Saving 8 seconds of Linux boot time by setting ahci.mask_port_map

【2025-08-11】→ Gary Laski

ATA ports cost precious boot time due to being probed by libATA. The probing process takes around 1 second (empirically) per port, which in my case, resulted in 8 seconds of boot time wasted.

I stumbled upon this while writing a script to determine the longest steps in boot as reported by dmesg logs. Six of the top ten longest steps were due to these SATA link down messages:

$ ./main
...
Time waste 1.0398389999999997:
	[    3.429959] ata2: SATA link down (SStatus 0 SControl 0)
	[    4.469798] ata3: failed to resume link (SControl 0)
Time waste 1.039834:
	[    5.509962] ata4: SATA link down (SStatus 0 SControl 0)
	[    6.549796] ata5: failed to resume link (SControl 0)
Time waste 1.0398329999999998:
	[    8.629970] ata7: SATA link down (SStatus 0 SControl 0)
	[    9.669803] ata8: failed to resume link (SControl 0)
Time waste 1.0398300000000003:
	[    6.549974] ata5: SATA link down (SStatus 0 SControl 0)
	[    7.589804] ata6: failed to resume link (SControl 0)
Time waste 1.0398080000000007:
	[    7.589983] ata6: SATA link down (SStatus 0 SControl 0)
	[    8.629791] ata7: failed to resume link (SControl 0)
Time waste 1.0398070000000001:
	[    2.389970] ata1: SATA link down (SStatus 0 SControl 0)
	[    3.429777] ata2: failed to resume link (SControl 0)

SATA ports are initialized by libATA and the module ahci, and the probing of all 8 ports takes around 8 seconds.

$ dmesg | less
[    1.339435] libata version 3.00 loaded.
[    1.347914] ahci 0000:02:00.1: version 3.0
[    1.348046] ahci 0000:02:00.1: SSS flag set, parallel bus scan disabled
[    1.348351] ahci 0000:02:00.1: AHCI vers 0001.0301, 32 command slots, 6 Gbps, SATA mode
[    1.348531] ahci 0000:02:00.1: 8/8 ports implemented (port mask 0xff)
[    1.348687] ahci 0000:02:00.1: flags: 64bit ncq sntf stag pm led clo only pmp pio slum part sxs deso sadm sds apst 
[    1.352143] scsi host0: ahci
-- repeats --
[    1.354126] scsi host7: ahci
[    1.354316] ata1: SATA max UDMA/133 abar m131072@0xfce80000 port 0xfce80100 irq 64 lpm-pol 4
-- repeats --
[    1.355252] ata8: SATA max UDMA/133 abar m131072@0xfce80000 port 0xfce80480 irq 64 lpm-pol 4
...
[    2.389787] ata1: failed to resume link (SControl 0)
[    2.389970] ata1: SATA link down (SStatus 0 SControl 0)
-- repeats --
[    9.669803] ata8: failed to resume link (SControl 0)
[    9.669983] ata8: SATA link down (SStatus 0 SControl 0)
...
[   21.964553]

The modinfo for ahci gives a kernel parameter that masks controller ports, which can disable this probing behavior:

$ modinfo ahci | less
...
parm:          mask_port_map:32-bits port map masks to ignore controllers ports. 
               Valid values are: "<mask>" to apply the same mask to all AHCI controller devices, 
               and "<pci_dev>=<mask>,<pci_dev>=<mask>,..." to specify different masks for the controllers specified, 
               where <pci_dev> is the PCI ID of an AHCI controller in the form "domain:bus:dev.func" (charp)

Since I am not using any of these ports, I tried setting the mask value to all 0's:

ahci.mask_port_map=0

Unfortunately, this did not work. There's a catch - the mask value of 0 is reserved for disabling the mask, which I found documented in the kernel source. I'm not sure why this cases exists - there's already a vaule that conveys this information - 0xFFFFFFFF - which is the inital value as shown in the earlier log:

 8/8 ports implemented (port mask 0xff)

There's likely a valid reason for this behavior known to the ahci developers, but it's worth commenting on. For those curious, here's the function in question:

drivers/ata/ahci.h:

/*
 * Return true if a port should be ignored because it is excluded from
 * the host port map.
 */
static inline bool ahci_ignore_port(struct ahci_host_priv *hpriv,
				    unsigned int portid)
{
	if (portid >= hpriv->nports)
		return true;
	/* mask_port_map not set means that all ports are available */
	if (!hpriv->mask_port_map)
		return false;
	return !(hpriv->mask_port_map & (1 << portid));
}

Ultimately, I ended up setting 1 bit higher than any of the ports I need (9th bit): ahci.mask_port_map=0x100. This disables the ports, skipping the 8 second probing step, and reducing boot time from 21 to 13 seconds.

$ dmesg | less
...
[    1.344934] libata version 3.00 loaded.
[    1.349827] ahci 0000:02:00.1: version 3.0
[    1.349883] ahci 0000:02:00.1: masking port_map 0xff -> 0x0
[    1.350121] ahci 0000:02:00.1: SSS flag set, parallel bus scan disabled
[    1.350320] ahci 0000:02:00.1: AHCI vers 0001.0301, 32 command slots, 6 Gbps, SATA mode
[    1.350526] ahci 0000:02:00.1: 0/8 ports implemented (port mask 0x0)
[    1.350673] ahci 0000:02:00.1: flags: 64bit ncq sntf stag pm led clo only pmp pio slum part sxs deso sadm sds apst 
[    1.351795] scsi host0: ahci
-- repeats --
[    1.354037] scsi host7: ahci
[    1.354228] ata1: DUMMY
-- repeats --
[    1.355077] ata8: DUMMY
...
[   13.577978]