4 lore raw

PATCH 5/9 gfs2: Don't demote a glock until its revokes are written

In reply to Sasha Levin

From: Bob Peterson <[email protected]>

[ Upstream commit df5db5f9ee112e76b5202fbc331f990a0fc316d6 ]

Before this patch, run_queue would demote glocks based on whether there are any more holders. But if the glock has pending revokes that haven't been written to the media, giving up the glock might end in file system corruption if the revokes never get written due to io errors, node crashes and fences, etc. In that case, another node will replay the metadata blocks associated with the glock, but because the revoke was never written, it could replay that block even though the glock had since been granted to another node who might have made changes.

This patch changes the logic in run_queue so that it never demotes a glock until its count of pending revokes reaches zero.

---
 fs/gfs2/glock.c | 3 +++
 1 file changed, 3 insertions(+)
fs/gfs2/glock.c +3 -0
index efd44d5645d83..adc1a97cfe96d 100644
@@ -548,6 +548,9 @@ __acquires(&gl->gl_lockref.lock)
 			goto out_unlock;
 		if (nonblock)
 			goto out_sched;
+		smp_mb();
+		if (atomic_read(&gl->gl_revokes) != 0)
+			goto out_sched;
 		set_bit(GLF_DEMOTE_IN_PROGRESS, &gl->gl_flags);
 		GLOCK_BUG_ON(gl, gl->gl_demote_state == LM_ST_EXCLUSIVE);
 		gl->gl_target = gl->gl_demote_state;
Signature · 1 line
2.20.1
author maintainer
#5 lore raw

PATCH 6/9 x86/boot: Use unsigned comparison for addresses

In reply to Sasha Levin

From: Arvind Sankar <[email protected]>

[ Upstream commit 81a34892c2c7c809f9c4e22c5ac936ae673fb9a2 ]

The load address is compared with LOAD_PHYSICAL_ADDR using a signed comparison currently (using jge instruction).

When loading a 64-bit kernel using the new efi32_pe_entry() point added by:

97aa276579b2 ("efi/x86: Add true mixed mode entry point into .compat section")

using Qemu with -m 3072, the firmware actually loads us above 2Gb, resulting in a very early crash.

Use the JAE instruction to perform a unsigned comparison instead, as physical addresses should be considered unsigned.

---
 arch/x86/boot/compressed/head_32.S | 2 +-
 arch/x86/boot/compressed/head_64.S | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)
arch/x86/boot/compressed/head_32.S +1 -1
index fd0b6a272dd5b..7532f6f536774 100644
@@ -170,7 +170,7 @@ preferred_addr:
 	notl	%eax
 	andl    %eax, %ebx
 	cmpl	$LOAD_PHYSICAL_ADDR, %ebx
-	jge	1f
+	jae	1f
 #endif
 	movl	$LOAD_PHYSICAL_ADDR, %ebx
 1:
arch/x86/boot/compressed/head_64.S +2 -2
index 9e3a183561a9a..3fac2d133e4ec 100644
@@ -104,7 +104,7 @@ ENTRY(startup_32)
 	notl	%eax
 	andl	%eax, %ebx
 	cmpl	$LOAD_PHYSICAL_ADDR, %ebx
-	jge	1f
+	jae	1f
 #endif
 	movl	$LOAD_PHYSICAL_ADDR, %ebx
 1:
@@ -339,7 +339,7 @@ preferred_addr:
 	notq	%rax
 	andq	%rax, %rbp
 	cmpq	$LOAD_PHYSICAL_ADDR, %rbp
-	jge	1f
+	jae	1f
 #endif
 	movq	$LOAD_PHYSICAL_ADDR, %rbp
 1:
Signature · 1 line
2.20.1
author maintainer
#2 lore raw

PATCH 3/9 selftests/x86/ptrace_syscall_32: Fix no-vDSO segfault

In reply to Sasha Levin

From: Andy Lutomirski <[email protected]>

[ Upstream commit 630b99ab60aa972052a4202a1ff96c7e45eb0054 ]

If AT_SYSINFO is not present, don't try to call a NULL pointer.

---
 tools/testing/selftests/x86/ptrace_syscall.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
tools/testing/selftests/x86/ptrace_syscall.c +6 -2
index 1e3da137a8bb9..5390c827a359e 100644
@@ -413,8 +413,12 @@ int main()
 
 #if defined(__i386__) && (!defined(__GLIBC__) || __GLIBC__ > 2 || __GLIBC_MINOR__ >= 16)
 	vsyscall32 = (void *)getauxval(AT_SYSINFO);
-	printf("[RUN]\tCheck AT_SYSINFO return regs\n");
-	test_sys32_regs(do_full_vsyscall32);
+	if (vsyscall32) {
+		printf("[RUN]\tCheck AT_SYSINFO return regs\n");
+		test_sys32_regs(do_full_vsyscall32);
+	} else {
+		printf("[SKIP]\tAT_SYSINFO is not available\n");
+	}
 #endif
 
 	test_ptrace_syscall_restart();
Signature · 1 line
2.20.1
author maintainer
#8 lore raw

PATCH 9/9 btrfs: track reloc roots based on their commit root bytenr

In reply to Sasha Levin

From: Josef Bacik <[email protected]>

[ Upstream commit ea287ab157c2816bf12aad4cece41372f9d146b4 ]

We always search the commit root of the extent tree for looking up back references, however we track the reloc roots based on their current bytenr.

This is wrong, if we commit the transaction between relocating tree blocks we could end up in this code in build_backref_tree

  if (key.objectid == key.offset) {
	  /*
	   * Only root blocks of reloc trees use backref
	   * pointing to itself.
	   */
	  root = find_reloc_root(rc, cur->bytenr);
	  ASSERT(root);
	  cur->root = root;
	  break;

}

find_reloc_root() is looking based on the bytenr we had in the commit root, but if we've COWed this reloc root we will not find that bytenr, and we will trip over the ASSERT(root).

Fix this by using the commit_root->start bytenr for indexing the commit root. Then we change the __update_reloc_root() caller to be used when we switch the commit root for the reloc root during commit.

This fixes the panic I was seeing when we started throttling relocation for delayed refs.

---
 fs/btrfs/relocation.c | 17 +++++++----------
 1 file changed, 7 insertions(+), 10 deletions(-)
fs/btrfs/relocation.c +7 -10
index 97bc85ca508c2..8d98e8e248b7a 100644
@@ -1296,7 +1296,7 @@ static int __must_check __add_reloc_root(struct btrfs_root *root)
 	if (!node)
 		return -ENOMEM;
 
-	node->bytenr = root->node->start;
+	node->bytenr = root->commit_root->start;
 	node->data = root;
 
 	spin_lock(&rc->reloc_root_tree.lock);
@@ -1328,10 +1328,11 @@ static void __del_reloc_root(struct btrfs_root *root)
 	if (rc && root->node) {
 		spin_lock(&rc->reloc_root_tree.lock);
 		rb_node = tree_search(&rc->reloc_root_tree.rb_root,
-				      root->node->start);
+				      root->commit_root->start);
 		if (rb_node) {
 			node = rb_entry(rb_node, struct mapping_node, rb_node);
 			rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
+			RB_CLEAR_NODE(&node->rb_node);
 		}
 		spin_unlock(&rc->reloc_root_tree.lock);
 		if (!node)
@@ -1349,7 +1350,7 @@ static void __del_reloc_root(struct btrfs_root *root)
  * helper to update the 'address of tree root -> reloc tree'
  * mapping
  */
-static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
+static int __update_reloc_root(struct btrfs_root *root)
 {
 	struct rb_node *rb_node;
 	struct mapping_node *node = NULL;
@@ -1357,7 +1358,7 @@ static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
 
 	spin_lock(&rc->reloc_root_tree.lock);
 	rb_node = tree_search(&rc->reloc_root_tree.rb_root,
-			      root->node->start);
+			      root->commit_root->start);
 	if (rb_node) {
 		node = rb_entry(rb_node, struct mapping_node, rb_node);
 		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
@@ -1369,7 +1370,7 @@ static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
 	BUG_ON((struct btrfs_root *)node->data != root);
 
 	spin_lock(&rc->reloc_root_tree.lock);
-	node->bytenr = new_bytenr;
+	node->bytenr = root->node->start;
 	rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
 			      node->bytenr, &node->rb_node);
 	spin_unlock(&rc->reloc_root_tree.lock);
@@ -1519,6 +1520,7 @@ int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
 	}
 
 	if (reloc_root->commit_root != reloc_root->node) {
+		__update_reloc_root(reloc_root);
 		btrfs_set_root_node(root_item, reloc_root->node);
 		free_extent_buffer(reloc_root->commit_root);
 		reloc_root->commit_root = btrfs_root_node(reloc_root);
@@ -4717,11 +4719,6 @@ int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
 	BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
 	       root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
 
-	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
-		if (buf == root->node)
-			__update_reloc_root(root, cow->start);
-	}
-
 	level = btrfs_header_level(buf);
 	if (btrfs_header_generation(buf) <=
 	    btrfs_root_last_snapshot(&root->root_item))
Signature · 1 line
2.20.1
author maintainer
#6 lore raw

PATCH 7/9 locking/lockdep: Avoid recursion in lockdep_count_{for,back}ward_deps()

In reply to Sasha Levin

From: Boqun Feng <[email protected]>

[ Upstream commit 25016bd7f4caf5fc983bbab7403d08e64cba3004 ]

Qian Cai reported a bug when PROVE_RCU_LIST=y, and read on /proc/lockdep triggered a warning:

[ ] DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)
...
[ ] Call Trace:

  [ ]  lock_is_held_type+0x5d/0x150
  [ ]  ? rcu_lockdep_current_cpu_online+0x64/0x80
  [ ]  rcu_read_lock_any_held+0xac/0x100
  [ ]  ? rcu_read_lock_held+0xc0/0xc0
  [ ]  ? __slab_free+0x421/0x540
  [ ]  ? kasan_kmalloc+0x9/0x10
  [ ]  ? __kmalloc_node+0x1d7/0x320
  [ ]  ? kvmalloc_node+0x6f/0x80
  [ ]  __bfs+0x28a/0x3c0
  [ ]  ? class_equal+0x30/0x30
  [ ]  lockdep_count_forward_deps+0x11a/0x1a0

The warning got triggered because lockdep_count_forward_deps() call __bfs() without current->lockdep_recursion being set, as a result a lockdep internal function (__bfs()) is checked by lockdep, which is unexpected, and the inconsistency between the irq-off state and the state traced by lockdep caused the warning.

Apart from this warning, lockdep internal functions like __bfs() should always be protected by current->lockdep_recursion to avoid potential deadlocks and data inconsistency, therefore add the current->lockdep_recursion on-and-off section to protect __bfs() in both lockdep_count_forward_deps() and lockdep_count_backward_deps()

---
 kernel/locking/lockdep.c | 4 ++++
 1 file changed, 4 insertions(+)
kernel/locking/lockdep.c +4 -0
index d7f425698a4a1..9f56e3fac795a 100644
@@ -1241,9 +1241,11 @@ unsigned long lockdep_count_forward_deps(struct lock_class *class)
 	this.class = class;
 
 	raw_local_irq_save(flags);
+	current->lockdep_recursion = 1;
 	arch_spin_lock(&lockdep_lock);
 	ret = __lockdep_count_forward_deps(&this);
 	arch_spin_unlock(&lockdep_lock);
+	current->lockdep_recursion = 0;
 	raw_local_irq_restore(flags);
 
 	return ret;
@@ -1268,9 +1270,11 @@ unsigned long lockdep_count_backward_deps(struct lock_class *class)
 	this.class = class;
 
 	raw_local_irq_save(flags);
+	current->lockdep_recursion = 1;
 	arch_spin_lock(&lockdep_lock);
 	ret = __lockdep_count_backward_deps(&this);
 	arch_spin_unlock(&lockdep_lock);
+	current->lockdep_recursion = 0;
 	raw_local_irq_restore(flags);
 
 	return ret;
Signature · 1 line
2.20.1
author maintainer
#7 lore raw

PATCH 8/9 btrfs: remove a BUG_ON() from merge_reloc_roots()

In reply to Sasha Levin

From: Josef Bacik <[email protected]>

[ Upstream commit 7b7b74315b24dc064bc1c683659061c3d48f8668 ]

This was pretty subtle, we default to reloc roots having 0 root refs, so if we crash in the middle of the relocation they can just be deleted. If we successfully complete the relocation operations we'll set our root refs to 1 in prepare_to_merge() and then go on to merge_reloc_roots().

At prepare_to_merge() time if any of the reloc roots have a 0 reference still, we will remove that reloc root from our reloc root rb tree, and then clean it up later.

However this only happens if we successfully start a transaction. If we've aborted previously we will skip this step completely, and only have reloc roots with a reference count of 0, but were never properly removed from the reloc control's rb tree.

This isn't a problem per-se, our references are held by the list the reloc roots are on, and by the original root the reloc root belongs to. If we end up in this situation all the reloc roots will be added to the dirty_reloc_list, and then properly dropped at that point. The reloc control will be free'd and the rb tree is no longer used.

There were two options when fixing this, one was to remove the BUG_ON(), the other was to make prepare_to_merge() handle the case where we couldn't start a trans handle.

IMO this is the cleaner solution. I started with handling the error in prepare_to_merge(), but it turned out super ugly. And in the end this BUG_ON() simply doesn't matter, the cleanup was happening properly, we were just panicing because this BUG_ON() only matters in the success case. So I've opted to just remove it and add a comment where it was.

---
 fs/btrfs/relocation.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)
fs/btrfs/relocation.c +15 -1
index b106d365257d3..97bc85ca508c2 100644
@@ -2457,7 +2457,21 @@ void merge_reloc_roots(struct reloc_control *rc)
 			free_reloc_roots(&reloc_roots);
 	}
 
-	BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
+	/*
+	 * We used to have
+	 *
+	 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
+	 *
+	 * here, but it's wrong.  If we fail to start the transaction in
+	 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
+	 * have actually been removed from the reloc_root_tree rb tree.  This is
+	 * fine because we're bailing here, and we hold a reference on the root
+	 * for the list that holds it, so these roots will be cleaned up when we
+	 * do the reloc_dirty_list afterwards.  Meanwhile the root->reloc_root
+	 * will be cleaned up on unmount.
+	 *
+	 * The remaining nodes will be cleaned up by free_reloc_control.
+	 */
 }
 
 static void free_block_list(struct rb_root *blocks)
Signature · 1 line
2.20.1
author maintainer
#3 lore raw

PATCH 4/9 libata: Remove extra scsi_host_put() in ata_scsi_add_hosts()

In reply to Sasha Levin

From: John Garry <[email protected]>

[ Upstream commit 1d72f7aec3595249dbb83291ccac041a2d676c57 ]

If the call to scsi_add_host_with_dma() in ata_scsi_add_hosts() fails, then we may get use-after-free KASAN warns:

==================================================================

BUG: KASAN: use-after-free in kobject_put+0x24/0x180
Read of size 1 at addr ffff0026b8c80364 by task swapper/0/1
CPU: 1 PID: 1 Comm: swapper/0 Tainted: G        W         5.6.0-rc3-00004-g5a71b206ea82-dirty #1765
Hardware name: Huawei TaiShan 200 (Model 2280)/BC82AMDD, BIOS 2280-V2 CS V3.B160.01 02/24/2020
Call trace:
dump_backtrace+0x0/0x298
show_stack+0x14/0x20
dump_stack+0x118/0x190
print_address_description.isra.9+0x6c/0x3b8
__kasan_report+0x134/0x23c
kasan_report+0xc/0x18
__asan_load1+0x5c/0x68
kobject_put+0x24/0x180
put_device+0x10/0x20
scsi_host_put+0x10/0x18
ata_devres_release+0x74/0xb0
release_nodes+0x2d0/0x470
devres_release_all+0x50/0x78
really_probe+0x2d4/0x560
driver_probe_device+0x7c/0x148
device_driver_attach+0x94/0xa0
__driver_attach+0xa8/0x110
bus_for_each_dev+0xe8/0x158
driver_attach+0x30/0x40
bus_add_driver+0x220/0x2e0
driver_register+0xbc/0x1d0
__pci_register_driver+0xbc/0xd0
ahci_pci_driver_init+0x20/0x28
do_one_initcall+0xf0/0x608
kernel_init_freeable+0x31c/0x384
kernel_init+0x10/0x118
ret_from_fork+0x10/0x18

Allocated by task 5:
save_stack+0x28/0xc8
__kasan_kmalloc.isra.8+0xbc/0xd8
kasan_kmalloc+0xc/0x18
__kmalloc+0x1a8/0x280
scsi_host_alloc+0x44/0x678
ata_scsi_add_hosts+0x74/0x268
ata_host_register+0x228/0x488
ahci_host_activate+0x1c4/0x2a8
ahci_init_one+0xd18/0x1298
local_pci_probe+0x74/0xf0
work_for_cpu_fn+0x2c/0x48
process_one_work+0x488/0xc08
worker_thread+0x330/0x5d0
kthread+0x1c8/0x1d0
ret_from_fork+0x10/0x18

Freed by task 5:
save_stack+0x28/0xc8
__kasan_slab_free+0x118/0x180
kasan_slab_free+0x10/0x18
slab_free_freelist_hook+0xa4/0x1a0
kfree+0xd4/0x3a0
scsi_host_dev_release+0x100/0x148
device_release+0x7c/0xe0
kobject_put+0xb0/0x180
put_device+0x10/0x20
scsi_host_put+0x10/0x18
ata_scsi_add_hosts+0x210/0x268
ata_host_register+0x228/0x488
ahci_host_activate+0x1c4/0x2a8
ahci_init_one+0xd18/0x1298
local_pci_probe+0x74/0xf0
work_for_cpu_fn+0x2c/0x48
process_one_work+0x488/0xc08
worker_thread+0x330/0x5d0
kthread+0x1c8/0x1d0
ret_from_fork+0x10/0x18

There is also refcount issue, as well:

WARNING: CPU: 1 PID: 1 at lib/refcount.c:28 refcount_warn_saturate+0xf8/0x170

The issue is that we make an erroneous extra call to scsi_host_put() for that host:

So in ahci_init_one()->ata_host_alloc_pinfo()->ata_host_alloc(), we setup a device release method - ata_devres_release() - which intends to release the SCSI hosts:

static void ata_devres_release(struct device *gendev, void *res)

{
	...
	for (i = 0; i < host->n_ports; i++) {
		struct ata_port *ap = host->ports[i];

		if (!ap)
			continue;

		if (ap->scsi_host)
			scsi_host_put(ap->scsi_host);

	}
	...
}

However in the ata_scsi_add_hosts() error path, we also call scsi_host_put() for the SCSI hosts.

Fix by removing the the scsi_host_put() calls in ata_scsi_add_hosts() and leave this to ata_devres_release().

Fixes f31871951b38 ("libata: separate out ata_host_alloc() and ata_host_register()")
---
 drivers/ata/libata-scsi.c | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
drivers/ata/libata-scsi.c +3 -6
index f5eb102a2cf76..c4f2b563c9f03 100644
@@ -4444,22 +4444,19 @@ int ata_scsi_add_hosts(struct ata_host *host, struct scsi_host_template *sht)
 		 */
 		shost->max_host_blocked = 1;
 
-		rc = scsi_add_host_with_dma(ap->scsi_host,
-						&ap->tdev, ap->host->dev);
+		rc = scsi_add_host_with_dma(shost, &ap->tdev, ap->host->dev);
 		if (rc)
-			goto err_add;
+			goto err_alloc;
 	}
 
 	return 0;
 
- err_add:
-	scsi_host_put(host->ports[i]->scsi_host);
  err_alloc:
 	while (--i >= 0) {
 		struct Scsi_Host *shost = host->ports[i]->scsi_host;
 
+		/* scsi_host_put() is in ata_devres_release() */
 		scsi_remove_host(shost);
-		scsi_host_put(shost);
 	}
 	return rc;
 }
Signature · 1 line
2.20.1

Sasha Levin

author maintainer
#0 lore raw

PATCH 1/9 irqchip/versatile-fpga: Handle chained IRQs properly

From: Sungbo Eo <[email protected]>

[ Upstream commit 486562da598c59e9f835b551d7cf19507de2d681 ]

Enclose the chained handler with chained_irq_{enter,exit}(), so that the muxed interrupts get properly acked.

This patch also fixes a reboot bug on OX820 SoC, where the jiffies timer interrupt is never acked. The kernel waits a clock tick forever in calibrate_delay_converge(), which leads to a boot hang.

Fixes c41b16f8c9d9 ("ARM: integrator/versatile: consolidate FPGA IRQ handling code")
---
 drivers/irqchip/irq-versatile-fpga.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)
drivers/irqchip/irq-versatile-fpga.c +10 -2
index 37dd4645bf188..66502bdefdf78 100644
@@ -5,6 +5,7 @@
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/irqchip.h>
+#include <linux/irqchip/chained_irq.h>
 #include <linux/irqchip/versatile-fpga.h>
 #include <linux/irqdomain.h>
 #include <linux/module.h>
@@ -67,12 +68,16 @@ static void fpga_irq_unmask(struct irq_data *d)
 
 static void fpga_irq_handle(struct irq_desc *desc)
 {
+	struct irq_chip *chip = irq_desc_get_chip(desc);
 	struct fpga_irq_data *f = irq_desc_get_handler_data(desc);
-	u32 status = readl(f->base + IRQ_STATUS);
+	u32 status;
+
+	chained_irq_enter(chip, desc);
 
+	status = readl(f->base + IRQ_STATUS);
 	if (status == 0) {
 		do_bad_IRQ(desc);
-		return;
+		goto out;
 	}
 
 	do {
@@ -81,6 +86,9 @@ static void fpga_irq_handle(struct irq_desc *desc)
 		status &= ~(1 << irq);
 		generic_handle_irq(irq_find_mapping(f->domain, irq));
 	} while (status);
+
+out:
+	chained_irq_exit(chip, desc);
 }
 
 /*
Signature · 1 line
2.20.1
author maintainer
#1 lore raw

PATCH 2/9 sched: Avoid scale real weight down to zero

From: Michael Wang <[email protected]>

[ Upstream commit 26cf52229efc87e2effa9d788f9b33c40fb3358a ]

During our testing, we found a case that shares no longer working correctly, the cgroup topology is like:

/sys/fs/cgroup/cpu/A (shares=102400)
/sys/fs/cgroup/cpu/A/B (shares=2)
/sys/fs/cgroup/cpu/A/B/C (shares=1024)

/sys/fs/cgroup/cpu/D (shares=1024)
/sys/fs/cgroup/cpu/D/E (shares=1024)
/sys/fs/cgroup/cpu/D/E/F (shares=1024)

The same benchmark is running in group C & F, no other tasks are running, the benchmark is capable to consumed all the CPUs.

We suppose the group C will win more CPU resources since it could enjoy all the shares of group A, but it's F who wins much more.

The reason is because we have group B with shares as 2, since A->cfs_rq.load.weight == B->se.load.weight == B->shares/nr_cpus, so A->cfs_rq.load.weight become very small.

And in calc_group_shares() we calculate shares as:

  load = max(scale_load_down(cfs_rq->load.weight), cfs_rq->avg.load_avg);
  shares = (tg_shares * load) / tg_weight;

Since the 'cfs_rq->load.weight' is too small, the load become 0 after scale down, although 'tg_shares' is 102400, shares of the se which stand for group A on root cfs_rq become 2.

While the se of D on root cfs_rq is far more bigger than 2, so it wins the battle.

Thus when scale_load_down() scale real weight down to 0, it's no longer telling the real story, the caller will have the wrong information and the calculation will be buggy.

This patch add check in scale_load_down(), so the real weight will be >= MIN_SHARES after scale, after applied the group C wins as expected.

---
 kernel/sched/sched.h | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
kernel/sched/sched.h +7 -1
index 15c08752926b0..819bd5fb02647 100644
@@ -73,7 +73,13 @@ static inline void update_idle_core(struct rq *rq) { }
 #ifdef CONFIG_64BIT
 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT)
 # define scale_load(w)		((w) << SCHED_FIXEDPOINT_SHIFT)
-# define scale_load_down(w)	((w) >> SCHED_FIXEDPOINT_SHIFT)
+# define scale_load_down(w) \
+({ \
+	unsigned long __w = (w); \
+	if (__w) \
+		__w = max(2UL, __w >> SCHED_FIXEDPOINT_SHIFT); \
+	__w; \
+})
 #else
 # define NICE_0_LOAD_SHIFT	(SCHED_FIXEDPOINT_SHIFT)
 # define scale_load(w)		(w)
Signature · 1 line
2.20.1