OSDN Git Service

qmiga/qemu.git
2 years agolinux-user/hexagon: Implement setup_sigtramp
Richard Henderson [Wed, 29 Sep 2021 13:05:35 +0000 (09:05 -0400)]
linux-user/hexagon: Implement setup_sigtramp

Continue to initialize the words on the stack, as documented.
However, use the off-stack trampoline.

Reviewed-by: Taylor Simpson <tsimpson@quicinc.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210929130553.121567-9-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
2 years agolinux-user/cris: Implement setup_sigtramp
Richard Henderson [Wed, 29 Sep 2021 13:05:34 +0000 (09:05 -0400)]
linux-user/cris: Implement setup_sigtramp

Split out setup_sigreturn so that we can continue to
initialize the words on the stack, as documented.
However, use the off-stack trampoline.

Cc: Edgar E. Iglesias <edgar.iglesias@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210929130553.121567-8-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
2 years agolinux-user/alpha: Implement setup_sigtramp
Richard Henderson [Wed, 29 Sep 2021 13:05:33 +0000 (09:05 -0400)]
linux-user/alpha: Implement setup_sigtramp

Create and record the two signal trampolines.
Use them when the guest does not use ka_restorer.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210929130553.121567-7-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
2 years agolinux-user/arm: Implement setup_sigtramp
Richard Henderson [Wed, 29 Sep 2021 13:05:32 +0000 (09:05 -0400)]
linux-user/arm: Implement setup_sigtramp

Mirror what the kernel does in arch/arm/kernel/signal.h,
using the old sigframe struct in the rt sigframe struct.

Update the trampoline code to match the kernel: this uses
sp-relative accesses rather than pc-relative.

Copy the code into frame->retcode from the trampoline page.
This minimises the different cases wrt arm vs thumb vs fdpic.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20210929130553.121567-6-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
2 years agolinux-user/arm: Drop "_v2" from symbols in signal.c
Richard Henderson [Wed, 29 Sep 2021 13:05:31 +0000 (09:05 -0400)]
linux-user/arm: Drop "_v2" from symbols in signal.c

Since we no longer support "v1", there's no need to distinguish "v2".

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210929130553.121567-5-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
2 years agolinux-user/arm: Drop v1 signal frames
Richard Henderson [Wed, 29 Sep 2021 13:05:30 +0000 (09:05 -0400)]
linux-user/arm: Drop v1 signal frames

Version 2 signal frames are used from 2.6.12 and since cbc14e6f286,
we have set UNAME_MINIMUM_RELEASE to 2.6.32.

Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210929130553.121567-4-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
2 years agolinux-user/aarch64: Implement setup_sigtramp
Richard Henderson [Wed, 29 Sep 2021 13:05:29 +0000 (09:05 -0400)]
linux-user/aarch64: Implement setup_sigtramp

Create and record the rt signal trampoline.
Use it when the guest does not use SA_RESTORER.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210929130553.121567-3-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
2 years agolinux-user: Add infrastructure for a signal trampoline page
Richard Henderson [Wed, 29 Sep 2021 13:05:28 +0000 (09:05 -0400)]
linux-user: Add infrastructure for a signal trampoline page

Allocate a page to hold the signal trampoline(s).
Invoke a guest-specific hook to fill in the contents
of the page before marking it read-execute again.

Reviewed-by: Max Filippov <jcmvbkbc@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20210929130553.121567-2-richard.henderson@linaro.org>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
2 years agoMerge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20210930' into...
Peter Maydell [Thu, 30 Sep 2021 20:16:54 +0000 (21:16 +0100)]
Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20210930' into staging

target-arm queue:
 * allwinner-h3: Switch to SMC as PSCI conduit
 * arm: tcg: Adhere to SMCCC 1.3 section 5.2
 * xlnx-zcu102, xlnx-versal-virt: Support BBRAM and eFUSE devices
 * gdbstub related code cleanups
 * Don't put FPEXC and FPSID in org.gnu.gdb.arm.vfp XML
 * Use _init vs _new convention in bus creation function names
 * sabrelite: Connect SPI flash CS line to GPIO3_19

# gpg: Signature made Thu 30 Sep 2021 16:11:20 BST
# gpg:                using RSA key E1A5C593CD419DE28E8315CF3C2525ED14360CDE
# gpg:                issuer "peter.maydell@linaro.org"
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>" [ultimate]
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* remotes/pmaydell/tags/pull-target-arm-20210930: (22 commits)
  hw/arm: sabrelite: Connect SPI flash CS line to GPIO3_19
  ide: Rename ide_bus_new() to ide_bus_init()
  qbus: Rename qbus_create() to qbus_new()
  qbus: Rename qbus_create_inplace() to qbus_init()
  pci: Rename pci_root_bus_new_inplace() to pci_root_bus_init()
  ipack: Rename ipack_bus_new_inplace() to ipack_bus_init()
  scsi: Replace scsi_bus_new() with scsi_bus_init(), scsi_bus_init_named()
  target/arm: Don't put FPEXC and FPSID in org.gnu.gdb.arm.vfp XML
  target/arm: Move gdbstub related code out of helper.c
  target/arm: Fix coding style issues in gdbstub code in helper.c
  configs: Don't include 32-bit-only GDB XML in aarch64 linux configs
  docs/system/arm: xlnx-versal-virt: BBRAM and eFUSE Usage
  hw/arm: xlnx-zcu102: Add Xilinx eFUSE device
  hw/arm: xlnx-zcu102: Add Xilinx BBRAM device
  hw/arm: xlnx-versal-virt: Add Xilinx eFUSE device
  hw/arm: xlnx-versal-virt: Add Xilinx BBRAM device
  hw/nvram: Introduce Xilinx battery-backed ram
  hw/nvram: Introduce Xilinx ZynqMP eFuse device
  hw/nvram: Introduce Xilinx Versal eFuse device
  hw/nvram: Introduce Xilinx eFuse QOM
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agoMerge remote-tracking branch 'remotes/bonzini-gitlab/tags/for-upstream' into staging
Peter Maydell [Thu, 30 Sep 2021 16:38:30 +0000 (17:38 +0100)]
Merge remote-tracking branch 'remotes/bonzini-gitlab/tags/for-upstream' into staging

* SGX implementation for x86
* Miscellaneous bugfixes
* Fix dependencies from ROMs to qtests

# gpg: Signature made Thu 30 Sep 2021 14:30:35 BST
# gpg:                using RSA key F13338574B662389866C7682BFFBD25F78C7AE83
# gpg:                issuer "pbonzini@redhat.com"
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>" [full]
# gpg:                 aka "Paolo Bonzini <pbonzini@redhat.com>" [full]
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4  E2F7 7E15 100C CD36 69B1
#      Subkey fingerprint: F133 3857 4B66 2389 866C  7682 BFFB D25F 78C7 AE83

* remotes/bonzini-gitlab/tags/for-upstream: (33 commits)
  meson_options.txt: Switch the default value for the vnc option to 'auto'
  build-sys: add HAVE_IPPROTO_MPTCP
  memory: Add tracepoint for dirty sync
  memory: Name all the memory listeners
  target/i386: Fix memory leak in sev_read_file_base64()
  tests: qtest: bios-tables-test depends on the unpacked edk2 ROMs
  meson: unpack edk2 firmware even if --disable-blobs
  target/i386: Add the query-sgx-capabilities QMP command
  target/i386: Add HMP and QMP interfaces for SGX
  docs/system: Add SGX documentation to the system manual
  sgx-epc: Add the fill_device_info() callback support
  i440fx: Add support for SGX EPC
  q35: Add support for SGX EPC
  i386: acpi: Add SGX EPC entry to ACPI tables
  i386/pc: Add e820 entry for SGX EPC section(s)
  hw/i386/pc: Account for SGX EPC sections when calculating device memory
  hw/i386/fw_cfg: Set SGX bits in feature control fw_cfg accordingly
  Adjust min CPUID level to 0x12 when SGX is enabled
  i386: Propagate SGX CPUID sub-leafs to KVM
  i386: kvm: Add support for exposing PROVISIONKEY to guest
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agoMerge remote-tracking branch 'remotes/dg-gitlab/tags/ppc-for-6.2-20210930' into staging
Peter Maydell [Thu, 30 Sep 2021 15:13:04 +0000 (16:13 +0100)]
Merge remote-tracking branch 'remotes/dg-gitlab/tags/ppc-for-6.2-20210930' into staging

ppc patch queue for 2021-09-30

Here's the next batch of ppc related patches for qemu-6.2.  Highlights
are:
 * Fixes for several TCG math instructions from the El Dorado Institute
 * A number of improvements to the powernv machine type
 * Support for a new DEVICE_UNPLUG_GUEST_ERROR QAPI event from Daniel
   Barboza
 * Support for the new FORM2 PAPR NUMA representation.  This allows
   more specific NUMA distances, as well as asymmetric configurations
 * Fix for 64-bit decrementer (used on MicroWatt CPUs)
 * Assorted fixes and cleanups
 * A number of updates to MAINTAINERS

Note that the DEVICE_UNPLUG_GUEST_ERROR stuff includes changes to
files outside my normal area, but has suitable Acks.

The MAINTAINERS updates are mostly about marking minor platforms
unmaintained / orphaned, and moving some pieces away from myself and
Greg.  As we move onto other projects, we're going to need to drop
more of the ppc maintainership, though we're hoping we can avoid too
abrupt a change.

# gpg: Signature made Thu 30 Sep 2021 06:42:41 BST
# gpg:                using RSA key 75F46586AE61A66CC44E87DC6C38CACA20D9B392
# gpg: Good signature from "David Gibson <david@gibson.dropbear.id.au>" [full]
# gpg:                 aka "David Gibson (Red Hat) <dgibson@redhat.com>" [full]
# gpg:                 aka "David Gibson (ozlabs.org) <dgibson@ozlabs.org>" [full]
# gpg:                 aka "David Gibson (kernel.org) <dwg@kernel.org>" [unknown]
# Primary key fingerprint: 75F4 6586 AE61 A66C C44E  87DC 6C38 CACA 20D9 B392

* remotes/dg-gitlab/tags/ppc-for-6.2-20210930: (44 commits)
  MAINTAINERS: Demote sPAPR from "Supported" to "Maintained"
  MAINTAINERS: Add information for OpenPIC
  MAINTAINERS: Remove David & Greg as reviewers/co-maintainers of powernv
  MAINTAINERS: Orphan obscure ppc platforms
  MAINTAINERS: Remove David & Greg as reviewers for a number of boards
  MAINTAINERS: Remove machine specific files from ppc TCG CPUs entry
  spapr/xive: Fix kvm_xive_source_reset trace event
  spapr_numa.c: fixes in spapr_numa_FORM2_write_rtas_tables()
  hw/intc: openpic: Clean up the styles
  hw/intc: openpic: Drop Raven related codes
  hw/intc: openpic: Correct the reset value of IPIDR for FSL chipset
  target/ppc: Fix 64-bit decrementer
  target/ppc: Convert debug to trace events (decrementer and IRQ)
  spapr_numa.c: handle auto NUMA node with no distance info
  spapr_numa.c: FORM2 NUMA affinity support
  spapr: move FORM1 verifications to post CAS
  spapr_numa.c: rename numa_assoc_array to FORM1_assoc_array
  spapr_numa.c: parametrize FORM1 macros
  spapr_numa.c: scrap 'legacy_numa' concept
  spapr_numa.c: split FORM1 code into helpers
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agomeson_options.txt: Switch the default value for the vnc option to 'auto'
Thomas Huth [Fri, 3 Sep 2021 08:13:55 +0000 (10:13 +0200)]
meson_options.txt: Switch the default value for the vnc option to 'auto'

There is no reason why VNC should always be enabled and not be set to
the default value. We already switched the setting in the "configure"
script in commit 3a6a1256d4 ("configure: Allow vnc to get disabled with
--without-default-features"), so let's do that in meson_options.txt now,
too.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20210903081358.956267-3-thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agobuild-sys: add HAVE_IPPROTO_MPTCP
Marc-André Lureau [Tue, 7 Sep 2021 12:19:13 +0000 (16:19 +0400)]
build-sys: add HAVE_IPPROTO_MPTCP

The QAPI schema shouldn't rely on C system headers #define, but on
configure-time project #define, so we can express the build condition in
a C-independent way.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20210907121943.3498701-3-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agomemory: Add tracepoint for dirty sync
Peter Xu [Tue, 17 Aug 2021 01:37:06 +0000 (21:37 -0400)]
memory: Add tracepoint for dirty sync

Trace at memory_region_sync_dirty_bitmap() for log_sync() or global_log_sync()
on memory regions.  One trace line should suffice when it finishes, so as to
estimate the time used for each log sync process.

Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20210817013706.30986-1-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agomemory: Name all the memory listeners
Peter Xu [Tue, 17 Aug 2021 01:35:52 +0000 (21:35 -0400)]
memory: Name all the memory listeners

Provide a name field for all the memory listeners.  It can be used to identify
which memory listener is which.

Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210817013553.30584-2-peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agotarget/i386: Fix memory leak in sev_read_file_base64()
Peter Maydell [Fri, 20 Aug 2021 16:56:50 +0000 (17:56 +0100)]
target/i386: Fix memory leak in sev_read_file_base64()

In sev_read_file_base64() we call g_file_get_contents(), which
allocates memory for the file contents.  We then base64-decode the
contents (which allocates another buffer for the decoded data), but
forgot to free the memory for the original file data.

Use g_autofree to ensure that the file data is freed.

Fixes: Coverity CID 1459997
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20210820165650.2839-1-peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agotests: qtest: bios-tables-test depends on the unpacked edk2 ROMs
Paolo Bonzini [Thu, 23 Sep 2021 10:55:29 +0000 (06:55 -0400)]
tests: qtest: bios-tables-test depends on the unpacked edk2 ROMs

Skip the test if bzip2 is not available, and run it after they are
uncompressed.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210923105529.3845741-2-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agomeson: unpack edk2 firmware even if --disable-blobs
Paolo Bonzini [Thu, 23 Sep 2021 10:55:28 +0000 (06:55 -0400)]
meson: unpack edk2 firmware even if --disable-blobs

The edk2 firmware blobs are needed to run bios-tables-test.  Unpack
them if any UEFI-enabled target is selected, so that the test can run.
This is a bit more than is actually necessary, since bios-tables-test
does not run for all UEFI-enabled targets, but it is the easiest
way to write this logic.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20210923105529.3845741-1-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agotarget/i386: Add the query-sgx-capabilities QMP command
Yang Zhong [Fri, 10 Sep 2021 10:22:57 +0000 (18:22 +0800)]
target/i386: Add the query-sgx-capabilities QMP command

Libvirt can use query-sgx-capabilities to get the host
sgx capabilities to decide how to allocate SGX EPC size to VM.

Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210910102258.46648-3-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agotarget/i386: Add HMP and QMP interfaces for SGX
Yang Zhong [Fri, 10 Sep 2021 10:22:56 +0000 (18:22 +0800)]
target/i386: Add HMP and QMP interfaces for SGX

The QMP and HMP interfaces can be used by monitor or QMP tools to retrieve
the SGX information from VM side when SGX is enabled on Intel platform.

Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210910102258.46648-2-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agodocs/system: Add SGX documentation to the system manual
Sean Christopherson [Mon, 19 Jul 2021 11:21:36 +0000 (19:21 +0800)]
docs/system: Add SGX documentation to the system manual

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-34-yang.zhong@intel.com>
[Convert to reStructuredText, and adopt the standard === --- ~~~ headings
 suggested for example by Linux. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agosgx-epc: Add the fill_device_info() callback support
Yang Zhong [Mon, 19 Jul 2021 11:21:35 +0000 (19:21 +0800)]
sgx-epc: Add the fill_device_info() callback support

Since there is no fill_device_info() callback support, and when we
execute "info memory-devices" command in the monitor, the segfault
will be found.

This patch will add this callback support and "info memory-devices"
will show sgx epc memory exposed to guest. The result as below:

qemu) info memory-devices
Memory device [sgx-epc]: ""
  memaddr: 0x180000000
  size: 29360128
  memdev: /objects/mem1
Memory device [sgx-epc]: ""
  memaddr: 0x181c00000
  size: 10485760
  memdev: /objects/mem2

Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-33-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi440fx: Add support for SGX EPC
Sean Christopherson [Mon, 19 Jul 2021 11:21:24 +0000 (19:21 +0800)]
i440fx: Add support for SGX EPC

Enable SGX EPC virtualization, which is currently only support by KVM.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-22-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoq35: Add support for SGX EPC
Sean Christopherson [Mon, 19 Jul 2021 11:21:23 +0000 (19:21 +0800)]
q35: Add support for SGX EPC

Enable SGX EPC virtualization, which is currently only support by KVM.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-21-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: acpi: Add SGX EPC entry to ACPI tables
Sean Christopherson [Mon, 19 Jul 2021 11:21:22 +0000 (19:21 +0800)]
i386: acpi: Add SGX EPC entry to ACPI tables

The ACPI Device entry for SGX EPC is essentially a hack whose primary
purpose is to provide software with a way to autoprobe SGX support,
e.g. to allow software to implement SGX support as a driver.  Details
on the individual EPC sections are not enumerated through ACPI tables,
i.e. software must enumerate the EPC sections via CPUID.  Furthermore,
software expects to see only a single EPC Device in the ACPI tables
regardless of the number of EPC sections in the system.

However, several versions of Windows do rely on the ACPI tables to
enumerate the address and size of the EPC.  So, regardless of the number
of EPC sections exposed to the guest, create exactly *one* EPC device
with a _CRS entry that spans the entirety of all EPC sections (which are
guaranteed to be contiguous in Qemu).

Note, NUMA support for EPC memory is intentionally not considered as
enumerating EPC NUMA information is not yet defined for bare metal.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-20-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386/pc: Add e820 entry for SGX EPC section(s)
Sean Christopherson [Mon, 19 Jul 2021 11:21:21 +0000 (19:21 +0800)]
i386/pc: Add e820 entry for SGX EPC section(s)

Note that SGX EPC is currently guaranteed to reside in a single
contiguous chunk of memory regardless of the number of EPC sections.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-19-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agohw/i386/pc: Account for SGX EPC sections when calculating device memory
Sean Christopherson [Mon, 19 Jul 2021 11:21:20 +0000 (19:21 +0800)]
hw/i386/pc: Account for SGX EPC sections when calculating device memory

Add helpers to detect if SGX EPC exists above 4g, and if so, where SGX
EPC above 4g ends.  Use the helpers to adjust the device memory range
if SGX EPC exists above 4g.

For multiple virtual EPC sections, we just put them together physically
contiguous for the simplicity because we don't support EPC NUMA affinity
now. Once the SGX EPC NUMA support in the kernel SGX driver, we will
support this in the future.

Note that SGX EPC is currently hardcoded to reside above 4g.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-18-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agohw/i386/fw_cfg: Set SGX bits in feature control fw_cfg accordingly
Sean Christopherson [Mon, 19 Jul 2021 11:21:19 +0000 (19:21 +0800)]
hw/i386/fw_cfg: Set SGX bits in feature control fw_cfg accordingly

Request SGX an SGX Launch Control to be enabled in FEATURE_CONTROL
when the features are exposed to the guest. Our design is the SGX
Launch Control bit will be unconditionally set in FEATURE_CONTROL,
which is unlike host bios.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-17-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoAdjust min CPUID level to 0x12 when SGX is enabled
Sean Christopherson [Mon, 19 Jul 2021 11:21:18 +0000 (19:21 +0800)]
Adjust min CPUID level to 0x12 when SGX is enabled

SGX capabilities are enumerated through CPUID_0x12.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-16-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Propagate SGX CPUID sub-leafs to KVM
Sean Christopherson [Mon, 19 Jul 2021 11:21:17 +0000 (19:21 +0800)]
i386: Propagate SGX CPUID sub-leafs to KVM

The SGX sub-leafs are enumerated at CPUID 0x12.  Indices 0 and 1 are
always present when SGX is supported, and enumerate SGX features and
capabilities.  Indices >=2 are directly correlated with the platform's
EPC sections.  Because the number of EPC sections is dynamic and user
defined, the number of SGX sub-leafs is "NULL" terminated.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-15-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: kvm: Add support for exposing PROVISIONKEY to guest
Sean Christopherson [Mon, 19 Jul 2021 11:21:16 +0000 (19:21 +0800)]
i386: kvm: Add support for exposing PROVISIONKEY to guest

If the guest want to fully use SGX, the guest needs to be able to
access provisioning key. Add a new KVM_CAP_SGX_ATTRIBUTE to KVM to
support provisioning key to KVM guests.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-14-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Update SGX CPUID info according to hardware/KVM/user input
Sean Christopherson [Mon, 19 Jul 2021 11:21:15 +0000 (19:21 +0800)]
i386: Update SGX CPUID info according to hardware/KVM/user input

Expose SGX to the guest if and only if KVM is enabled and supports
virtualization of SGX.  While the majority of ENCLS can be emulated to
some degree, because SGX uses a hardware-based root of trust, the
attestation aspects of SGX cannot be emulated in software, i.e.
ultimately emulation will fail as software cannot generate a valid
quote/report.  The complexity of partially emulating SGX in Qemu far
outweighs the value added, e.g. an SGX specific simulator for userspace
applications can emulate SGX for development and testing purposes.

Note, access to the PROVISIONKEY is not yet advertised to the guest as
KVM blocks access to the PROVISIONKEY by default and requires userspace
to provide additional credentials (via ioctl()) to expose PROVISIONKEY.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-13-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Add feature control MSR dependency when SGX is enabled
Sean Christopherson [Mon, 19 Jul 2021 11:21:14 +0000 (19:21 +0800)]
i386: Add feature control MSR dependency when SGX is enabled

SGX adds multiple flags to FEATURE_CONTROL to enable SGX and Flexible
Launch Control.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-12-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Add get/set/migrate support for SGX_LEPUBKEYHASH MSRs
Sean Christopherson [Mon, 19 Jul 2021 11:21:13 +0000 (19:21 +0800)]
i386: Add get/set/migrate support for SGX_LEPUBKEYHASH MSRs

On real hardware, on systems that supports SGX Launch Control, those
MSRs are initialized to digest of Intel's signing key; on systems that
don't support SGX Launch Control, those MSRs are not available but
hardware always uses digest of Intel's signing key in EINIT.

KVM advertises SGX LC via CPUID if and only if the MSRs are writable.
Unconditionally initialize those MSRs to digest of Intel's signing key
when CPU is realized and reset to reflect the fact. This avoids
potential bug in case kvm_arch_put_registers() is called before
kvm_arch_get_registers() is called, in which case guest's virtual
SGX_LEPUBKEYHASH MSRs will be set to 0, although KVM initializes those
to digest of Intel's signing key by default, since KVM allows those MSRs
to be updated by Qemu to support live migration.

Save/restore the SGX Launch Enclave Public Key Hash MSRs if SGX Launch
Control (LC) is exposed to the guest. Likewise, migrate the MSRs if they
are writable by the guest.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Kai Huang <kai.huang@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-11-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Add SGX CPUID leaf FEAT_SGX_12_1_EAX
Sean Christopherson [Mon, 19 Jul 2021 11:21:12 +0000 (19:21 +0800)]
i386: Add SGX CPUID leaf FEAT_SGX_12_1_EAX

CPUID leaf 12_1_EAX is an Intel-defined feature bits leaf enumerating
the platform's SGX capabilities that may be utilized by an enclave, e.g.
whether or not an enclave can gain access to the provision key.
Currently there are six capabilities:

   - INIT: set when the enclave has has been initialized by EINIT.  Cannot
           be set by software, i.e. forced to zero in CPUID.
   - DEBUG: permits a debugger to read/write into the enclave.
   - MODE64BIT: the enclave runs in 64-bit mode
   - PROVISIONKEY: grants has access to the provision key
   - EINITTOKENKEY: grants access to the EINIT token key, i.e. the
                    enclave can generate EINIT tokens
   - KSS: Key Separation and Sharing enabled for the enclave.

Note that the entirety of CPUID.0x12.0x1, i.e. all registers, enumerates
the allowed ATTRIBUTES (128 bits), but only bits 31:0 are directly
exposed to the user (via FEAT_12_1_EAX).  Bits 63:32 are currently all
reserved and bits 127:64 correspond to the allowed XSAVE Feature Request
Mask, which is calculated based on other CPU features, e.g. XSAVE, MPX,
AVX, etc... and is not exposed to the user.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-10-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Add SGX CPUID leaf FEAT_SGX_12_0_EBX
Sean Christopherson [Mon, 19 Jul 2021 11:21:11 +0000 (19:21 +0800)]
i386: Add SGX CPUID leaf FEAT_SGX_12_0_EBX

CPUID leaf 12_0_EBX is an Intel-defined feature bits leaf enumerating
the platform's SGX extended capabilities.  Currently there is a single
capabilitiy:

   - EXINFO: record information about #PFs and #GPs in the enclave's SSA

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-9-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Add SGX CPUID leaf FEAT_SGX_12_0_EAX
Sean Christopherson [Mon, 19 Jul 2021 11:21:10 +0000 (19:21 +0800)]
i386: Add SGX CPUID leaf FEAT_SGX_12_0_EAX

CPUID leaf 12_0_EAX is an Intel-defined feature bits leaf enumerating
the CPU's SGX capabilities, e.g. supported SGX instruction sets.
Currently there are four enumerated capabilities:

    - SGX1 instruction set, i.e. "base" SGX
    - SGX2 instruction set for dynamic EPC management
    - ENCLV instruction set for VMM oversubscription of EPC
    - ENCLS-C instruction set for thread safe variants of ENCLS

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-8-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Add primary SGX CPUID and MSR defines
Sean Christopherson [Mon, 19 Jul 2021 11:21:09 +0000 (19:21 +0800)]
i386: Add primary SGX CPUID and MSR defines

Add CPUID defines for SGX and SGX Launch Control (LC), as well as
defines for their associated FEATURE_CONTROL MSR bits.  Define the
Launch Enclave Public Key Hash MSRs (LE Hash MSRs), which exist
when SGX LC is present (in CPUID), and are writable when SGX LC is
enabled (in FEATURE_CONTROL).

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-7-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agovl: Add sgx compound properties to expose SGX EPC sections to guest
Sean Christopherson [Tue, 28 Sep 2021 08:40:58 +0000 (10:40 +0200)]
vl: Add sgx compound properties to expose SGX EPC sections to guest

Because SGX EPC is enumerated through CPUID, EPC "devices" need to be
realized prior to realizing the vCPUs themselves, i.e. long before
generic devices are parsed and realized.  From a virtualization
perspective, the CPUID aspect also means that EPC sections cannot be
hotplugged without paravirtualizing the guest kernel (hardware does
not support hotplugging as EPC sections must be locked down during
pre-boot to provide EPC's security properties).

So even though EPC sections could be realized through the generic
-devices command, they need to be created much earlier for them to
actually be usable by the guest.  Place all EPC sections in a
contiguous block, somewhat arbitrarily starting after RAM above 4g.
Ensuring EPC is in a contiguous region simplifies calculations, e.g.
device memory base, PCI hole, etc..., allows dynamic calculation of the
total EPC size, e.g. exposing EPC to guests does not require -maxmem,
and last but not least allows all of EPC to be enumerated in a single
ACPI entry, which is expected by some kernels, e.g. Windows 7 and 8.

The new compound properties command for sgx like below:
 ......
 -object memory-backend-epc,id=mem1,size=28M,prealloc=on \
 -object memory-backend-epc,id=mem2,size=10M \
 -M sgx-epc.0.memdev=mem1,sgx-epc.1.memdev=mem2

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-6-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoi386: Add 'sgx-epc' device to expose EPC sections to guest
Sean Christopherson [Mon, 19 Jul 2021 11:21:07 +0000 (19:21 +0800)]
i386: Add 'sgx-epc' device to expose EPC sections to guest

SGX EPC is enumerated through CPUID, i.e. EPC "devices" need to be
realized prior to realizing the vCPUs themselves, which occurs long
before generic devices are parsed and realized.  Because of this,
do not allow 'sgx-epc' devices to be instantiated after vCPUS have
been created.

The 'sgx-epc' device is essentially a placholder at this time, it will
be fully implemented in a future patch along with a dedicated command
to create 'sgx-epc' devices.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-5-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoqom: Add memory-backend-epc ObjectOptions support
Yang Zhong [Mon, 19 Jul 2021 11:21:06 +0000 (19:21 +0800)]
qom: Add memory-backend-epc ObjectOptions support

Add the new 'memory-backend-epc' user creatable QOM object in
the ObjectOptions to support SGX since v6.1, or the sgx backend
object cannot bootup.

Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-4-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agohostmem: Add hostmem-epc as a backend for SGX EPC
Sean Christopherson [Mon, 19 Jul 2021 11:21:05 +0000 (19:21 +0800)]
hostmem: Add hostmem-epc as a backend for SGX EPC

EPC (Enclave Page Cahe) is a specialized type of memory used by Intel
SGX (Software Guard Extensions).  The SDM desribes EPC as:

    The Enclave Page Cache (EPC) is the secure storage used to store
    enclave pages when they are a part of an executing enclave. For an
    EPC page, hardware performs additional access control checks to
    restrict access to the page. After the current page access checks
    and translations are performed, the hardware checks that the EPC
    page is accessible to the program currently executing. Generally an
    EPC page is only accessed by the owner of the executing enclave or
    an instruction which is setting up an EPC page.

Because of its unique requirements, Linux manages EPC separately from
normal memory.  Similar to memfd, the device /dev/sgx_vepc can be
opened to obtain a file descriptor which can in turn be used to mmap()
EPC memory.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-3-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agoKconfig: Add CONFIG_SGX support
Yang Zhong [Mon, 19 Jul 2021 11:21:34 +0000 (19:21 +0800)]
Kconfig: Add CONFIG_SGX support

Add new CONFIG_SGX for sgx support in the Qemu, and the Kconfig
default enable sgx in the i386 platform.

Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Message-Id: <20210719112136.57018-32-yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agomemory: Add RAM_PROTECTED flag to skip IOMMU mappings
Sean Christopherson [Mon, 19 Jul 2021 11:21:04 +0000 (19:21 +0800)]
memory: Add RAM_PROTECTED flag to skip IOMMU mappings

Add a new RAMBlock flag to denote "protected" memory, i.e. memory that
looks and acts like RAM but is inaccessible via normal mechanisms,
including DMA.  Use the flag to skip protected memory regions when
mapping RAM for DMA in VFIO.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Yang Zhong <yang.zhong@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2 years agohw/arm: sabrelite: Connect SPI flash CS line to GPIO3_19
Xuzhou Cheng [Mon, 27 Sep 2021 14:28:25 +0000 (22:28 +0800)]
hw/arm: sabrelite: Connect SPI flash CS line to GPIO3_19

The Linux spi-imx driver does not work on QEMU. The reason is that the
state of m25p80 loops in STATE_READING_DATA state after receiving
RDSR command, the new command is ignored. Before sending a new command,
CS line should be pulled high to make the state of m25p80 back to IDLE.

Currently the SPI flash CS line is connected to the SPI controller, but
on the real board, it's connected to GPIO3_19. This matches the ecspi1
device node in the board dts.

ecspi1 node in imx6qdl-sabrelite.dtsi:
  &ecspi1 {
          cs-gpios = <&gpio3 19 GPIO_ACTIVE_LOW>;
          pinctrl-names = "default";
          pinctrl-0 = <&pinctrl_ecspi1>;
          status = "okay";

          flash: m25p80@0 {
                  compatible = "sst,sst25vf016b", "jedec,spi-nor";
                  spi-max-frequency = <20000000>;
                  reg = <0>;
          };
  };

Should connect the SSI_GPIO_CS to GPIO3_19 when adding a spi-nor to
spi1 on sabrelite machine.

Verified this patch on Linux v5.14.

Logs:
  # echo "01234567899876543210" > test
  # mtd_debug erase /dev/mtd0 0x0 0x1000
  Erased 4096 bytes from address 0x00000000 in flash
  # mtd_debug write /dev/mtdblock0 0x0 20 test
  Copied 20 bytes from test to address 0x00000000 in flash
  # mtd_debug read /dev/mtdblock0 0x0 20 test_out
  Copied 20 bytes from address 0x00000000 in flash to test_out
  # cat test_out
  01234567899876543210#

Signed-off-by: Xuzhou Cheng <xuzhou.cheng@windriver.com>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Reviewed-by: Bin Meng <bin.meng@windriver.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20210927142825.491-1-xchengl.cn@gmail.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agoide: Rename ide_bus_new() to ide_bus_init()
Peter Maydell [Thu, 23 Sep 2021 12:11:53 +0000 (13:11 +0100)]
ide: Rename ide_bus_new() to ide_bus_init()

The function ide_bus_new() does an in-place initialization.  Rename
it to ide_bus_init() to follow our _init vs _new convention.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Corey Minyard <cminyard@mvista.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Acked-by: John Snow <jsnow@redhat.com> (Feel free to merge.)
Message-id: 20210923121153.23754-7-peter.maydell@linaro.org

2 years agoqbus: Rename qbus_create() to qbus_new()
Peter Maydell [Thu, 23 Sep 2021 12:11:52 +0000 (13:11 +0100)]
qbus: Rename qbus_create() to qbus_new()

Rename the "allocate and return" qbus creation function to
qbus_new(), to bring it into line with our _init vs _new convention.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Corey Minyard <cminyard@mvista.com>
Message-id: 20210923121153.23754-6-peter.maydell@linaro.org

2 years agoqbus: Rename qbus_create_inplace() to qbus_init()
Peter Maydell [Thu, 23 Sep 2021 12:11:51 +0000 (13:11 +0100)]
qbus: Rename qbus_create_inplace() to qbus_init()

Rename qbus_create_inplace() to qbus_init(); this is more in line
with our usual naming convention for functions that in-place
initialize objects.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-id: 20210923121153.23754-5-peter.maydell@linaro.org

2 years agopci: Rename pci_root_bus_new_inplace() to pci_root_bus_init()
Peter Maydell [Thu, 23 Sep 2021 12:11:50 +0000 (13:11 +0100)]
pci: Rename pci_root_bus_new_inplace() to pci_root_bus_init()

Rename the pci_root_bus_new_inplace() function to
pci_root_bus_init(); this brings the bus type in to line with a
"_init for in-place init, _new for allocate-and-return" convention.
To do this we need to rename the implementation-internal function
that was using the pci_root_bus_init() name to
pci_root_bus_internal_init().

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-id: 20210923121153.23754-4-peter.maydell@linaro.org

2 years agoipack: Rename ipack_bus_new_inplace() to ipack_bus_init()
Peter Maydell [Thu, 23 Sep 2021 12:11:49 +0000 (13:11 +0100)]
ipack: Rename ipack_bus_new_inplace() to ipack_bus_init()

Rename ipack_bus_new_inplace() to ipack_bus_init(), to bring it in to
line with a "_init for in-place init, _new for allocate-and-return"
convention.  Drop the 'name' argument, because the only caller does
not pass in a name.  If a future caller does need to specify the bus
name, we should create an ipack_bus_init_named() function at that
point.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-id: 20210923121153.23754-3-peter.maydell@linaro.org

2 years agoscsi: Replace scsi_bus_new() with scsi_bus_init(), scsi_bus_init_named()
Peter Maydell [Thu, 23 Sep 2021 12:11:48 +0000 (13:11 +0100)]
scsi: Replace scsi_bus_new() with scsi_bus_init(), scsi_bus_init_named()

The function scsi_bus_new() creates a new SCSI bus; callers can
either pass in a name argument to specify the name of the new bus, or
they can pass in NULL to allow the bus to be given an automatically
generated unique name.  Almost all callers want to use the
autogenerated name; the only exception is the virtio-scsi device.

Taking a name argument that should almost always be NULL is an
easy-to-misuse API design -- it encourages callers to think perhaps
they should pass in some standard name like "scsi" or "scsi-bus".  We
don't do this anywhere for SCSI, but we do (incorrectly) do it for
other bus types such as i2c.

The function name also implies that it will return a newly allocated
object, when it in fact does in-place allocation.  We more commonly
name such functions foo_init(), with foo_new() being the
allocate-and-return variant.

Replace all the scsi_bus_new() callsites with either:
 * scsi_bus_init() for the usual case where the caller wants
   an autogenerated bus name
 * scsi_bus_init_named() for the rare case where the caller
   needs to specify the bus name

and document that for the _named() version it's then the caller's
responsibility to think about uniqueness of bus names.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 20210923121153.23754-2-peter.maydell@linaro.org

2 years agotarget/arm: Don't put FPEXC and FPSID in org.gnu.gdb.arm.vfp XML
Peter Maydell [Tue, 21 Sep 2021 16:29:00 +0000 (17:29 +0100)]
target/arm: Don't put FPEXC and FPSID in org.gnu.gdb.arm.vfp XML

Currently we send VFP XML which includes D0..D15 or D0..D31, plus
FPSID, FPSCR and FPEXC.  The upstream GDB tolerates this, but its
definition of this XML feature does not include FPSID or FPEXC.  In
particular, for M-profile cores there are no FPSID or FPEXC
registers, so advertising those is wrong.

Move FPSID and FPEXC into their own bit of XML which we only send for
A and R profile cores.  This brings our definition of the XML
org.gnu.gdb.arm.vfp feature into line with GDB's own (at least for
non-Neon cores...) and means we don't claim to have FPSID and FPEXC
on M-profile.

(It seems unlikely to me that any gdbstub users really care about
being able to look at FPEXC and FPSID; but we've supplied them to gdb
for a decade and it's not hard to keep doing so.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210921162901.17508-5-peter.maydell@linaro.org

2 years agotarget/arm: Move gdbstub related code out of helper.c
Peter Maydell [Tue, 21 Sep 2021 16:28:59 +0000 (17:28 +0100)]
target/arm: Move gdbstub related code out of helper.c

Currently helper.c includes some code which is part of the arm
target's gdbstub support.  This code has a better home: in gdbstub.c
and gdbstub64.c.  Move it there.

Because aarch64_fpu_gdb_get_reg() and aarch64_fpu_gdb_set_reg() move
into gdbstub64.c, this means that they're now compiled only for
TARGET_AARCH64 rather than always.  That is the only case when they
would ever be used, but it does mean that the ifdef in
arm_cpu_register_gdb_regs_for_features() needs to be adjusted to
match.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210921162901.17508-4-peter.maydell@linaro.org

2 years agotarget/arm: Fix coding style issues in gdbstub code in helper.c
Peter Maydell [Tue, 21 Sep 2021 16:28:58 +0000 (17:28 +0100)]
target/arm: Fix coding style issues in gdbstub code in helper.c

We're going to move this code to a different file; fix the coding
style first so checkpatch doesn't complain.  This includes deleting
the spurious 'break' statements after returns in the
vfp_gdb_get_reg() function.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210921162901.17508-3-peter.maydell@linaro.org

2 years agoconfigs: Don't include 32-bit-only GDB XML in aarch64 linux configs
Peter Maydell [Tue, 21 Sep 2021 16:28:57 +0000 (17:28 +0100)]
configs: Don't include 32-bit-only GDB XML in aarch64 linux configs

The aarch64-linux QEMU usermode binaries can never run 32-bit
code, so they do not need to include the GDB XML for it.
(arm_cpu_register_gdb_regs_for_features() will not use these
XML files if the CPU has ARM_FEATURE_AARCH64, so we will not
advertise to gdb that we have them.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210921162901.17508-2-peter.maydell@linaro.org

2 years agodocs/system/arm: xlnx-versal-virt: BBRAM and eFUSE Usage
Tong Ho [Fri, 17 Sep 2021 05:24:00 +0000 (22:24 -0700)]
docs/system/arm: xlnx-versal-virt: BBRAM and eFUSE Usage

Add BBRAM and eFUSE usage to the Xilinx Versal Virt board
document.

Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-10-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agohw/arm: xlnx-zcu102: Add Xilinx eFUSE device
Tong Ho [Fri, 17 Sep 2021 05:23:59 +0000 (22:23 -0700)]
hw/arm: xlnx-zcu102: Add Xilinx eFUSE device

Connect the support for ZynqMP eFUSE one-time field-programmable
bit array.

The command argument:
  -drive if=pflash,index=3,...
Can be used to optionally connect the bit array to a
backend storage, such that field-programmed values
in one invocation can be made available to next
invocation.

The backend storage must be a seekable binary file, and
its size must be 768 bytes or larger. A file with all
binary 0's is a 'blank'.

Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-9-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agohw/arm: xlnx-zcu102: Add Xilinx BBRAM device
Tong Ho [Fri, 17 Sep 2021 05:23:58 +0000 (22:23 -0700)]
hw/arm: xlnx-zcu102: Add Xilinx BBRAM device

Connect the support for Xilinx ZynqMP Battery-Backed RAM (BBRAM)

The command argument:
  -drive if=pflash,index=2,...
Can be used to optionally connect the bbram to a backend
storage, such that field-programmed values in one
invocation can be made available to next invocation.

The backend storage must be a seekable binary file, and
its size must be 36 bytes or larger. A file with all
binary 0's is a 'blank'.

Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-8-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agohw/arm: xlnx-versal-virt: Add Xilinx eFUSE device
Tong Ho [Fri, 17 Sep 2021 05:23:57 +0000 (22:23 -0700)]
hw/arm: xlnx-versal-virt: Add Xilinx eFUSE device

Connect the support for Versal eFUSE one-time field-programmable
bit array.

The command argument:
  -drive if=pflash,index=1,...
Can be used to optionally connect the bit array to a
backend storage, such that field-programmed values
in one invocation can be made available to next
invocation.

The backend storage must be a seekable binary file, and
its size must be 3072 bytes or larger. A file with all
binary 0's is a 'blank'.

Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-7-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agohw/arm: xlnx-versal-virt: Add Xilinx BBRAM device
Tong Ho [Fri, 17 Sep 2021 05:23:56 +0000 (22:23 -0700)]
hw/arm: xlnx-versal-virt: Add Xilinx BBRAM device

Connect the support for Versal Battery-Backed RAM (BBRAM)

The command argument:
  -drive if=pflash,index=0,...
Can be used to optionally connect the bbram to a backend
storage, such that field-programmed values in one
invocation can be made available to next invocation.

The backend storage must be a seekable binary file, and
its size must be 36 bytes or larger. A file with all
binary 0's is a 'blank'.

Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-6-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agohw/nvram: Introduce Xilinx battery-backed ram
Tong Ho [Fri, 17 Sep 2021 05:23:55 +0000 (22:23 -0700)]
hw/nvram: Introduce Xilinx battery-backed ram

This device is present in Versal and ZynqMP product
families to store a 256-bit encryption key.

Co-authored-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Co-authored-by: Sai Pavan Boddu <sai.pavan.boddu@xilinx.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Signed-off-by: Sai Pavan Boddu <sai.pavan.boddu@xilinx.com>
Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-5-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agohw/nvram: Introduce Xilinx ZynqMP eFuse device
Tong Ho [Fri, 17 Sep 2021 05:23:54 +0000 (22:23 -0700)]
hw/nvram: Introduce Xilinx ZynqMP eFuse device

This implements the Xilinx ZynqMP eFuse, an one-time
field-programmable non-volatile storage device.  There is
only one such device in the Xilinx ZynqMP product family.

Co-authored-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Co-authored-by: Sai Pavan Boddu <sai.pavan.boddu@xilinx.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Signed-off-by: Sai Pavan Boddu <sai.pavan.boddu@xilinx.com>
Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-4-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agohw/nvram: Introduce Xilinx Versal eFuse device
Tong Ho [Fri, 17 Sep 2021 05:23:53 +0000 (22:23 -0700)]
hw/nvram: Introduce Xilinx Versal eFuse device

This implements the Xilinx Versal eFuse, an one-time
field-programmable non-volatile storage device.  There is
only one such device in the Xilinx Versal product family.

This device has two separate mmio interfaces, a controller
and a flatten readback.

The controller provides interfaces for field-programming,
configuration, control, and status.

The flatten readback is a cache to provide a byte-accessible
read-only interface to efficiently read efuse array.

Co-authored-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Co-authored-by: Sai Pavan Boddu <sai.pavan.boddu@xilinx.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Signed-off-by: Sai Pavan Boddu <sai.pavan.boddu@xilinx.com>
Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-3-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agohw/nvram: Introduce Xilinx eFuse QOM
Tong Ho [Fri, 17 Sep 2021 05:23:52 +0000 (22:23 -0700)]
hw/nvram: Introduce Xilinx eFuse QOM

This introduces the QOM for Xilinx eFuse, an one-time
field-programmable storage bit array.

The actual mmio interface to the array varies by device
families and will be provided in different change-sets.

Co-authored-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Co-authored-by: Sai Pavan Boddu <sai.pavan.boddu@xilinx.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Signed-off-by: Sai Pavan Boddu <sai.pavan.boddu@xilinx.com>
Signed-off-by: Tong Ho <tong.ho@xilinx.com>
Message-id: 20210917052400.1249094-2-tong.ho@xilinx.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agoarm: tcg: Adhere to SMCCC 1.3 section 5.2
Alexander Graf [Thu, 16 Sep 2021 15:54:04 +0000 (17:54 +0200)]
arm: tcg: Adhere to SMCCC 1.3 section 5.2

The SMCCC 1.3 spec section 5.2 says

  The Unknown SMC Function Identifier is a sign-extended value of (-1)
  that is returned in the R0, W0 or X0 registers. An implementation must
  return this error code when it receives:

    * An SMC or HVC call with an unknown Function Identifier
    * An SMC or HVC call for a removed Function Identifier
    * An SMC64/HVC64 call from AArch32 state

To comply with these statements, let's always return -1 when we encounter
an unknown HVC or SMC call.

Signed-off-by: Alexander Graf <agraf@csgraf.de>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agoallwinner-h3: Switch to SMC as PSCI conduit
Alexander Graf [Mon, 20 Sep 2021 20:39:31 +0000 (22:39 +0200)]
allwinner-h3: Switch to SMC as PSCI conduit

The Allwinner H3 SoC uses Cortex-A7 cores which support virtualization.
However, today we are configuring QEMU to use HVC as PSCI conduit.

That means HVC calls get trapped into QEMU instead of the guest's own
emulated CPU and thus break the guest's ability to execute virtualization.

Fix this by moving to SMC as conduit, freeing up HYP completely to the VM.

Signed-off-by: Alexander Graf <agraf@csgraf.de>
Message-id: 20210920203931.66527-1-agraf@csgraf.de
Fixes: 740dafc0ba0 ("hw/arm: add Allwinner H3 System-on-Chip")
Reviewed-by: Niek Linnenbank <nieklinnenbank@gmail.com>
Tested-by: Niek Linnenbank <nieklinnenbank@gmail.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agoMerge remote-tracking branch 'remotes/ericb/tags/pull-nbd-2021-09-27-v2' into staging
Peter Maydell [Thu, 30 Sep 2021 10:28:30 +0000 (11:28 +0100)]
Merge remote-tracking branch 'remotes/ericb/tags/pull-nbd-2021-09-27-v2' into staging

nbd patches for 2021-09-27

- Vladimir Sementsov-Ogievskiy: Rework coroutines of qemu NBD client
  to improve reconnect support
- Eric Blake: Relax server in regards to NBD_OPT_LIST_META_CONTEXT
- Vladimir Sementsov-Ogievskiy: Plumb up 64-bit bulk-zeroing support
  in block layer, in preparation for future NBD spec extensions
- Nir Soffer: Default to writeback cache in qemu-nbd

# gpg: Signature made Wed 29 Sep 2021 22:07:58 BST
# gpg:                using RSA key 71C2CC22B1C4602927D2F3AAA7A16B4A2527436A
# gpg: Good signature from "Eric Blake <eblake@redhat.com>" [full]
# gpg:                 aka "Eric Blake (Free Software Programmer) <ebb9@byu.net>" [full]
# gpg:                 aka "[jpeg image of size 6874]" [full]
# Primary key fingerprint: 71C2 CC22 B1C4 6029 27D2  F3AA A7A1 6B4A 2527 436A

* remotes/ericb/tags/pull-nbd-2021-09-27-v2:
  block/nbd: check that received handle is valid
  block/nbd: drop connection_co
  block/nbd: refactor nbd_recv_coroutines_wake_all()
  block/nbd: move nbd_recv_coroutines_wake_all() up
  block/nbd: nbd_channel_error() shutdown channel unconditionally
  nbd/client-connection: nbd_co_establish_connection(): fix non set errp
  nbd/server: Allow LIST_META_CONTEXT without STRUCTURED_REPLY
  block/io: allow 64bit discard requests
  block: use int64_t instead of int in driver discard handlers
  block: make BlockLimits::max_pdiscard 64bit
  block/io: allow 64bit write-zeroes requests
  block: use int64_t instead of int in driver write_zeroes handlers
  block: make BlockLimits::max_pwrite_zeroes 64bit
  block: use int64_t instead of uint64_t in copy_range driver handlers
  block: use int64_t instead of uint64_t in driver write handlers
  block: use int64_t instead of uint64_t in driver read handlers
  qcow2: check request on vmstate save/load path
  block/io: bring request check to bdrv_co_(read,write)v_vmstate
  qemu-nbd: Change default cache mode to writeback

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agoMAINTAINERS: Demote sPAPR from "Supported" to "Maintained"
David Gibson [Tue, 21 Sep 2021 04:01:12 +0000 (14:01 +1000)]
MAINTAINERS: Demote sPAPR from "Supported" to "Maintained"

qemu/KVM on Power is no longer my primary job responsibility, nor Greg
Kurz'.  I still have some time for upstream maintenance, but it's no longer
accurate to say that I'm paid to do so.  Therefore, reduce sPAPR (the
"pseries" machine type) from Supported to Maintained.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
2 years agoMAINTAINERS: Add information for OpenPIC
David Gibson [Tue, 21 Sep 2021 03:34:35 +0000 (13:34 +1000)]
MAINTAINERS: Add information for OpenPIC

The OpenPIC interrupt controller was once the de facto standard on ppc
machines.  In qemu it's now only used on some Macintosh and the
Freescale e500 machine.  It has no listed maintainer, and as far as I
know, no-one who's really familiar with it any more.

Since I'm moving away from the area, I no longer have capacity to do even
minimal maintenance of it under the auspices of the ppc targets in general.

Therefore, this patch lists the main part of openpic, and marks it as
"Odd Fixes" to be looked after by Mark Cave-Ayland who handles the
Macintosh targets.  The openpic_kvm variant is only used on e500, so
add it to the files for that machine type (itself already Orphaned).

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
2 years agoMAINTAINERS: Remove David & Greg as reviewers/co-maintainers of powernv
David Gibson [Tue, 21 Sep 2021 03:20:24 +0000 (13:20 +1000)]
MAINTAINERS: Remove David & Greg as reviewers/co-maintainers of powernv

With our interests moving to other areas, Greg and myself no longer have
capacity to be regular reviewers of code for the powernv machine type, let
alone co-maintainers.  Additionally, not being IBM employees, we don't have
easy access to the hardware information we'd need for good review.

Therefore, remove our names as reviewers and/or co-maintainers of the
powernv machine type, and the related XIVE interrupt controller.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Acked-by: Cédric Le Goater <clg@kaod.org>
2 years agoMAINTAINERS: Orphan obscure ppc platforms
David Gibson [Tue, 21 Sep 2021 03:02:17 +0000 (13:02 +1000)]
MAINTAINERS: Orphan obscure ppc platforms

There are a nunber of old embedded ppc machine types which have been little
changed and in "Odd Fixes" state for a long time.  With both myself and
Greg Kurz moving toward other areas, we no longer have the capacity to
keep reviewing and maintaining even the rare patches that come in for those
platforms.

Therefore, remove our names as reviewers and mark these platforms as
orphaned.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
2 years agoMAINTAINERS: Remove David & Greg as reviewers for a number of boards
David Gibson [Tue, 21 Sep 2021 03:13:45 +0000 (13:13 +1000)]
MAINTAINERS: Remove David & Greg as reviewers for a number of boards

Greg and I are moving towards other areas and no longer have capacity to
act as regular reviewers for several of the secondary ppc machine types.
So, remove ourselves as reviewers for Macintosh, PReP, sam460ex and
pegasos2 in MAINTAINERS.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Acked-by: BALATON Zoltan <balaton@eik.bme.hu>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
2 years agoMAINTAINERS: Remove machine specific files from ppc TCG CPUs entry
David Gibson [Mon, 27 Sep 2021 04:17:45 +0000 (14:17 +1000)]
MAINTAINERS: Remove machine specific files from ppc TCG CPUs entry

Currently the PowerPC TCG CPUs entry in MAINTAINERS lists all of hw/ppc/
and include/hw/ppc.  Nearly all the files in those places are related to
specific ppc machine types, rather than to the actual CPUs however.  Those
machine types list their own files separately, often overlapping with this.
For greater clarity, remove these misleading entries from the TCG CPUs
stanza, leaving just hw/ppc/ppc.c and hw/ppc/ppc_booke.c which are the only
ones common to a wide range of PPC TCG cpus each.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr/xive: Fix kvm_xive_source_reset trace event
Cédric Le Goater [Wed, 22 Sep 2021 07:02:05 +0000 (09:02 +0200)]
spapr/xive: Fix kvm_xive_source_reset trace event

The trace event was placed in the wrong routine. Move it under
kvmppc_xive_source_reset_one().

Fixes: 4e960974d4ee ("xive: Add trace events")
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20210922070205.1235943-1-clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr_numa.c: fixes in spapr_numa_FORM2_write_rtas_tables()
Daniel Henrique Barboza [Wed, 22 Sep 2021 12:28:52 +0000 (09:28 -0300)]
spapr_numa.c: fixes in spapr_numa_FORM2_write_rtas_tables()

This patch has a handful of modifications for the recent added
FORM2 support:

- to not allocate more than the necessary size in 'distance_table'.
At this moment the array is oversized due to allocating uint32_t for
all elements, when most of them fits in an uint8_t. Fix it by
changing the array to uint8_t and allocating the exact size;

- use stl_be_p() to store the uint32_t at the start of 'distance_table';

- use sizeof(uint32_t) to skip the uint32_t length when populating the
distances;

- use the NUMA_DISTANCE_MIN macro from sysemu/numa.h to avoid hardcoding
the local distance value.

Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210922122852.130054-2-danielhb413@gmail.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agohw/intc: openpic: Clean up the styles
Bin Meng [Sat, 18 Sep 2021 03:26:53 +0000 (11:26 +0800)]
hw/intc: openpic: Clean up the styles

Correct the multi-line comment format. No functional changes.

Signed-off-by: Bin Meng <bin.meng@windriver.com>
Message-Id: <20210918032653.646370-3-bin.meng@windriver.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agohw/intc: openpic: Drop Raven related codes
Bin Meng [Sat, 18 Sep 2021 03:26:52 +0000 (11:26 +0800)]
hw/intc: openpic: Drop Raven related codes

There is no machine that uses Motorola MCP750 (aka Raven) model.
Drop the related codes.

While we are here, drop the mentioning of Intel GW80314 I/O
companion chip in the comments as it has been obsolete for years,
and correct a typo too.

Signed-off-by: Bin Meng <bin.meng@windriver.com>
Message-Id: <20210918032653.646370-2-bin.meng@windriver.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agohw/intc: openpic: Correct the reset value of IPIDR for FSL chipset
Bin Meng [Sat, 18 Sep 2021 03:26:51 +0000 (11:26 +0800)]
hw/intc: openpic: Correct the reset value of IPIDR for FSL chipset

The reset value of IPIDR should be zero for Freescale chipset, per
the following 2 manuals I checked:

- P2020RM (https://www.nxp.com/webapp/Download?colCode=P2020RM)
- P4080RM (https://www.nxp.com/webapp/Download?colCode=P4080RM)

Currently it is set to 1, which leaves the IPI enabled on core 0
after power-on reset. Such may cause unexpected interrupt to be
delivered to core 0 if the IPI is triggered from core 0 to other
cores later.

Fixes: ffd5e9fe0276 ("openpic: Reset IRQ source private members")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/584
Signed-off-by: Bin Meng <bin.meng@windriver.com>
Message-Id: <20210918032653.646370-1-bin.meng@windriver.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agotarget/ppc: Fix 64-bit decrementer
Cédric Le Goater [Mon, 20 Sep 2021 06:12:03 +0000 (08:12 +0200)]
target/ppc: Fix 64-bit decrementer

The current way the mask is built can overflow with a 64-bit decrementer.
Use sextract64() to extract the signed values and remove the logic to
handle negative values which has become useless.

Cc: Luis Fernando Fujita Pires <luis.pires@eldorado.org.br>
Fixes: a8dafa525181 ("target/ppc: Implement large decrementer support for TCG")
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20210920061203.989563-5-clg@kaod.org>
Reviewed-by: Luis Pires <luis.pires@eldorado.org.br>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agotarget/ppc: Convert debug to trace events (decrementer and IRQ)
Cédric Le Goater [Mon, 20 Sep 2021 06:12:02 +0000 (08:12 +0200)]
target/ppc: Convert debug to trace events (decrementer and IRQ)

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20210920061203.989563-4-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr_numa.c: handle auto NUMA node with no distance info
Daniel Henrique Barboza [Mon, 20 Sep 2021 17:49:47 +0000 (14:49 -0300)]
spapr_numa.c: handle auto NUMA node with no distance info

numa_complete_configuration() in hw/core/numa.c always adds a NUMA node
for the pSeries machine if none was specified, but without node distance
information for the single node created.

NUMA FORM1 affinity code didn't rely on numa_state information to do its
job, but FORM2 does. As is now, this is the result of a pSeries guest
with NUMA FORM2 affinity when no NUMA nodes is specified:

$ numactl -H
available: 1 nodes (0)
node 0 cpus: 0
node 0 size: 16222 MB
node 0 free: 15681 MB
No distance information available.

This can be amended in spapr_numa_FORM2_write_rtas_tables(). We're
enforcing that the local distance (the distance to the node to itself) is
always 10. This allows for the proper creation of the NUMA distance tables,
fixing the output of 'numactl -H' in the guest:

$ numactl -H
available: 1 nodes (0)
node 0 cpus: 0
node 0 size: 16222 MB
node 0 free: 15685 MB
node distances:
node   0
  0:  10

CC: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-8-danielhb413@gmail.com>
Acked-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr_numa.c: FORM2 NUMA affinity support
Daniel Henrique Barboza [Mon, 20 Sep 2021 17:49:46 +0000 (14:49 -0300)]
spapr_numa.c: FORM2 NUMA affinity support

The main feature of FORM2 affinity support is the separation of NUMA
distances from ibm,associativity information. This allows for a more
flexible and straightforward NUMA distance assignment without relying on
complex associations between several levels of NUMA via
ibm,associativity matches. Another feature is its extensibility. This base
support contains the facilities for NUMA distance assignment, but in the
future more facilities will be added for latency, performance, bandwidth
and so on.

This patch implements the base FORM2 affinity support as follows:

- the use of FORM2 associativity is indicated by using bit 2 of byte 5
of ibm,architecture-vec-5. A FORM2 aware guest can choose to use FORM1
or FORM2 affinity. Setting both forms will default to FORM2. We're not
advertising FORM2 for pseries-6.1 and older machine versions to prevent
guest visible changes in those;

- ibm,associativity-reference-points has a new semantic. Instead of
being used to calculate distances via NUMA levels, it's now used to
indicate the primary domain index in the ibm,associativity domain of
each resource. In our case it's set to {0x4}, matching the position
where we already place logical_domain_id;

- two new RTAS DT artifacts are introduced: ibm,numa-lookup-index-table
and ibm,numa-distance-table. The index table is used to list all the
NUMA logical domains of the platform, in ascending order, and allows for
spartial NUMA configurations (although QEMU ATM doesn't support that).
ibm,numa-distance-table is an array that contains all the distances from
the first NUMA node to all other nodes, then the second NUMA node
distances to all other nodes and so on;

- get_max_dist_ref_points(), get_numa_assoc_size() and get_associativity()
now checks for OV5_FORM2_AFFINITY and returns FORM2 values if the guest
selected FORM2 affinity during CAS.

Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-7-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr: move FORM1 verifications to post CAS
Daniel Henrique Barboza [Mon, 20 Sep 2021 17:49:45 +0000 (14:49 -0300)]
spapr: move FORM1 verifications to post CAS

FORM2 NUMA affinity is prepared to deal with empty (memory/cpu less)
NUMA nodes. This is used by the DAX KMEM driver to locate a PAPR SCM
device that has a different latency than the original NUMA node from the
regular memory. FORM2 is also able  to deal with asymmetric NUMA
distances gracefully, something that our FORM1 implementation doesn't
do.

Move these FORM1 verifications to a new function and wait until after
CAS, when we're sure that we're sticking with FORM1, to enforce them.

Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-6-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr_numa.c: rename numa_assoc_array to FORM1_assoc_array
Daniel Henrique Barboza [Mon, 20 Sep 2021 17:49:44 +0000 (14:49 -0300)]
spapr_numa.c: rename numa_assoc_array to FORM1_assoc_array

Introducing a new NUMA affinity, FORM2, requires a new mechanism to
switch between affinity modes after CAS. Also, we want FORM2 data
structures and functions to be completely separated from the existing
FORM1 code, allowing us to avoid adding new code that inherits the
existing complexity of FORM1.

The idea of switching values used by the write_dt() functions in
spapr_numa.c was already introduced in the previous patch, and
the same approach will be used when dealing with the FORM1 and FORM2
arrays.

We can accomplish that by that by renaming the existing numa_assoc_array
to FORM1_assoc_array, which now is used exclusively to handle FORM1 affinity
data. A new helper get_associativity() is then introduced to be used by the
write_dt() functions to retrieve the current ibm,associativity array of
a given node, after considering affinity selection that might have been
done during CAS. All code that was using numa_assoc_array now needs to
retrieve the array by calling this function.

This will allow for an easier plug of FORM2 data later on.

Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-5-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr_numa.c: parametrize FORM1 macros
Daniel Henrique Barboza [Mon, 20 Sep 2021 17:49:43 +0000 (14:49 -0300)]
spapr_numa.c: parametrize FORM1 macros

The next preliminary step to introduce NUMA FORM2 affinity is to make
the existing code independent of FORM1 macros and values, i.e.
MAX_DISTANCE_REF_POINTS, NUMA_ASSOC_SIZE and VCPU_ASSOC_SIZE. This patch
accomplishes that by doing the following:

- move the NUMA related macros from spapr.h to spapr_numa.c where they
are used. spapr.h gets instead a 'NUMA_NODES_MAX_NUM' macro that is used
to refer to the maximum number of NUMA nodes, including GPU nodes, that
the machine can support;

- MAX_DISTANCE_REF_POINTS and NUMA_ASSOC_SIZE are renamed to
FORM1_DIST_REF_POINTS and FORM1_NUMA_ASSOC_SIZE. These FORM1 specific
macros are used in FORM1 init functions;

- code that uses MAX_DISTANCE_REF_POINTS now retrieves the
max_dist_ref_points value using get_max_dist_ref_points().
NUMA_ASSOC_SIZE is replaced by get_numa_assoc_size() and VCPU_ASSOC_SIZE
is replaced by get_vcpu_assoc_size(). These functions are used by the
generic device tree functions and h_home_node_associativity() and will
allow them to switch between FORM1 and FORM2 without changing their core
logic.

Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-4-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr_numa.c: scrap 'legacy_numa' concept
Daniel Henrique Barboza [Mon, 20 Sep 2021 17:49:42 +0000 (14:49 -0300)]
spapr_numa.c: scrap 'legacy_numa' concept

When first introduced, 'legacy_numa' was a way to refer to guests that
either wouldn't be affected by associativity domain calculations, namely
the ones with only 1 NUMA node, and pre 5.2 guests that shouldn't be
affected by it because it would be an userspace change. Calling these
cases 'legacy_numa' was a convenient way to label these cases.

We're about to introduce a new NUMA affinity, FORM2, and this concept
of 'legacy_numa' is now a bit misleading because, although it is called
'legacy' it is in fact a FORM1 exclusive contraint.

This patch removes spapr_machine_using_legacy_numa() and open code the
conditions in each caller. While we're at it, move the chunk inside
spapr_numa_FORM1_affinity_init() that sets all numa_assoc_array domains
with 'node_id' to spapr_numa_define_FORM1_domains(). This chunk was
being executed if !pre_5_2_numa_associativity and num_nodes => 1, the
same conditions in which spapr_numa_define_FORM1_domains() is called
shortly after.

Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-3-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr_numa.c: split FORM1 code into helpers
Daniel Henrique Barboza [Mon, 20 Sep 2021 17:49:41 +0000 (14:49 -0300)]
spapr_numa.c: split FORM1 code into helpers

The upcoming FORM2 NUMA affinity will support asymmetric NUMA topologies
and doesn't need be concerned with all the legacy support for older
pseries FORM1 guests.

We're also not going to calculate associativity domains based on numa
distance (via spapr_numa_define_associativity_domains) since the
distances will be written directly into new DT properties.

Let's split FORM1 code into its own functions to allow for easier
insertion of FORM2 logic later on.

Reviewed-by: Greg Kurz <groug@kaod.org>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210920174947.556324-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agotarget/ppc: Check privilege level based on PSR and LPCR[HR] in tlbie[l]
Matheus Ferst [Fri, 17 Sep 2021 11:47:51 +0000 (08:47 -0300)]
target/ppc: Check privilege level based on PSR and LPCR[HR] in tlbie[l]

PowerISA v3.0B made tlbie[l] hypervisor privileged when PSR=0 and HR=1.
To allow the check at translation time, we'll use the HR bit of LPCR to
check the MMU mode instead of the PATE.HR.

Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
Message-Id: <20210917114751.206845-3-matheus.ferst@eldorado.org.br>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agotarget/ppc: add LPCR[HR] to DisasContext and hflags
Matheus Ferst [Fri, 17 Sep 2021 11:47:50 +0000 (08:47 -0300)]
target/ppc: add LPCR[HR] to DisasContext and hflags

Add a Host Radix field (hr) in DisasContext with LPCR[HR] value to allow
us to decide between Radix and HPT while validating instructions
arguments. Note that PowerISA v3.1 does not require LPCR[HR] and PATE.HR
to match if the thread is in ultravisor/hypervisor real addressing mode,
so ctx->hr may be invalid if ctx->hv and ctx->dr are set.

Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210917114751.206845-2-matheus.ferst@eldorado.org.br>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agotarget/ppc: Replace debug messages by asserts for unknown IRQ pins
Cédric Le Goater [Mon, 20 Sep 2021 06:12:01 +0000 (08:12 +0200)]
target/ppc: Replace debug messages by asserts for unknown IRQ pins

If an unknown pin of the IRQ controller is raised, something is very
wrong in the QEMU model. It is better to abort.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20210920061203.989563-3-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agotarget/ppc: Convert debug to trace events (exceptions)
Cédric Le Goater [Mon, 20 Sep 2021 06:12:00 +0000 (08:12 +0200)]
target/ppc: Convert debug to trace events (exceptions)

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Message-Id: <20210920061203.989563-2-clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agomemory_hotplug.c: send DEVICE_UNPLUG_GUEST_ERROR in acpi_memory_hotplug_write()
Daniel Henrique Barboza [Tue, 7 Sep 2021 00:47:55 +0000 (21:47 -0300)]
memory_hotplug.c: send DEVICE_UNPLUG_GUEST_ERROR in acpi_memory_hotplug_write()

MEM_UNPLUG_ERROR is deprecated since the introduction of
DEVICE_UNPLUG_GUEST_ERROR. Keep emitting both while the deprecation of
MEM_UNPLUG_ERROR is pending.

CC: Michael S. Tsirkin <mst@redhat.com>
CC: Igor Mammedov <imammedo@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210907004755.424931-8-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agospapr: use DEVICE_UNPLUG_GUEST_ERROR to report unplug errors
Daniel Henrique Barboza [Tue, 7 Sep 2021 00:47:54 +0000 (21:47 -0300)]
spapr: use DEVICE_UNPLUG_GUEST_ERROR to report unplug errors

Linux Kernel 5.12 is now unisolating CPU DRCs in the device_removal
error path, signalling that the hotunplug process wasn't successful.
This allow us to send a DEVICE_UNPLUG_GUEST_ERROR in drc_unisolate_logical()
to signal this error to the management layer.

We also have another error path in spapr_memory_unplug_rollback() for
configured LMB DRCs. Kernels older than 5.13 will not unisolate the LMBs
in the hotunplug error path, but it will reconfigure them. Let's send
the DEVICE_UNPLUG_GUEST_ERROR event in that code path as well to cover the
case of older kernels.

Acked-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210907004755.424931-7-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agoqapi/qdev.json: add DEVICE_UNPLUG_GUEST_ERROR QAPI event
Daniel Henrique Barboza [Tue, 7 Sep 2021 00:47:53 +0000 (21:47 -0300)]
qapi/qdev.json: add DEVICE_UNPLUG_GUEST_ERROR QAPI event

At this moment we only provide one event to report a hotunplug error,
MEM_UNPLUG_ERROR. As of Linux kernel 5.12 and QEMU 6.0.0, the pseries
machine is now able to report unplug errors for other device types, such
as CPUs.

Instead of creating a (device_type)_UNPLUG_ERROR for each new device,
create a generic DEVICE_UNPLUG_GUEST_ERROR event that can be used by all
guest side unplug errors in the future. This event has a similar API as
the existing DEVICE_DELETED event, always providing the QOM path of the
device and dev->id if there's any.

With this new generic event, MEM_UNPLUG_ERROR is now marked as deprecated.

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20210907004755.424931-6-danielhb413@gmail.com>
[dwg: Correct missing ')' in stubs/qdev.c]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
2 years agoMerge remote-tracking branch 'remotes/vivier/tags/q800-for-6.2-pull-request' into...
Peter Maydell [Wed, 29 Sep 2021 20:20:49 +0000 (21:20 +0100)]
Merge remote-tracking branch 'remotes/vivier/tags/q800-for-6.2-pull-request' into staging

Q800 Pull request 20210929

NuBus cleanup and improvement

# gpg: Signature made Wed 29 Sep 2021 10:27:51 BST
# gpg:                using RSA key CD2F75DDC8E3A4DC2E4F5173F30C38BD3F2FBE3C
# gpg:                issuer "laurent@vivier.eu"
# gpg: Good signature from "Laurent Vivier <lvivier@redhat.com>" [full]
# gpg:                 aka "Laurent Vivier <laurent@vivier.eu>" [full]
# gpg:                 aka "Laurent Vivier (Red Hat) <lvivier@redhat.com>" [full]
# Primary key fingerprint: CD2F 75DD C8E3 A4DC 2E4F  5173 F30C 38BD 3F2F BE3C

* remotes/vivier/tags/q800-for-6.2-pull-request:
  q800: configure nubus available slots for Quadra 800
  q800: wire up nubus IRQs
  nubus: add support for slot IRQs
  nubus-bridge: make slot_available_mask a qdev property
  nubus-bridge: embed the NubusBus object directly within nubus-bridge
  nubus: move NubusBus from mac-nubus-bridge to nubus-bridge
  mac-nubus-bridge: rename MacNubusState to MacNubusBridge
  nubus-bridge: introduce separate NubusBridge structure
  nubus: move nubus to its own 32-bit address space
  nubus-device: add romfile property for loading declaration ROMs
  nubus-device: remove nubus_register_rom() and nubus_register_format_block()
  macfb: don't register declaration ROM
  nubus: generate bus error when attempting to access empty slots
  nubus: add trace-events for empty slot accesses
  nubus: implement BusClass get_dev_path()
  nubus: move slot bitmap checks from NubusDevice realize() to BusClass check_address()
  nubus: use bitmap to manage available slots
  nubus-device: expose separate super slot memory region
  nubus-device: rename slot_nb variable to slot
  nubus: add comment indicating reference documents

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2 years agoblock/nbd: check that received handle is valid
Vladimir Sementsov-Ogievskiy [Thu, 2 Sep 2021 10:38:05 +0000 (13:38 +0300)]
block/nbd: check that received handle is valid

If we don't have active request, that waiting for this handle to be
received, we should report an error.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210902103805.25686-6-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
2 years agoblock/nbd: drop connection_co
Vladimir Sementsov-Ogievskiy [Thu, 2 Sep 2021 10:38:04 +0000 (13:38 +0300)]
block/nbd: drop connection_co

OK, that's a big rewrite of the logic.

Pre-patch we have an always running coroutine - connection_co. It does
reply receiving and reconnecting. And it leads to a lot of difficult
and unobvious code around drained sections and context switch. We also
abuse bs->in_flight counter which is increased for connection_co and
temporary decreased in points where we want to allow drained section to
begin. One of these place is in another file: in nbd_read_eof() in
nbd/client.c.

We also cancel reconnect and requests waiting for reconnect on drained
begin which is not correct. And this patch fixes that.

Let's finally drop this always running coroutine and go another way:
do both reconnect and receiving in request coroutines.

The detailed list of changes below (in the sequence of diff hunks).

1. receiving coroutines are woken directly from nbd_channel_error, when
   we change s->state

2. nbd_co_establish_connection_cancel(): we don't have drain_begin now,
   and in nbd_teardown_connection() all requests should already be
   finished (and reconnect is done from request). So
   nbd_co_establish_connection_cancel() is called from
   nbd_cancel_in_flight() (to cancel the request that is doing
   nbd_co_establish_connection()) and from reconnect_delay_timer_cb()
   (previously we didn't need it, as reconnect delay only should cancel
   active requests not the reconnection itself). But now reconnection
   itself is done in the separate thread (we now call
   nbd_client_connection_enable_retry() in nbd_open()), and we need to
   cancel the requests that wait in nbd_co_establish_connection()
   now).

2A. We do receive headers in request coroutine. But we also should
   dispatch replies for other pending requests. So,
   nbd_connection_entry() is turned into nbd_receive_replies(), which
   does reply dispatching while it receives other request headers, and
   returns when it receives the requested header.

3. All old staff around drained sections and context switch is dropped.
   In details:
   - we don't need to move connection_co to new aio context, as we
     don't have connection_co anymore
   - we don't have a fake "request" of connection_co (extra increasing
     in_flight), so don't care with it in drain_begin/end
   - we don't stop reconnection during drained section anymore. This
     means that drain_begin may wait for a long time (up to
     reconnect_delay). But that's an improvement and more correct
     behavior see below[*]

4. In nbd_teardown_connection() we don't have to wait for
   connection_co, as it is dropped. And cleanup for s->ioc and nbd_yank
   is moved here from removed connection_co.

5. In nbd_co_do_establish_connection() we now should handle
   NBD_CLIENT_CONNECTING_NOWAIT: if new request comes when we are in
   NBD_CLIENT_CONNECTING_NOWAIT, it still should call
   nbd_co_establish_connection() (who knows, maybe the connection was
   already established by another thread in the background). But we
   shouldn't wait: if nbd_co_establish_connection() can't return new
   channel immediately the request should fail (we are in
   NBD_CLIENT_CONNECTING_NOWAIT state).

6. nbd_reconnect_attempt() is simplified: it's now easier to wait for
   other requests in the caller, so here we just assert that fact.
   Also delay time is now initialized here: we can easily detect first
   attempt and start a timer.

7. nbd_co_reconnect_loop() is dropped, we don't need it. Reconnect
   retries are fully handle by thread (nbd/client-connection.c), delay
   timer we initialize in nbd_reconnect_attempt(), we don't have to
   bother with s->drained and friends. nbd_reconnect_attempt() now
   called from nbd_co_send_request().

8. nbd_connection_entry is dropped: reconnect is now handled by
   nbd_co_send_request(), receiving reply is now handled by
   nbd_receive_replies(): all handled from request coroutines.

9. So, welcome new nbd_receive_replies() called from request coroutine,
   that receives reply header instead of nbd_connection_entry().
   Like with sending requests, only one coroutine may receive in a
   moment. So we introduce receive_mutex, which is locked around
   nbd_receive_reply(). It also protects some related fields. Still,
   full audit of thread-safety in nbd driver is a separate task.
   New function waits for a reply with specified handle being received
   and works rather simple:

   Under mutex:
     - if current handle is 0, do receive by hand. If another handle
       received - switch to other request coroutine, release mutex and
       yield. Otherwise return success
     - if current handle == requested handle, we are done
     - otherwise, release mutex and yield

10: in nbd_co_send_request() we now do nbd_reconnect_attempt() if
    needed. Also waiting in free_sema queue we now wait for one of two
    conditions:
    - connectED, in_flight < MAX_NBD_REQUESTS (so we can start new one)
    - connectING, in_flight == 0, so we can call
      nbd_reconnect_attempt()
    And this logic is protected by s->send_mutex

    Also, on failure we don't have to care of removed s->connection_co

11. nbd_co_do_receive_one_chunk(): now instead of yield() and wait for
    s->connection_co we just call new nbd_receive_replies().

12. nbd_co_receive_one_chunk(): place where s->reply.handle becomes 0,
    which means that handling of the whole reply is finished. Here we
    need to wake one of coroutines sleeping in nbd_receive_replies().
    If none are sleeping - do nothing. That's another behavior change: we
    don't have endless recv() in the idle time. It may be considered as
    a drawback. If so, it may be fixed later.

13. nbd_reply_chunk_iter_receive(): don't care about removed
    connection_co, just ping in_flight waiters.

14. Don't create connection_co, enable retry in the connection thread
    (we don't have own reconnect loop anymore)

15. We now need to add a nbd_co_establish_connection_cancel() call in
    nbd_cancel_in_flight(), to cancel the request that is doing a
    connection attempt.

[*], ok, now we don't cancel reconnect on drain begin. That's correct:
    reconnect feature leads to possibility of long-running requests (up
    to reconnect delay). Still, drain begin is not a reason to kill
    long requests. We should wait for them.

    This also means, that we can again reproduce a dead-lock, described
    in 8c517de24a8a1dcbeb54e7e12b5b0fda42a90ace.
    Why we are OK with it:
    1. Now this is not absolutely-dead dead-lock: the vm is unfrozen
       after reconnect delay. Actually 8c517de24a8a1dc fixed a bug in
       NBD logic, that was not described in 8c517de24a8a1dc and led to
       forever dead-lock. The problem was that nobody woke the free_sema
       queue, but drain_begin can't finish until there is a request in
       free_sema queue. Now we have a reconnect delay timer that works
       well.
    2. It's not a problem of the NBD driver, but of the ide code,
       because it does drain_begin under the global mutex; the problem
       doesn't reproduce when using scsi instead of ide.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210902103805.25686-5-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: grammar and comment tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>
2 years agoblock/nbd: refactor nbd_recv_coroutines_wake_all()
Vladimir Sementsov-Ogievskiy [Thu, 2 Sep 2021 10:38:03 +0000 (13:38 +0300)]
block/nbd: refactor nbd_recv_coroutines_wake_all()

Split out nbd_recv_coroutine_wake_one(), as it will be used
separately.
Rename the function and add a possibility to wake only first found
sleeping coroutine.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210902103805.25686-4-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: grammar tweak]
Signed-off-by: Eric Blake <eblake@redhat.com>
2 years agoblock/nbd: move nbd_recv_coroutines_wake_all() up
Vladimir Sementsov-Ogievskiy [Thu, 2 Sep 2021 10:38:02 +0000 (13:38 +0300)]
block/nbd: move nbd_recv_coroutines_wake_all() up

We are going to use it in nbd_channel_error(), so move it up. Note,
that we are going also refactor and rename
nbd_recv_coroutines_wake_all() in future anyway, so keeping it where it
is and making forward declaration doesn't make real sense.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20210902103805.25686-3-vsementsov@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
2 years agoblock/nbd: nbd_channel_error() shutdown channel unconditionally
Vladimir Sementsov-Ogievskiy [Thu, 2 Sep 2021 10:38:01 +0000 (13:38 +0300)]
block/nbd: nbd_channel_error() shutdown channel unconditionally

Don't rely on connection being totally broken in case of -EIO. Safer
and more correct is to just shut down the channel anyway, since we
change the state and plan on reconnecting.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20210902103805.25686-2-vsementsov@virtuozzo.com>
[eblake: grammar tweaks]
Signed-off-by: Eric Blake <eblake@redhat.com>