esp: accumulate SCSI commands for PDMA transfers in cmdbuf instead of pdma_buf

ESP SCSI commands are already accumulated in cmdbuf and so there is no need to
keep a separate pdma_buf buffer. Accumulate SCSI commands for PDMA transfers in
cmdbuf instead of pdma_buf so update cmdlen accordingly and change pdma_origin
for PDMA transfers to CMD which allows the PDMA origin to be removed.

This commit also removes a stray memcpy() from get_cmd() which is a no-op because
cmdlen is always zero at the start of a command.

Notionally the removal of pdma_buf from vmstate_esp_pdma also breaks migration
compatibility for the PDMA subsection until its complete removal by the end of
the series.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: move pdma_len and TC logic into esp_pdma_read()/esp_pdma_write()

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: use pdma_origin directly in esp_pdma_read()/esp_pdma_write()

This is the first step in removing get_pdma_buf() from esp.c.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: introduce esp_pdma_read() and esp_pdma_write() functions

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: remove minlen restriction in handle_ti

The limiting of DMA transfers to the maximum size of the available data is already
handled by esp_do_dma() and do_dma_pdma_cb().

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: remove dma_left from ESPState

The ESP device already keeps track of the remaining bytes left to transfer via
its TC (transfer counter) register which is decremented for each byte that
is transferred across the SCSI bus.

Switch the transfer logic to use the value of TC instead of dma_left and then
remove dma_left completely, adding logic to the vmstate_esp post_load() function
to transfer the old dma_left value to the TC register during migration from
older versions.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: remove dma_counter from ESPState

The value of dma_counter is set once at the start of the transfer and remains
the same until the transfer is complete. This allows the check in esp_transfer_data
to be simplified since dma_left will always be non-zero until the transfer is
completed.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: apply transfer length adjustment when STC is zero at TC load time

Perform the length adjustment whereby a value of 0 in the STC represents
a transfer length of 0x10000 at the point where the TC is loaded at the
start of a DMA command rather than just when a TI (Transfer Information)
command is executed. This better matches the description as given in the
datasheet.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: introduce esp_get_stc()

This function simplifies reading the STC register value without having to manually
shift each individual 8-bit value.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: introduce esp_get_tc() and esp_set_tc()

These functions simplify reading and writing the TC register value without having to
manually shift each individual 8-bit value.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: determine transfer direction directly from SCSI phase

The transfer direction is currently determined by checking the sign of ti_size
but as this series progresses ti_size can be zero at the end of the transfer.

Use the SCSI phase to determine the transfer direction as used in other SCSI
controller implementations.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: add PDMA trace events

This will become more useful later when trying to debug mixed FIFO and PDMA
requests.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: fix esp_reg_read() trace event

Move the trace event to the end of the function so that it correctly reports
the returned value if it doesn't come directly from the rregs array.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: add trace event when receiving a TI command

This enables us to determine whether the command being issued is for a DMA or a
non-DMA transfer.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: add vmstate_esp version to embedded ESPState

The QOM object representing ESPState is currently embedded within both the
SYSBUS_ESP and PCI_ESP devices with migration state handled by embedding
vmstate_esp within each device using VMSTATE_STRUCT.

Since the vmstate_esp fields are embedded directly within the migration
stream, the incoming vmstate_esp version_id is lost. The only version information
available is that from vmstate_sysbus_esp_scsi and vmstate_esp_pci_scsi, but
those versions represent their respective devices and not that of the underlying
ESPState.

Resolve this by adding a new version-dependent field in vmstate_sysbus_esp_scsi
and vmstate_esp_pci_scsi which stores the vmstate_esp version_id field within
ESPState to be used to allow migration from older QEMU versions.

Finally bump the vmstate_esp version to 5 to cover the upcoming ESPState changes
within this patch series.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: QOMify the internal ESP device state

Make this new QOM device state a child device of both the sysbus-esp and esp-pci
implementations.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: rename existing ESP QOM type to SYSBUS_ESP

The existing ESP QOM type currently represents a sysbus device with an embedded
ESP state. Rename the type to SYSBUS_ESP accordingly.

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

esp: checkpatch fixes

Signed-off-by: Mark Cave-Ayland <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Laurent Vivier <[email protected]>
Message-Id: <20210304221103[email protected]>

Merge remote-tracking branch 'remotes/cohuck-gitlab/tags/s390x-20210305' into staging

some accumulated s390x fixes

# gpg: Signature made Fri 05 Mar 2021 15:50:00 GMT
# gpg:                using RSA key C3D0D66DC3624FF6A8C018CEDECF6B93C6F02FAF
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Cornelia Huck <[email protected]>" [unknown]
# gpg:                 aka "Cornelia Huck <[email protected]>" [full]
# gpg:                 aka "Cornelia Huck <[email protected]>" [full]
# gpg:                 aka "Cornelia Huck <[email protected]>" [unknown]
# gpg:                 aka "Cornelia Huck <[email protected]>" [unknown]
# Primary key fingerprint: C3D0 D66D C362 4FF6 A8C0  18CE DECF 6B93 C6F0 2FAF

* remotes/cohuck-gitlab/tags/s390x-20210305:
  target/s390x/kvm: Simplify debug code
  vfio-ccw: Do not read region ret_code after write
  css: SCHIB measurement block origin must be aligned
  virtio-ccw: commands on revision-less devices
  s390x/pci: restore missing Query PCI Function CLP data
  hw/s390x: fix build for virtio-9p-ccw
  target/s390x/arch_dump: Fix warning for the name field in the PT_NOTE section
  s390x/cpu_model: disallow unpack for --only-migratable

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2021-03-05' into staging

QAPI patches patches for 2021-03-05

# gpg: Signature made Fri 05 Mar 2021 14:42:18 GMT
# gpg:                using RSA key 354BC8B3D7EB2A6B68674E5F3870B400EB918653
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Markus Armbruster <[email protected]>" [full]
# gpg:                 aka "Markus Armbruster <[email protected]>" [full]
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867  4E5F 3870 B400 EB91 8653

* remotes/armbru/tags/pull-qapi-2021-03-05:
  qapi: Fix parse errors for removal of null from schema language
  qapi: Remove QMP events and commands from user-mode builds
  qga: Utilize QAPI_LIST_APPEND in qmp_guest_network_get_interfaces
  error: Fix "Converting to ERRP_GUARD()" doc on "valid at return"

Signed-off-by: Peter Maydell <[email protected]>

qapi: Fix parse errors for removal of null from schema language

Commit 9d55380b5a "qapi: Remove null from schema language" (v4.2.0)
neglected to update two error messages.  Do that now.

Signed-off-by: Markus Armbruster <[email protected]>
Message-Id: <20210224101442.1837475[email protected]>
Reviewed-by: Eric Blake <[email protected]>
Reviewed-by: John Snow <[email protected]>

qapi: Remove QMP events and commands from user-mode builds

We removed the QMP loop in user-mode builds in commit 1935e0e4e09
("qapi/meson: Remove QMP from user-mode emulation"), now commands
and events code is unreachable.

Signed-off-by: Philippe Mathieu-Daudé <[email protected]>
Message-Id: <20210224171642.3242293[email protected]>
Acked-by: Paolo Bonzini <[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

qga: Utilize QAPI_LIST_APPEND in qmp_guest_network_get_interfaces

I found another spot that can benefit from using our macros instead of
open-coding qapi list creation.

Signed-off-by: Eric Blake <[email protected]>
Message-Id: <20210205171634.1491258[email protected]>
Reviewed-by: Markus Armbruster <[email protected]>
Signed-off-by: Markus Armbruster <[email protected]>

error: Fix "Converting to ERRP_GUARD()" doc on "valid at return"

Setting errp = NULL is wrong: the automatic error propagation still
propagates the dangling pointer _auto_errp_prop.local_err.  We need to
set *errp = NULL to clear the dangling pointer.

Signed-off-by: Markus Armbruster <[email protected]>
Message-Id: <20210125132635.1253219[email protected]>
Reviewed-by: Vladimir Sementsov-Ogievskiy <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>

Merge remote-tracking branch 'remotes/alistair/tags/pull-riscv-to-apply-20210304' into staging

RISC-V PR for 6.0

This PR is a collection of RISC-V patches:
 - Improvements to SiFive U OTP
 - Upgrade OpenSBI to v0.9
 - Support the QMP dump-guest-memory
 - Add support for the SiFive SPI controller (sifive_u)
 - Initial RISC-V system documentation
 - A fix for the Goldfish RTC
 - MAINTAINERS updates
 - Support for high PCIe memory in the virt machine

# gpg: Signature made Thu 04 Mar 2021 14:44:31 GMT
# gpg:                using RSA key F6C4AC46D4934868D3B8CE8F21E10D29DF977054
# gpg: Good signature from "Alistair Francis <[email protected]>" [full]
# Primary key fingerprint: F6C4 AC46 D493 4868 D3B8  CE8F 21E1 0D29 DF97 7054

* remotes/alistair/tags/pull-riscv-to-apply-20210304:
  hw/riscv: virt: Map high mmio for PCIe
  hw/riscv: virt: Limit RAM size in a 32-bit system
  hw/riscv: virt: Drop the 'link_up' parameter of gpex_pcie_init()
  hw/riscv: Drop 'struct MemmapEntry'
  MAINTAINERS: Add a SiFive machine section
  goldfish_rtc: re-arm the alarm after migration
  docs/system: riscv: Add documentation for sifive_u machine
  docs/system: Add RISC-V documentation
  docs/system: Sort targets in alphabetical order
  hw/riscv: sifive_u: Change SIFIVE_U_GEM_IRQ to decimal value
  hw/riscv: sifive_u: Add QSPI2 controller and connect an SD card
  hw/riscv: sifive_u: Add QSPI0 controller and connect a flash
  hw/ssi: Add SiFive SPI controller support
  hw/block: m25p80: Add various ISSI flash information
  hw/block: m25p80: Add ISSI SPI flash support
  target-riscv: support QMP dump-guest-memory
  roms/opensbi: Upgrade from v0.8 to v0.9
  hw/misc: sifive_u_otp: Use error_report() when block operation fails
  target/riscv: Declare csr_ops[] with a known size

Signed-off-by: Peter Maydell <[email protected]>

hw/riscv: virt: Map high mmio for PCIe

Some peripherals require 64-bit PCI address, so let's map the high
mmio space for PCIe.

For RV32, the address is hardcoded to below 4 GiB from the highest
accessible physical address. For RV64, the base address depends on
top of RAM and is aligned to its size which is using 16 GiB for now.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210220144807[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/riscv: virt: Limit RAM size in a 32-bit system

RV32 supports 34-bit physical address hence the maximum RAM size
should be limited. Limit the RAM size to 10 GiB, which leaves
some room for PCIe high mmio space.

For 32-bit host, this is not needed as machine->ram_size cannot
represent a RAM size that big. Use a #if size test to only do
the size limitation for the 64-bit host.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210220144807[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/riscv: virt: Drop the 'link_up' parameter of gpex_pcie_init()

`link_up` is never used in gpex_pcie_init(). Drop it.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210220144807[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/riscv: Drop 'struct MemmapEntry'

There is already a MemMapEntry type defined in hwaddr.h. Let's drop
the RISC-V defined `struct MemmapEntry` and use the existing one.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Message-id: 20210220144807[email protected]
Signed-off-by: Alistair Francis <[email protected]>

MAINTAINERS: Add a SiFive machine section

Signed-off-by: Alistair Francis <[email protected]>
Acked-by: Bin Meng <[email protected]>
Acked-by: Palmer Dabbelt <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Message-id: 6bc077e5ae4a9512c8adf81ae194718f2f17c402.1612836645[email protected]

goldfish_rtc: re-arm the alarm after migration

After a migration the clock offset is updated, but we also
need to re-arm the alarm if needed.

Signed-off-by: Laurent Vivier <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Message-id: 20201220112615[email protected]
Signed-off-by: Alistair Francis <[email protected]>

docs/system: riscv: Add documentation for sifive_u machine

This adds detailed documentation for RISC-V `sifive_u` machine,
including the following information:

- Supported devices
- Hardware configuration information
- Boot options
- Machine-specific options
- Running Linux kernel
- Running VxWorks kernel
- Running U-Boot, and with an alternate configuration

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

docs/system: Add RISC-V documentation

Add RISC-V system emulator documentation for generic information.
`Board-specific documentation` and `RISC-V CPU features` are only
a placeholder and will be added in the future.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

docs/system: Sort targets in alphabetical order

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/riscv: sifive_u: Change SIFIVE_U_GEM_IRQ to decimal value

All other peripherals' IRQs are in the format of decimal value.
Change SIFIVE_U_GEM_IRQ to be consistent.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/riscv: sifive_u: Add QSPI2 controller and connect an SD card

This adds the QSPI2 controller to the SoC, and connects an SD
card to it. The generation of corresponding device tree source
fragment is also added.

Specify machine property `msel` to 11 to boot the same upstream
U-Boot SPL and payload image for the SiFive HiFive Unleashed board.
Note subsequent payload is stored in the SD card image.

$ qemu-system-riscv64 -nographic -M sifive_u,msel=11 -smp 5 -m 8G \
    -bios u-boot-spl.bin -drive file=sdcard.img,if=sd

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/riscv: sifive_u: Add QSPI0 controller and connect a flash

This adds the QSPI0 controller to the SoC, and connects an ISSI
25WP256 flash to it. The generation of corresponding device tree
source fragment is also added.

Since the direct memory-mapped mode is not supported by the SiFive
SPI model, the <reg> property does not populate the second group
which represents the memory mapped address of the SPI flash.

With this commit, upstream U-Boot for the SiFive HiFive Unleashed
board can boot on QEMU 'sifive_u' out of the box. This allows users
to develop and test the recommended RISC-V boot flow with a real
world use case: ZSBL (in QEMU) loads U-Boot SPL from SPI flash to
L2LIM, then U-Boot SPL loads the payload from SPI flash that is
combined with OpenSBI fw_dynamic firmware and U-Boot proper.

Specify machine property `msel` to 6 to allow booting from the SPI
flash. U-Boot spl is directly loaded via `-bios`, and subsequent
payload is stored in the SPI flash image. Example command line:

$ qemu-system-riscv64 -nographic -M sifive_u,msel=6 -smp 5 -m 8G \
    -bios u-boot-spl.bin -drive file=spi-nor.img,if=mtd

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/ssi: Add SiFive SPI controller support

This adds the SiFive SPI controller model for the FU540 SoC.
The direct memory-mapped SPI flash mode is unsupported.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/block: m25p80: Add various ISSI flash information

This updates the flash information table to include various ISSI
flashes that are supported by upstream U-Boot and Linux kernel.

Signed-off-by: Bin Meng <[email protected]>
Acked-by: Alistair Francis <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/block: m25p80: Add ISSI SPI flash support

This adds the ISSI SPI flash support. The number of dummy cycles in
fast read, fast read dual output and fast read quad output commands
is currently using the default 8. Likewise, the same default value
is used for fast read dual/quad I/O command. Per the datasheet [1],
the number of dummy cycles is configurable, but this is not modeled
at present.

For flash whose size is larger than 16 MiB, the sequence of 3-byte
address along with EXTADD bit in the bank address register (BAR) is
not supported. We assume that guest software always uses op codes
with 4-byte address sequence. Fortunately, this is the case for both
U-Boot and Linux spi-nor drivers.

QPI (Quad Peripheral Interface) that supports 2-cycle instruction
has different default values for dummy cycles of fast read family
commands, and is unsupported at the time being.

[1] http://www.issi.com/WW/pdf/25LP-WP256.pdf

Signed-off-by: Bin Meng <[email protected]>
Acked-by: Alistair Francis <[email protected]>
Message-id: 20210126060007[email protected]
Signed-off-by: Alistair Francis <[email protected]>

target-riscv: support QMP dump-guest-memory

Add the support needed for creating prstatus elf notes. This allows
us to use QMP dump-guest-memory.

Now ELF notes of RISC-V only contain prstatus elf notes.

Signed-off-by: Yifei Jiang <[email protected]>
Signed-off-by: Mingwang Li <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Reviewed-by: Andrew Jones <[email protected]>
Reviewed-by: Palmer Dabbelt <[email protected]>
Message-id: 20210201124458[email protected]
Signed-off-by: Alistair Francis <[email protected]>

roms/opensbi: Upgrade from v0.8 to v0.9

Upgrade OpenSBI from v0.8 to v0.9 and the pre-built bios images.

The v0.9 release includes the following commits:

35bc810 docs/platform: Update QEMU parameter for fw_payload
78afe11 config.mk: Update QEMU run command for generic and sifive fu540 platforms
ec3e5b1 docs/platform: sifive_fu540: Update U-Boot instructions
7d61a68 README.md: fix markdown link formatting
a5f9104 lib/utils: fdt: Update FDT expand size to 1024 for reserved memory node
ec1abf6 include: sbi_bitops: Remove dead shift assignment in ffs/fls
8e47649 lib: Add sbi_strncmp implementation
2845d2d lib: utils: Add a macro in libfdt_env.h for strncmp
2cfd2fc lib: utils: Use strncmp in fdt_parse_hart_id()
937caee lib: sbi_misaligned_ldst: Determine transformed instruction length correctly
4b18a2a firmware: fw_base: Improve exception stack setup in trap handler
9d56961 lib: sbi_trap: Fix hstatus.SPVP update in sbi_trap_redirect()
d7f87d9 platform: kendryte/k210: fixup FDT
e435ba0 lib: sbi_init: Avoid thundering hurd problem with coldboot_lock
4f3bad6 lib: sbi: Handle the case where MTVAL has illegal instruction address
7b0b289 lib: sbi: Remove redundant SBI_HART_HAS_PMP feature
74d1db7 lib: sbi: Improve PMP CSR detection and progamming
2c341f7 lib: sbi: Detect and print MHPM counters at boot-time
162d453 include: sbi: Few cosmetic changes in riscv_encoding.h
ebc8ebc lib: sbi: Improve HPM CSR read/write emulation
dcb10c0 lib: sbi: Don't handle VS-mode ecall in sbi_trap_handler()
bef63d6 include: Rename ECALL defines to match latest RISC-V spec
c1c7c3e lib: sbi_trap: Allow M-mode to M-mode ECALLs
6734304 lib: sbi: Allow specifying start mode to sbi_hsm_hart_start() API
7ccf6bf lib: sbi: Allow specifying mode in sbi_hart_pmp_check_addr() API
9f935a4 lib: utils: Improve fdt_cpu_fixup() implementation
172fa16 lib: sbi: Ensure coldboot HART supports next privilege mode
aaeca7e platform: generic: Don't mark non-MMU HARTs as invalid
7701ea1 lib: sbi: Fix PMP CSR detection
79bf80b lib: sbi_scratch: typo scatch
a04c465 makefile: fix clean directive
af4b50f Makefile: Build ELF, BIN and LD script in platform build directory
6ca0969 firmware: Add common FW_FDT_PATH compile-time option
9c07c51 firmware: Remove FW_PAYLOAD_FDT_PATH compile-time option
e9a4bfb Makefile: Allow padding zeros when converting DTB to C source
a0f2d4a platform: kendryte/k210: Add some padding for FDT fixups
dbeeacb include: sbi: Remove redundant includes from sbi_platform.h
a12d46a include: sbi: Remove pmp_region callbacks from sbi_platform_operations
a126886 lib: sbi: Configure PMP late in coldboot and warmboot path
f81d6f6 lib: sbi: Remove redundant hartid parameter from sbi_hart_init()
8b65005 include: sbi: Make hartmask pointer const in sbi_hartmask_test_hart()
b1678af lib: sbi: Add initial domain support
e73b92d lib: sbi: Extend sbi_hsm_hart_started_mask() for domains
3a30d2c lib: sbi: Extend sbi_hsm_hart_start() for domains
530e95b lib: sbi: Optimize sbi_hsm_hart_started_mask() implementation
3e20037 lib: sbi: Extend sbi_system_reset() for domains
5edbb7c lib: utils: Update fdt_reserved_memory_fixup() to use current domain
5fd99db lib: utils: Update fdt_cpu_fixup() to use current domain
e856462 lib: sbi: Remove redundant sbi_hart_pmp_xyz() functions
c10c30b lib: sbi: Configure PMP based on domain memory regions
c347408 lib: sbi: Display domain details in boot prints
fdf5d5c docs: Add initial documentation for domain support
74c0ea1 lib: utils: Implement "ranges" property parsing
bf21632 lib: sbi: Detect PMP granularity and number of address bits
a809f40 lib: sbi: Improve boot time print with additional PMP information
914f81f Makefile: Add option to use toolchain default ABI and ISA string
48616b3 lib: sbi: Improve boot prints in cold boot sequence
781cafd docs: fix a typo error
54a7734 include: sbi: Add SBI SRST extension related defines
c4acc60 include: sbi: Remove opensbi specific reset type defines
da07479 platform: Remove dummy system reset functions
5c429ae lib: sbi: Improve system reset platform operations
548d03e lib: sbi: Implement System Reset (SRST) SBI extension
2677324 firmware: fw_base: Optimize trap handler for RV32 systems
8d2edc4 lib: sbi: Fix sbi_hart_switch_mode() for u-mode
3d921fa lib: sbi: Fix typo in sbi_domain_finalize()
4e37022 lib: sbi: Fix domain_count check in sbi_domain_finalize()
c709d40 lib: sbi: Auto start domain only if boot HART within limits
c1f6d89 include: sbi: Use lower bits for domain memory region permissions
62ea4f4 lib: sbi: Override domain boot HART when coldboot HART assigned to it
555e737 lib: sbi: Add error prints in sbi_domain_finalize()
9b65dca include: sbi: Add domains_init() platform operation
c0d2baa docs: Add domain device tree binding documentation
ba741ea lib: utils: Add helper routines to populate domains from FDT
4fffb53 platform: generic: Populate domains from FDT
e7da0b4 lib: utils/libfdt: Upgrade to v1.6.0 release
2179777 lib: utils: Allow FDT domain iteration functions to fail
7baccfc lib: sbi: Add function to register new domain
6fc1986 lib: utils: Remove fdt_domain_get() function
a029bd9 lib: sbi: Remove domain_get() platform callback function
7dcb1e1 lib: sbi: Fix sign-extension in sbi_misaligned_load_handler()
80bc506 lib: sbi: Replace args with trap registers in ecall handler
b7df5e4 lib: sbi: Introduce sbi_trap_exit() API
12394a2 lib: sbi: Allow custom local TLB flush function
0d49c3b lib: utils: Fix shakti uart implementation
db56341 lib: sbi: Allow platforms to provide root domain memory regions
e884416 include: sbi: No need to pack struct sbi_trap_regs
386eba2 include: sbi: No need to pack struct sbi_scratch
1bbf361 include: sbi: Don't pack struct sbi_platform and sbi_platform_operations
da5293f platform: template: Fix compile error
234ed8e include: Bump-up version to 0.9

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-Id: 20210119234438[email protected]
Signed-off-by: Alistair Francis <[email protected]>

hw/misc: sifive_u_otp: Use error_report() when block operation fails

At present when blk_pread() / blk_pwrite() fails, a guest error
is logged, but this is not really a guest error. Change to use
error_report() instead.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 1611026585[email protected]
Signed-off-by: Alistair Francis <[email protected]>

target/riscv: Declare csr_ops[] with a known size

csr_ops[] is currently declared with an unknown size in cpu.h.
Since the array size is known, let's do a complete declaration.

Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Alistair Francis <[email protected]>
Message-id: 1611024723[email protected]
Signed-off-by: Alistair Francis <[email protected]>

target/s390x/kvm: Simplify debug code

We already have the 'run' variable holding 'cs->kvm_run' value.

Signed-off-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Message-Id: <20210303182219.1631042[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>

Merge remote-tracking branch 'remotes/kraxel/tags/ui-20210304-pull-request' into staging

ui/console: message surface tweaks.
ui/cocoa: bugfixes and cleanups.

# gpg: Signature made Thu 04 Mar 2021 08:36:53 GMT
# gpg:                using RSA key A0328CFFB93A17A79901FE7D4CB6D8EED3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <[email protected]>" [full]
# gpg:                 aka "Gerd Hoffmann <[email protected]>" [full]
# gpg:                 aka "Gerd Hoffmann (private) <[email protected]>" [full]
# Primary key fingerprint: A032 8CFF B93A 17A7 9901  FE7D 4CB6 D8EE D3E8 7138

* remotes/kraxel/tags/ui-20210304-pull-request:
  virtio-gpu: Do not distinguish the primary console
  ui/console: Pass placeholder surface to displays
  ui/console: Add placeholder flag to message surface
  ui/cocoa: Replace fprintf with error_report
  configure: Improve OpenGL dependency detections
  ui/cocoa: Fix stride resolution of pixman image
  ui/gtk: vte: fix sending multiple characeters
  ui/cocoa: Remove the uses of full screen APIs

Signed-off-by: Peter Maydell <[email protected]>

Merge remote-tracking branch 'remotes/dgilbert-gitlab/tags/pull-virtiofs-20210304' into staging

virtiofs minor security fix

Fix xattrmap to drop remapped security.capability capabilities.

Signed-off-by: Dr. David Alan Gilbert <[email protected]>
# gpg: Signature made Thu 04 Mar 2021 10:36:45 GMT
# gpg:                using RSA key 45F5C71B4A0CB7FB977A9FA90516331EBC5BFDE7
# gpg: Good signature from "Dr. David Alan Gilbert (RH2) <[email protected]>" [full]
# Primary key fingerprint: 45F5 C71B 4A0C B7FB 977A  9FA9 0516 331E BC5B FDE7

* remotes/dgilbert-gitlab/tags/pull-virtiofs-20210304:
  virtiofs: drop remapped security.capability xattr as needed

Signed-off-by: Peter Maydell <[email protected]>

virtiofs: drop remapped security.capability xattr as needed

On Linux, the 'security.capability' xattr holds a set of
capabilities that can change when an executable is run, giving
a limited form of privilege escalation to those programs that
the writer of the file deemed worthy.

Any write causes the 'security.capability' xattr to be dropped,
stopping anyone from gaining privilege by modifying a blessed
file.

Fuse relies on the daemon to do this dropping, and in turn the
daemon relies on the host kernel to drop the xattr for it.  However,
with the addition of -o xattrmap, the xattr that the guest
stores its capabilities in is now not the same as the one that
the host kernel automatically clears.

Where the mapping changes 'security.capability', explicitly clear
the remapped name to preserve the same behaviour.

This bug is assigned CVE-2021-20263.

Signed-off-by: Dr. David Alan Gilbert <[email protected]>
Reviewed-by: Vivek Goyal <[email protected]>

vfio-ccw: Do not read region ret_code after write

A pwrite() call returns the number of bytes written (or -1 on error),
and vfio-ccw compares this number with the size of the region to
determine if an error had occurred or not.

If they are not equal, this is a failure and the errno is used to
determine exactly how things failed. An errno of zero is possible
(though unlikely) in this situation and would be translated to a
successful operation.

If they ARE equal, the ret_code field is read from the region to
determine how to proceed. While the kernel sets the ret_code field
as necessary, the region and thus this field is not "written back"
to the user. So the value can only be what it was initialized to,
which is zero.

So, let's convert an unexpected length with errno of zero to a
return code of -EFAULT, and explicitly set an expected length to
a return code of zero. This will be a little safer and clearer.

Suggested-by: Matthew Rosato <[email protected]>
Signed-off-by: Eric Farman <[email protected]>
Message-Id: <20210303160739.2179378[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>

css: SCHIB measurement block origin must be aligned

The Measurement Block Origin inside the SCHIB is used when
Measurement Block format 1 is in used and must be aligned
on 64 bytes otherwise an operand exception is recognized
when issuing the Modify Sub CHannel (MSCH) instruction.

Signed-off-by: Pierre Morel <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Message-Id: <1613741973[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>

virtio-ccw: commands on revision-less devices

The virtio standard specifies that any non-transitional device must
reject commands prior to revision setting (which we do). Devices
that are transitional need to assume revision 0 (legacy) if the
driver sends a non-revision-setting command first in order to
support legacy drivers. We neglected to do the latter.

Fortunately, nearly everything worked as intended anyway; the only
problem was not properly rejecting revision setting after some other
command had been issued. Easy to fix by setting revision to 0 if
we see a non-revision command on a legacy-capable revision-less
device.

Found by code inspection, not observed in the wild.

Signed-off-by: Cornelia Huck <[email protected]>
Reviewed-by: Thomas Huth <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Acked-by: Halil Pasic <[email protected]>
Message-Id: <20210216111830.1087847[email protected]>

s390x/pci: restore missing Query PCI Function CLP data

Some CLP response data was accidentally dropped when fixing endianness
issues with the Query PCI Function CLP response.  All of these values are
sent as 0s to the guest for emulated devices, so the impact is only
observed on passthrough devices.

Fixes: a4e2fff1b104 ("s390x/pci: fix endianness issues")
Signed-off-by: Matthew Rosato <[email protected]>
Message-Id: <1613681609[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>

hw/s390x: fix build for virtio-9p-ccw

Commit  2c44220d05 ("meson: convert hw/arch*"), which migrated the old
Makefile.objs to meson.build  accidentally excluded virtio-ccw-9p.c and
thus the virtio-9p-ccw device from the build (and potentially also
included the file virtio-ccw-blk.c twice in the source set). And since
CONFIG_VIRTFS can't be used the way it was used here (see commit
2c9dce0196 ("meson: do not use CONFIG_VIRTFS")), the preconditions have
to be written differently.

Let's fix this!

Signed-off-by: Halil Pasic <[email protected]>
Fixes: 2c44220d05 ("meson: convert hw/arch*")
Reported-by: Jakob Naucke <[email protected]>
Cc: [email protected]
Reviewed-by: Thomas Huth <[email protected]>
Message-Id: <20210218034059.1096078[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>

target/s390x/arch_dump: Fix warning for the name field in the PT_NOTE section

There is a compiler warning with GCC 9.3 when compiling with
the -fsanitize=thread compiler flag:

 In function 'strncpy',
    inlined from 's390x_write_elf64_notes' at ../target/s390x/arch_dump.c:219:9:
 /usr/include/x86_64-linux-gnu/bits/string_fortified.h:106:10: error:
  '__builtin_strncpy' specified bound 8 equals destination size
  [-Werror=stringop-truncation]

Since the name should always be NUL-terminated, let's use g_strlcpy() to
silence this warning. And while we're at it, also add an assert() to make
sure that the provided names always fit the size field (which is fine for
the current callers, the function is called once with "CORE" and once with
"LINUX" as a name).

Signed-off-by: Thomas Huth <[email protected]>
Reviewed-by: Christian Borntraeger <[email protected]>
Message-Id: <20210205093921[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>

s390x/cpu_model: disallow unpack for --only-migratable

Secure execution (aka protected virtualization) guests cannot be
migrated at the moment. If the unpack facility is provided in the cpu
model, a guest may choose to transition to secure mode, making the
guest unmigratable at that point in time. If the machine was explicitly
started with --only-migratable, we would get a failure only when the
guest actually tries to transition; instead, explicitly disallow the
unpack facility if --only-migratable was specified to avoid late
surprises.

Signed-off-by: Christian Borntraeger <[email protected]>
Reviewed-by: David Hildenbrand <[email protected]>
Reviewed-by: Halil Pasic <[email protected]>
Message-Id: <20210125135332[email protected]>
Signed-off-by: Cornelia Huck <[email protected]>

virtio-gpu: Do not distinguish the primary console

In the past, virtio-gpu set NULL as the surface for the secondary
consoles to hide its window. The distinction is now handled in
ui/console and the display backends and virtio-gpu does no longer
have to do that.

Signed-off-by: Akihiko Odaki <[email protected]>
Message-Id: <20210225101316[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

ui/console: Pass placeholder surface to displays

ui/console used to accept NULL as graphic console surface, but its
semantics was inconsistent among displays:
- cocoa and gtk-egl perform NULL dereference.
- egl-headless, spice and spice-egl do nothing.
- gtk releases underlying resources.
- sdl2-2d and sdl2-gl destroys the window.
- vnc shows a message, "Display output is not active."

Fortunately, only virtio-gpu and virtio-gpu-3d assign NULL so
we can study them to figure out the desired behavior. They assign
NULL *except* for the primary display when the device is realized,
reset, or its scanout is disabled. This effectively destroys
windows for the (uninitialized) secondary displays.

To implement the consistent behavior of display device
realization/reset, this change embeds it to the operation
switching the surface. When NULL was given as a new surface when
switching, ui/console will instead passes a placeholder down
to each display listeners.

sdl destroys the window for a secondary console if its surface is a
placeholder. The other displays simply shows the placeholder.

Signed-off-by: Akihiko Odaki <[email protected]>
Message-Id: <20210225101316[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

ui/console: Add placeholder flag to message surface

The surfaces created with former qemu_create_message_surface
did not display the content from the guest and always contained
simple messages describing the reason.

A display backend may want to hide the window showing such a
surface. This change renames the function to
qemu_create_placeholder_surface, and adds "placeholder" flag; the
display can check the flag to decide to do anything special like
hiding the window.

Signed-off-by: Akihiko Odaki <[email protected]>
Message-Id: <20210225101316[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

ui/cocoa: Replace fprintf with error_report

Signed-off-by: Akihiko Odaki <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Message-Id: <20210223131106[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

configure: Improve OpenGL dependency detections

This has the following visible changes:

- GBM is required only for OpenGL dma-buf.
- X11 is explicitly required by gtk-egl.
- EGL is now mandatory for the OpenGL displays.

The last one needs some detailed description. Before this change,
EGL was tested only for OpenGL dma-buf with the check of
EGL_MESA_image_dma_buf_export. However, all of the OpenGL
displays depend on EGL and EGL_MESA_image_dma_buf_export is always
defined by epoxy's EGL interface.
Therefore, it makes more sense to always check the presence of EGL
and say the OpenGL displays are available along with OpenGL dma-buf
if it is present.

Signed-off-by: Akihiko Odaki <[email protected]>
Message-Id: <20210223060307[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

ui/cocoa: Fix stride resolution of pixman image

A display can receive an image which its stride is greater than its
width. In fact, when a guest requests virtio-gpu to scan out a
smaller part of an image, virtio-gpu passes it to a display as an
image which its width represents the one of the part and its stride
equals to the one of the whole image.

This change makes ui/cocoa to cover such cases.

Signed-off-by: Akihiko Odaki <[email protected]>
Message-Id: <20210222144012[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

ui/gtk: vte: fix sending multiple characeters

When using the GTK UI with libvte, multicharacter keystrokes are not
sent correctly (such as arrow keys). gd_vc_in should check the
CharBackend's can_receive instead of assuming multiple characters can be
received. This is not an issue for e.g. the SDL UI because
qemu_chr_be_write is called with len=1 for each character (SDL sends
more than once keystroke).

Modify gd_vc_in to call qemu_chr_be_write multiple times if necessary.

Buglink: https://bugs.launchpad.net/qemu/+bug/1407808
Signed-off-by: Zack Marvel <[email protected]>
Message-Id: <20210221170613[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

pc,virtio,pci: bug fixes

Fixes all over the place. Specifically this fixes
a bug which made windows guests lose device config
(such as the configured fixed IP) after upgrading
to the new QEMU.

Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Tue 02 Mar 2021 14:19:51 GMT
# gpg:                using RSA key 5D09FD0871C8F85B94CA8A0D281F0DB8D28D5469
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>" [full]
# gpg:                 aka "Michael S. Tsirkin <[email protected]>" [full]
# Primary key fingerprint: 0270 606B 6F3C DF3D 0B17  0970 C350 3912 AFBE 8E67
#      Subkey fingerprint: 5D09 FD08 71C8 F85B 94CA  8A0D 281F 0DB8 D28D 5469

* remotes/mst/tags/for_upstream:
  vhost: simplify vhost_dev_init() fail_busyloop label
  hw/pci: Have safer pcie_bus_realize() by checking error path
  virtio-net: handle zero mac for a vdpa peer
  i386/acpi: restore device paths for pre-5.1 vms

Signed-off-by: Peter Maydell <[email protected]>

ui/cocoa: Remove the uses of full screen APIs

The detections of [NSView -enterFullScreen:] and
[NSView -exitFullScreen:] were wrong. A detection is coded as:
[NSView respondsToSelector:@selector(exitFullScreenModeWithOptions:)]
but it should be:
[NSView instancesRespondToSelector:@selector(exitFullScreenModeWithOptions:)]

Because of those APIs were not detected, ui/cocoa always falled
back to a borderless window whose frame matches the screen to
implement fullscreen behavior.

The code using [NSView -enterFullScreen:] and
[NSView -exitFullScreen:] will be used if you fix the detections,
but its behavior is undesirable; the full screen view stretches
the video, changing the aspect ratio, even if zooming is disabled.

This change removes the code as it does nothing good.

Signed-off-by: Akihiko Odaki <[email protected]>
Message-Id: <20210220013138[email protected]>
Signed-off-by: Gerd Hoffmann <[email protected]>

Merge remote-tracking branch 'remotes/bonzini-gitlab/tags/for-upstream' into staging

* fix --enable-fuzzing linker failures (Alexander)
* target/i386: Add bus lock debug exception support (Chenyi)
* update documentation for preferred boolean option syntax (Daniel)
* make SCSI io_timeout configurable (Hannes)
* fix handling of guest recoverable SCSI errors (myself)
* misc fixes (Pavel, Zheng Zhan Liang, Zihao)
* fix installation of binaries with entitlements (Akihiko)

# gpg: Signature made Thu 25 Feb 2021 14:41:56 GMT
# gpg:                using RSA key F13338574B662389866C7682BFFBD25F78C7AE83
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Paolo Bonzini <[email protected]>" [full]
# gpg:                 aka "Paolo Bonzini <[email protected]>" [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: (29 commits)
  tcg/i386: rdpmc: fix the the condtions
  chardev: do not use short form boolean options in non-QemuOpts character device descriptions
  vl: deprecate -writeconfig
  target/i386: Add bus lock debug exception support
  qom/object.c: Fix typo
  target/i386: update to show preferred boolean syntax for -cpu
  docs: update to show preferred boolean syntax for -cpu
  docs: update to show preferred boolean syntax for -vnc
  docs: update to show preferred boolean syntax for -chardev
  qemu-options: update to show preferred boolean syntax for -vnc
  qemu-options: update to show preferred boolean syntax for -incoming
  qemu-options: update to show preferred boolean syntax for -netdev
  qemu-options: update to show preferred boolean syntax for -spice
  qemu-options: update to show preferred boolean syntax for -chardev
  gdbstub: use preferred boolean option syntax
  char: don't fail when client is not connected
  scsi: drop 'result' argument from command_complete callback
  scsi-disk: pass guest recoverable errors through even for rerror=stop
  scsi-disk: pass SCSI status to scsi_handle_rw_error
  scsi: introduce scsi_sense_from_errno()
  ...

Signed-off-by: Peter Maydell <[email protected]>

vhost: simplify vhost_dev_init() fail_busyloop label

Requiring a conditional for every goto is tedious:

  if (busyloop_timeout) {
      goto fail_busyloop;
  } else {
      goto fail;
  }

Move the conditional to into the fail_busyloop label so that it's safe
to jump to this label unconditionally.

This change makes the migrate_add_blocker() error case more consistent.
It jumped to fail_busyloop unconditionally whereas the memslots limits
error case was conditional.

Signed-off-by: Stefan Hajnoczi <[email protected]>
Message-Id: <20210222114931[email protected]>
Reviewed-by: Stefano Garzarella <[email protected]>
Reviewed-by: Philippe Mathieu-Daudé <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

hw/pci: Have safer pcie_bus_realize() by checking error path

While pci_bus_realize() currently does not use the Error* argument,
it would be an error to leave pcie_bus_realize() setting bus->flags
if pci_bus_realize() had failed.

Fix by using a local Error* and return early (propagating the error)
if pci_bus_realize() failed.

Reported-by: Markus Armbruster <[email protected]>
Signed-off-by: Philippe Mathieu-Daudé <[email protected]>
Message-Id: <20210201153700[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

virtio-net: handle zero mac for a vdpa peer

Some mlx vdpa devices with kernels at least up to 5.11 currently present
0 as their MAC address.  This is because they have not been
  pre-configured with a MAC: they have a learning bridge and only learn
the MAC once guest is up.  Kernel patches and tools to allow programming
the MAC from host are being developed. For now - since these
combinations exist in the field - let's detect zero mac and just try to
proceed with the mac from the qemu command line.

This makes the guest use this MAC to send packets in turn teaching
the MAC to the card, and things work.

TODO:
report the actual MAC from QEMU commad line in the info message.
TODO:
detect that a (non-zero) hardware MAC does not match QEMU command line
and fail init.

Signed-off-by: Cindy Lu <[email protected]>
Message-Id: <20210225165506[email protected]>

mst: rewritten code comments, message printed and the commit log.

Cc: Eli Cohen <[email protected]>
Cc: Parav Pandit <[email protected]>
Tested-by: Adrian Moreno <[email protected]>
Tested-by: Sean Mooney <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>

i386/acpi: restore device paths for pre-5.1 vms

After fixing the _UID value for the primary PCI root bridge in
af1b80ae it was discovered that this change updates Windows
configuration in an incompatible way causing network configuration
failure unless DHCP is used. More details provided on the list:

https://lists.gnu.org/archive/html/qemu-devel/2021-02/msg08484.html

This change reverts the _UID update from 1 to 0 for q35 and i440fx
VMs before version 5.2 to maintain the original behaviour when
upgrading.

Cc: [email protected]
Cc: [email protected]
Reported-by: Thomas Lamprecht <[email protected]>
Suggested-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Vitaly Cheptsov <[email protected]>
Message-Id: <20210301195919[email protected]>
Tested-by: Thomas Lamprecht <[email protected]>
Reviewed-by: Igor Mammedov <[email protected]>
Reviewed-by: Michael S. Tsirkin <[email protected]>
Signed-off-by: Michael S. Tsirkin <[email protected]>
Fixes: af1b80ae56c9 ("i386/acpi: fix inconsistent QEMU/OVMF device paths")

Merge remote-tracking branch 'remotes/stsquad/tags/pull-testing-updates-240221-1' into staging

Testing tweaks (build, docs, bumps)

  - expose cross compiler info in meson pretty print
  - bump Fedora to 33
  - "graceful" handling of missing virgl config
  - updates to the container documentation
  - move CODING_STYLE.rst into developer manual

# gpg: Signature made Wed 24 Feb 2021 11:08:03 GMT
# gpg:                using RSA key 6685AE99E75167BCAFC8DF35FBD0DB095A9E2A44
# gpg: Good signature from "Alex Bennée (Master Work Key) <[email protected]>" [full]
# Primary key fingerprint: 6685 AE99 E751 67BC AFC8  DF35 FBD0 DB09 5A9E 2A44

* remotes/stsquad/tags/pull-testing-updates-240221-1:
  docs: move CODING_STYLE into the developer documentation
  docs/devel: add forward reference to check-tcg
  docs/devel: update the container based tests
  docs/devel: expand on use of containers to build tests
  tests/acceptance: allow a "graceful" failing for virtio-gpu test
  docker: Bump Fedora images to release 33
  meson.build: expose TCG cross compiler information in summary

Signed-off-by: Peter Maydell <[email protected]>

tcg/i386: rdpmc: fix the the condtions

Signed-off-by: Zheng Zhan Liang <[email protected]>
Message-Id: <20210225054756[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

chardev: do not use short form boolean options in non-QemuOpts character device descriptions

Options such as "-gdb" or "-serial" accept a part-QemuOpts part-parsed-by-hand
character device description.  Do not use short form boolean options in the
QemuOpts part.

Signed-off-by: Paolo Bonzini <[email protected]>

vl: deprecate -writeconfig

The functionality of -writeconfig is limited and the code
does not even try to detect cases where it prints incorrect
syntax (for example if values have a quote in them, since
qemu_config_parse does not support any kind of escaping)
so remove it.

Signed-off-by: Paolo Bonzini <[email protected]>

Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

pc,virtio,pci: fixes, tests

Fixes all over the place, a new test.

Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Tue 23 Feb 2021 16:00:29 GMT
# gpg:                using RSA key 5D09FD0871C8F85B94CA8A0D281F0DB8D28D5469
# gpg:                issuer "[email protected]"
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>" [full]
# gpg:                 aka "Michael S. Tsirkin <[email protected]>" [full]
# Primary key fingerprint: 0270 606B 6F3C DF3D 0B17  0970 C350 3912 AFBE 8E67
#      Subkey fingerprint: 5D09 FD08 71C8 F85B 94CA  8A0D 281F 0DB8 D28D 5469

* remotes/mst/tags/for_upstream:
  qtest/acpi/bios-tables-test: update acpi tables
  acpi: add test case for -no-hpet
  i386: acpi: Don't build HPET ACPI entry if HPET is disabled
  hw/i386: declare ACPI mother board resource for MMCONFIG region
  acpi: add test case for smm unsupported -machine smm=off
  acpi: set fadt.smi_cmd to zero when SMM is not supported
  acpi/core: always set SCI_EN when SMM isn't supported
  ich9, piix4: add property, smm-compat, to keep compatibility of SMM
  qtest: update tests/qtest/bios-tables-test-allowed-diff.h
  checkpatch: don't emit warning on newly created acpi data files
  tests/data/acpi/virt/DSDT.pxb: update with _CCA
  acpi/gpex: Fix cca attribute check for pxb device
  acpi: Allow pxb DSDT acpi table changes
  pcie: don't set link state active if the slot is empty
  failover: really display a warning when the primary device is not found
  virtio-net: add missing object_unref()
  pci: cleanup failover sanity check

Signed-off-by: Peter Maydell <[email protected]>

target/i386: Add bus lock debug exception support

Bus lock debug exception is a feature that can notify the kernel by
generate an #DB trap after the instruction acquires a bus lock when
CPL>0. This allows the kernel to enforce user application throttling or
mitigations.

This feature is enumerated via CPUID.(EAX=7,ECX=0).ECX[bit 24].

Signed-off-by: Chenyi Qiang <[email protected]>
Message-Id: <20210202090224[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

qom/object.c: Fix typo

A simple typo (noticed by inspection).

Signed-off-by: Doug Evans <[email protected]>
Message-Id: <000000000000530c7105bb191b33@google.com>
Signed-off-by: Paolo Bonzini <[email protected]>
Signed-off-by: Doug Evans <[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

target/i386: update to show preferred boolean syntax for -cpu

The preferred syntax is to use "foo=on|off", rather than a bare
"+foo" or "-foo"

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

docs: update to show preferred boolean syntax for -cpu

The preferred syntax is to use "foo=on|off", rather than a bare
"+foo" or "-foo"

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

docs: update to show preferred boolean syntax for -vnc

The preferred syntax is to use "foo=on|off", rather than a bare
"foo" or "nofoo".

The on|off syntax has been supported since -vnc switched to use
QemuOpts in commit 4db14629c38611061fc19ec6927405923de84f08

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

docs: update to show preferred boolean syntax for -chardev

The preferred syntax is to use "foo=on|off", rather than a bare
"foo" or "nofoo".

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

qemu-options: update to show preferred boolean syntax for -vnc

The preferred syntax is to use "foo=on|off", rather than a bare
"foo" or "nofoo".

The on|off syntax has been supported since -vnc switched to use
QemuOpts in commit 4db14629c38611061fc19ec6927405923de84f08

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

qemu-options: update to show preferred boolean syntax for -incoming

The preferred syntax is to use "foo=on|off", rather than a bare
"foo" or "nofoo".

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

qemu-options: update to show preferred boolean syntax for -netdev

The preferred syntax is to use "foo=on|off", rather than a bare
"foo" or "nofoo".

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

qemu-options: update to show preferred boolean syntax for -spice

The preferred syntax is to use "foo=on|off", rather than a bare
"foo" or "nofoo".

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

qemu-options: update to show preferred boolean syntax for -chardev

The preferred syntax is to use "foo=on|off", rather than a bare
"foo" or "nofoo".

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

gdbstub: use preferred boolean option syntax

Signed-off-by: Daniel P. Berrangé <[email protected]>
Message-Id: <20210216191027[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

char: don't fail when client is not connected

This patch checks that ioc is not null before
using it in tcp socket tcp_chr_add_watch function.

The failure occurs in replay mode of the execution,
when monitor and serial port are tcp servers,
and there are no clients connected to them:

-monitor tcp:127.0.0.1:8081,server,nowait
-serial tcp:127.0.0.1:8082,server,nowait

Signed-off-by: Pavel Dovgalyuk <[email protected]>
Reviewed-by: Marc-André Lureau <[email protected]>
Message-Id: <161284977034.741841.12565530923825663110.stgit@pasha-ThinkPad-X280>
Signed-off-by: Paolo Bonzini <[email protected]>

scsi: drop 'result' argument from command_complete callback

The command complete callback has a SCSIRequest as the first argument,
and the status field of that structure is identical to the 'status'
argument. So drop the argument from the callback.

Signed-off-by: Hannes Reinecke <[email protected]>
Message-Id: <20201116184041[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

scsi-disk: pass guest recoverable errors through even for rerror=stop

Right now, recoverable sense values are only passed directly to the
guest only for rerror=report.  However, when rerror/werror are 'stop'
we still don't want the host to be involved on every UNIT ATTENTION
(especially considered that the QMP event will not have enough information
to act on the report).

Signed-off-by: Paolo Bonzini <[email protected]>

scsi-disk: pass SCSI status to scsi_handle_rw_error

Instead of fishing it from *r->status, just pass the SCSI status
as a positive value of the second parameter and an errno as a
negative value.

Signed-off-by: Paolo Bonzini <[email protected]>

scsi: introduce scsi_sense_from_errno()

The new function is an extension of the switch statement in scsi-disk.c
which also includes the errno cases only found in sg_io_sense_from_errno.
This allows us to consolidate the errno handling.

Extracted from a patch by Hannes Reinecke <[email protected]>.

Signed-off-by: Paolo Bonzini <[email protected]>

scsi-disk: do not complete requests early for rerror/werror=ignore

When requested to ignore errors, just do nothing and let the
request complete normally.  This means that the request will
be accounted correctly.

This is what commit 40dce4ee61 ("scsi-disk: fix rerror/werror=ignore",
2018-10-19) was supposed to do:

Fixes: 40dce4ee61 ("scsi-disk: fix rerror/werror=ignore", 2018-10-19)
Signed-off-by: Paolo Bonzini <[email protected]>

scsi-disk: move scsi_handle_rw_error earlier

Remove the forward declaration.

Signed-off-by: Paolo Bonzini <[email protected]>

virtio-scsi: don't process IO on fenced dataplane

If virtio_scsi_dataplane_start fails, there is a small window when it drops the
aio lock (in aio_wait_bh_oneshot) and the dataplane's AIO handler can
still run during that window.

This is done after the dataplane was marked as fenced, thus we use this flag
to avoid it doing any IO.

Signed-off-by: Maxim Levitsky <[email protected]>
Message-Id: <20201217150040[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

scsi: allow user to set werror as report

'enospc' is the default for -drive, but qemu allows user to set
drive option werror. If werror of scsi-generic is set to 'report'
by user, qemu will not allow vm to start.

This patch allow user to set werror as 'report' for scsi-generic.

Signed-off-by: Zihao Chang <[email protected]>
Reviewed-by: Fam Zheng <[email protected]>
Message-Id: <20201103061240[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

scsi: add tracing for SG_IO commands

Add tracepoints for SG_IO commands to allow for debugging
of SG_IO commands.

Signed-off-by: Hannes Reinecke <[email protected]>
Message-Id: <20201116183114[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

scsi: make io_timeout configurable

The current code sets an infinite timeout on SG_IO requests,
causing the guest to stall if the host experiences a frame
loss.
This patch adds an 'io_timeout' parameter for SCSIDevice to
make the SG_IO timeout configurable, and also shortens the
default timeout to 30 seconds to avoid infinite stalls.

Signed-off-by: Hannes Reinecke <[email protected]>
Message-Id: <20201116183114[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

multiprocess: move feature to meson_options.txt

While at it, improve the description of the feature in the summary and
help message.

Signed-off-by: Paolo Bonzini <[email protected]>

configure: fix --enable-fuzzing linker failures

With --enable-fuzzing, QEMU_CFLAGS include -fsanitize=fuzzer-no-link.
This should allow us to build non-fuzzer binaries using objects
instrumented for fuzzing. However, to do that, we also need to link with
-fsanitize=fuzzer-no-link. We were not doing that.

Reported-by: Li Qiang <[email protected]>,
Signed-off-by: Alexander Bulekov <[email protected]>
Message-Id: <20210221174510[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

hvf: Sign the code after installation

Before this change, the code signed during the build was installed
directly.

However, the signature gets invalidated because meson modifies the code
to fix dynamic library install names during the install process.

It also prevents meson to strip the code because the pre-signed file is
not marked as an executable (although it is somehow able to perform the
modification described above).

With this change, the unsigned code will be installed and modified by
meson first, and a script signs it later.

Signed-off-by: Akihiko Odaki <[email protected]>
Message-Id: <20210225000614[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>