Jamin Lin [Thu, 22 May 2025 02:33:02 +0000 (10:33 +0800)]
hw/intc/aspeed: Set impl.min_access_size to 4
This patch explicitly sets ".impl.min_access_size = 4" to match the
declared ".valid.min_access_size = 4", enforcing stricter access size
checking and preventing inconsistent partial accesses to the interrupt
controller registers.
Jamin Lin [Thu, 15 May 2025 08:10:00 +0000 (16:10 +0800)]
test/qtest/hace: Add tests for AST2700
The HACE models in AST2600 and AST2700 are nearly identical. Based on the
AST2600 test cases, new tests have been added for AST2700.
Implemented test functions for SHA-256, SHA-384, SHA-512, and MD5.
Added scatter-gather and accumulation test variants.
For AST2700, the HACE controller base address starts at "0x12070000", and
the DRAM start address is "0x4_00000000".
Jamin Lin [Thu, 15 May 2025 08:09:59 +0000 (16:09 +0800)]
test/qtest/hace: Support to validate 64-bit hmac key buffer addresses
Added "key" and "key_hi" fields to "AspeedMasks" for 64-bit addresses test.
Updated "aspeed_test_addresses" to validate "HACE_HASH_KEY_BUFF" and
"HACE_HASH_KEY_BUFF_HI".
Ensured correct masking of 64-bit addresses by checking both lower and upper
32-bit registers.
Jamin Lin [Thu, 15 May 2025 08:09:58 +0000 (16:09 +0800)]
test/qtest/hace: Support to test upper 32 bits of digest and source addresses
Added "src_hi" and "dest_hi" fields to "AspeedMasks" for 64-bit addresses test.
Updated "aspeed_test_addresses" to validate "HACE_HASH_SRC_HI" and
"HACE_HASH_DIGEST_HI".
Ensured correct masking of 64-bit addresses by checking both lower and upper
32-bit registers.
Jamin Lin [Thu, 15 May 2025 08:09:57 +0000 (16:09 +0800)]
test/qtest/hace: Support 64-bit source and digest addresses for AST2700
Added "HACE_HASH_SRC_HI" and "HACE_HASH_DIGEST_HI", "HACE_HASH_KEY_BUFF_HI"
registers to store upper 32 bits.
Updated "write_regs" to handle 64-bit source and digest addresses.
Jamin Lin [Thu, 15 May 2025 08:09:56 +0000 (16:09 +0800)]
test/qtest/hace: Update source data and digest data type to 64-bit
Currently, the hash data source and digest result buffer addresses are set to
32-bit. However, the AST2700 CPU is a 64-bit Cortex-A35 architecture, and its
DRAM base address is also 64-bit.
To support AST2700, update the hash data source address and digest result buffer
address to use 64-bit addressing.
Jamin Lin [Thu, 15 May 2025 08:09:55 +0000 (16:09 +0800)]
test/qtest/hace: Add tests for AST1030
The HACE model in AST2600 and AST1030 is identical. Referencing the AST2600
test cases, new tests have been created for AST1030.
Implemented test functions for SHA-256, SHA-384, SHA-512, and MD5.
Added scatter-gather and accumulation test variants.
For AST1030, the HACE controller base address starts at "0x7e6d0000", and the
SDRAM start address is "0x0".
Jamin Lin [Thu, 15 May 2025 08:09:54 +0000 (16:09 +0800)]
test/qtest/hace: Add SHA-384 tests for AST2600
Introduced "test_sha384_ast2600" to validate SHA-384 hashing.
Added "test_sha384_sg_ast2600" for scatter-gather SHA-384 verification.
Implemented "test_sha384_accum_ast2600" to test SHA-384 accumulation.
Registered new test cases in "main" to ensure execution.
Jamin Lin [Thu, 15 May 2025 08:09:53 +0000 (16:09 +0800)]
test/qtest/hace: Add SHA-384 test cases for ASPEED HACE model
Introduced SHA-384 test functions to verify hashing operations.
Extended support for scatter-gather ("_sg") and accumulation ("_accum") tests.
Updated test result vectors for SHA-384 validation.
Jamin Lin [Thu, 15 May 2025 08:09:52 +0000 (16:09 +0800)]
test/qtest/hace: Adjust test address range for AST1030 due to SRAM limitations
The digest_addr is set to "src_addr + 0x1000000", where src_addr is the DRAM
base address. However, the value 0x1000000 (16MB) is too large because the
AST1030 does not support DRAM, and its SRAM size is only 768KB.
A range size of 0x10000 (64KB) is sufficient for HACE test cases, as the test
vector size does not exceed 64KB.
Updates:
1. Direct Access Mode
Update digest_addr to "src_addr + 0x10000" in the following functions:
aspeed_test_md5
aspeed_test_sha256
aspeed_test_sha512
2. Scatter-Gather (SG) Mode
Update source address for different SG buffer addresses in the following
functions:
src_addr1 = src_addr + 0x10000
src_addr2 = src_addr + 0x20000
src_addr3 = src_addr + 0x30000
digest_addr = src_addr + 0x40000
aspeed_test_sha256_sg
aspeed_test_sha512_sg
3. ACC Mode Update
Update the SG List start address: src_addr + 0x10000
Update the SG List buffer size to 0x30000 (192KB).
Jamin Lin [Thu, 15 May 2025 08:09:51 +0000 (16:09 +0800)]
test/qtest/hace: Specify explicit array sizes for test vectors and hash results
To enhance code readability and prevent potential buffer overflows or unintended
size assumptions, this commit updates all fixed-size array declarations to use
explicit array sizes.
Jamin Lin [Thu, 15 May 2025 08:09:50 +0000 (16:09 +0800)]
test/qtest: Introduce a new aspeed-hace-utils.c to place common testcases
The test cases for the ASPEED HACE model were originally placed in
aspeed_hace-test.c. However, this test file only supports ARM32. To enable
compatibility with all ASPEED SoCs, including the AST2700, which uses the
AArch64 architecture, this update introduces a new source file,
aspeed-hace-utils.c.
All common APIs and test cases have been moved from aspeed_hace-test.c to
aspeed-hace-utils.c to facilitate reuse across different ASPEED SoCs.
As a result, these test cases can now be reused for AST2700 and future ASPEED
SoC testing.
Jamin Lin [Thu, 15 May 2025 08:09:48 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Support to dump plaintext and digest for better debugging
1. Added "hace_hexdump()" to dump a contiguous buffer using qemu_hexdump.
2. Added "hace_iov_hexdump()" to flatten and dump scatter-gather source vectors.
3. Introduced a new trace event: "aspeed_hace_hexdump".
Jamin Lin [Thu, 15 May 2025 08:09:46 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Support DMA 64 bits dram address
According to the AST2700 design, the data source address is 64-bit, with
R_HASH_SRC_HI storing bits [63:32] and R_HASH_SRC storing bits [31:0].
Similarly, the digest address is 64-bit, with R_HASH_DEST_HI storing bits
[63:32] and R_HASH_DEST storing bits [31:0].
To maintain compatibility with older SoCs such as the AST2600, the AST2700 HW
automatically set bit 34 of the 64-bit sg_addr. As a result, the firmware
only needs to provide a 32-bit sg_addr containing bits [31:0]. This is
sufficient for the AST2700, as it uses a DRAM offset rather than a DRAM
address.
Introduce a has_dma64 class attribute and set it to true for the AST2700.
Jamin Lin [Thu, 15 May 2025 08:09:45 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Add support for source, digest, key buffer 64 bit addresses
According to the AST2700 design, the data source address is 64-bit, with
R_HASH_SRC_HI storing bits [63:32] and R_HASH_SRC storing bits [31:0].
Similarly, the digest address is 64-bit, with R_HASH_DIGEST_HI storing bits
[63:32] and R_HASH_DIGEST storing bits [31:0]. The HMAC key buffer address is also
64-bit, with R_HASH_KEY_BUFF_HI storing bits [63:32] and R_HASH_KEY_BUFF storing
bits [31:0].
The AST2700 supports a maximum DRAM size of 8 GB, with a DRAM addressable range
from 0x0_0000_0000 to 0x1_FFFF_FFFF. Since this range fits within 34 bits, only
bits [33:0] are needed to store the DRAM offset. To optimize address storage,
the high physical address bits [1:0] of the source, digest and key buffer
addresses are stored as dram_offset bits [33:32].
To achieve this, a src_hi_mask with a mask value of 0x3 is introduced, ensuring
that src_addr_hi consists of bits [1:0]. The final src_addr is computed as
(src_addr_hi[1:0] << 32) | src_addr[31:0], representing the DRAM offset within
bits [33:0].
Similarly, a dest_hi_mask with a mask value of 0x3 is introduced to ensure that
dest_addr_hi consists of bits [1:0]. The final dest_addr is calculated as
(dest_addr_hi[1:0] << 32) | dest_addr[31:0], representing the DRAM offset within
bits [33:0].
Additionally, a key_hi_mask with a mask value of 0x3 is introduced to ensure
that key_buf_addr_hi consists of bits [1:0]. The final key_buf_addr is
determined as (key_buf_addr_hi[1:0] << 32) | key_buf_addr[31:0], representing
the DRAM offset within bits [33:0].
This approach eliminates the need to reduce the high part of the DRAM physical
address for DMA operations. Previously, this was calculated as
(high physical address bits [7:0] - 4), since the DRAM start address is
0x4_00000000, making the high part address [7:0] - 4.
Jamin Lin [Thu, 15 May 2025 08:09:44 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Move register size to instance class and dynamically allocate regs
Dynamically allocate the register array by removing the hardcoded
ASPEED_HACE_NR_REGS macro.
To support different register sizes across SoC variants, introduce a new
"nr_regs" class attribute and replace the static "regs" array with dynamically
allocated memory.
Add a new "aspeed_hace_unrealize" function to properly free the allocated "regs"
memory during device cleanup.
Remove the bounds checking in the MMIO read/write handlers since the
MemoryRegion size now matches the (register array size << 2).
This commit updates the VMState fields accordingly. The VMState version was
already bumped in a previous patch of this series, so no further version change
is needed.
Jamin Lin [Thu, 15 May 2025 08:09:43 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Support accumulative mode for direct access mode
Enable accumulative mode for direct access mode operations. In direct access
mode, only a single source buffer is used, so the "iovec" count is set to 1.
If "acc_mode" is enabled:
1. Accumulate "total_req_len" with the current request length ("plen").
2. Check for padding and determine whether this is the final request.
Jamin Lin [Thu, 15 May 2025 08:09:42 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Rename R_HASH_DEST to R_HASH_DIGEST and introduce 64-bit hash digest address helper
Renaming R_HASH_DEST to R_HASH_DIGEST for better semantic clarity.
The AST2700 CPU, based on the Cortex-A35, features a 64-bit DRAM address space.
To prepare for future AST2700 support, this change introduces a new helper
function hash_get_digest_addr() to encapsulate digest address extraction logic
and improve code readability.
Jamin Lin [Thu, 15 May 2025 08:09:41 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Introduce 64-bit hash source address helper function
The AST2700 CPU, based on the Cortex-A35, is a 64-bit processor, and its DRAM
address space is also 64-bit. To support future AST2700 updates, the source
hash buffer address data type is being updated to 64-bit.
Introduces the "hash_get_source_addr()" helper function to extract the source hash
buffer address.
Jamin Lin [Thu, 15 May 2025 08:09:40 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Extract accumulation-mode hash execution into helper function
To improve code readability and maintainability of do_hash_operation(), this
commit introduces a new helper function: hash_execute_acc_mode().
This function encapsulates the full flow for accumulation mode, including
context initialization, update, conditional finalization, and digest writeback
with I/O vector unmapping.
Jamin Lin [Thu, 15 May 2025 08:09:38 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Extract digest write and iov unmap into helper function
To improve code readability and maintainability of do_hash_operation(), this
commit introduces a new helper function: hash_write_digest_and_unmap_iov().
The helper consolidates the final digest writeback and subsequent unmapping of
the I/O vectors into a single routine.
Jamin Lin [Thu, 15 May 2025 08:09:37 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Extract SG-mode hash buffer setup into helper function
To improve code readability and maintainability of do_hash_operation(), this
commit introduces a new helper function: hash_prepare_sg_iov().
This function handles scatter-gather (SG) mode setup, including SG list
parsing, address mapping, and optional accumulation mode support with
padding detection.
Jamin Lin [Thu, 15 May 2025 08:09:36 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Extract direct mode hash buffer setup into helper function
To improve code readability and maintainability of do_hash_operation(), this
commit introduces a new helper function: hash_prepare_direct_iov().
This function encapsulates the logic for setting up the I/O vector (iov)
in direct mode (non-scatter-gather).
Jamin Lin [Thu, 15 May 2025 08:09:35 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Ensure HASH_IRQ is always set to prevent firmware hang
Currently, if the program encounters an unsupported algorithm, it does not set
the HASH_IRQ bit in the status register and send an interrupt to indicate
command completion. As a result, the FW gets stuck waiting for a completion
signal from the HACE module.
Additionally, in do_hash_operation, if an error occurs within the conditional
statement, the HASH_IRQ bit is not set in the status register. This causes the
firmware to continuously send HASH commands, as it is unaware that the HACE
model has completed processing the command.
To fix this, the HASH_IRQ bit in the status register must always be set to
ensure that the firmware receives an interrupt from the HACE module, preventing
it from getting stuck or repeatedly sending HASH commands.
Jamin Lin [Thu, 15 May 2025 08:09:34 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Improve readability and consistency in variable naming
Currently, users define multiple local variables within different if-statements.
To improve readability and maintain consistency in variable naming, rename the
variables accordingly.
Introduced "sg_addr" to clearly indicate the scatter-gather mode buffer address.
Jamin Lin [Thu, 15 May 2025 08:09:33 +0000 (16:09 +0800)]
hw/misc/aspeed_hace: Remove unused code for better readability
In the previous design of the hash framework, accumulative hashing was not
supported. To work around this limitation, commit 5cd7d85 introduced an
iov_cache array to store all the hash data from firmware.
Once the ASPEED HACE model collected all the data, it passed the iov_cache to
the hash API to calculate the final digest.
However, with commit e3c0752, the hash framework now supports accumulative
hashing. This allows us to refactor the ASPEED HACE model, removing redundant
logic and simplifying the implementation for better readability and
maintainability.
As a result, the iov_count variable is no longer needed—it was previously used
to track how many cached entries were used for hashing.
To maintain VMSTATE compatibility after removing this field, the VMSTATE_VERSION
is bumped to 2
This cleanup follows significant changes in commit 4c1d0af4a28d, making the
model more readable.
- Deleted "iov_cache" and "iov_count" from "AspeedHACEState".
- Removed "reconstruct_iov" function and related logic.
- Simplified "do_hash_operation" by eliminating redundant checks.
Fabiano Rosas [Fri, 9 May 2025 17:50:47 +0000 (14:50 -0300)]
tests/qtest/ast2700-smc-test: Fix leak
ASAN spotted a leak of the memory used to hold the tmp_path:
Direct leak of 35 byte(s) in 1 object(s) allocated from:
#0 0x55e29aa96da9 in malloc ../projects/compiler-rt/lib/asan/asan_malloc_linux.cpp:69:3
#1 0x7fe0cfb26518 in g_malloc ../glib/gmem.c:106
#2 0x7fe0cfb4146e in g_strconcat ../glib/gstrfuncs.c:629
#3 0x7fe0cfb0a78f in g_get_tmp_name ../glib/gfileutils.c:1742
#4 0x7fe0cfb0b00b in g_file_open_tmp ../glib/gfileutils.c:1802
#5 0x55e29ab53961 in test_ast2700_evb ../tests/qtest/ast2700-smc-test.c:20:10
#6 0x55e29ab53803 in main ../tests/qtest/ast2700-smc-test.c:65:5
#7 0x7fe0cf7bd24c in __libc_start_main ../csu/libc-start.c:308
#8 0x55e29a9f7759 in _start ../sysdeps/x86_64/start.S:120
Signed-off-by: Fabiano Rosas <farosas@suse.de> Reviewed-by: Jamin Lin <jamin_lin@aspeedtech.com>
Message-ID: <20250509175047.26066-1-farosas@suse.de> Signed-off-by: Cédric Le Goater <clg@redhat.com>
* tag 'audio-pull-request' of https://gitlab.com/marcandre.lureau/qemu:
audio: Reset rate control when adding bytes
alsaaudio: Set try-poll to false by default
audio: add float sample endianness converters
audio/mixeng: remove unnecessary pointer type casts
hw/audio/asc: replace g_malloc0() with g_malloc()
hw/audio/asc: fix SIGSEGV in asc_realize()
audio: fix size calculation in AUD_get_buffer_size_out()
audio: fix SIGSEGV in AUD_get_buffer_size_out()
tests/functional: use 'none' audio driver for q800 tests
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
* tag 'ui-pull-request' of https://gitlab.com/marcandre.lureau/qemu:
ui/gtk-egl: Render guest content with padding in fixed-scale mode
ui/gtk-gl-area: Render guest content with padding in fixed-scale mode
ui/gtk: Consider scaling when propagating ui info
ui/gtk: Don't update scale in fixed scale mode in gtk-egl.c
ui/sdl: Consider scaling in mouse event handling
ui/gtk: Update scales in fixed-scale mode when rendering GL area
gtk/ui: Introduce helper gd_update_scale
ui/gtk: Use consistent naming for variables in different coordinates
ui/gtk: Document scale and coordinate handling
ui/vdagent: remove migration blocker
ui/vdagent: add migration support
ui/vdagent: factor out clipboard peer registration
ui/vdagent: keep "connected" state
ui/vdagent: replace Buffer with GByteArray
ui/clipboard: delay clipboard update when not running
ui/clipboard: add vmstate_cbinfo
ui/clipboard: split out QemuClipboardContent
ui/clipboard: use int for selection field
ui/gtk: warn if setting the clipboard failed
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Akihiko Odaki [Mon, 17 Mar 2025 08:05:29 +0000 (17:05 +0900)]
audio: Reset rate control when adding bytes
Commit 90320051ea99 ("spiceaudio: add a pcm_ops buffer_get_free
function") caused to emit messages saying "Resetting rate control"
frequently when the guest generates no frames.
audio_rate_peek_bytes() resets the rate control when frames < 0 ||
frames > 65536 where frames is the rate-limited number of frames.
Resetting when frames < 0 is sensible as the number simply doesn't make
sense.
There is a problem when frames > 65536. It implies the guest stopped
generating frames for a while so it makes sense to reset the rate
control when the guest resumed generating frames. However, the
commit mentioned earlier broke this assumption by letting spiceaudio
call audio_rate_peek_bytes() whether the guest is generating frames or
not.
Reset the rate control in audio_rate_add_bytes(), which is called only
when actually adding frames, according to the previous call to
audio_rate_peek_bytes() to avoid frequent rate control resets even when
the guest generates no frame.
BALATON Zoltan [Sun, 16 Mar 2025 00:20:46 +0000 (01:20 +0100)]
alsaaudio: Set try-poll to false by default
Quoting Volker Rümelin: "try-poll=on tells the ALSA backend to try to
use an event loop instead of the audio timer. This works most of the
time. But the poll event handler in the ALSA backend has a bug. For
example, if the guest can't provide enough audio frames in time, the
ALSA buffer is only partly full and the event handler will be called
again and again on every iteration of the main loop. This increases
the processor load and the guest has less processor time to provide
new audio frames in time. I have two examples where a guest can't
recover from this situation and the guest seems to hang."
One reproducer I've found is booting MorphOS demo iso on
qemu-system-ppc -machine pegasos2 -audio alsa which should play a
startup sound but instead it freezes. Even when it does not hang it
plays choppy sound. Volker suggested using command line to set
try-poll=off saying: "The try-poll=off arguments are typically
necessary, because the alsa backend has a design issue with
try-poll=on. If the guest can't provide enough audio frames, it's
really unhelpful to ask for new audio frames on every main loop
iteration until the guest can provide enough audio frames. Timer based
playback doesn't have that problem."
But users cannot easily find this option and having a non-working
default is really unhelpful so to make life easier just set it to
false by default which works until the issue with the alsa backend can
be fixed.
Volker Rümelin [Thu, 15 May 2025 05:44:29 +0000 (07:44 +0200)]
audio: add float sample endianness converters
Commit ed2a4a7941 ("audio: proper support for float samples in
mixeng") added support for float audio samples. As there were no
audio frontend devices with float support at that time, the code
was limited to native endian float samples.
When nobody was paying attention, an audio device that supports
floating point samples crept in with commit eb9ad377bb
("virtio-sound: handle control messages and streams").
Add code for the audio subsystem to convert float samples to the
correct endianness.
The type punning code was taken from the PipeWire project.
Volker Rümelin [Thu, 15 May 2025 05:44:26 +0000 (07:44 +0200)]
hw/audio/asc: fix SIGSEGV in asc_realize()
AUD_open_out() may fail and return NULL. This may then lead to
a segmentation fault in memset() below. The memset() behaviour
is undefined if the pointer to the destination object is a null
pointer.
Volker Rümelin [Thu, 15 May 2025 05:44:25 +0000 (07:44 +0200)]
audio: fix size calculation in AUD_get_buffer_size_out()
The buffer size calculated by AUD_get_buffer_size_out() is often
incorrect. sw->hw->samples * sw->hw->info.bytes_per_frame is the
size of the mixing engine buffer in audio frames multiplied by
the size of one frame of the audio backend. Due to resampling or
format conversion, the size of the frontend buffer can differ
significantly.
Return the correct buffer size when the mixing engine is used.
Volker Rümelin [Thu, 15 May 2025 05:44:24 +0000 (07:44 +0200)]
audio: fix SIGSEGV in AUD_get_buffer_size_out()
As far as the emulated audio devices are concerned the pointer
returned by AUD_open_out() is an opaque handle. This includes
the NULL pointer. In this case, AUD_get_buffer_size_out() should
return a sensible buffer size instead of triggering a segmentation
fault. All other public AUD_*_out() and audio_*_out() functions
handle this case.
Volker Rümelin [Thu, 15 May 2025 05:44:23 +0000 (07:44 +0200)]
tests/functional: use 'none' audio driver for q800 tests
Since commit ac13a6b3fd ("audio: add Apple Sound Chip (ASC)
emulation") the Quadra 800 machine has an audio device. It is
not guaranteed that the default audio driver of the audio
subsystem will work correctly on all host systems. Therefore,
the 'none' audio driver should be used in all q800 tests.
Weifeng Liu [Sun, 11 May 2025 07:33:19 +0000 (15:33 +0800)]
ui/gtk-egl: Render guest content with padding in fixed-scale mode
Scaling was not respected when rendering frames in gtk-egl.c (used if
gl=on and X11 mode). To fix this, add fields x and y to struct egl_fb
for x offset and y offset so we can add padding to window.
Weifeng Liu [Sun, 11 May 2025 07:33:18 +0000 (15:33 +0800)]
ui/gtk-gl-area: Render guest content with padding in fixed-scale mode
In fixed-scale mode (zoom-to-fit=false), we expect that scale should not
change, meaning that if window size is larger than guest surface,
padding is supposed to be added to preserve the scale. However, in
OpenGL mode (gl=on), guest surface is always painted to the whole canvas
without any padding. This change tries to fix this bug by adding
appropriate padding when drawing surfaces.
Weifeng Liu [Sun, 11 May 2025 07:33:17 +0000 (15:33 +0800)]
ui/gtk: Consider scaling when propagating ui info
The ui width and height sent to guest is supposed to be in buffer
coordinate. Hence conversion is required.
If scaling (global window scale and zooming scale) is not respected in
non-free-scale mode, window size could keep changing because of the
existence of the iteration of the following steps:
1. In resize event or configure event, a size larger (or smaller) than
the currently used one might be calculated due to not considering
scaling.
2. On reception of the display size change event in guest, the guest
might decide to do a mode setting and use the larger (or smaller)
mode.
3. When the new guest scan-out command arrives, QEMU would request the
window size to change to fit the new buffer size. This will trigger a
resize event or a configure event, making us go back to step 1.
Weifeng Liu [Sun, 11 May 2025 07:33:16 +0000 (15:33 +0800)]
ui/gtk: Don't update scale in fixed scale mode in gtk-egl.c
Scale shouldn't be changed until user explicitly requests it in fixed
scale mode (full-screen=false and free-scale=false). Use function
gd_update_scale to complete scale updating instead.
Weifeng Liu [Sun, 11 May 2025 07:33:15 +0000 (15:33 +0800)]
ui/sdl: Consider scaling in mouse event handling
When using sdl display backend, if the window is scaled, incorrect mouse
positions will be reported since scaling is not properly handled. Fix it
by transforming the positions from window coordinate to guest buffer
coordinate.
Weifeng Liu [Sun, 11 May 2025 07:33:14 +0000 (15:33 +0800)]
ui/gtk: Update scales in fixed-scale mode when rendering GL area
When gl=on, scale_x and scale_y were set to 1 on startup that didn't
reflect the real situation of the scan-out in free scale mode, resulting
in incorrect cursor coordinates to be sent when moving the mouse
pointer. Simply updating the scales before rendering the image fixes
this issue.
Weifeng Liu [Sun, 11 May 2025 07:33:11 +0000 (15:33 +0800)]
ui/gtk: Document scale and coordinate handling
The existence of multiple scaling factors forces us to deal with various
coordinate systems and this would be confusing. It would be beneficial
to define the concepts clearly and use consistent representation for
variables in different coordinates.
During post-load of migration, virtio will notify of fe_open state.
However vdagent code will handle this as a reconnection. This will
trigger a connection reset/caps with the agent.
Check if the state actually changed before resetting the connection.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Buffer is slightly more advanced than GByteArray, since it has a
cursor/position. But vdagent code doesn't need it. This simplify a bit
the code, and migration state.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Each clipboard owner will have to save its QemuClipboardInfo and
reregister its owned clipboard after loading. (the global cbinfo has
only pointers to owners, so it can't restore the relation with its owner
if it was to handle migration)
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
* tag 'for-upstream' of https://repo.or.cz/qemu/kevin:
file-posix: Probe paths and retry SG_IO on potential path errors
file-posix: allow BLKZEROOUT with -t writeback
qemu-img: fix offset calculation in bench
scsi-disk: Advertise FUA support by default
scsi-disk: Add native FUA write support
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Stefan Hajnoczi [Thu, 22 May 2025 18:48:55 +0000 (14:48 -0400)]
Merge tag 'misc-next-pull-request' of https://gitlab.com/berrange/qemu into staging
Misc VNC, I/O, Crypto & checkpatch changes
* Fix VNC tight encoding with 8/16-bpp formats with
mixed endian server/client
* Fix VNC non-tight encoding with mixed endian server/client
* Drop built-in AES impl from non-TCG usage, requiring
nettle/gcrypt/gnutls
* Fix validation of SPDX-License-Identifier in new files
* Mandate SPDX-License-Identifier in Rust source
* Reject license boilerplate in new files
* Add full control over TCP keep alive setting for sockets
* tag 'misc-next-pull-request' of https://gitlab.com/berrange/qemu: (23 commits)
scripts/checkpatch.pl: mandate SPDX tag for Rust src files
util/qemu-sockets: Introduce inet socket options controlling TCP keep-alive
util/qemu-sockets: Refactor inet_parse() to use QemuOpts
util/qemu-sockets: Add support for keep-alive flag to passive sockets
util/qemu-sockets: Refactor success and failure paths in inet_listen_saddr()
util/qemu-sockets: Refactor setting client sockopts into a separate function
io: Fix partial struct copy in qio_dns_resolver_lookup_sync_inet()
scripts/checkpatch: reject license boilerplate on new files
scripts/checkpatch: reimplement mandate for SPDX-License-Identifier
scripts/checkpatch: use new hook for MAINTAINERS update check
scripts/checkpatch: expand pattern for matching makefiles
scripts/checkpatch: use new hook for file permissions check
scripts/checkpatch: use new hook for ACPI test data check
scripts/checkpatch: introduce tracking of file start/end
scripts/checkpatch.pl: fix various indentation mistakes
Revert "scripts: mandate that new files have SPDX-License-Identifier"
crypto: fully drop built-in cipher provider
tests: fix skipping cipher tests when AES is not available
tests: skip legacy qcow2 encryption test if AES is not available
tests: skip encrypted secret tests if AES is not available
...
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Kevin Wolf [Thu, 22 May 2025 13:08:03 +0000 (15:08 +0200)]
file-posix: Probe paths and retry SG_IO on potential path errors
When scsi-block is used on a host multipath device, it runs into the
problem that the kernel dm-mpath doesn't know anything about SCSI or
SG_IO and therefore can't decide if a SG_IO request returned an error
and needs to be retried on a different path. Instead of getting working
failover, an error is returned to scsi-block and handled according to
the configured error policy. Obviously, this is not what users want,
they want working failover.
QEMU can parse the SG_IO result and determine whether this could have
been a path error, but just retrying the same request could just send it
to the same failing path again and result in the same error.
With a kernel that supports the DM_MPATH_PROBE_PATHS ioctl on dm-mpath
block devices (queued in the device mapper tree for Linux 6.16), we can
tell the kernel to probe all paths and tell us if any usable paths
remained. If so, we can now retry the SG_IO ioctl and expect it to be
sent to a working path.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250522130803.34738-1-kwolf@redhat.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Hanna Czenczek <hreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Stefan Hajnoczi [Thu, 17 Apr 2025 21:10:53 +0000 (17:10 -0400)]
file-posix: allow BLKZEROOUT with -t writeback
The Linux BLKZEROOUT ioctl is only invoked when BDRV_O_NOCACHE is set
because old kernels did not invalidate the page cache. In that case
mixing BLKZEROOUT with buffered I/O could lead to corruption.
However, Linux 4.9 commit 22dd6d356628 ("block: invalidate the page
cache when issuing BLKZEROOUT") made BLKZEROOUT coherent with the page
cache.
I have checked that Linux 4.9+ kernels are shipped at least as far back
as Debian 10 (buster), openSUSE Leap 15.2, and RHEL/CentOS 8.
Use BLKZEROOUT with buffered I/O, mostly so `qemu-img ... -t
writeback` can offload write zeroes.
Cc: Paolo Bonzini <pbonzini@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20250417211053.98700-1-stefanha@redhat.com> Reviewed-by: Eric Blake <eblake@redhat.com> Reviewed-by: Kevin Wolf <kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Denis Rastyogin [Tue, 6 May 2025 14:13:37 +0000 (17:13 +0300)]
qemu-img: fix offset calculation in bench
This error was discovered by fuzzing qemu-img.
The current offset calculation leads to an EIO error
in block/block-backend.c: blk_check_byte_request():
if (offset > len || len - offset < bytes) {
return -EIO;
}
This triggers the error message:
"qemu-img: Failed request: Input/output error".
Example of the issue:
offset: 260076
len: 260096
bytes: 4096
This fix ensures that offset remains within a valid range.
Signed-off-by: Denis Rastyogin <gerben@altlinux.org>
Message-ID: <20250506141410.100119-1-gerben@altlinux.org>
[kwolf: Fixed up integer overflow] Signed-off-by: Kevin Wolf <kwolf@redhat.com>
With the default TCP stack configuration, it could be even 2 hours
before the connection times out due to the other side not being
reachable. However, in some cases, the application needs to be aware of
a connection issue much sooner.
This is the case, for example, for postcopy live migration. If there is
no traffic from the migration destination guest (server-side) to the
migration source guest (client-side), the destination keeps waiting for
pages indefinitely and does not switch to the postcopy-paused state.
This can happen, for example, if the destination QEMU instance is
started with the '-S' command line option and the machine is not started
yet, or if the machine is idle and produces no new page faults for
not-yet-migrated pages.
This patch introduces new inet socket parameters that control count,
idle period, and interval of TCP keep-alive packets before the
connection is considered broken. These parameters are available on
systems where the respective TCP socket options are defined, that
includes Linux, Windows, macOS, but not OpenBSD. Additionally, macOS
defines TCP_KEEPIDLE as TCP_KEEPALIVE instead, so the patch supplies its
own definition.
The default value for all is 0, which means the system configuration is
used.
Signed-off-by: Juraj Marcin <jmarcin@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Juraj Marcin [Wed, 21 May 2025 13:52:34 +0000 (15:52 +0200)]
util/qemu-sockets: Refactor inet_parse() to use QemuOpts
Currently, the inet address parser cannot handle multiple options where
one is prefixed with the name of the other. For example, with the
'keep-alive-idle' option added, the current parser cannot parse
'127.0.0.1:5000,keep-alive-idle=60,keep-alive' correctly. Instead, it
fails with "error parsing 'keep-alive' flag '-idle=60,keep-alive'".
To resolve these issues, this patch rewrites the inet address parsing
using the QemuOpts parser, which the inet_parse_flag() function tries to
mimic. This new parser supports all previously supported options and on
top of that the 'numeric' flag is now also supported. The only
difference is, the new parser produces an error if an unknown option is
passed, instead of silently ignoring it.
Signed-off-by: Juraj Marcin <jmarcin@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Juraj Marcin [Wed, 21 May 2025 13:52:33 +0000 (15:52 +0200)]
util/qemu-sockets: Add support for keep-alive flag to passive sockets
Commit aec21d3175 (qapi: Add InetSocketAddress member keep-alive)
introduces the keep-alive flag, which enables the SO_KEEPALIVE socket
option, but only on client-side sockets. However, this option is also
useful for server-side sockets, so they can check if a client is still
reachable or drop the connection otherwise.
This patch enables the SO_KEEPALIVE socket option on passive server-side
sockets if the keep-alive flag is enabled. This socket option is then
inherited by active server-side sockets communicating with connected
clients.
Signed-off-by: Juraj Marcin <jmarcin@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Juraj Marcin [Wed, 21 May 2025 13:52:32 +0000 (15:52 +0200)]
util/qemu-sockets: Refactor success and failure paths in inet_listen_saddr()
To get a listening socket, we need to first create a socket, try binding
it to a certain port, and lastly starting listening to it. Each of these
operations can fail due to various reasons, one of them being that the
requested address/port is already in use. In such case, the function
tries the same process with a new port number.
This patch refactors the port number loop, so the success path is no
longer buried inside the 'if' statements in the middle of the loop. Now,
the success path is not nested and ends at the end of the iteration
after successful socket creation, binding, and listening. In case any of
the operations fails, it either continues to the next iteration (and the
next port) or jumps out of the loop to handle the error and exits the
function.
Signed-off-by: Juraj Marcin <jmarcin@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Juraj Marcin [Wed, 21 May 2025 13:52:31 +0000 (15:52 +0200)]
util/qemu-sockets: Refactor setting client sockopts into a separate function
This is done in preparation for enabling the SO_KEEPALIVE support for
server sockets and adding settings for more TCP keep-alive socket
options.
Signed-off-by: Juraj Marcin <jmarcin@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Juraj Marcin [Wed, 21 May 2025 13:52:30 +0000 (15:52 +0200)]
io: Fix partial struct copy in qio_dns_resolver_lookup_sync_inet()
Commit aec21d3175 (qapi: Add InetSocketAddress member keep-alive)
introduces the keep-alive flag, but this flag is not copied together
with other options in qio_dns_resolver_lookup_sync_inet().
This patch fixes this issue and also prevents future ones by copying the
entire structure first and only then overriding a few attributes that
need to be different.
Fixes: aec21d31756c (qapi: Add InetSocketAddress member keep-alive) Signed-off-by: Juraj Marcin <jmarcin@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
scripts/checkpatch: reject license boilerplate on new files
The previous commit mandates use of SPDX-License-Identifier on common
source files, and encourages it on all other files.
Some contributors are none the less still also including the license
boilerplate text. This is redundant and will potentially cause
trouble if inconsistent with the SPDX declaration.
Match common boilerplate text blurbs and report them as invalid,
for newly added files.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Cédric Le Goater <clg@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
scripts/checkpatch: reimplement mandate for SPDX-License-Identifier
Going forward we want all newly created source files to have an
SPDX-License-Identifier tag present.
Initially mandate this for C, Python, Perl, Shell source files,
as well as JSON (QAPI) and Makefiles, while encouraging users
to consider it for other file types.
The new attempt at detecting missing SPDX-License-Identifier relies
on the hooks for relying triggering logic at the end of scanning a
new file in the diff.
Tested-by: Cédric Le Goater <clg@redhat.com> Reviewed-by: Cédric Le Goater <clg@redhat.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
scripts/checkpatch: use new hook for MAINTAINERS update check
When seeing a new/deleted/renamed file we check to see if MAINTAINERS
is updated, but we don't give the user a list of files affected, as
we don't want to repeat the same warning many times over.
Using the new file list hook, we can give a single warning at the
end with a list of filenames included.
Reviewed-by: Cédric Le Goater <clg@redhat.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
scripts/checkpatch: expand pattern for matching makefiles
The current regex matches Makefile & Makefile.objs, but the latter is
no longer used, anjd we're missing coverage of Makefile.include and
Makefile.target. Expand the pattern to match any suffix.
Reviewed-by: Cédric Le Goater <clg@redhat.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
scripts/checkpatch: use new hook for file permissions check
The file permissions check is the kind of check intended to be performed
in the new start of file hook.
Reviewed-by: Cédric Le Goater <clg@redhat.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
scripts/checkpatch: introduce tracking of file start/end
Some checks want to be performed either at the start of a new file
within a patch, or at the end. This is complicated by the fact that
the information relevant to the check may be spread across multiple
lines. It is further complicated by a need to support both git and
non-git diffs, and special handling for renames where there might
not be any patch hunks.
To handle this more sanely, introduce explicit tracking of file
start/end, taking account of git metadata, and calling a hook
function at each transition.
Reviewed-by: Cédric Le Goater <clg@redhat.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
scripts/checkpatch.pl: fix various indentation mistakes
Various checks in the code were under-indented relative to other
surrounding code. Some places used 4-space indents instead of
single tab, while other places simply used too few tabs.
Reviewed-by: Cédric Le Goater <clg@redhat.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The logic in this commit was flawed in two critical ways
* It always failed to report SPDX validation on the last newly
added file. IOW, it only worked if at least 2 new files were
added in a commit
* If an existing file change, followed a new file change, in
the commit and the existing file context/changed lines
included SPDX-License-Identifier, it would incorrectly
associate this with the previous newly added file.
Simply reverting this commit will make it significantly easier to
understand the improved logic in the following commit.
Reported-by: Peter Maydell <peter.maydell@linaro.org> Reviewed-by: Cédric Le Goater <clg@redhat.com> Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
When originally creating the internal crypto cipher APIs, they were
wired up to use the built-in D3DES and AES implementations, as a way
to gracefully transition to the new APIs without introducing an
immediate hard dep on any external crypto libraries for the VNC
password auth (D3DES) or the qcow2 encryption (AES).
In the 6.1.0 release we dropped the built-in D3DES impl, and also
the XTS mode for the AES impl, leaving only AES with ECB/CBC modes.
The rational was that with the system emulators, it is expected that
3rd party crypto libraries will be available.
The qcow2 LUKS impl is preferred to the legacy raw AES impl, and by
default that requires AES in XTS mode, limiting the usefulness of
the built-in cipher provider.
The built-in AES impl has known timing attacks and is only suitable
for use cases where a security boundary is already not expected to
be provided (TCG).
Providing a built-in cipher impl thus potentially misleads users,
should they configure a QEMU without any crypto library, and try
to use it with the LUKS backend, even if that requires a non-default
configuration choice.
Complete what we started in 6.1.0 and purge the remaining AES
support.
Use of either gnutls, nettle, or libcrypt is now mandatory for any
cipher support, except for TCG impls.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
ui/vnc: fix tight palette pixel encoding for 8/16-bpp formats
When sending a tight rectangle with the palette filter, if the client
format was 8/16bpp, the colours on big endian hosts are not set as
we're sending the wrong bytes. We must first cast the 32-bit colour
to a 16/8-bit value, and then send the result.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
ui/vnc: take account of client byte order in pixman format
The set_pixel_conversion() method is responsible for determining whether
the VNC client pixel format matches the server format, and thus whether
we can use the fast path "copy" impl for sending pixels, or must use
the generic impl with bit swizzling.
The VNC server format is set at build time to VNC_SERVER_FB_FORMAT,
which corresponds to PIXMAN_x8r8g8b8.
The qemu_pixman_get_format() method is then responsible for converting
the VNC pixel format into a pixman format.
The VNC client pixel shifts are relative to the associated endianness.
The pixman formats are always relative to the host native endianness.
The qemu_pixman_get_format() method does not take into account the
VNC client endianness, and is thus returning a pixman format that is
only valid with the host endianness matches that of the VNC client.
This has been broken since pixman was introduced to the VNC server:
connecting from a LE client to a QEMU on a BE server, or the
reverse.
The bug was masked, however, because almost all VNC clients will
advertize support for the "tight" encoding and the QEMU VNC server
will prefer "tight" if advertized.
The tight_pack24 method is responsible for taking a set of pixels
which have already been converted into client endianness and then
repacking them into the TPIXEL format which the RFB spec defines
as
"TPIXEL is only 3 bytes long, where the first byte is the
red component, the second byte is the green component,
and the third byte is the blue component of the pixel
color value"
IOW, the TPIXEL format is fixed on the wire, regardless of what
the VNC client declare as its endianness.
Since the VNC pixel encoding code was failing to honour the endian
flag of the client, the tight_pack24 method was always operating
on data in native endianness. Its impl cancelled out the VNC pixel
encoding bug.
With the VNC pixel encoding code now fixed, the tight_pack24 method
needs to take into account that it is operating on data in client
endianness, not native endianness. It thus may need to invert the
pixel shifts.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
ui/vnc.c: replace big endian flag with byte order value
It will make it easier to do certain comparisons in future if we
store G_BIG_ENDIAN/G_LITTLE_ENDIAN directly, instead of a boolean
flag, as we can then compare directly to the G_BYTE_ORDER constant.
Reviewed-by: BALATON Zoltan <balaton@eik.bme.hu> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Alberto Faria [Fri, 2 May 2025 12:11:15 +0000 (13:11 +0100)]
scsi-disk: Advertise FUA support by default
Allow the guest to submit FUA requests directly, instead of forcing it
to emulate them using a regular flush.
Signed-off-by: Alberto Faria <afaria@redhat.com>
Message-ID: <20250502121115.3613717-3-afaria@redhat.com> Reviewed-by: Kevin Wolf <kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Alberto Faria [Fri, 2 May 2025 12:11:14 +0000 (13:11 +0100)]
scsi-disk: Add native FUA write support
Simply propagate the FUA flag on write requests to the driver. The block
layer will emulate it if necessary.
Signed-off-by: Alberto Faria <afaria@redhat.com>
Message-ID: <20250502121115.3613717-2-afaria@redhat.com> Reviewed-by: Kevin Wolf <kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Stefan Hajnoczi [Wed, 21 May 2025 14:25:55 +0000 (10:25 -0400)]
Merge tag 'migration-20250520-pull-request' of https://gitlab.com/peterx/qemu into staging
Migration pull
- Peter's vmstate static checker entry for recent hpet change
- Zhijian's rdma test patch to properly skip tests when locked_vm too low,
and ipv6 test
- Fabiano's few CI changes
- Prasad's last patches to enable postcopy on precopy-multifd
- Maciej's fix to disable multifd zerocopy for device states
- Peter's HMP change to "info migrate", plus a small fix on cap set
* tag 'migration-20250520-pull-request' of https://gitlab.com/peterx/qemu:
migration/hmp: Add "info migrate -a", reorg the dump
migration: Allow caps to be set when preempt or multifd cap enabled
migration/multifd: Don't send device state packets with zerocopy flag
tests/qtest/migration: add postcopy tests with multifd
migration: enable multifd and postcopy together
migration: write zero pages when postcopy enabled
ci: Reduce the size of artifacts for build-previous-qemu
ci: Fix build-previous-qemu when the version tag is absent
ci: Re-enable python subtests in qtest migration suite
qtest/migration/rdma: Add test for rdma migration with ipv6
qtest/migration/rdma: Enforce RLIMIT_MEMLOCK >= 128MB requirement
scripts/vmstate-static-checker.py: Add new hpet entry for num_timers
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Peter Xu [Tue, 13 May 2025 21:08:17 +0000 (17:08 -0400)]
migration/hmp: Add "info migrate -a", reorg the dump
A new parameter "-a" is added to "info migrate" to dump all info, while
when not specified it only dumps the important ones. When at it, reorg
everything to make it easier to read for human.
The general rule is:
- Put important things at the top
- Reuse a single line when things are very relevant, hence reducing lines
needed to show the results
- Remove almost useless ones (e.g. "normal_bytes", while we also have
both "page size" and "normal" pages)
- Regroup things, so that related fields will show together
- etc.
Before this change, it looks like (one example of a completed case):
globals:
store-global-state: on
only-migratable: off
send-configuration: on
send-section-footer: on
send-switchover-start: on
clear-bitmap-shift: 18
Migration status: completed
total time: 122952 ms
downtime: 76 ms
setup: 15 ms
transferred ram: 130825923 kbytes
throughput: 8717.68 mbps
remaining ram: 0 kbytes
total ram: 16777992 kbytes
duplicate: 997263 pages
normal: 32622225 pages
normal bytes: 130488900 kbytes
dirty sync count: 10
page size: 4 kbytes
multifd bytes: 117134260 kbytes
pages-per-second: 169431
postcopy request count: 5835
precopy ram: 15 kbytes
postcopy ram: 13691151 kbytes
After this change, sample output (default, no "-a" specified):
Status: active
Time (ms): total=3040, setup=4, exp_down=300
RAM info:
Throughput (mbps): 10.51
Sizes (KB): psize=4, total=4211528,
transferred=3979, remain=4206452,
precopy=3978, multifd=0, postcopy=0
Pages: normal=992, zero=277, rate_per_sec=320
Others: dirty_syncs=1
Globals:
store-global-state: on
only-migratable: off
send-configuration: on
send-section-footer: on
send-switchover-start: on
clear-bitmap-shift: 18
XBZRLE: size=67108864, transferred=0, pages=0, miss=188451
miss_rate=0.00, encode_rate=0.00, overflow=0
CPU Throttle (%): 0
Dirty-limit Throttle (us): 0
Dirty-limit Ring Full (us): 0
Postcopy Blocktime (ms): 0
Postcopy vCPU Blocktime: ...
Reviewed-by: Dr. David Alan Gilbert <dave@treblig.org> Tested-by: Mario Casquero <mcasquer@redhat.com>
[peterx: print "," too in 1st line of RAM info] Signed-off-by: Peter Xu <peterx@redhat.com>
Peter Xu [Tue, 13 May 2025 21:33:16 +0000 (17:33 -0400)]
migration: Allow caps to be set when preempt or multifd cap enabled
With commit 82137e6c8c ("migration: enforce multifd and postcopy preempt to
be set before incoming"), and if postcopy preempt / multifd is enabled, one
cannot setup any capability because these checks would always fail.
(qemu) migrate_set_capability xbzrle off
Error: Postcopy preempt must be set before incoming starts
To fix it, check existing cap and only raise an error if the specific cap
changed.
Fixes: 82137e6c8c ("migration: enforce multifd and postcopy preempt to be set before incoming") Reviewed-by: Dr. David Alan Gilbert <dave@treblig.org> Reviewed-by: Juraj Marcin <jmarcin@redhat.com> Signed-off-by: Peter Xu <peterx@redhat.com>
migration/multifd: Don't send device state packets with zerocopy flag
If zerocopy is enabled for multifd then QIO_CHANNEL_WRITE_FLAG_ZERO_COPY
flag is forced into all multifd channel write calls via p->write_flags
that was setup in multifd_nocomp_send_setup().
However, device state packets aren't compatible with zerocopy - the data
buffer isn't getting kept pinned until multifd channel flush.
Make sure to mask that QIO_CHANNEL_WRITE_FLAG_ZERO_COPY flag in a multifd
send thread if the data being sent is device state.
Prasad Pandit [Mon, 12 May 2025 12:51:23 +0000 (18:21 +0530)]
migration: enable multifd and postcopy together
Enable Multifd and Postcopy migration together.
The migration_ioc_process_incoming() routine checks
magic value sent on each channel and helps to properly
setup multifd and postcopy channels.
The Precopy and Multifd threads work during the initial
guest RAM transfer. When migration moves to the Postcopy
phase, the multifd threads cease to send data on multifd
channels and Postcopy threads on the destination
request/pull data from the source side.
projects / thirdparty / qemu.git / log
