lk/linux-kernel-test
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

[MIPS] SMTC: Instant IPI replay.

Browse Source 
SMTC pseudo-interrupts between TCs are deferred and queued if the target
TC is interrupt-inhibited (IXMT). In the first SMTC prototypes, these
queued IPIs were serviced on return to user mode, or on entry into the
kernel idle loop. The INSTANT_REPLAY option dispatches them as part of
local_irq_restore() processing, which adds runtime overhead (hence the
option to turn it off), but ensures that IPIs are handled promptly even
under heavy I/O interrupt load.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This commit is contained in:
Ralf Baechle
2007-01-20 00:18:01 +00:00
parent 9ee79a3d37
commit ac8be95504
3 changed files with 70 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
arch/mips/Kconfig
View File

@@ -1568,6 +1568,20 @@ config MIPS_MT_FPAFF
depends on MIPS_MT
default y
config MIPS_MT_SMTC_INSTANT_REPLAY
bool "Low-latency Dispatch of Deferred SMTC IPIs"
depends on MIPS_MT_SMTC
default y
help
SMTC pseudo-interrupts between TCs are deferred and queued
if the target TC is interrupt-inhibited (IXMT). In the first
SMTC prototypes, these queued IPIs were serviced on return
to user mode, or on entry into the kernel idle loop. The
INSTANT_REPLAY option dispatches them as part of local_irq_restore()
processing, which adds runtime overhead (hence the option to turn
it off), but ensures that IPIs are handled promptly even under
heavy I/O interrupt load.
config MIPS_VPE_LOADER_TOM
bool "Load VPE program into memory hidden from linux"
depends on MIPS_VPE_LOADER

56
arch/mips/kernel/smtc.c
View File

@@ -1017,6 +1017,33 @@ void setup_cross_vpe_interrupts(void)
* SMTC-specific hacks invoked from elsewhere in the kernel.
*/
void smtc_ipi_replay(void)
{
/*
* To the extent that we've ever turned interrupts off,
* we may have accumulated deferred IPIs. This is subtle.
* If we use the smtc_ipi_qdepth() macro, we'll get an
* exact number - but we'll also disable interrupts
* and create a window of failure where a new IPI gets
* queued after we test the depth but before we re-enable
* interrupts. So long as IXMT never gets set, however,
* we should be OK: If we pick up something and dispatch
* it here, that's great. If we see nothing, but concurrent
* with this operation, another TC sends us an IPI, IXMT
* is clear, and we'll handle it as a real pseudo-interrupt
* and not a pseudo-pseudo interrupt.
*/
if (IPIQ[smp_processor_id()].depth > 0) {
struct smtc_ipi *pipi;
extern void self_ipi(struct smtc_ipi *);
while ((pipi = smtc_ipi_dq(&IPIQ[smp_processor_id()]))) {
self_ipi(pipi);
smtc_cpu_stats[smp_processor_id()].selfipis++;
}
}
}
void smtc_idle_loop_hook(void)
{
#ifdef SMTC_IDLE_HOOK_DEBUG
@@ -1113,29 +1140,14 @@ void smtc_idle_loop_hook(void)
if (pdb_msg != &id_ho_db_msg[0])
printk("CPU%d: %s", smp_processor_id(), id_ho_db_msg);
#endif /* SMTC_IDLE_HOOK_DEBUG */
/*
* To the extent that we've ever turned interrupts off,
* we may have accumulated deferred IPIs. This is subtle.
* If we use the smtc_ipi_qdepth() macro, we'll get an
* exact number - but we'll also disable interrupts
* and create a window of failure where a new IPI gets
* queued after we test the depth but before we re-enable
* interrupts. So long as IXMT never gets set, however,
* we should be OK: If we pick up something and dispatch
* it here, that's great. If we see nothing, but concurrent
* with this operation, another TC sends us an IPI, IXMT
* is clear, and we'll handle it as a real pseudo-interrupt
* and not a pseudo-pseudo interrupt.
*/
if (IPIQ[smp_processor_id()].depth > 0) {
struct smtc_ipi *pipi;
extern void self_ipi(struct smtc_ipi *);
if ((pipi = smtc_ipi_dq(&IPIQ[smp_processor_id()])) != NULL) {
self_ipi(pipi);
smtc_cpu_stats[smp_processor_id()].selfipis++;
}
}
/*
* Replay any accumulated deferred IPIs. If "Instant Replay"
* is in use, there should never be any.
*/
#ifndef CONFIG_MIPS_MT_SMTC_INSTANT_REPLAY
smtc_ipi_replay();
#endif /* CONFIG_MIPS_MT_SMTC_INSTANT_REPLAY */
}
void smtc_soft_dump(void)