Optimizations to the AT keyboard driver to reduce unnecessary delays during the FreeBSD boot process #1826

bllgg · 2025-08-28T01:58:45Z

Summary of changes

Introduced a new loader tunable hw.atkbd.fast_delay to /boot/loader.conf.
- When enabled, the driver uses atkbd_short_delay, which applies significantly shorter delays in the following routines:
  - empty_both_buffers()
  - reset_aux_dev()
  - test_controller()
  - reset_kbd()
- New short delays are up to 1000× smaller (for buffer resets) and about 40× smaller (for controller and keyboard resets).
- Default behaviour remains with shorter delays, and hw.atkbd.short_delay should be set to 0, if someone needs longer delays for the old, slower hardware
Added a separate conf file into the /sys/amd64/conf named BOOT_PROF to enable the TSlogs

Notes:

This patch is part of the ongoing GSoC 2025 work on FreeBSD boot process optimisation.

github-actions · 2025-08-28T01:58:52Z

Thank you for taking the time to contribute to FreeBSD!
There are a few issues that need to be fixed:

Missing Signed-off-by lines^{a14d561, 2a346c8, 8647a0b, 1167a7d, 656f7f4, 9e792f7, 14b61b2, ef083de, 3008f30, f48b1a3, 08b8843, def2868, a9577ca, 372c80f, 4c8104d, 7119cd6, 3d612c3, f2936e4, fa0d7e7, 9fe27ed, ecc8869, 458007b, 0862497, 20a8b23, d962160, 30fb0f4, 5daf8ed, 317eb5d, 54b1dbf, 0873dc6, 1525338}
Real email address is needed^59f8f5d
Expected Signed-off-by: , got Signed-off-by: ^{a242431, e29bfa5}

Please review CONTRIBUTING.md, then update and push your branch again.

tsoome · 2025-08-28T06:50:51Z

sys/dev/atkbdc/atkbdc.c


-#define MAXKBDC		1		/* XXX */
+#define MAXKBDC				1	/* XXX */
+#define MILISECOND_MULTIPLIER		1000	/* micro second values will be multiplied to convert to mili seconds */


typo, 'milli'. also, be mindful about line lengths, in general, we want to fit into 80 chars

tsoome · 2025-08-28T06:53:46Z

sys/dev/atkbdc/atkbdc.c

 			(bus_space_write_1((k)->iot, (k)->ioh1, 0, (d)))

+#define ATKBD_DELAY(x)	(atkbd_short_delay ? (x) : (x) * MILISECOND_MULTIPLIER)
+#define RESET_DELAY(x)	(atkbd_short_delay ? (x) * MILISECOND_MULTIPLIER : (x) * MILISECOND_MULTIPLIER / RESET_DELAY_DIVISION_FACTOR)


with long macro body we use \ to continue on next line and next line will be indented

tsoome · 2025-08-28T06:54:38Z

sys/dev/atkbdc/atkbdc.c


-    int delay_us = 5000;
-    int max_wait_us = KBD_RESETDELAY*1000*KBD_MAXWAIT;
+    int delay_us = RESET_DELAY(KBD_RESETDELAY); /* if the atkbd_short_delay is activated, delay will be shorten */


line length there too

concussious

Wow this is cool! Please put these settings in atkbdc(4) when you get the chance.

This reverts commit 65059dd. lindebugfs does he vast majority of its pseudofs initialization nearly everywhere but pseudofs, so let's defer this to post-brsnching.

We need to build OpenSSL when bootstrapping certctl. On MacOS ktls_enable() isn't defined anywhere, so without this the build fails. We don't need KTLS in the bootstrap library, so just disable it. Reviewed by: khorben, ngie Differential Revision: https://reviews.freebsd.org/D52341

When LIB_PACKAGE was added, MIT Kerberos was updated to use it but Heimdal was not, so it still used PACKAGE=kerberos-lib. Since we deleted kerberos-lib-all.ucl, this caused update-packages to fail when WITHOUT_MITKRB5 is set. Change the Heimdal libraries to use LIB_PACKAGE by setting this in kerberos5/lib/Makefile.inc, and remove PACKAGE=kerberos-lib from the individual Makefiles for each library. This means Heimdal gets the same set of packages as MIT Kerberos, except for kerberos-kdc which we don't create for Heimdal. Fixes: 929f596 ("packages: Improve handling of -lib packages") Reported by: jlduran Reviewed by: jlduran, cy Differential Revision: https://reviews.freebsd.org/D52371

Mark the unused space after pv_memattr as reserved. This doesn't change the size of the struct as this padding was already inserted by the compiler. Sponsored by: Arm Ltd

To allow for possible future use add reserved fields to struct mdproc. Sponsored by: Arm Ltd

Originally in e234027 the function argument was a "const void *data" which we changed to "const int" in FreeBSD. The Linux version simply cast the const void * to (enum rtw89_rf_path) both losing the const and changing the type. When later the function argument became a const union rtw89_fw_element_arg in 5b760ea it seems we tried to undo the older FreeBSD specific part and initially mismerged it leading to the const void *; but when fixing the type we did not undo the __DECONST but did a mechanical copy. Reviewed by: bz Co-authored-by: bz (commit message) Obtained from: CheriBSD Sponsored by: AFRL, DARPA Fixes: 6d67aab ("rtw89: update Realtek's rtw89 driver.") Differential Revision: https://reviews.freebsd.org/D52359

This will be used later to extend cpu feature detection Reviewed by: imp (earlier version) Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52356

This will also be used as a tunable to control features in a later change. Reviewed by: imp Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52357

Allow the user to enable/disable supported cpu features/errata from a known tunable. This will allow the user to disable features, e.g. to work around broken firmware. It can also be used to enable workarounds for rare errata. Reviewed by: imp Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52358

Add a function that can check if a given midr is within a range of revisions. This will be used to check if a CPU is affected by a known erratum. Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52187

- Use the passed in midr rather than reading from pcpu - Use the new midr_check_var_part_range function to check if the midr is within the range of affected CPUs Reviewed by: alc Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52188

Use the new midr_check_var_part_range function to check if running on an affected Neoverse-N1 CPU. Reviewed by: emaste Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52189

When adding an interface with an IP address to a bridge, or assigning an IP address to an interface which is in a bridge, and member_ifaddrs=1, print a warning so users are informed this is deprecated. Also add "(deprecated)" to the sysctl description. MFC after: 9 hours Reviewed by: markj Differential Revision: https://reviews.freebsd.org/D52335

We aren't ready to release with this set to 0: there's no solution for running DHCP on a bridge from a devd hook, there are missing edge cases like network boot with the boot interface in a bridge, and third-party tools (e.g., vm-bhyve) still direct users to use the incorrect configuration and/or don't support the correct configuration. Based on popular vote, we'll set this back to 0 in both main and for stable/15. MFC after: 9 hours Reviewed by: markj Differential Revision: https://reviews.freebsd.org/D52336

It's unused. Reviewed by: emaste Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52191

Add pmap_s1_invalidate_all_kernel to remove the need to check for the kernel_pmap when it's passed in. While here replace pmap calls to cpu_tlb_flushID with pmap_s1_invalidate_all_kernel as they are identical. Reviewed by: kib Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52183

Replace cpu_tlb_flushID with pmap_s1_invalidate_all_kernel in initarm. They are identical in the SMP case. Reviewed by: alc Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52184

The Arm Cortex-A55, Cortex-A76, and Cortex-A510 CPUs have errata that require multiple TLBI, DSB instructions to workaround. Add support to pmap to implement these. As it appears that all affected TLBI calls are via pmap.c this should be sufficient. As all variants of this erratum are Category-B (rare) require the user to enable it at boot time. Reviewed by: alc Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52190

Otherwise files are handled in the order returned by find(1). This in turn results in inconsistent ordering of files in the METALOG mtree file produced by "make installworld -DNO_ROOT", which results in non-reproducible filesystem images. Reviewed by: gshapiro, emaste MFC after: 2 weeks Sponsored by: The FreeBSD Foundation Sponsored by: Klara, Inc. Differential Revision: https://reviews.freebsd.org/D52349

Add kqueue tracking to jails, inspired by how it's done with processes. EVFILT_JAIL takes a jail ID, and tracks with NOTE_JAIL_SET, NOTE_JAIL_ATTACH, NOTE_JAIL_REMOVE, and NOTE_JAIL_CHILD. It also uses the NOTE_TRACK mechanism that EVFILT_PROC uses, using the same result flags (NOTE_CHILD and NOTE_TRACKERR). Relnotes: yes Differential Revision: https://reviews.freebsd.org/D51940

While here, advertise the IFCAP_HWSTATS capability to avoid the net stack from double counting it. Co-authored-by: zlei Reviewed by: zlei MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D44816

…r of octets sent When ALTQ is enabled, this driver does "hardware" accounting and soft accounting at the same time. Prefer the "hardware" one to make the logic simpler. Reviewed by: zlei MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D44817

Similar to process descriptors, jail desriptors are allow jail administration using the file descriptor interface instead of JIDs. They come from and can be used by jail_set(2) and jail_get(2), and there are two new system calls, jail_attach_jd(2) and jail_remove_jd(2). Reviewed by: bz, brooks Relnotes: yes Differential Revision: https://reviews.freebsd.org/D43696

Add extra defines required for bc-7.1.0 to the Makefiles.

This reverts commit d549de7.

When freeing the last reference of the net80211 node the net80211 node_free() code may directly call into the crypto code to delete the keys. While we still holding the wiphy lock this would lead to a recursion on the non-recursive wiphy lock. Defer freeing the reference until we are back under the net80211 com lock. Reported by: Mark Phillips (mark freebsdfoundation.org) on 15.0-ALPHA1 MFC after: 3 days

StartImage() may return additional data from failure. This data has text message followed by optional binary blob. Print out the text message (if present) and free the data. See 7.4.2 EFI_BOOT_SERVICES.StartImage() page 199 UEFI_Spec_Final_2.11.pdf. Reviewed by: imp

MFC after: 3 days

Separate ossl's existing AES-NI GCM implementation into a common ossl_aes_gcm.c and add conditionals to switch between OpenSSL's AES-NI and POWER8 GCM routines depending on the architecture. Since the existing AVX-512 implementation is less agnostic, move it into a separate ossl_aes_gcm_avx512.c. Reviewed by: markj MFC after: 1 month Differential Revision: https://reviews.freebsd.org/D44274

PR: 286720 Reviewed by: zlei MFC after: 1 week Differential Revision: https://reviews.freebsd.org/D50491

Reported by: markj, bz Approved by: markj (mentor) Fixes: 2b74ff5 ("ichwd: introduce i6300esbwd watch dog driver") Differential Revision: https://reviews.freebsd.org/D52476

The new Framework 16 with ALC285 exhibits the same issue as the previous model. Therefore, we apply the same fix to the new model. Reviewed by: emaste, Daniel Schaefer <[email protected]> Approved by: lwhsu (mentor) MFC after: 3 days Sponsored by: The FreeBSD Foundation Sponsored by: Framework Computer Inc Differential Revision: https://reviews.freebsd.org/D52423

Reviewed by: bapt@ Approved by: bapt@ Differential Revision: https://reviews.freebsd.org/D52436

* Set mode of sudoers to 0440. Reviewed by: bapt@, jlduran@ Approved by: bapt@, jlduran@ Differential Revision: https://reviews.freebsd.org/D52438

To better comply with the cloud-init specification, we need to support the configuration of network-related parameters from the network-config file, which is common in most deployments. Reviewed by: bapt@ Approved by: bapt@ Differential Revision: https://reviews.freebsd.org/D52419

* Set mode of etc directory to 0755. * Use user.localbase sysctl instead of /usr/local. * Add test case for doas. * Set ${LOCALBASE} instead of /usr/local in nuageinit(7) man page. Reviewed by: bapt@ Approved by: bapt@ Differential Revision: https://reviews.freebsd.org/D52437

Approved by: bapt@

Reviewed by: bcr MFC after: 3 days

MFC after: 3 days Sponsored by: Klara, Inc. Reviewed by: markj Differential Revision: https://reviews.freebsd.org/D52489

I received a report that certain Linux application would crash on a message of a departure of an interface with FreeBSD name. Looks like dropping NL_RTM_DELLINK is a lesser evil than relay them with FreeBSD names. This reverts commit 554907b.

Use bitfields to pack tx_credits and plen into a single 32-bit word. Reviewed by: np Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D47759

This is intended to support control work requests that manipulate connection state but do not transmit data. Raw WRs are transmitted immediately if possible, otherwise they are queued to an mbufq in the toe pcb until sufficient credits are available. Raw WRs take precendence over transmitting socket data. Reviewed by: np Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D47761

Reviewed by: np Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D47762

In particular, the work request to update the TCB state when switching from plain TCP to TLS is now queued on the connection's offload transmit queue rather than over the per-port control queue. This also handles an unlikely edge case that a connection does not have sufficient credits to transmit other work requests synchronously such as programming the key in on-card memory or altering TCB fields. Reviewed by: np Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D47763

PR: 289447 MFC after: 1 week

and whitespace cleanup. MFC after: 3 days

MFC after: 3 days

Instead of using the EVFILT_PROC model of attempting to automatically register new events when a child jail is created, just give a single event when a child jail is created. As was already done with jail attach events, make a best-effort report of the added jail's id in kn_data. If the are multiple NOTE_JAIL_CHILD and/or NOTE_JAIL_ATTACH events, set the NOTE_JAIL_MULTI flag, and don't report anything in data, indicating that the caller will need to query the system state on their own. MFC after: 3 days

…ion with boot configuration

…S2 mouse drivers to identify the bottlenecks, mount process of the root file system, usb initialization and device initialization with TSLOG to identify the bottlenecks

…dev-atkbd-psm # Conflicts: # sys/dev/atkbdc/atkbdc.c

bllgg marked this pull request as ready for review August 28, 2025 02:40

cperciva self-assigned this Aug 28, 2025

bllgg requested a review from bsdimp as a code owner August 28, 2025 05:02

tsoome reviewed Aug 28, 2025

View reviewed changes

concussious reviewed Aug 28, 2025

View reviewed changes

kevans91 and others added 24 commits September 4, 2025 07:51

Revert "pseudofs: defer initialization until first mount"

d346229

This reverts commit 65059dd. lindebugfs does he vast majority of its pseudofs initialization nearly everywhere but pseudofs, so let's defer this to post-brsnching.

arm64: Make the padding in struct ms_page explicit

b21629d

Mark the unused space after pv_memattr as reserved. This doesn't change the size of the struct as this padding was already inserted by the compiler. Sponsored by: Arm Ltd

arm64: Add padding to struct mdproc

59ddbef

To allow for possible future use add reserved fields to struct mdproc. Sponsored by: Arm Ltd

arm64: Add a macro to create cpu_feat

b3605a7

This will be used later to extend cpu feature detection Reviewed by: imp (earlier version) Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52356

arm64: Add a sysctl to see if features are enabled

9204a31

This will also be used as a tunable to control features in a later change. Reviewed by: imp Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52357

arm64: Add a function to check a range of CPU revs

c76b024

Add a function that can check if a given midr is within a range of revisions. This will be used to check if a CPU is affected by a known erratum. Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52187

arm64: Update the Neoverse-N1 1542419 errata check

efb1c4b

Use the new midr_check_var_part_range function to check if running on an affected Neoverse-N1 CPU. Reviewed by: emaste Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52189

arm64: Remove CPU_MATCH_RAW

f6d4c9c

It's unused. Reviewed by: emaste Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52191

arm64: Replace cpu_tlb_flushID in initarm

fa23bfb

Replace cpu_tlb_flushID with pmap_s1_invalidate_all_kernel in initarm. They are identical in the SMP case. Reviewed by: alc Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D52184

Like for -P et al, state that -deElL are syscons only options.

d5dd6ad

vtnet: Do "hardware" accounting for the total number of received octets

a14d561

While here, advertise the IFCAP_HWSTATS capability to avoid the net stack from double counting it. Co-authored-by: zlei Reviewed by: zlei MFC after: 2 weeks Differential Revision: https://reviews.freebsd.org/D44816

stesser and others added 30 commits September 11, 2025 16:42

usr.bin/gh-bc: update for version 7.1.0

2d06844

Add extra defines required for bc-7.1.0 to the Makefiles.

Revert "libc: Remove readdir_r(3)"

d20c825

This reverts commit d549de7.

sys/efi.h: Some style(9) tweaks

e584c46

MFC after: 3 days

git-blame-ignore-revs: sys/efi.h style(9) tweaks

dd14103

MFC after: 3 days

bnxt: Fix the request length in bnxt_hwrm_func_backing_store_cfg()

5ca390f

PR: 286720 Reviewed by: zlei MFC after: 1 week Differential Revision: https://reviews.freebsd.org/D50491

ichwd: address unused function warning by marking as __unused

3449a3a

Reported by: markj, bz Approved by: markj (mentor) Fixes: 2b74ff5 ("ichwd: introduce i6300esbwd watch dog driver") Differential Revision: https://reviews.freebsd.org/D52476

nuageinit: Remove duplicate description of users.{user}.groups

9f3330f

Reviewed by: bapt@ Approved by: bapt@ Differential Revision: https://reviews.freebsd.org/D52436

nuageinit: chmod sudoers directory instead of chmod (again) sudoers file

a5cc9b7

* Set mode of sudoers to 0440. Reviewed by: bapt@, jlduran@ Approved by: bapt@, jlduran@ Differential Revision: https://reviews.freebsd.org/D52438

nuageinit: Add me to copyright list

1855506

Approved by: bapt@

backlight(9): add cross-ref to backlight(8)

a74c8b8

Reviewed by: bcr MFC after: 3 days

uptime: Mention libxo support in manual page

55bd607

MFC after: 3 days Sponsored by: Klara, Inc. Reviewed by: markj Differential Revision: https://reviews.freebsd.org/D52489

cxgbe tom: Halve the size of offload transmit software descriptors

5a38857

Use bitfields to pack tx_credits and plen into a single 32-bit word. Reviewed by: np Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D47759

cxgbe tom: Export alloc_raw_mbuf for use in other TOE drivers

a7aab22

Reviewed by: np Sponsored by: Chelsio Communications Differential Revision: https://reviews.freebsd.org/D47762

lib/msun: cpow{,f,l}(CMLX(0.0, 0.0)) should return 1.0

8b41ba8

PR: 289447 MFC after: 1 week

sys/bus.h: Some style(9) fixes

2223875

and whitespace cleanup. MFC after: 3 days

git-blame-ignore-revs: sys/bus.h style(9) fixes

64dca4e

MFC after: 3 days

BOOT_PROF: Reduced the dealy in AT keyboard and PS2 mouse initializat…

317eb5d

…ion with boot configuration

BOOT_PROF: Instrument the initialization process of AT keyboard and P…

54b1dbf

…S2 mouse drivers to identify the bottlenecks, mount process of the root file system, usb initialization and device initialization with TSLOG to identify the bottlenecks

BOOT_PROF: Updated the man doc for atkbdc

0873dc6

Merge remote-tracking branch 'origin/bllgg-dev-atkbd-psm' into bllgg-…

1525338

…dev-atkbd-psm # Conflicts: # sys/dev/atkbdc/atkbdc.c

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Optimizations to the AT keyboard driver to reduce unnecessary delays during the FreeBSD boot process #1826

Optimizations to the AT keyboard driver to reduce unnecessary delays during the FreeBSD boot process #1826

bllgg commented Aug 28, 2025 •

edited

Loading

Uh oh!

github-actions bot commented Aug 28, 2025 •

edited

Loading

Uh oh!

tsoome Aug 28, 2025

Uh oh!

tsoome Aug 28, 2025

Uh oh!

tsoome Aug 28, 2025

Uh oh!

concussious left a comment

Uh oh!

Uh oh!

Optimizations to the AT keyboard driver to reduce unnecessary delays during the FreeBSD boot process #1826

Are you sure you want to change the base?

Optimizations to the AT keyboard driver to reduce unnecessary delays during the FreeBSD boot process #1826

Conversation

bllgg commented Aug 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary of changes

Notes:

Uh oh!

github-actions bot commented Aug 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tsoome Aug 28, 2025

Choose a reason for hiding this comment

Uh oh!

tsoome Aug 28, 2025

Choose a reason for hiding this comment

Uh oh!

tsoome Aug 28, 2025

Choose a reason for hiding this comment

Uh oh!

concussious left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

bllgg commented Aug 28, 2025 •

edited

Loading

github-actions bot commented Aug 28, 2025 •

edited

Loading