Barry K. Nathan reported the following lockdep warning:
[ 197.343948] BUG: warning at kernel/lockdep.c:1856/trace_hardirqs_on()
[ 197.345928] [<c010329b>] show_trace_log_lvl+0x5b/0x105
[ 197.346359] [<c0103896>] show_trace+0x1b/0x20
[ 197.346759] [<c01038ed>] dump_stack+0x1f/0x24
[ 197.347159] [<c012efa2>] trace_hardirqs_on+0xfb/0x185
[ 197.348873] [<c029b009>] _spin_unlock_irq+0x24/0x2d
[ 197.350620] [<e09034e8>] do_tx_done+0x171/0x179 [ns83820]
[ 197.350895] [<e090445c>] ns83820_irq+0x149/0x20b [ns83820]
[ 197.351166] [<c013b4b8>] handle_IRQ_event+0x1d/0x52
[ 197.353216] [<c013c6c2>] handle_level_irq+0x97/0xe1
[ 197.355157] [<c01048c3>] do_IRQ+0x8b/0xac
[ 197.355612] [<c0102d9d>] common_interrupt+0x25/0x2c
this is caused because the ns83820 driver re-enables irq flags
in hardirq context.
While legal in theory, in practice it should only be done if the
hardware is really old and has some very high overhead in its ISR.
(such as PIO IDE)
For modern hardware, running ISRs with irqs enabled is discouraged,
because 1) new hardware is fast enough to not cause latency problems
2) allowing the nesting of hardware interrupts only 'spreads out'
the handling of the current ISR, causing extra cachemisses that would
otherwise not happen. Furthermore, on architectures where ISRs share
the kernel stacks, enabling interrupts in ISRs introduces a much
higher kernel-stack-nesting and thus kernel-stack-overflow risk.
3) not managing irq-flags via the _irqsave / _irqrestore variants
is dangerous: it's easy to forget whether one function nests inside
another, and irq flags might be mismanaged.
In the few cases where re-enabling interrupts in an ISR is considered
useful (and unavoidable), it has to be taught to the lock validator
explicitly (because the lock validator needs the "no ISR ever enables
hardirqs" artificial simplification to keep the IRQ/softirq locking
dependencies manageable).
This teaching is done via the explicit use local_irq_enable_in_hardirq().
On a stock kernel this maps to local_irq_enable(). If the lock validator
is enabled then this does not enable interrupts.
Now, the analysis of drivers/net/ns83820.c's irq flags use: the
irq-enabling in irq context seems intentional, but i dont think it's
justified. Furthermore, the driver suffers from problem #3 above too,
in ns83820_tx_timeout() it disables irqs via local_irq_save(), but
then it calls do_tx_done() which does a spin_unlock_irq(),
re-enabling for a function that does not expect it! While currently
this bug seems harmless (only some debug printout seems to be
affected by it), it's nevertheless something to be fixed.
So this patch makes the ns83820 ISR irq-flags-safe, and cleans up
do_tx_done() use and locking to avoid the ns83820_tx_timeout() bug.
From: Arjan van de Ven <arjan@linux.intel.com>
ns83820_mib_isr takes the misc_lock in IRQ context. All other places that
do this in the ISR already use _irqsave versions, make this consistent at
least. At some point in the future someone should audit the driver to see
if all _irqsave's in the ISR can go away, this is generally an iffy/fragile
proposition though; for now get it safe, simple and consistent.
From: Arjan van de Ven <arjan@linux.intel.com>
ok this is a real driver deadlock:
The ns83820 driver enabled interrupts (by unlocking the misc_lock with
_irq) while still holding the rx_info.lock, which is required to be irq
safe since it's used in the ISR like this:
writel(1, dev->base + IER);
spin_unlock_irq(&dev->misc_lock);
kick_rx(ndev);
spin_unlock_irq(&dev->rx_info.lock);
This is can cause a deadlock if an irq was pending at the first
spin_unlock_irq already, or if one would hit during kick_rx().
Simply remove the first _irq solves this
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
For messages prior to register_netdev(), prefer dev_printk() because
that prints out both our driver name and our [PCI | whatever] bus id.
Updates: 8139{cp,too}, b44, bnx2, cassini, {eepro,epic}100, fealnx,
hamachi, ne2k-pci, ns83820, pci-skeleton, r8169.
Signed-off-by: Jeff Garzik <jeff@garzik.org>
this changes if() BUG(); constructs to BUG_ON() which is
cleaner, contains unlikely() and can better optimized away.
Signed-off-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Replace all occurences of 0xff.. in calls to function pci_set_dma_mask()
and pci_set_consistant_dma_mask() with the corresponding DMA_xBIT_MASK from
linux/dma-mapping.h.
Signed-off-by: Matthias Gehre <M.Gehre@gmx.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
They deal with wrapping correctly and are nicer to read. Also make
jiffies-holding variables unsigned long.
Signed-off-by: Marcelo Feitoza Parisi <marcelo@feitoza.com.br>
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Jeff Garzik <jgarzik@pobox.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
To help in reducing the number of include dependencies, several files were
touched as they were getting needed headers indirectly for stuff they use.
Thanks also to Alan Menegotto for pointing out that net/dccp/proto.c had
linux/dccp.h include twice.
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch kills include/linux/eeprom.h .
Rationale:
- it was only used by one single driver
- even this driver didn't do anything useful with it
- most of this file are non-inline and non-static functions (sic)
This removes include/linux/eeprom.h and cleans drivers/net/ns83820.c up.
If you think eeprom.h should be used more extensively, please consider:
- the code has to be moved from the header file to a .c file
- the currently empty write function has to be implemented
- ns83820.c or any other driver should actually use it
Noone did any of these during the more than 3 years eeprom.h already
exists...
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
- added typedef unsigned int __nocast gfp_t;
- replaced __nocast uses for gfp flags with gfp_t - it gives exactly
the same warnings as far as sparse is concerned, doesn't change
generated code (from gcc point of view we replaced unsigned int with
typedef) and documents what's going on far better.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Fix implicit nocast warnings in ns83820 code, including __nocast:
drivers/net/ns83820.c:603:46: warning: implicit cast to nocast type
Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Fix implicit nocast warnings in ns83820 code:
drivers/net/ns83820.c:603:46: warning: implicit cast to nocast type
Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Use schedule_timeout_interruptible() instead of
set_current_state()/schedule_timeout() to reduce kernel size.
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Cc: Jeff Garzik <jgarzik@pobox.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Many drivers use skb->tail unnecessarily.
In these situations, the code roughly looks like:
dev = dev_alloc_skb(...);
[optional] skb_reserve(skb, ...);
... skb->tail ...
But even if the skb_reserve() happens, skb->data equals
skb->tail. So it doesn't make any sense to use anything
other than skb->data in these cases.
Another case was the s2io.c driver directly mucking with
the skb->data and skb->tail pointers. It really just wanted
to do an skb_reserve(), so that's what the code was changed
to do instead.
Another reason I'm making this change as it allows some SKB
cleanups I have planned simpler to merge. In those cleanups,
skb->head, skb->tail, and skb->end pointers are removed, and
replaced with skb->head_room and skb->tail_room integers.
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Jeff Garzik <jgarzik@pobox.com>
Use the DMA_{64,32}BIT_MASK constants from dma-mapping.h when calling
pci_set_dma_mask() or pci_set_consistent_dma_mask()
This patch includes dma-mapping.h explicitly because it caused errors
on some architectures otherwise.
See http://marc.theaimsgroup.com/?t=108001993000001&r=1&w=2 for details
Signed-off-by: Tobias Klauser <tklauser@nuerscht.ch>
Signed-off-by: Domen Puncer <domen@coderock.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Ingo Molnar