Message ID | 20240831113223.9627-1-jdamato@fastly.com (mailing list archive) |
---|---|
State | Changes Requested |
Delegated to: | Netdev Maintainers |
Headers | show |
Series | [net] net: napi: Make napi_defer_irqs u32 | expand |
On Sat, Aug 31, 2024 at 1:32 PM Joe Damato <jdamato@fastly.com> wrote: > > In commit 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > napi_defer_irqs was added to net_device and napi_defer_irqs_count was > added to napi_struct, both as type int. > > This value never goes below zero. Change the type for both from int to > u32, and add an overflow check to sysfs to limit the value to S32_MAX. > > Before this patch: > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > $ cat /sys/class/net/eth4/napi_defer_hard_irqs > -2147483647 > > After this patch: > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > bash: line 0: echo: write error: Numerical result out of range > > Fixes: 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > Cc: stable@kernel.org > Cc: Eric Dumazet <edumazet@google.com> > Suggested-by: Jakub Kicinski <kuba@kernel.org> > Signed-off-by: Joe Damato <jdamato@fastly.com> > --- I do not think this deserves a change to stable trees. Signed or unsigned, what is the issue ? Do you really need one extra bit ? /sys/class/net/XXXXX/tx_queue_len has a similar behavior.
On Mon, Sep 02, 2024 at 03:01:28PM +0200, Eric Dumazet wrote: > On Sat, Aug 31, 2024 at 1:32 PM Joe Damato <jdamato@fastly.com> wrote: > > > > In commit 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > > napi_defer_irqs was added to net_device and napi_defer_irqs_count was > > added to napi_struct, both as type int. > > > > This value never goes below zero. Change the type for both from int to > > u32, and add an overflow check to sysfs to limit the value to S32_MAX. > > > > Before this patch: > > > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > > $ cat /sys/class/net/eth4/napi_defer_hard_irqs > > -2147483647 > > > > After this patch: > > > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > > bash: line 0: echo: write error: Numerical result out of range > > > > Fixes: 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > > Cc: stable@kernel.org > > Cc: Eric Dumazet <edumazet@google.com> > > Suggested-by: Jakub Kicinski <kuba@kernel.org> > > Signed-off-by: Joe Damato <jdamato@fastly.com> > > --- > > I do not think this deserves a change to stable trees. OK, I can send any other revisions to -next, instead. > Signed or unsigned, what is the issue ? > > Do you really need one extra bit ? I made the maximum S32_MAX because the practical limit has always been S32_MAX. Any larger values overflow. Keeping it at S32_MAX does not change anything about existing behavior, which was my goal. Would you prefer if it was U32_MAX instead? Or are you asking me to leave it the way it is?
On Mon, Sep 2, 2024 at 6:29 PM Joe Damato <jdamato@fastly.com> wrote: > > On Mon, Sep 02, 2024 at 03:01:28PM +0200, Eric Dumazet wrote: > > On Sat, Aug 31, 2024 at 1:32 PM Joe Damato <jdamato@fastly.com> wrote: > > > > > > In commit 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > > > napi_defer_irqs was added to net_device and napi_defer_irqs_count was > > > added to napi_struct, both as type int. > > > > > > This value never goes below zero. Change the type for both from int to > > > u32, and add an overflow check to sysfs to limit the value to S32_MAX. > > > > > > Before this patch: > > > > > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > > > $ cat /sys/class/net/eth4/napi_defer_hard_irqs > > > -2147483647 > > > > > > After this patch: > > > > > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > > > bash: line 0: echo: write error: Numerical result out of range > > > > > > Fixes: 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > > > Cc: stable@kernel.org > > > Cc: Eric Dumazet <edumazet@google.com> > > > Suggested-by: Jakub Kicinski <kuba@kernel.org> > > > Signed-off-by: Joe Damato <jdamato@fastly.com> > > > --- > > > > I do not think this deserves a change to stable trees. > > OK, I can send any other revisions to -next, instead. > > > Signed or unsigned, what is the issue ? > > > > Do you really need one extra bit ? > > I made the maximum S32_MAX because the practical limit has always > been S32_MAX. Any larger values overflow. Keeping it at S32_MAX does > not change anything about existing behavior, which was my goal. > > Would you prefer if it was U32_MAX instead? > > Or are you asking me to leave it the way it is? I think this would target net-next at most, please lets avoid hassles for stable teams.
On Mon, Sep 02, 2024 at 03:01:28PM +0200, Eric Dumazet wrote: > On Sat, Aug 31, 2024 at 1:32 PM Joe Damato <jdamato@fastly.com> wrote: > > > > In commit 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > > napi_defer_irqs was added to net_device and napi_defer_irqs_count was > > added to napi_struct, both as type int. > > > > This value never goes below zero. Change the type for both from int to > > u32, and add an overflow check to sysfs to limit the value to S32_MAX. > > > > Before this patch: > > > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > > $ cat /sys/class/net/eth4/napi_defer_hard_irqs > > -2147483647 > > > > After this patch: > > > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > > bash: line 0: echo: write error: Numerical result out of range > > > > Fixes: 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > > Cc: stable@kernel.org > > Cc: Eric Dumazet <edumazet@google.com> > > Suggested-by: Jakub Kicinski <kuba@kernel.org> > > Signed-off-by: Joe Damato <jdamato@fastly.com> > > --- > > I do not think this deserves a change to stable trees. > > Signed or unsigned, what is the issue ? > > Do you really need one extra bit ? > > /sys/class/net/XXXXX/tx_queue_len has a similar behavior. Sorry, Eric, I'm not following. Are you asking me to allow u32_max for napi_defer_hard_irqs the same way tx_queue_len does and avoid overflow that way: include/linux/netdevice.h: unsigned int tx_queue_len; dev_change_tx_queue_len(..., unsigned long new_len): if (new_len != (unsigned int)new_len) return -ERANGE;
On Mon, Sep 02, 2024 at 07:00:48PM +0200, Eric Dumazet wrote: > On Mon, Sep 2, 2024 at 6:29 PM Joe Damato <jdamato@fastly.com> wrote: > > > > On Mon, Sep 02, 2024 at 03:01:28PM +0200, Eric Dumazet wrote: > > > On Sat, Aug 31, 2024 at 1:32 PM Joe Damato <jdamato@fastly.com> wrote: > > > > > > > > In commit 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > > > > napi_defer_irqs was added to net_device and napi_defer_irqs_count was > > > > added to napi_struct, both as type int. > > > > > > > > This value never goes below zero. Change the type for both from int to > > > > u32, and add an overflow check to sysfs to limit the value to S32_MAX. > > > > > > > > Before this patch: > > > > > > > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > > > > $ cat /sys/class/net/eth4/napi_defer_hard_irqs > > > > -2147483647 > > > > > > > > After this patch: > > > > > > > > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' > > > > bash: line 0: echo: write error: Numerical result out of range > > > > > > > > Fixes: 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") > > > > Cc: stable@kernel.org > > > > Cc: Eric Dumazet <edumazet@google.com> > > > > Suggested-by: Jakub Kicinski <kuba@kernel.org> > > > > Signed-off-by: Joe Damato <jdamato@fastly.com> > > > > --- > > > > > > I do not think this deserves a change to stable trees. > > > > OK, I can send any other revisions to -next, instead. > > > > > Signed or unsigned, what is the issue ? > > > > > > Do you really need one extra bit ? > > > > I made the maximum S32_MAX because the practical limit has always > > been S32_MAX. Any larger values overflow. Keeping it at S32_MAX does > > not change anything about existing behavior, which was my goal. > > > > Would you prefer if it was U32_MAX instead? > > > > Or are you asking me to leave it the way it is? > > I think this would target net-next at most, please lets avoid hassles > for stable teams. Sure, that's fine with me. I'm just not sure what you meant by your comment about the extra bit and what you are asking me to make the maximum limit? I have no preference. I just want to prevent overflow and then make the per-NAPI stuff compatible with existing sysfs code as much as possible.
diff --git a/Documentation/networking/net_cachelines/net_device.rst b/Documentation/networking/net_cachelines/net_device.rst index 70c4fb9d4e5c..d68f37f5b1f8 100644 --- a/Documentation/networking/net_cachelines/net_device.rst +++ b/Documentation/networking/net_cachelines/net_device.rst @@ -98,7 +98,7 @@ unsigned_int num_rx_queues unsigned_int real_num_rx_queues - read_mostly get_rps_cpu struct_bpf_prog* xdp_prog - read_mostly netif_elide_gro() unsigned_long gro_flush_timeout - read_mostly napi_complete_done -int napi_defer_hard_irqs - read_mostly napi_complete_done +u32 napi_defer_hard_irqs - read_mostly napi_complete_done unsigned_int gro_max_size - read_mostly skb_gro_receive unsigned_int gro_ipv4_max_size - read_mostly skb_gro_receive rx_handler_func_t* rx_handler read_mostly - __netif_receive_skb_core diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index 607009150b5f..39eafd2e2368 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -356,7 +356,7 @@ struct napi_struct { unsigned long state; int weight; - int defer_hard_irqs_count; + u32 defer_hard_irqs_count; unsigned long gro_bitmask; int (*poll)(struct napi_struct *, int); #ifdef CONFIG_NETPOLL @@ -2091,7 +2091,7 @@ struct net_device { unsigned int real_num_rx_queues; struct netdev_rx_queue *_rx; unsigned long gro_flush_timeout; - int napi_defer_hard_irqs; + u32 napi_defer_hard_irqs; unsigned int gro_max_size; unsigned int gro_ipv4_max_size; rx_handler_func_t __rcu *rx_handler; diff --git a/net/core/net-sysfs.c b/net/core/net-sysfs.c index 444f23e74f8e..b34d731524d5 100644 --- a/net/core/net-sysfs.c +++ b/net/core/net-sysfs.c @@ -32,6 +32,7 @@ #ifdef CONFIG_SYSFS static const char fmt_hex[] = "%#x\n"; static const char fmt_dec[] = "%d\n"; +static const char fmt_uint[] = "%u\n"; static const char fmt_ulong[] = "%lu\n"; static const char fmt_u64[] = "%llu\n"; @@ -425,6 +426,9 @@ NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong); static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val) { + if (val > S32_MAX) + return -ERANGE; + WRITE_ONCE(dev->napi_defer_hard_irqs, val); return 0; } @@ -438,7 +442,7 @@ static ssize_t napi_defer_hard_irqs_store(struct device *dev, return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs); } -NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec); +NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_uint); static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t len)
In commit 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") napi_defer_irqs was added to net_device and napi_defer_irqs_count was added to napi_struct, both as type int. This value never goes below zero. Change the type for both from int to u32, and add an overflow check to sysfs to limit the value to S32_MAX. Before this patch: $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' $ cat /sys/class/net/eth4/napi_defer_hard_irqs -2147483647 After this patch: $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs' bash: line 0: echo: write error: Numerical result out of range Fixes: 6f8b12d661d0 ("net: napi: add hard irqs deferral feature") Cc: stable@kernel.org Cc: Eric Dumazet <edumazet@google.com> Suggested-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Joe Damato <jdamato@fastly.com> --- Documentation/networking/net_cachelines/net_device.rst | 2 +- include/linux/netdevice.h | 4 ++-- net/core/net-sysfs.c | 6 +++++- 3 files changed, 8 insertions(+), 4 deletions(-)