[kvm-unit-tests PATCH v10 0/9] s390x: Testing the Channel Subsystem I/O

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi All,

Goal of the series is to have a framework to test Channel-Subsystem I/O with
QEMU/KVM.
  
To be able to support interrupt for CSS I/O and for SCLP we need to modify
the interrupt framework to allow re-entrant interruptions.
  
We add a registration for IRQ callbacks to the test program to define its own
interrupt handler. We need to do special work under interrupt like acknowledging
the interrupt.
  
This series presents three tests:
- Enumeration:
        The CSS is enumerated using the STSCH instruction recursively on all
        potentially existing channels.
        Keeping the first channel found as a reference for future use.
        Checks STSCH
 
- Enable:
        If the enumeration succeeded the tests enables the reference
        channel with MSCH and verifies with STSCH that the channel is
        effectively enabled, retrying a predefined count on failure
	to enable the channel
        Checks MSCH       
 
- Sense:
        If the channel is enabled this test sends a SENSE_ID command
        to the reference channel, analyzing the answer and expecting
        the Control unit type being 0x3832, a.k.a. virtio-ccw.
        Checks SSCH(READ) and IO-IRQ

Note:
- The following 5 patches are general usage and may be pulled first:
  s390x: saving regs for interrupts
  s390x: I/O interrupt registration
  s390x: export the clock get_clock_ms() utility
  s390x: clock and delays calculations
  s390x: define function to wait for interrupt

- These 4 patches are really I/O oriented:
  s390x: Library resources for CSS tests
  s390x: css: stsch, enumeration test
  s390x: css: msch, enable test
  s390x: css: ssch/tsch with sense and interrupt

Regards,
Pierre

Pierre Morel (9):
  s390x: saving regs for interrupts
  s390x: I/O interrupt registration
  s390x: export the clock get_clock_ms() utility
  s390x: clock and delays calculations
  s390x: define function to wait for interrupt
  s390x: Library resources for CSS tests
  s390x: css: stsch, enumeration test
  s390x: css: msch, enable test
  s390x: css: ssch/tsch with sense and interrupt

 lib/s390x/asm/arch_def.h |  14 ++
 lib/s390x/asm/time.h     |  50 +++++++
 lib/s390x/css.h          | 286 +++++++++++++++++++++++++++++++++++++++
 lib/s390x/css_dump.c     | 152 +++++++++++++++++++++
 lib/s390x/css_lib.c      | 277 +++++++++++++++++++++++++++++++++++++
 lib/s390x/interrupt.c    |  23 +++-
 lib/s390x/interrupt.h    |   8 ++
 s390x/Makefile           |   3 +
 s390x/css.c              | 162 ++++++++++++++++++++++
 s390x/cstart64.S         |  41 +++++-
 s390x/intercept.c        |  11 +-
 s390x/unittests.cfg      |   4 +
 12 files changed, 1018 insertions(+), 13 deletions(-)
 create mode 100644 lib/s390x/asm/time.h
 create mode 100644 lib/s390x/css.h
 create mode 100644 lib/s390x/css_dump.c
 create mode 100644 lib/s390x/css_lib.c
 create mode 100644 lib/s390x/interrupt.h
 create mode 100644 s390x/css.c

-- 
2.25.1

from v9 to v10:

- postpone and removed from series 
  - s390x: Use PSW bits definitions in cstart
  - s390x: Move control register bit definitions and add AFP to them
  reason: gcc compiler for RedHat-7 does not allow UL suffix for
          macro declarations.
  (Thomas)

- postpone and removed from series 
  -s390x-retrieve-decimal-and-hexadecimal-kernel-par.patch
  reason: must be reworked and since we can do without an argument
          I prefer to separate it from this series.
  (Janosch, Andrew)

- several word spelling, better comments
  (Connie, Thomas)

- Extensively reworked ssch/tsch with sense and interrupt
  - added a dedicated ISC to the channel subsystem test
    when enabling a channel
  - reworked the ISC enabling in CR6
  - reworked the test on residual count
  - moved the IRQ handler and the start function inside the
    library to share with following CSS tests.
  - added more report_info to the IRQ handler
  - rename some functions, start_single_ccw, css_irq_io,
    css_residual_count
  (Connie, Pierre)

from v8 to v9:

- rename PSW_EXCEPTION_MASK to PSW_MASK_ON_EXCEPTION
  (Thomas)

- changed false max microseconds in delay calculation
  (Thomas)

- fix bug in decimal parameter calculation
  (Thomas)

- fix bug in msch inline assembly
  (Thomas)

- add report_abort() for unprobable result of tsch
  in I/O IRQ
  (Thomas)

- use of existing lctlg() wrapper instead of new function
  (Thomas)

from v7 to v8

* Suppress ccw-pong specific remote device
  (Thomas, Janosch)

* use virtio-net-ccw as device instead of a specific
  device: no more need for QEMU patch
  (Connie)

* Add kernel parameter to access a different device.
  (Connie)

* Add tests on subschannel reading, length, garbage.
  (Connie)

* Several naming changes and reorganizations of definitions.
  (Connie)

* Take wrapping into account for delay calculation

* Align CCW1 on 8 bytes boundary

* Reorganize the first three patches
  (Janosch)

from v6 to v7

* s390x: saving regs for interrupts
- macro name modificatio for SAVE_REGS_STACK
  (David)
- saving the FPC
  (David)

* s390x: Use PSW bits definitions in cstart
- suppress definition for PSW_RESET_MASK
  use PSW_EXCEPTION_MASK | PSW_MASK_SHORT_PSW
  (David)

* s390x: Library resources for CSS tests
* s390x: css: stsch, enumeration test
  move library definitions from stsch patche
  to the Library patch
  add the library to the s390 Makefile here
  (Janosch)
  
* s390x: css: msch, enable test
  Add retries when enable fails
  (Connie)
  Re-introduce the patches for delay implementation
  to add a delay between retries

* s390x: define function to wait for interrupt
  Changed name from wfi to wait_for_interrupt
  (Janosch)

* s390x: css: ssch/tsch with sense and interrupt
  add a flag parameter to ssch and use it to add
  SLI (Suppress Length Indication) flag for SENSE_ID
  (Connie)


from v5 to v6
- Added comments for IRQ handling in
  s390x: saving regs for interrupts
  (Janosch) 

- fixed BUG on reset_psw, added PSW_MASK_PSW_SHORT
  and PSW_RESET_MASK

- fixed several lines over 80 chars

- fixed licenses, use GPL V2 (no more LGPL)

- replacing delay() with wfi() (Wait For Interrupt)
  during the css tests
  (suggested by Connie's comments)

- suppressing delay() induces suppressing the patch
  "s390x: export the clock get_clock_ms() utility"
  which is already reviewed but can be picked from
  the v5 series.

- changed the logic of the tests, the 4 css tests
  are always run one after the other so no need to 
  re-run enumeration and enabling at the begining
  of each tests, it has alredy been done.
  This makes code simpler.

from v4 to v5
- add a patch to explicitely define the initial_cr0
  value
  (Janosch)
- add RB from Janosh on interrupt registration
- several formating, typo correction and removing
  unnecessary initialization in "linrary resources..."
  (Janosch)
- several formating and typo corrections on
  "stsch enumeration test"
  (Connie)
- reworking the msch test
  (Connie)
- reworking of ssch test, pack the sense-id structure
  (Connie)

from v3 to v4
- add RB from David and Thomas for patchs 
  (3) irq registration and (4) clock export
- rework the PSW bit definitions
  (Thomas)
- Suppress undef DEBUG from css_dump
  (Thomas)
- rework report() functions using new scheme
  (Thomas)
- suppress un-necessary report_info()
- more spelling corrections
- add a loop around enable bit testing
  (Connie)
- rework IRQ testing
  (Connie)
- Test data addresses to be under 2G
  (Connie)

from v2 to v3:
- Rework spelling
  (Connie)
- More descriptions
  (Connie)
- use __ASSEMBLER__ preprocessing to keep
  bits definitions and C structures in the same file
  (David)
- rename the new file clock.h as time.h
  (Janosch, David?)
- use registration for the IO interruption
  (David, Thomas)
- test the SCHIB to verify it has really be modified
  (Connie)
- Lot of simplifications in the tests
  (Connie)

from v1 to v2:
- saving floating point registers (David, Janosh)
- suppress unused PSW bits defintions (Janosh)
- added Thomas reviewed-by
- style and comments modifications (Connie, Janosh)
- moved get_clock_ms() into headers and use it (Thomas)
- separate header and library utility from tests
- Suppress traces, separate tests, make better usage of reports




[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux