- 01 Apr, 2015 20 commits
-
-
Deng-Cheng Zhu authored
With sched_clock being ready, it makes sense to add the option of IRQ time accounting -- When we have a fast enough sched_clock, IRQ time accounting will be enabled (see sched_clock_register). Signed-off-by:
Deng-Cheng Zhu <dengcheng.zhu@imgtec.com> Cc: linux-mips@linux-mips.org Cc: macro@linux-mips.org Patchwork: https://patchwork.linux-mips.org/patch/9489/Signed-off-by:
Ralf Baechle <ralf@linux-mips.org>
-
Deng-Cheng Zhu authored
Use sb1250 hpt for sched_clock source. This implementation will give high resolution cputime accounting. Signed-off-by:
-
Deng-Cheng Zhu authored
This is to make checkpatch.pl happy for the next patch. It would otherwise say -- ERROR: Do not include the paragraph about writing to the Free Software Foundation's mailing address from the sample GPL notice. The FSF has changed addresses in the past, and may do so again. Linux already includes a copy of the GPL. Signed-off-by:
-
Deng-Cheng Zhu authored
This is to prepare for the upcoming read_sched_clock implementation, which will also need to get cycles from the high precision timer. Signed-off-by:
-
Deng-Cheng Zhu authored
Use jz4740 timer counter for sched_clock source. This implementation will give high resolution cputime accounting. Signed-off-by:
-
Deng-Cheng Zhu authored
Use txx9 up-counter for sched_clock source. This implementation will give high resolution cputime accounting. Signed-off-by:
-
Deng-Cheng Zhu authored
Use ip27 hub real time counter for sched_clock source. This implementation will give high resolution cputime accounting. Signed-off-by:
-
Deng-Cheng Zhu authored
Use DEC I/O ASIC's free-running counter for sched_clock source. This implementation will give high resolution cputime accounting. Acked-by:
Maciej W. Rozycki <macro@linux-mips.org> Signed-off-by:
-
Deng-Cheng Zhu authored
This is to make checkpatch.pl happy for the next patch. It would otherwise say -- ERROR: Do not include the paragraph about writing to the Free Software Foundation's mailing address from the sample GPL notice. The FSF has changed addresses in the past, and may do so again. Linux already includes a copy of the GPL. Acked-by:
-
Deng-Cheng Zhu authored
Use the ZBbus cycle counter for sched_clock source. This implementation will give high resolution cputime accounting. Signed-off-by:
-
Deng-Cheng Zhu authored
This is to make checkpatch.pl happy for the next patch. It would otherwise say -- ERROR: Do not include the paragraph about writing to the Free Software Foundation's mailing address from the sample GPL notice. The FSF has changed addresses in the past, and may do so again. Linux already includes a copy of the GPL. Signed-off-by:
-
Deng-Cheng Zhu authored
Use c0 count register for sched_clock source. This implementation will give high resolution cputime accounting. Signed-off-by:
-
Deng-Cheng Zhu authored
This will provide sched_clock interface to implement individual read_sched_clock(). Not for CAVIUM_OCTEON_SOC as it defines its own sched_clock() directly (not using the sched_clock_register interface). Signed-off-by:
-
Deng-Cheng Zhu authored
GENERIC_SCHED_CLOCK can be selected by architectures other than ARM. The current dependencies of CLKSRC_VERSATILE make it possible that other architectures will have CLKSRC_VERSATILE available in configuration once they select GENERIC_SCHED_CLOCK, whereas this clock source should be solely available to ARM in reality. This patch adds one more dependency to CLKSRC_VERSATILE to fix the issue. Signed-off-by:
-
Deng-Cheng Zhu authored
With the correct cmpxchg64 on 32-bit platforms, we can now add the config HAVE_VIRT_CPU_ACCOUNTING_GEN into arch/mips/Kconfig. Signed-off-by:
-
Deng-Cheng Zhu authored
This is in preparation of adding HAVE_VIRT_CPU_ACCOUNTING_GEN support in the next patch. Without having cmpxchg64 to use the generic implementation, kernel linking will complain: kernel/built-in.o: In function `cputime_adjust': cputime.c:(.text+0x33748): undefined reference to `__cmpxchg_called_with_bad_pointer' cputime.c:(.text+0x33810): undefined reference to `__cmpxchg_called_with_bad_pointer' Signed-off-by:
-
Deng-Cheng Zhu authored
We have HIGH_RES_TIMERS to support SCHED_HRTICK. But SCHED_HRTICK is in kernel/Kconfig.hz where HZ values unsuitable for MIPS are defined. So we simply add this config in arch/mips/Kconfig as opposed to including the whole kernel/Kconfig.hz. Signed-off-by:
-
David Daney authored
Now it is supported, so let people select it. [ralf@linux-mips.org: Folded in fix for bogus CONFIG_ kconfig symbol prefix. Issue reported by Valentin Rothberg <valentinrothberg@gmail.com>.] Signed-off-by:
David Daney <david.daney@cavium.com> Signed-off-by:
Leonid Rosenboim <lrosenboim@caviumnetworks.com> Signed-off-by:
Aleksey Makarov <aleksey.makarov@auriga.com> Cc: linux-mips@linux-mips.org Cc: linux-kernel@vger.kernel.org Patchwork: https://patchwork.linux-mips.org/patch/9592/Signed-off-by:
-
David Daney authored
Needed for little-endian ioport access. This fixes NOR flash in little-endian mode Signed-off-by:
-
David Daney authored
Compensate for the differences in the layout of in-memory bootloader information as seen from little-endian mode. Signed-off-by:
-
- 31 Mar, 2015 20 commits
-
-
Markos Chandras authored
Start the GIC counter before we try to determine its frequency. Signed-off-by:
Markos Chandras <markos.chandras@imgtec.com> Cc: linux-kernel@vger.kernel.org Cc: linux-mips@linux-mips.org Patchwork: https://patchwork.linux-mips.org/patch/9596/Signed-off-by:
-
Markos Chandras authored
Start the GIC counter after configuring the clocksource since there are no guarantees the counter will be running after a CPU reset. Signed-off-by:
-
Markos Chandras authored
We add new functions to start and stop the GIC counter since there are no guarantees the counter will be running after a CPU reset. The GIC counter is stopped by setting the 29th bit on the GIC Config register and it is started by clearing that bit. Signed-off-by:
-
James Hogan authored
Implement KGDB IO operations for MIPS Fast Debug Channel (FDC). This can be enabled via Kconfig, which also allows the channel number to be chosen. The magic sysrq hack is implemented in the TTY driver, detecting just ^C for the KGDB channel, and ^O followed by a letter for the FDC console channel. The KGDB operations are reasonably efficient thanks to the flush callback, with a 4 byte buffer being used in both directions to allow up to 4 bytes to be encoded per FDC word. Reading of data for KGDB will discard any data received on other channels, which clearly isn't ideal, but given that there is a single FIFO shared between channels we can't do much better. Signed-off-by:
James Hogan <james.hogan@imgtec.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Jason Wessel <jason.wessel@windriver.com> Cc: linux-mips@linux-mips.org Cc: kgdb-bugreport@lists.sourceforge.net Cc: linux-kernel@vger.kernel.org Patchwork: https://patchwork.linux-mips.org/patch/9147/Signed-off-by:
-
James Hogan authored
Add support for early console of MIPS Fast Debug Channel (FDC) on channel 1 with a call very early from the MIPS setup_arch(). Signed-off-by:
-
James Hogan authored
Add TTY driver and consoles for the MIPS EJTAG Fast Debug Channel (FDC), which is found on the per-CPU MIPS Common Device Mapped Memory (CDMM) bus. The FDC is a per-CPU device which is used to communicate with an EJTAG probe. RX and TX FIFOs exist, containing 32-bits of data and 4-bit channel numbers. 16 general data streams are implemented on this for TTY and console use by encoding up to 4 bytes on each 32-bit FDC word. The TTY devices are named e.g. /dev/ttyFDC3c2 for channel 2 of the FDC attached to logical CPU 3. These can be used for getting the kernel log, a login prompt, or as a GDB remote transport, all over EJTAG and without needing a serial port. It can have an interrupt to notify of when incoming data is available in the RX FIFO or when the TX FIFO is no longer full. The detection of this interrupt occurs in architecture / platform code, but it may be shared with the timer and/or performance counter interrupt. Due to the per-CPU nature of the hardware, all outgoing TTY data is written out from a kthread which is pinned to the appropriate CPU. The console is not bound to a specific CPU, so output will appear on the chosen channel on whichever CPU the code is executing on. Enable with e.g. console=fdc1 in kernel arguments. /dev/console is bound to the same channel on the boot CPU's FDC if it exists. Signed-off-by:
-
James Hogan authored
On certain cores (namely proAptiv and P5600) incoming data via a Fast Debug Channel (FDC) while the core is blocked on a wait instruction will cause the wait not to wake up even when another interrupt is received. This makes an idle target stop as soon as you send FDC data to it, until the debug probe interrupts it and restarts the wait instruction. This is worked around by avoiding using r4k_wait on these cores if CONFIG_MIPS_EJTAG_FDC_TTY is enabled (which would imply the user intends to use the FDC). [ralf@linux-mips.org: Fix conflict.] Signed-off-by:
-
James Hogan authored
Implement the weak get_c0_fdc_int() function for Malta. The Fast Debug Channel (FDC) interrupt is obtained mainly depending on whether a GIC is present. Vectored external interrupt mode isn't yet supported. Signed-off-by:
-
James Hogan authored
Add a function to the MIPS GIC driver for retrieving the Fast Debug Channel (FDC) interrupt number, similar to the existing ones for the timer and perf counter interrupts. This will be used by platform implementations of get_c0_fdc_int() if a GIC is present. A workaround exists for interAptiv and proAptiv which claim to be able to route the FDC interrupt but don't seem to be able to in practice (at least on Malta). [ralf@linux-mips.org: Fix conflict.] Signed-off-by:
-
James Hogan authored
Treat the Fast Debug Channel (FDC) interrupt the same as the timer and performance counter interrupts. Like them, the FDC IRQ is also per-VPE, and also doesn't use a per-CPU device ID yet. Per-CPU device IDs don't seem to work with IRQF_SHARED which is needed for compatibility with cores which don't route the FDC IRQ through the GIC. For hardware which routes FDC IRQs through the GIC this is something that could be added later. Signed-off-by:
-
James Hogan authored
Read the CPU IRQ line reportedly used for the Fast Debug Channel (FDC) interrupt from the IntCtl register and store it in cp0_fdc_irq where platform implementations of the new weak platform function get_c0_fdc_int() can refer to it. [ralf@linux-mips.org: Fixed conflict.] Signed-off-by:
-
James Hogan authored
Add architectural field definitions relating to the Fast Debug Channel (FDC) interrupt, namely the pending bit in Cause and the field in IntCtl to specify which CPU IRQ line the FDC interrupt is routed to. Signed-off-by:
-
James Hogan authored
Implement mips_cdmm_phys_base() for Malta, returning the physical base address 0x1fc10000 which is "typically unused". This allows the Common Device Memory Map (CDMM) region to be mapped, and devices in that region (such as the Fast Debug Channel (FDC) hardware for communication over EJTAG) to be discovered. Signed-off-by:
-
James Hogan authored
Add MIPS Common Device Memory Map (CDMM) support in the form of a bus in the standard Linux device model. Each device attached via CDMM is discoverable via an 8-bit type identifier and may contain a number of blocks of memory mapped registers in the CDMM region. IRQs are expected to be handled separately. Due to the per-cpu (per-VPE for MT cores) nature of the CDMM devices, all the driver callbacks take place from workqueues which are run on the right CPU for the device in question, so that the driver doesn't need to be as concerned about which CPU it is running on. Callbacks also exist for when CPUs are taken offline, so that any per-CPU resources used by the driver can be disabled so they don't get forcefully migrated. CDMM devices are created as children of the CPU device they are attached to. Any existing CDMM configuration by the bootloader will be inherited, however platforms wishing to enable CDMM should implement the weak mips_cdmm_phys_base() function (see asm/cdmm.h) so that the bus driver knows where it should put the CDMM region in the physical address space if the bootloader hasn't already enabled it. A mips_cdmm_early_probe() function is also provided to allow early boot or particularly low level code to set up the CDMM region and probe for a specific device type, for example early console or KGDB IO drivers for the EJTAG Fast Debug Channel (FDC) CDMM device. Signed-off-by:
-
James Hogan authored
Add architectural definitions and probing for the MIPS Common Device Memory Map (CDMM) region. When supported and enabled at a particular physical address, this region allows some number of per-CPU devices to be discovered and controlled via MMIO. A bit exists in Config3 to determine whether the feature is present, and a CDMMBase CP0 register allows the region to be enabled at a particular physical address. [ralf@linux-mips.org: Sort conflict with other patches.] Signed-off-by:
-
James Hogan authored
Before release 2 of the architecture there weren't separate interrupt pending bits for the local CPU interrupts (timer & perf counter overflow), so when they were connected to the same interrupt line the timer handler had to call the performance counter handler before knowing whether a timer interrupt was actually pending. Now another CPU local interrupt, for the Fast Debug Channel (FDC), can also be routed to an arbitrary interrupt line. It isn't scalable to keep adding cross-calls between handlers for these cases of shared interrupt lines, especially since the FDC could in theory share its interrupt line with the performance counter, timer, or both. Fortunately since release 2 of the architecture separate interrupt pending bits do exist in the Cause register. This allows local interrupts which share an interrupt line to have separate handlers using IRQF_SHARED. Unfortunately they can't easily have their own irqchip as there is no generic way to individually mask them. Enable this sharing to happen by removing the special case for when the perf count shares an IRQ with the timer. cp0_perfcount_irq and cp0_compare_irq can then be set to the same value with shared interrupt handlers registered for both of them. Pre-R2 code should be unaffected. cp0_perfcount_irq will always be -1 and the timer handler will contnue to call into the perf counter handler. Signed-off-by:
-
James Hogan authored
When requesting the performance counter overflow interrupt, pass flags which are compatible with the cevt-r4k driver, in particular IRQF_SHARED so that the two handlers can share the same IRQ. This is possible since release 2 of the architecture where there are separate pending interrupt bits for the timer interrupt and the performance counter interrupt. This will be necessary since the FDC interrupt can also be arbitrarily routed to a CPU interrupt, possibly sharing with the timer, the performance counters, or both, and it isn't scalable to have all the handlers able to call other handlers that may be on the same IRQ line. Signed-off-by:
-
James Hogan authored
When requesting the performance counter overflow interrupt, pass flags which are compatible with the cevt-r4k driver, in particular IRQF_SHARED so that the two handlers can share the same IRQ. This is possible since release 2 of the architecture where there are separate pending interrupt bits for the timer interrupt and the performance counter interrupt. This will be necessary since the FDC interrupt can also be arbitrarily routed to a CPU interrupt, possibly sharing with the timer, the performance counters, or both, and it isn't scalable to have all the handlers able to call other handlers that may be on the same IRQ line. Shared handlers must also have a unique device pointer so they can be individually removed, so &mipspmu is now passed in for that instead of NULL. Signed-off-by:
-
Ralf Baechle authored
The attempt to get gcc to generate best possible code turned c0_compare_interrupt() into a bit of Italian pasta code. Tweak for sanity. Signed-off-by: -
James Hogan authored
Make the cevt-r4k interrupt handler shared so that other interrupt handlers (specifically the performance counter overflow handler and fast debug channel interrupt handler) can share the same interrupt line. This simply imvolves returning IRQ_NONE when no timer interrupt has been handled to allow other handlers to run, and passing IRQF_SHARED when setting up the IRQ handler so that other handlers (with compatible flags) can be registered. Signed-off-by:
-
