July 2026 - SeaBIOS - mail.coreboot.org

[PATCH v3 0/3] atiext: Convert tables to C and fixes for radeon
by Chad Jablonski July 31, 2026

July 31, 2026

Changes from v2: - Remove mention of unrelated in-flight QEMU patch - Split size and ddc type fixes into separate patches - Add QEMU cmd for testing Changes from v1: - Convert tables to C (Kevin O'Connor) - Use actual header size for size field via sizeof Chad Jablonski (2): atiext: Add size to bios table for radeon Linux driver (KMS) atiext: Add DDC type to tables for radeon Linux driver (KMS) Kevin O'Connor (1): atiext: Implement ATI tables in atiext.c vgasrc/ati-tables.S | 43 ---------------------------- vgasrc/atiext.c | 68 +++++++++++++++++++++++++++++++++++++++++++++ vgasrc/vgaentry.S | 6 +++- 3 files changed, 73 insertions(+), 44 deletions(-) delete mode 100644 vgasrc/ati-tables.S -- 2.54.0

3 7

[PATCH] floppy: Fix failed media sense return code
by Kevin O'Connor July 31, 2026

July 31, 2026

An error in floppy_media_sense() returned error code 0xc0. However, this appears to be a typo when porting the original bochs bios code, as it uses 0x0c. It appears this typo was introduced in the first SeaBIOS commit (f076a3ee). Change the code to return 0x0c (DISK_RET_EBADTRACK) on an error. Signed-off-by: Kevin O'Connor <kevin(a)koconnor.net> --- src/hw/floppy.c | 2 +- src/std/disk.h | 1 - 2 files changed, 1 insertion(+), 2 deletions(-) diff --git a/src/hw/floppy.c b/src/hw/floppy.c index ebed54f..a51711a 100644 --- a/src/hw/floppy.c +++ b/src/hw/floppy.c @@ -515,7 +515,7 @@ floppy_media_sense(struct drive_s *drive_gf) // Attempt media sense. for (stype=1; ; stype++) { if (stype >= ARRAY_SIZE(FloppyInfo)) - return DISK_RET_EMEDIA; + return DISK_RET_EBADTRACK; if (stype==ftype || (GET_GLOBAL(FloppyInfo[stype].floppy_size) != GET_GLOBAL(FloppyInfo[ftype].floppy_size)) diff --git a/src/std/disk.h b/src/std/disk.h index 6397108..68ccecc 100644 --- a/src/std/disk.h +++ b/src/std/disk.h @@ -17,7 +17,6 @@ #define DISK_RET_ELOCKED 0xb1 #define DISK_RET_ENOTREMOVABLE 0xb2 #define DISK_RET_ETOOMANYLOCKS 0xb4 -#define DISK_RET_EMEDIA 0xC0 #define DISK_RET_ENOTREADY 0xAA -- 2.53.0

1 1

[PATCH] floppy: detect an empty drive via the disk change line
by Christian Quante July 29, 2026

July 29, 2026

floppy_prep() recalibrates and then calls floppy_media_sense(), which issues READ ID once per data rate. A drive with no medium spins no diskette and therefore produces no index pulses, so a real controller never completes READ ID and every attempt waits out FLOPPY_IRQ_TIMEOUT. Booting with an empty drive, or an INT 13h read of an empty A:, costs about ten seconds. The disk change line answers the same question immediately. It latches when the medium is removed and is cleared only by a step pulse with a diskette in the drive. A recalibrate leaves the head on cylinder 0 and need not step at all, so seek to cylinder 1 and read DIR again. Linux's floppy driver has done exactly this since 2.0 and never issues READ ID. Only the 1.2MB 5.25" drive and the 3.5" drives have the change line (CMOS types 2..5); on the older 5.25" types the pin is not driven and reads back as a permanent "changed". Restrict the probe to those types, and report any uncertainty as "medium present", so that the code can only ever fall back to the current behaviour. Measured with qemu-system-x86_64, "-machine pc -device floppy, drive-type=144" and no medium, INT 13h AH=02h on drive A: before 0.11 s, AH = 0x20 (controller failure) after 0.11 s, AH = 0xC0 (no media) QEMU completes READ ID successfully on an empty drive, which hides the timeout; a fix for that is on qemu-devel [1]. With it applied, so that READ ID never completes as on real hardware, the same read takes 10.12 s before this change and 0.11 s after it. Booting from a diskette is unaffected, including the cold start where the change line is set for a freshly inserted medium. [1] https://lore.kernel.org/qemu-devel/20260710113134.43012-1-christian@quante.… Signed-off-by: Christian Quante <christian(a)quante.one> --- Notes for reviewers, not for the commit message: How the numbers were measured: a 512-byte boot sector issues INT 13h AH=02h on drive A: and then exits the VM through isa-debug-exit, so the time is what the guest waits, not what the harness waits. The drive is "-device floppy,drive-type=144" with no medium. Both BIOS images were built from master (c2a33ad9), which has not touched src/hw/floppy.c since rel-1.17.0. What motivated this: OS/2 2.11 cannot detect that a diskette has been removed, because QEMU answers READ ID on an empty drive. With the QEMU side fixed so that READ ID never completes, OS/2's driver issues one READ ID, waits 2.55 s, resets the controller and correctly reports "no diskette" -- but SeaBIOS then spends 10 s in floppy_media_sense() on every boot. This patch removes that cost. The two changes are useful independently: on today's unmodified QEMU this one already turns AH = 0x20 (controller failure) into AH = 0xC0 (no media). Tested: boot from a diskette; INT 13h read of a present diskette; cold start with a freshly inserted medium, where the change line is set and a naive probe would report the drive as empty; and an empty drive. Also exercised through PC-DOS 7, IBM DOS 5.02, Windows for Workgroups 3.11 and OS/2 2.11. Not tested on real hardware -- I have none with a floppy drive. The gating on CMOS types 2..5 and the "when in doubt, say present" fallback are meant to make that safe, but a second opinion on drives without a change line would be welcome. One open question: DISK_RET_EMEDIA (0xC0) is what floppy_media_sense() already returns when it exhausts the data rates, so I kept it. A real AT BIOS times the command out and returns 0x80 (timeout), which DOS renders as "Not ready reading drive A" rather than "General failure". If 0x80 is the better answer here I am happy to change it. src/hw/floppy.c | 51 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 51 insertions(+) diff --git a/src/hw/floppy.c b/src/hw/floppy.c index 9e6647d4..5bdd6541 100644 --- a/src/hw/floppy.c +++ b/src/hw/floppy.c @@ -27,6 +27,9 @@ #define PORT_FD_DATA 0x03f5 #define PORT_FD_DIR 0x03f7 +// Digital Input Register (read) +#define FLOPPY_DIR_DSKCHG 0x80 + #define FLOPPY_SIZE_CODE 0x02 // 512 byte sectors #define FLOPPY_DATALEN 0xff // Not used - because size code is 0x02 #define FLOPPY_MOTOR_TICKS 37 // ~2 seconds @@ -442,6 +445,48 @@ floppy_drive_readid(u8 floppyid, u8 data_rate, u8 head) return 0; } +// Only 1.2MB 5.25" and the 3.5" drives have a disk change line. On the older +// 5.25" types the pin is not driven and reads back as a permanent "changed". +static int +floppy_has_changeline(u8 ftype) +{ + return ftype >= 2 && ftype <= 5; +} + +// Determine whether a diskette is present, without waiting for a command to +// time out. The disk change line latches when the medium is removed and is +// cleared only by a step pulse with a diskette in the drive. A recalibrate +// leaves the head on cylinder 0 and need not step at all, so step to cylinder +// 1 and look again. Returns 0 only when the drive is known to be empty; any +// uncertainty is reported as "present", leaving the caller to find out. +static int +floppy_media_present(struct drive_s *drive_gf, u8 floppyid) +{ + if (!floppy_has_changeline(GET_GLOBALFLAT(drive_gf->floppy_type))) + return 1; + + // The caller has just recalibrated this drive, so it is selected in the + // DOR and the DIR reflects its change line. + if (!(inb(PORT_FD_DIR) & FLOPPY_DIR_DSKCHG)) + return 1; + + u8 param[2]; + param[0] = floppyid; + param[1] = 1; + int ret = floppy_drive_pio(floppyid, FC_SEEK, param); + if (ret) + return 1; + int present = !(inb(PORT_FD_DIR) & FLOPPY_DIR_DSKCHG); + + // Put the head back on cylinder 0 and fix up the BDA. + ret = floppy_drive_recal(floppyid); + if (ret) + return 1; + + dprintf(2, "Floppy_media_present %d: %d\n", floppyid, present); + return present; +} + static int floppy_media_sense(struct drive_s *drive_gf) { @@ -498,6 +543,12 @@ floppy_prep(struct drive_s *drive_gf, u8 cylinder) if (ret) return ret; + // An empty drive gives no index pulses, so READ ID would never + // complete and media sensing would wait out its timeout once per + // data rate. Ask the disk change line instead. + if (!floppy_media_present(drive_gf, floppyid)) + return DISK_RET_EMEDIA; + // Sense media. ret = floppy_media_sense(drive_gf); if (ret) -- 2.53.0

3 18

[PATCH v2 0/2] atiext: Convert tables to C and fix for radeon
by Chad Jablonski July 28, 2026

July 28, 2026

Full testing requires the QEMU ati-vga CP engine patch series under review on qemu-devel: https://lore.kernel.org/qemu-devel/20260719161536.774061-1-chad@jablonski.x… Changes from v1: - Convert tables to C (Kevin O'Connor) - Use actual header size for size field via sizeof Chad Jablonski (1): atiext: Fix bios tables for radeon KMS driver Kevin O'Connor (1): atiext: Implement ATI tables in atiext.c vgasrc/ati-tables.S | 43 ------------------------------ vgasrc/atiext.c | 65 +++++++++++++++++++++++++++++++++++++++++++++ vgasrc/vgaentry.S | 6 ++++- 3 files changed, 70 insertions(+), 44 deletions(-) delete mode 100644 vgasrc/ati-tables.S -- 2.54.0

3 6

[PATCH] ati-vga: Fix bios tables for radeon KMS driver
by Chad Jablonski July 22, 2026

July 22, 2026

This fixes two issues when running ati-vga with model=rv100 under QEMU using a KMS driver. First, the radeon KMS driver does an additional check when looking up COMBIOS table entries. It checks a header size field to validate that the entry falls in the header region. Prior to this patch that size field is zero which causes the driver to fallback to another method (CLOCK_CNTL registers) also resulting in bad values and a division by zero. 0x60 matches what I've observed in the bios of a Radeon QY (RV100) (PCI 1002:5159). But I don't think the value is actually important other than needing to be large enough to encompass the table entries used here. Second, the connector info section needs a non-zero value in the DDC type field. Zero causes the radeon driver to fail to initialize the DDC bus and then no connectors are found. DDC_DVI (0x2) is used here so that the GPIO_DVI_DDC register is used for the radeon. From what I have seen r128 drivers ignore this field entirely. Tested on Debian Squeeze (Linux 2.6.32) and Debian Trixie (Linux 6.12.94). Signed-off-by: Chad Jablonski <chad(a)jablonski.xyz> --- To fully test this you'll need the ati-vga CP engine series under review on qemu-devel: https://lore.kernel.org/qemu-devel/20260719161536.774061-1-chad@jablonski.x… That series is focused on UMS only but this unlocks KMS! --- vgasrc/ati-tables.S | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/vgasrc/ati-tables.S b/vgasrc/ati-tables.S index cdbde2fa..d7a8e22d 100644 --- a/vgasrc/ati-tables.S +++ b/vgasrc/ati-tables.S @@ -12,6 +12,8 @@ // main info .org 0x50 _ati_main: + .org 0x50 + 0x06 + .byte 0x60 // COMBIOS header size .org 0x50 + 0x30 .word _ati_pll .org 0x50 + 0x50 @@ -37,7 +39,7 @@ _ati_pll: _ati_connector: .byte 0x10 // one chip .byte 0x01 // one connector - .word 0x3000 // type DVI-I + .word 0x3200 // type DVI-I and DVI DDC .word 0 // end of list .org 0x200 -- 2.54.0

3 3

[PATCH v4 1/1] e820: Add E820_SOFT_RESERVED type support
by fanhuang July 8, 2026

July 8, 2026

Recognize E820 type 0xEFFFFFFF (Soft Reserved) emitted by QEMU and propagate it to the guest E820 map. Without this, SeaBIOS would silently drop the entry and the guest kernel could not honor the reservation. Signed-off-by: FangSheng Huang <FangSheng.Huang(a)amd.com> --- Changes since v3: - Drop the pciinit.c change (64-bit PCI MMIO window adjustment). The patch is now strictly E820 type recognition and pass-through. How tested: qemu-system-x86_64 -m 8G,maxmem=24G,slots=4 \ -bios out/bios.bin \ -object memory-backend-ram,size=8G,id=m0 \ -object memory-backend-ram,size=8G,id=spm0 \ -numa node,cpus=0-7,nodeid=0,memdev=m0 \ -numa node,nodeid=1 \ -device sp-mem,id=spm-dev0,memdev=spm0,node=1 SeaBIOS debug log: qemu/e820: addr 0x0000000280000000 len 0x0000000200000000 [soft-reserved] Guest /proc/iomem: 280000000-47fffffff : Soft Reserved Companion QEMU device (merged into QEMU master; will ship in 11.1): the sp-mem series (10 commits), e83b01083631..c31b39063fe9, starting with "hw/mem: add sp-mem device for Specific Purpose Memory". src/e820map.h | 11 ++++++----- src/fw/paravirt.c | 5 +++++ 2 files changed, 11 insertions(+), 5 deletions(-) diff --git a/src/e820map.h b/src/e820map.h index 07ce16ec..c23ccb0d 100644 --- a/src/e820map.h +++ b/src/e820map.h @@ -3,11 +3,12 @@ #include "types.h" // u64 -#define E820_RAM 1 -#define E820_RESERVED 2 -#define E820_ACPI 3 -#define E820_NVS 4 -#define E820_UNUSABLE 5 +#define E820_RAM 1 +#define E820_RESERVED 2 +#define E820_ACPI 3 +#define E820_NVS 4 +#define E820_UNUSABLE 5 +#define E820_SOFT_RESERVED 0xEFFFFFFF struct e820entry { u64 start; diff --git a/src/fw/paravirt.c b/src/fw/paravirt.c index e5d4eca0..7b03efc3 100644 --- a/src/fw/paravirt.c +++ b/src/fw/paravirt.c @@ -768,6 +768,11 @@ static int qemu_early_e820(void) dprintf(1, "qemu/e820: addr 0x%016llx len 0x%016llx [reserved]\n", table.address, table.length); break; + case E820_SOFT_RESERVED: + e820_add(table.address, table.length, table.type); + dprintf(1, "qemu/e820: addr 0x%016llx len 0x%016llx [soft-reserved]\n", + table.address, table.length); + break; case E820_RAM: e820_add(table.address, table.length, table.type); dprintf(1, "qemu/e820: addr 0x%016llx len 0x%016llx [RAM]\n", -- 2.34.1

3 2

regression: cannot boot from nvme anymore, disabling 64bit pci io window helps
by Marek Marczykowski-Górecki July 7, 2026

July 7, 2026

Hi, After an update some time ago, a VM wit 8GB RAM cannot boot from nvme anymore. QEMU cmdline: /usr/bin/qemu-system-x86_64 \ -chardev ringbuf,id=serial0,logfile=serial0,logappend=on \ -serial chardev:serial0 \ -m 8192 \ -enable-kvm \ -vnc :2,share=force-shared \ -blockdev driver=file,node-name=hd0-file,filename=disk.qcow2,cache.no-flush=on,discard=unmap \ -blockdev driver=qcow2,node-name=hd0,file=hd0-file,cache.no-flush=on,discard=unmap \ -device nvme,id=hd0-device,drive=hd0,serial=hd0 \ -chardev stdio,id=seabios -device isa-debugcon,iobase=0x402,chardev=seabios \ -bios /home/user/seabios/out/bios.bin There is nothing fancy on the disk, just standard grub2 setup, but also I believe it doesn't matter, as it doesn't even reach grub. When it fails, the seabios log contains: PCI: 64: 0000038000000000 - 0000038040000000 PCI: map device bdf=00:04.0 bar 0, addr 38000000000, size 00004000 [mem] ... Searching bootorder for: /pci@i0cf8/isa@1/fdc@03f0/floppy@0 ATA controller 1 at 1f0/3f4/0 (irq 14 dev 9) ATA controller 2 at 170/374/0 (irq 15 dev 9) Searching bootorder for: HALT Can not map memory bar over 4Gig ... Boot failed: could not read the boot disk Working version has this instead: PCI: map device bdf=00:04.0 bar 0, addr febf0000, size 00004000 [mem] ... Booting from Hard Disk... Booting from 0000:7c00 00:04.0 is the nvme. I'm attaching full logs for both cases. Bisect shows: 96a8d130a8c2e908e357ce62cd713f2cc0b0a2eb is the first bad commit commit 96a8d130a8c2e908e357ce62cd713f2cc0b0a2eb (HEAD) Author: Gerd Hoffmann <kraxel(a)redhat.com> Date: Fri Sep 9 10:17:15 2022 +0200 be less conservative with the 64bit pci io window Current seabios code will only enable and use the 64bit pci io window in case it runs out of space in the 32bit pci mmio window below 4G. This patch will also enable the 64bit pci io window when (a) RAM above 4G is present, and (b) the physical address space size is known, and (c) seabios is running on a 64bit capable processor. This operates with the assumption that guests which are ok with memory above 4G most likely can handle mmio above 4G too. In case the 64bit pci io window is enabled also assign more memory to prefetchable pci bridge windows and the complete 64bit pci io window. The total mmio window size is 1/8 of the physical address space. Minimum bridge windows size is 1/256 of the total mmio window size. Signed-off-by: Gerd Hoffmann <kraxel(a)redhat.com> I've tried the recent "[PATCH] limit address space used for pci devices, part two" patch from the ML, but it doesn't help. OTOH, the series with "only enable 64bit pci io window when RAM >64G" does help. -- Best Regards, Marek Marczykowski-Górecki Invisible Things Lab

3 6