Maintain a per-CPU global "struct pt_regs *" variable which can be used instead
of passing regs around manually through all ~1800 interrupt handlers in the
Linux kernel.
The regs pointer is used in few places, but it potentially costs both stack
space and code to pass it around. On the FRV arch, removing the regs parameter
from all the genirq function results in a 20% speed up of the IRQ exit path
(ie: from leaving timer_interrupt() to leaving do_IRQ()).
Where appropriate, an arch may override the generic storage facility and do
something different with the variable. On FRV, for instance, the address is
maintained in GR28 at all times inside the kernel as part of general exception
handling.
Having looked over the code, it appears that the parameter may be handed down
through up to twenty or so layers of functions. Consider a USB character
device attached to a USB hub, attached to a USB controller that posts its
interrupts through a cascaded auxiliary interrupt controller. A character
device driver may want to pass regs to the sysrq handler through the input
layer which adds another few layers of parameter passing.
I've build this code with allyesconfig for x86_64 and i386. I've runtested the
main part of the code on FRV and i386, though I can't test most of the drivers.
I've also done partial conversion for powerpc and MIPS - these at least compile
with minimal configurations.
This will affect all archs. Mostly the changes should be relatively easy.
Take do_IRQ(), store the regs pointer at the beginning, saving the old one:
struct pt_regs *old_regs = set_irq_regs(regs);
And put the old one back at the end:
set_irq_regs(old_regs);
Don't pass regs through to generic_handle_irq() or __do_IRQ().
In timer_interrupt(), this sort of change will be necessary:
- update_process_times(user_mode(regs));
- profile_tick(CPU_PROFILING, regs);
+ update_process_times(user_mode(get_irq_regs()));
+ profile_tick(CPU_PROFILING);
I'd like to move update_process_times()'s use of get_irq_regs() into itself,
except that i386, alone of the archs, uses something other than user_mode().
Some notes on the interrupt handling in the drivers:
(*) input_dev() is now gone entirely. The regs pointer is no longer stored in
the input_dev struct.
(*) finish_unlinks() in drivers/usb/host/ohci-q.c needs checking. It does
something different depending on whether it's been supplied with a regs
pointer or not.
(*) Various IRQ handler function pointers have been moved to type
irq_handler_t.
Signed-Off-By: David Howells <dhowells@redhat.com>
(cherry picked from 1b16e7ac850969f38b375e511e3fa2f474a33867 commit)
The follow patch fixes a problem for Matt Taggart.
The Compaq system he had (dl380?) has a SmartArray device that exposes
the 53c1510 device in both RAID and "normal" modes. The difference
is in RAID mode, the smart array driver (IIRC) should claim the
device instead of sym2 driver. Patch below prevents sym2 from
claiming the device when the RAID "daughter board" is attached.
Signed-off-by: Grant Grundler <grundler@parisc-linux.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
This ugly hack was long overdue to die.
It was a way to print out Sparc interrupts in a more freindly format,
since IRQ numbers were arbitrary opaque 32-bit integers which vectored
into PIL levels. These 32-bit integers were not necessarily in the
0-->NR_IRQS range, but the PILs they vectored to were.
The idea now is that we will increase NR_IRQS a little bit and use a
virtual<-->real IRQ number mapping scheme similar to PowerPC.
That makes this IRQ printing hack irrelevant, and furthermore only a
handful of drivers actually used __irq_itoa() making it even less
useful.
Signed-off-by: David S. Miller <davem@davemloft.net>
Various scsi drivers use scsi_cmnd.buffer and scsi_cmnd.bufflen in their
queuecommand functions. Those fields are internal storage for the
midlayer only and are used to restore the original payload after
request_buffer and request_bufflen have been overwritten for EH. Using
the buffer and bufflen fields means they do very broken things in error
handling.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
When spinlock debugging is turned on, a struct completion grows beyond the
size allowed for the scsi_pointer. So move the struct completion back onto
the stack. The additional memory barriers are to keep us from completing
a random piece of kernel stack if the command happens to complete after
the error handling has finished.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
This patch enables clustering and sets max_sectors to 0xffff to enable
reading and writing of large blocks with tapes (and large transfers with
sg). This change is needed after the sg and st drivers started using
chained bios through scsi_request_async() in 2.6.16.
Signed-off-by: Kai Makisara <kai.makisara@kolumbus.fi>
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Use wait_for_completion_timeout() instead of using a timer (as
Christoph Hellwig did for aic7xxx).
That lets me eliminate the sym_eh_wait structure; the struct completion,
the old_done pointer and the to_do flag can be folded into the sym_ucmd
(which overrides the scsi_pointer in scsi_cmnd).
The sym_eh_done() function becomes much simpler as the timeout handling
is done in sym_eh_handler() directly.
The host_lock can be unlocked earlier, and I cache the host in
a local variable to make accesses to it quicker.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Undef SYM_OPT_HANDLE_DEVICE_QUEUEING.
Call sym_put_start_queue instead of sym_start_next_ccbs.
Turn asserts into checks that we can send the command to the adapter,
and return busy from queuecommand if we can't.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
- to_do was never set to SYM_EH_DO_COMPLETE, so remove that code
- move the spinlocks inside the common error handler code path
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
We had our own code (pci_get_base_address()) to get the bus address of
a BAR. We can get this using pcibios_resource_to_bus() instead.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Most of the Kconfig options for switching between IO Port and MMIO
operations use the opposite sense from sym2. Really, this option
should be set at a chipset level rather than per-driver.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Rather than print a list of targets at driver init time, print each
disabled target as we attempt to scan it.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
The NVRAM for both Tekram and Symbios boards allows the user to set the
speed and width for individual targets. I took that code out in March
2004 when we introduced Domain Validation, but it seems there's still
a legitimate need for it in some configurations.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Now that this constant has been added to dma-mapping.h, we don't need our
own definition
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
The upper layer doesn't send these down since 2.4.x (or 2.6 in
practice), so no need to handle it. Inline sym_setup_data_pointers
into its only caller so we can fail gracefully in the case we'd get
one neverless.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Allocate the lcb in slave_alloc and free it in slave_destroy. This allows
us to remove all the code that checks to see if it's already been allocated.
From: Christoph Hellwig <hch@lst.de>
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Enabling these features causes problems with some drives, so disable
them until they're debugged
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
sym2 version 2.2.1:
- Fix MMIO BAR detection (Thanks to Bob Picco)
- Fix odd-sized transfers with a wide bus (Thanks to Larry Stephens)
- Write posting fixes (Thanks to Thibaut Varene)
- Change one of the GFP_KERNEL allocations back into a GFP_ATOMIC
- Make CCB_BA() return a script-endian address
- Move range checks and disabling of devices from the queuecommand path
to slave_alloc()
- Remove a warning in sym_setup_cdb()
- Keep a pointer to the scsi_target instead of the scsi_dev in the tcb
- Remove a check for the upper layers passing an oversized cmd
- Replace CAM_REQ_ constants with the Linux DID_ constants
- Replace CAM_DIR_ constants with the Linux DMA_ constants
- Inline sym_read_parisc_pdc() on non-parisc systems
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
There's a slight bug in the routines in that if the period requires dt,
then the routine will unconditionally set it. DT may only be set if
Wide is also set, so this turns back on the wide bit.
For domain validation to work correctly, we need to observe the wide bit
absolutely.
Acked by: Matthew Wilcox <willy@debian.org>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
scsi_cmnd->serial_number_at_timeout doesn't serve any purpose
anymore. All serial_number == serial_number_at_timeout tests
are always true in abort callbacks. Kill the field. Also, as
->pid always equals ->serial_number and ->serial_number
doesn't have any special meaning anymore, update comments
above ->serial_number accordingly. Once we remove all uses of
this field from all lldd's, this field should go.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
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!
