SeaBIOS

seabios@seabios.org

10 participants
3770 discussions

Re: [PATCH 1/2] romfile: implement a generic loader
by Riku Viitanen 14 Feb '24

14 Feb '24

> I'd suggest to add a romfile_loadfile_low() I can definitely do that. > or add a 'struct zone_s' > parameter instead of passing in a function pointer. Sounds sensible. I see there's a malloc that takes zone as argument. > Also: What the trailing bytes are used for? Only for re-implementing the existing loadfile function, to reduce copy-pasted code.

1 0

[PATCH 1/2] romfile: implement a generic loader
by Riku Viitanen 14 Feb '24

14 Feb '24

romfile_loadfile_g: Based on romfile_loadfile but more flexible. User has to supply pointer to a malloc function and the number of trailing padding bytes. Thus, any memory region may be used. romfile_loadfile: It is now a wrapper around romfile_loadfile_g. Functionality is the same. Signed-off-by: Riku Viitanen <riku.viitanen(a)protonmail.com> --- src/romfile.c | 25 ++++++++++++++++++++----- src/romfile.h | 2 ++ 2 files changed, 22 insertions(+), 5 deletions(-) diff --git a/src/romfile.c b/src/romfile.c index b598274e..8bf95713 100644 --- a/src/romfile.c +++ b/src/romfile.c @@ -47,10 +47,12 @@ romfile_find(const char *name) return __romfile_findprefix(name, strlen(name) + 1, NULL); } -// Helper function to find, malloc_tmphigh, and copy a romfile. This -// function adds a trailing zero to the malloc'd copy. +// Generic function to find romfile, malloc (using provided function +// pointer), and copy a romfile. add_len specifies how many additional +// trailing bytes to reserve. The extra bytes will not be initialised. void * -romfile_loadfile(const char *name, int *psize) +romfile_loadfile_g(const char *name, int *psize, + void *(*malloc_fn)(), int add_len) { struct romfile_s *file = romfile_find(name); if (!file) @@ -60,7 +62,7 @@ romfile_loadfile(const char *name, int *psize) if (!filesize) return NULL; - char *data = malloc_tmphigh(filesize+1); + char *data = malloc_fn(filesize+add_len); if (!data) { warn_noalloc(); return NULL; @@ -74,7 +76,20 @@ romfile_loadfile(const char *name, int *psize) } if (psize) *psize = filesize; - data[filesize] = '\0'; + + return data; +} + +// Helper function to find, malloc_tmphigh, and copy a romfile. This +// function adds a trailing zero to the malloc'd copy. +void * +romfile_loadfile(const char *name, int *psize) +{ + char *data = romfile_loadfile_g(name, psize, &malloc_tmphigh, 1); + if (!data) + return NULL; + + data[*psize] = '\0'; return data; } diff --git a/src/romfile.h b/src/romfile.h index 3e0f8200..a320a5bc 100644 --- a/src/romfile.h +++ b/src/romfile.h @@ -13,6 +13,8 @@ struct romfile_s { void romfile_add(struct romfile_s *file); struct romfile_s *romfile_findprefix(const char *prefix, struct romfile_s *prev); struct romfile_s *romfile_find(const char *name); +void *romfile_loadfile_g(const char *name, int *psize, + void *(*malloc_fn)(), int add_len); void *romfile_loadfile(const char *name, int *psize); u64 romfile_loadint(const char *name, u64 defval); -- 2.43.0

2 1

[PATCH 2/2] vgahooks, optionroms: implement mxm 3.0 interrupts
by Riku Viitanen 10 Feb '24

10 Feb '24

VGAROMs on MXM graphics cards need certain int15h functions present. Tested on a HP EliteBook 8560w with coreboot and Quadro 2000M. A warning is displayed for 30 seconds and performance is nerfed: ERROR: Valid MXM Structure not found. POST halted for 30 seconds, P-state limited to P10... This patch implements the minimum required on this system (and implemented by the OEM BIOS): functions 0 and 1. These functions are specific to the MXM 3.0 Software Specification, earlier versions are not implemented due to lack of hardware. Documentation for versions 2.1 and 3.0 can be found freely online. Functions aren't specific to mainboards or GPUs (though some mainboards could need more functions implemented). The structure is mainboard-specific and is read from romfile "mxm-30-sis". It can be extracted from vendor BIOS by running those same interrupts. I wrote a tool to do it on Linux: https://codeberg.org/Riku_V/mxmdump/ Signed-off-by: Riku Viitanen <riku.viitanen(a)protonmail.com> --- src/optionroms.c | 9 +++++++ src/vgahooks.c | 69 ++++++++++++++++++++++++++++++++++++++++++++++++ src/vgahooks.h | 9 +++++++ 3 files changed, 87 insertions(+) create mode 100644 src/vgahooks.h diff --git a/src/optionroms.c b/src/optionroms.c index e906ab97..fcce7900 100644 --- a/src/optionroms.c +++ b/src/optionroms.c @@ -22,6 +22,7 @@ #include "string.h" // memset #include "util.h" // get_pnp_offset #include "tcgbios.h" // tpm_* +#include "vgahooks.h" // MXM30SIS static int EnforceChecksum, S3ResumeVga, RunPCIroms; @@ -463,6 +464,14 @@ vgarom_setup(void) RunPCIroms = romfile_loadint("etc/pci-optionrom-exec", 2); ScreenAndDebug = romfile_loadint("etc/screen-and-debug", 1); + // Load MXM 3.0 System Information Structure + void *mxm_sis = romfile_loadfile_g("mxm-30-sis", NULL, &malloc_low, 0); + if (mxm_sis) { + MXM30SIS = (u32)mxm_sis; + } else { + MXM30SIS = 0; + } + // Clear option rom memory memset((void*)BUILD_ROM_START, 0, rom_get_max() - BUILD_ROM_START); diff --git a/src/vgahooks.c b/src/vgahooks.c index 1f149532..a94840b2 100644 --- a/src/vgahooks.c +++ b/src/vgahooks.c @@ -18,8 +18,10 @@ #define VH_VIA 1 #define VH_INTEL 2 #define VH_SMI 3 +#define VH_MXM 4 int VGAHookHandlerType VARFSEG; +u32 MXM30SIS VARFSEG; static void handle_155fXX(struct bregs *regs) @@ -59,6 +61,7 @@ via_155f02(struct bregs *regs) dprintf(1, "Warning: VGA TV/CRT output type is hardcoded\n"); } + static void via_155f18(struct bregs *regs) { @@ -296,6 +299,69 @@ winent_mb6047_setup(struct pci_device *pci) SmiBootDisplay = 0x02; } +/**************************************************************** + * MXM VGA hooks + ****************************************************************/ + +// Function 0: Return Specification Support Level +static void +mxm_V30_F00(struct bregs *regs) +{ + regs->ax = 0x005f; // Success + regs->bl = 0x30; // MXM 3.0 + regs->cx = 0x0003; // Supported Functions + set_success(regs); +} + +// Function 1: Return a Pointer to the MXM Structure +static void +mxm_V30_F01(struct bregs *regs) +{ + switch (regs->cx) { + case 0x0030: + regs->ax = 0x005f; // Success + regs->es = GET_GLOBAL(MXM30SIS)/16; + regs->di = GET_GLOBAL(MXM30SIS)%16; + set_success(regs); + break; + default: + handle_155fXX(regs); + break; + } +} + +static void +mxm_V30(struct bregs *regs) +{ + switch (regs->bx) { + case 0xff00: mxm_V30_F00(regs); break; + case 0xff01: mxm_V30_F01(regs); break; + default: handle_155fXX(regs); break; + } +} + +static void +mxm_155f80(struct bregs *regs) +{ + // TODO: implement other versions, like 2.1 + mxm_V30(regs); +} + +static void +mxm_155f(struct bregs *regs) +{ + switch (regs->al) { + case 0x80: mxm_155f80(regs); break; + default: handle_155fXX(regs); break; + } +} + +void +mxm_setup(void) +{ + VGAHookHandlerType = VH_MXM; +} + /**************************************************************** * Entry and setup ****************************************************************/ @@ -313,6 +379,7 @@ handle_155f(struct bregs *regs) switch (htype) { case VH_VIA: via_155f(regs); break; case VH_INTEL: intel_155f(regs); break; + case VH_MXM: mxm_155f(regs); break; default: handle_155fXX(regs); break; } } @@ -352,4 +419,6 @@ vgahook_setup(struct pci_device *pci) via_setup(pci); else if (pci->vendor == PCI_VENDOR_ID_INTEL) intel_setup(pci); + else if (GET_GLOBAL(MXM30SIS)) + mxm_setup(); } diff --git a/src/vgahooks.h b/src/vgahooks.h new file mode 100644 index 00000000..f0c203af --- /dev/null +++ b/src/vgahooks.h @@ -0,0 +1,9 @@ +#ifndef __VGAHOOKS_H +#define __VGAHOOKS_H + +#include "types.h" // u32 + +extern u32 MXM30SIS; + + +#endif // vgahooks.h -- 2.43.0

1 0

[RFC v3 0/1] Add LBA 64bit support.
by Max Tottenham 26 Jan '24

26 Jan '24

Changes since V2: * Print READ CAPACITY(16) debug call at higher log level * Cleanup return statement for CMD_WRITE in src/hw/blockcmd.c:scsi_fill_cmd() * Change format specifier for printing drive->sectors to %llx * Minor commit log spelling/punctuation fixes Changes since V1: * De-duplicate blksize check and capacity logging * Fix off by 1 error for returned sector count (LBA addresses start from 0). Max Tottenham (1): Add LBA 64bit support for reads beyond 2TB. src/hw/blockcmd.c | 58 +++++++++++++++++++++++++++++++++++------------ src/hw/blockcmd.h | 27 ++++++++++++++++++++++ 2 files changed, 71 insertions(+), 14 deletions(-) -- 2.34.1

2 2

[RFC v2 0/1] Add LBA 64bit support.
by Max Tottenham 25 Jan '24

25 Jan '24

Changes since V1: * De-duplicate blksize check and capacity logging * Fix off by 1 error for returned sector count (LBA addresses start from 0). Max Tottenham (1): Add LBA 64bit support for reads beyond 2TB. src/hw/blockcmd.c | 61 +++++++++++++++++++++++++++++++++++------------ src/hw/blockcmd.h | 27 +++++++++++++++++++++ 2 files changed, 73 insertions(+), 15 deletions(-) -- 2.34.1

3 9

[RFC] Add LBA 64bit support for reads beyond 2TB.
by Max Tottenham 25 Jan '24

25 Jan '24

When booting from a >2TB drive/filesystem, it's possible what the kernel/bootloader may be updated and written out at an LBA address beyond what is normally accessible by the READ(10) SCSI commands. If this happens to the kernel grub will fail to boot the kernel as it will call into the BIOS with an LBA address >2TB, and the BIOS will return an error. Per the SCSI spec, >2TB drives should return 0XFFFFFFFF, and a READ CAPACITY(16) command should be issues to determine the full size of the drive, READ(16) commands can then be used in order to read data at LBA addresses beyond 2TB (64 bit LBA addresses). --- src/hw/blockcmd.c | 75 +++++++++++++++++++++++++++++++++++------------ src/hw/blockcmd.h | 27 +++++++++++++++++ 2 files changed, 84 insertions(+), 18 deletions(-) diff --git a/src/hw/blockcmd.c b/src/hw/blockcmd.c index 6b6fea97..4f5dc2d7 100644 --- a/src/hw/blockcmd.c +++ b/src/hw/blockcmd.c @@ -66,6 +66,23 @@ cdb_test_unit_ready(struct disk_op_s *op) return process_op(op); } +static int +cdb_read_capacity16(struct disk_op_s *op, struct cdbres_read_capacity_16 *data) +{ + struct cdb_sai_read_capacity_16 cmd; + memset(&cmd, 0, sizeof(cmd)); + cmd.command = CDB_CMD_SERVICE_ACTION_IN; + cmd.flags = CDB_CMD_SAI_READ_CAPACITY_16; + cmd.len = cpu_to_be32(sizeof(struct cdbres_read_capacity_16)); + op->command = CMD_SCSI; + op->count = 1; + op->buf_fl = data; + op->cdbcmd = &cmd; + op->blocksize = sizeof(*data); + return process_op(op); +} + + // Request capacity static int cdb_read_capacity(struct disk_op_s *op, struct cdbres_read_capacity *data) @@ -111,13 +128,23 @@ scsi_fill_cmd(struct disk_op_s *op, void *cdbcmd, int maxcdb) switch (op->command) { case CMD_READ: case CMD_WRITE: ; - struct cdb_rwdata_10 *cmd = cdbcmd; - memset(cmd, 0, maxcdb); - cmd->command = (op->command == CMD_READ ? CDB_CMD_READ_10 - : CDB_CMD_WRITE_10); - cmd->lba = cpu_to_be32(op->lba); - cmd->count = cpu_to_be16(op->count); - return GET_FLATPTR(op->drive_fl->blksize); + if (op->lba < 0xFFFFFFFFULL) { + struct cdb_rwdata_10 *cmd = cdbcmd; + memset(cmd, 0, maxcdb); + cmd->command = (op->command == CMD_READ ? CDB_CMD_READ_10 + : CDB_CMD_WRITE_10); + cmd->lba = cpu_to_be32(op->lba); + cmd->count = cpu_to_be16(op->count); + return GET_FLATPTR(op->drive_fl->blksize); + } else { + struct cdb_rwdata_16 *cmd = cdbcmd; + memset(cmd, 0, maxcdb); + cmd->command = (op->command == CMD_READ ? CDB_CMD_READ_16 + : CDB_CMD_WRITE_16); + cmd->lba = cpu_to_be64(op->lba); + cmd->count = cpu_to_be32(op->count); + return GET_FLATPTR(op->drive_fl->blksize); + } case CMD_SCSI: if (MODESEGMENT) return -1; @@ -331,18 +358,30 @@ scsi_drive_setup(struct drive_s *drive, const char *s, int prio) if (ret) return ret; - // READ CAPACITY returns the address of the last block. - // We do not bother with READ CAPACITY(16) because BIOS does not support - // 64-bit LBA anyway. - drive->blksize = be32_to_cpu(capdata.blksize); - if (drive->blksize != DISK_SECTOR_SIZE) { - dprintf(1, "%s: unsupported block size %d\n", s, drive->blksize); - return -1; - } - drive->sectors = (u64)be32_to_cpu(capdata.sectors) + 1; - dprintf(1, "%s blksize=%d sectors=%u\n" - , s, drive->blksize, (unsigned)drive->sectors); + if (be32_to_cpu(capdata.sectors) == 0xFFFFFFFFUL) { + dprintf(1, "%s: >2TB Detected trying READCAP(16)\n", s); + struct cdbres_read_capacity_16 capdata16; + ret = cdb_read_capacity16(&dop, &capdata16); + + drive->blksize = be32_to_cpu(capdata16.blksize); + if (drive->blksize != DISK_SECTOR_SIZE) { + dprintf(1, "%s: unsupported block size %d\n", s, drive->blksize); + return -1; + } + drive->sectors = be64_to_cpu(capdata16.sectors); + dprintf(1, "READCAP(16) %s blksize=%d sectors=%llX\n" + , s, drive->blksize, drive->sectors); + } else { + drive->blksize = be32_to_cpu(capdata.blksize); + if (drive->blksize != DISK_SECTOR_SIZE) { + dprintf(1, "%s: unsupported block size %d\n", s, drive->blksize); + return -1; + } + drive->sectors = (u64)be32_to_cpu(capdata.sectors) + 1; + dprintf(1, "%s blksize=%d sectors=%u\n" + , s, drive->blksize, (unsigned)drive->sectors); + } // We do not recover from USB stalls, so try to be safe and avoid // sending the command if the (obsolete, but still provided by QEMU) // fixed disk geometry page may not be supported. diff --git a/src/hw/blockcmd.h b/src/hw/blockcmd.h index f18543ed..2683186c 100644 --- a/src/hw/blockcmd.h +++ b/src/hw/blockcmd.h @@ -18,6 +18,16 @@ struct cdb_rwdata_10 { u8 pad[6]; } PACKED; +#define CDB_CMD_READ_16 0x88 +#define CDB_CMD_WRITE_16 0x8A +struct cdb_rwdata_16 { + u8 command; + u8 flags; + u64 lba; + u32 count; + u16 reserved_14; +} PACKED; + #define CDB_CMD_READ_CAPACITY 0x25 struct cdb_read_capacity { @@ -32,6 +42,23 @@ struct cdbres_read_capacity { u32 blksize; } PACKED; + +#define CDB_CMD_SERVICE_ACTION_IN 0x9E +#define CDB_CMD_SAI_READ_CAPACITY_16 0x10 +struct cdb_sai_read_capacity_16 { + u8 command; + u8 flags; + u64 lba; //marked as obsolete? + u32 len; + u16 reserved_14; +} PACKED; + +struct cdbres_read_capacity_16 { + u64 sectors; + u32 blksize; + u8 reserved_12[20]; +} PACKED; + #define CDB_CMD_TEST_UNIT_READY 0x00 #define CDB_CMD_INQUIRY 0x12 #define CDB_CMD_REQUEST_SENSE 0x03 -- 2.34.1

2 1

[PATCH] Add AHCI Power ON + ICC_ACTIVE into port setup code
by Andrej Kruták 25 Jan '24

25 Jan '24

Windows appears to put the AHCI port into 'Partial power management state' during reboot, the command puts it back into 'active state'. AHCI/1: link down 0x00000231 (SCR STAT register) -> AHCI/1: link up 0x00000133 Signed-off-by: Andrej Krutak andrej.krutak(a)sysgo.com --- src/hw/ahci.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/hw/ahci.c b/src/hw/ahci.c index 3fa845aa..4f0f640a 100644 --- a/src/hw/ahci.c +++ b/src/hw/ahci.c @@ -444,7 +444,8 @@ static int ahci_port_setup(struct ahci_port_s *port) ahci_port_writel(ctrl, pnr, PORT_CMD, cmd); /* spin up */ - cmd |= PORT_CMD_SPIN_UP; + cmd &= ~PORT_CMD_ICC_MASK; + cmd |= PORT_CMD_SPIN_UP | PORT_CMD_POWER_ON | PORT_CMD_ICC_ACTIVE; ahci_port_writel(ctrl, pnr, PORT_CMD, cmd); u32 end = timer_calc(AHCI_LINK_TIMEOUT); for (;;) { -- 2.34.1

2 1

[PATCH] esp-scsi: terminate DMA transfer when ESP data transfer completes
by Mark Cave-Ayland 25 Jan '24

25 Jan '24

When the ESP data transfer completes indicated by the STAT_TC flag being set, terminate the DMA transfer by issuing a DMA IDLE command. Otherwise in the case where the guest sends a reset followed by an ESP command, the DMA signal remains enabled and so the next SeaBIOS DMA transfer begins immediately when the next ESP command is received rather than waiting until the data is ready and the DMA command is issued. With this fix it is possible to boot a Windows XP ISO to the installer and complete a full installation within QEMU directly using SeaBIOS. Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk> --- src/hw/esp-scsi.c | 2 ++ 1 file changed, 2 insertions(+) [This fixes the Windows XP setup issue reported by Paolo when DC390 support was first added to QEMU (see https://lore.kernel.org/all/501A94D0.3080502@redhat.com/ for the original thread) way back in 2012.] diff --git a/src/hw/esp-scsi.c b/src/hw/esp-scsi.c index f19804f..8612069 100644 --- a/src/hw/esp-scsi.c +++ b/src/hw/esp-scsi.c @@ -147,6 +147,8 @@ esp_scsi_process_op(struct disk_op_s *op) /* At end of DMA TC is set again -> complete command. */ if (state == 1 && (stat & ESP_STAT_TC)) { state++; + /* Terminate esp_scsi_dma() command */ + outb(0, iobase + ESP_DMA_CMD); continue; } -- 2.39.2

2 1

Re: [PATCH 0/8] SeaVGABios/cbvga: Assume VGA compatible GPU in, text-mode
by segfault_bilibili 20 Jan '24

20 Jan '24

Hi Patrick, First, apologies for my bad English grammar and email composing skills. I tried your patch [7/8] to fix bugcheck 50 bsod on my Dell Latitude E6430 which is running Libreboot. In short, your patch initially did not fix the bsod, but after an one-line change of code it successfully fixed bsod, although I still don't know the full situation. According to my observation, _rom_header_size must be set to some value greater than 0x40, like 0x41. Under vendor BIOS, igdkmd64 is observed to have a "reattempt memmap" logic only if _rom_header_size > 0x40. After this reattempt of memmap completed, the driver was loaded correctly and it won't bsod. Otherwise, if _rom_header_size <= 0x40, such logic was seemed to be skipped, and then at certain breakpoint I observed that nothing but "??" representing unreadable bytes filled up windbg's memory peek window, because the vmaddr which was supposed to point to VBT actually pointed to invalid address. `!vtop` command confirmed this as well. In vgasrc/cbvga.c, related lines: > /* Increment rom header size */ > size +=3D (GET_FARVAR(0, vbt->hdr_vbt_size) + 512 - 1) >> 9; > SET_VGA(_rom_header_size, size); I set `size` to at least 0x41, like: size +=3D (GET_FARVAR(0, vbt->hdr_vbt_size) + 512 - 1) >> 9; if (size <= 0x40) size = 0x40 + 1; // newly added line SET_VGA(_rom_header_size, size); and then the bsod is gone. Cheers, segfault

1 0

trying to boot off vhost-user-fs-pci device?
by Michael Tokarev 29 Dec '23

29 Dec '23

Hi! While debugging an unrelated issue, I come to an interesting effect. When I run qemu with two devices, one vhost-user-fs-pci and another, say, virtio-blk, - both these devices have bootindex property. If I specify virtio-blk first and vhost-user-fs-pci second, everything works as expected and seabios boots the OS out of virtio-blk device. But if I specify vhost-user-fs-pci first, seabios will try to boot off virtiofs one, which isn't even a block device, and for which seabios does not have any driver. And in this case the boot just fails: Booting from Hard Disk.. Boot failed: could not read the boot disk next it tries pxe boot, cdrom, and fails. Does it make sense to try to boot off unsupported devices? Thanks, /mjt

2 1

Jump to page:

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

SeaBIOS