2.6.0-test9-mjb1
The patchset contains mainly scalability and NUMA stuff, and anything else that stops things from irritating me. It's meant to be pretty stable, not so much a testing ground for new stuff.
I'd be very interested in feedback from anyone willing to test on any platform, however large or small.
ftp://ftp.kernel.org/pub/linux/kernel/people/mbligh/2.6.0-test9/patch-2.6.0-test9-mjb1.bz2
Since 2.6.0-test8-mjb1 (~ = changed, + = added, - = dropped)
Notes:
Mostly a merge forwards.
Now in Linus' tree:
Dropped:
New:
+ autoswap Con Kolivas
Auto-tune swapiness
+ ext2_fix Andrew Morton
Fix a race in ext2
Pending:lotsa_sdsconfig_numasched
4/4 split
new kgdblist_of_lists
Hyperthreaded scheduler (Ingo Molnar) scheduler callers profiling (Anton or Bill Hartner) Child runs first (akpm)
Kexec
e1000 fixes
Update the lost timer ticks code pidmaps_nodepages (Dave Hansen)
Present in this patch:
early_printk Dave Hansen / Keith Mannthey
Allow printk before console_init
confighz Andrew Morton / Dave Hansen
Make HZ a config option of 100 Hz or 1000 Hz
config_page_offset Dave Hansen / Andrea
Make PAGE_OFFSET a config option
numameminfo Martin Bligh / Keith Mannthey
Expose NUMA meminfo information under /proc/meminfo.numa
sched_tunables Robert Love
Provide tunable parameters for the scheduler (+ NUMA scheduler)
partial_objrmap Dave McCracken
Object based rmap for filebacked pages.
spinlock_inlining Andrew Morton & Martin J. Bligh
Inline spinlocks for profiling. Made into a ugly config option by me.
lockmeter John Hawkes / Hanna Linder
Locking stats.
sched_interactive Ingo Molnar
Bugfix for interactive scheduler
local_balance_exec Martin J. Bligh
Modify balance_exec to use node-local queues when idle
tcp_speedup Martin J. Bligh
Speedup TCP (avoid double copy) as suggested by Linus
disable preempt Martin J. Bligh
I broke preempt somehow, temporarily disable it to stop accidents
ppc64 pci fix Anton Blanchard
Fix some ppc64 pci thing or other.
per_node_idt Zwane Mwaikambo
Per node IDT so we can do silly numbers of IO-APICs on NUMA-Q
aiofix2 Mingming Cao
fixed a bug in ioctx_alloc()
config_irqbal Keith Mannthey
Make irqbalance a config option
percpu_real_loadavg Dave Hansen / Martin J. Bligh
Tell me what the real load average is, and tell me per cpu.
nolock Dave McCracken
Nah, we don't like locks.
mbind_part1 Matt Dobson
Bind some memory for NUMA.
mbind_part2 Matt Dobson
Bind some more memory for NUMA.
per_node_rss Matt Dobson
Track which nodes tasks mem is on, so sched can be sensible.
pfn_to_nid Martin J. Bligh
Dance around the twisted rats nest of crap in i386 include.
gfp_node_strict Dave Hansen
Add a node strict binding as a gfp mask option
page_lock William Lee Irwin
Conditionally convert mapping->page_lock back to an rwlock
irqbal_fast Adam Litke
Balance IRQs more readily
kcg Adam Litke
Acylic call graphs from the kernel. Wheeeeeeeeeeeee!
numa_mem_equals Dave Hansen
mem= command line parameter NUMA awareness.
qlogic driver Qlogic
The qlogic driver
autoswap Con Kolivas
Auto-tune swapiness
ext2_fix Andrew Morton
Fix a race in ext2
-mjb Martin J. Bligh
Add a tag to the makefile
Quoted 23 lines · Martin J. Bligh
The patchset contains mainly scalability and NUMA stuff, and anything else that stops things from irritating me. It's meant to be pretty stable, not so much a testing ground for new stuff.
I'd be very interested in feedback from anyone willing to test on any platform, however large or small.
ftp://ftp.kernel.org/pub/linux/kernel/people/mbligh/2.6.0-test9/patch-2.6.0-test9-mjb1.bz2
Since 2.6.0-test8-mjb1 (~ = changed, + = added, - = dropped)
Notes:
Mostly a merge forwards.Now in Linus' tree:
Dropped:
New:
+ autoswap Con Kolivas
Auto-tune swapiness+
ext2_fixAndrew Morton
Fix a race in ext2
Quoted 11 lines
Pending:
lotsa_sdsconfig_numasched
4/4 split
new kgdblist_of_lists
Hyperthreaded scheduler (Ingo Molnar) scheduler callers profiling (Anton or Bill Hartner) Child runs first (akpm)
Kexec
e1000 fixes
Um... any e1000 fixes you have, please forward them to me and Intel rather than letting them languish in a tree.
Jeff
Quoted 4 lines · Jeff Garzik
Quoted 1 line · Martin J. Bligh
e1000 fixes
Um... any e1000 fixes you have, please forward them to me and Intel
rather than letting them languish in a tree.
There aren't any in there right now. The patches that Martin is referring to were probably a couple from Anton that got fixed and merged long, long ago. There's one that we keep around for ppc64, but it's not applicable to any other architectures and it's not really mainline material anyway.
I think Martin needs to update his list.
Signature · 2 lines
Dave Hansen [email protected]
Quoted 6 lines · Dave Hansen
Quoted 4 lines · Jeff Garzik
Quoted 1 line · Martin J. Bligh
e1000 fixes
Quoted 8 lines
Quoted 1 line
Um... any e1000 fixes you have, please forward them to me and Intel rather than letting them languish in a tree.
There aren't any in there right now. The patches that Martin is referring to were probably a couple from Anton that got fixed and merged long, long ago. There's one that we keep around for ppc64, but it's not applicable to any other architectures and it's not really mainline material anyway.
well, there's still this patch...
drivers/net/e1000/e1000_hw.c renamed from drivers/net/e1000/e1000_hw.c.orig +4 -4
@@ -4522,8 +4522,8 @@ uint32_t e1000_read_reg_io(struct e1000_hw *hw, uint32_t offset) { - uint32_t io_addr = hw->io_base; -uint32_t io_data = hw->io_base +4 ; +unsigned long io_addr = hw->io_base; +unsigned long io_data = hw->io_base +4 ;e1000_io_write (hw, io_addr, offset);return e1000_io_read (hw, io_data);
@@ -4542,8 +4542,8 @@ e1000_write_reg_io(struct e1000_hw *hw, uint32_t offset, uint32_t value) { - uint32_t io_addr = hw->io_base; -uint32_t io_data = hw->io_base +4 ; +unsigned long io_addr = hw->io_base; +unsigned long io_data = hw->io_base +4 ;e1000_io_write (hw, io_addr, offset);e1000_io_write (hw, io_data, value);
drivers/net/e1000/e1000_hw.h renamed from drivers/net/e1000/e1000_hw.h.orig +3 -3
@@ -317,9 +317,9 @@ void e1000_pci_clear_mwi(struct e1000_hwvoid e1000_read_pci_cfg (struct e1000_hw *hw,uint32_t reg,uint16_t * value);void e1000_write_pci_cfg (struct e1000_hw *hw,uint32_t reg,uint16_t * value);/* Port I/O is only supported on 82544 and newer */ -uint32_t e1000_io_read (struct e1000_hw *hw,uint32_t port); +uint32_t e1000_io_read (struct e1000_hw *hw,unsigned long port);uint32_t e1000_read_reg_io (struct e1000_hw *hw,uint32_t offset); -void e1000_io_write (struct e1000_hw *hw,uint32_t port,uint32_t value); +void e1000_io_write (struct e1000_hw *hw,unsigned long port,uint32_t value);void e1000_write_reg_io (struct e1000_hw *hw,uint32_t offset,uint32_t value);int32_t e1000_config_dsp_after_link_change (struct e1000_hw *hw,boolean_t link_up);int32_t e1000_set_d3_lplu_state (struct e1000_hw *hw,boolean_t active);
@@ -978,7 +978,7 @@ struct e1000_hw {e1000_ms_type master_slave;e1000_ms_type original_master_slave;e1000_ffe_config ffe_config_state; -uint32_t io_base; +unsigned long io_base;uint32_t phy_id;uint32_t phy_revision;uint32_t phy_addr;
drivers/net/e1000/e1000_main.c renamed from drivers/net/e1000/e1000_main.c.orig +2 -2
@@ -2621,13 +2621,13 @@ e1000_write_pci_cfg(struct e1000_hw *hw, }uint32_t -e1000_io_read (struct e1000_hw *hw,uint32_t port) +e1000_io_read (struct e1000_hw *hw,unsigned long port) {return inl (port); }void -e1000_io_write (struct e1000_hw *hw,uint32_t port,uint32_t value) +e1000_io_write (struct e1000_hw *hw,unsigned long port,uint32_t value) {outl (value, port); }
drivers/net/e1000/e1000_main.c e1000_write_pci_cfg(struct e1000_hw *hw, · line 2633
Quoted 7 lines · Jeff Garzik
well, there's still this patch...
voiddrivers/net/e1000/e1000_main.c renamed from drivers/net/e1000/e1000_main.c.orig +1
-1-e1000_io_write(struct e1000_hw *hw, uint32_t port, uint32_t value) +e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value) {outl (value, port); }
Whoops. I just went looking in the breakout directory and didn't see it in there. I wonder where it was hidden.
Anton, did this come from you? Did it stop some warnings or something?
Signature · 2 lines
Dave Hansen [email protected]
Quoted 14 lines · Dave Hansen
Quoted 7 lines · Jeff Garzik
well, there's still this patch...
voiddrivers/net/e1000/e1000_main.c renamed from drivers/net/e1000/e1000_main.c.orig +1
-1-e1000_io_write(struct e1000_hw *hw, uint32_t port, uint32_t value) +e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value){ outl(value, port); }Whoops. I just went looking in the breakout directory and didn't see it in there. I wonder where it was hidden.
Quoted 1 line
Anton, did this come from you? Did it stop some warnings or something?
"stop some warnings"? ;) It's obviously correct -- a port address _must_ be unsigned long. Anything less is uncivilized (and a bug).
Jeff
Quoted 5 lines
Whoops. I just went looking in the breakout directory and didn't see it in there. I wonder where it was hidden.
Anton, did this come from you? Did it stop some warnings or something?
I think it came from Scott. At one stage ppc64 had 64bit IO BARs, we have since switched to 32bit BARs but I kept the patch. As Jeff points out, the old behaviour is a bug.
Anton