Add support for PCIe ARI

PCIe Alternate RID Interpretation (ARI) is an optional feature that
allows devices to have up to 256 different functions.  It is
implemented by always setting the PCI slot number to 0 and
re-purposing the 5 bits used to encode the slot number to instead
contain the function number.  Combined with the original 3 bits
allocated for the function number, this allows for 256 functions.

This is enabled by default, but it's expected to be a no-op on currently
supported hardware.  It's a prerequisite for supporting PCI SR-IOV, and
I want the ARI support to go in early to help shake out any bugs in it.
ARI can be disabled by setting the tunable hw.pci.enable_ari=0.

Reviewed by:	kib
MFC after:	2 months
Sponsored by:	Sandvine Inc.
This commit is contained in:
rstone 2014-04-01 16:02:02 +00:00
parent 2d24e0ab16
commit c8c0939762
6 changed files with 242 additions and 17 deletions

View File

@ -354,6 +354,11 @@ SYSCTL_INT(_hw_pci, OID_AUTO, clear_buses, CTLFLAG_RDTUN, &pci_clear_buses, 0,
"Ignore firmware-assigned bus numbers.");
#endif
static int pci_enable_ari = 1;
TUNABLE_INT("hw.pci.enable_ari", &pci_enable_ari);
SYSCTL_INT(_hw_pci, OID_AUTO, enable_ari, CTLFLAG_RDTUN, &pci_enable_ari,
0, "Enable support for PCIe Alternative RID Interpretation");
static int
pci_has_quirk(uint32_t devid, int quirk)
{
@ -3454,6 +3459,19 @@ pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask)
#endif
}
static struct pci_devinfo *
pci_identify_function(device_t pcib, device_t dev, int domain, int busno,
int slot, int func, size_t dinfo_size)
{
struct pci_devinfo *dinfo;
dinfo = pci_read_device(pcib, domain, busno, slot, func, dinfo_size);
if (dinfo != NULL)
pci_add_child(dev, dinfo);
return (dinfo);
}
void
pci_add_children(device_t dev, int domain, int busno, size_t dinfo_size)
{
@ -3463,6 +3481,24 @@ pci_add_children(device_t dev, int domain, int busno, size_t dinfo_size)
int maxslots;
int s, f, pcifunchigh;
uint8_t hdrtype;
int first_func;
/*
* Try to detect a device at slot 0, function 0. If it exists, try to
* enable ARI. We must enable ARI before detecting the rest of the
* functions on this bus as ARI changes the set of slots and functions
* that are legal on this bus.
*/
dinfo = pci_identify_function(pcib, dev, domain, busno, 0, 0,
dinfo_size);
if (dinfo != NULL && pci_enable_ari)
PCIB_TRY_ENABLE_ARI(pcib, dinfo->cfg.dev);
/*
* Start looking for new devices on slot 0 at function 1 because we
* just identified the device at slot 0, function 0.
*/
first_func = 1;
KASSERT(dinfo_size >= sizeof(struct pci_devinfo),
("dinfo_size too small"));
@ -3475,14 +3511,13 @@ pci_add_children(device_t dev, int domain, int busno, size_t dinfo_size)
if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE)
continue;
if (hdrtype & PCIM_MFDEV)
pcifunchigh = PCI_FUNCMAX;
for (f = 0; f <= pcifunchigh; f++) {
dinfo = pci_read_device(pcib, domain, busno, s, f,
pcifunchigh = PCIB_MAXFUNCS(pcib);
for (f = first_func; f <= pcifunchigh; f++)
pci_identify_function(pcib, dev, domain, busno, s, f,
dinfo_size);
if (dinfo != NULL) {
pci_add_child(dev, dinfo);
}
}
/* For slots after slot 0 we need to check for function 0. */
first_func = 0;
}
#undef REG
}

View File

@ -56,6 +56,14 @@ static int pcib_suspend(device_t dev);
static int pcib_resume(device_t dev);
static int pcib_power_for_sleep(device_t pcib, device_t dev,
int *pstate);
static uint16_t pcib_ari_get_rid(device_t pcib, device_t dev);
static uint32_t pcib_read_config(device_t dev, u_int b, u_int s,
u_int f, u_int reg, int width);
static void pcib_write_config(device_t dev, u_int b, u_int s,
u_int f, u_int reg, uint32_t val, int width);
static int pcib_ari_maxslots(device_t dev);
static int pcib_ari_maxfuncs(device_t dev);
static int pcib_try_enable_ari(device_t pcib, device_t dev);
static device_method_t pcib_methods[] = {
/* Device interface */
@ -83,7 +91,8 @@ static device_method_t pcib_methods[] = {
DEVMETHOD(bus_teardown_intr, bus_generic_teardown_intr),
/* pcib interface */
DEVMETHOD(pcib_maxslots, pcib_maxslots),
DEVMETHOD(pcib_maxslots, pcib_ari_maxslots),
DEVMETHOD(pcib_maxfuncs, pcib_ari_maxfuncs),
DEVMETHOD(pcib_read_config, pcib_read_config),
DEVMETHOD(pcib_write_config, pcib_write_config),
DEVMETHOD(pcib_route_interrupt, pcib_route_interrupt),
@ -93,7 +102,8 @@ static device_method_t pcib_methods[] = {
DEVMETHOD(pcib_release_msix, pcib_release_msix),
DEVMETHOD(pcib_map_msi, pcib_map_msi),
DEVMETHOD(pcib_power_for_sleep, pcib_power_for_sleep),
DEVMETHOD(pcib_get_rid, pcib_get_rid),
DEVMETHOD(pcib_get_rid, pcib_ari_get_rid),
DEVMETHOD(pcib_try_enable_ari, pcib_try_enable_ari),
DEVMETHOD_END
};
@ -1802,28 +1812,95 @@ pcib_alloc_resource(device_t dev, device_t child, int type, int *rid,
}
#endif
/*
* If ARI is enabled on this downstream port, translate the function number
* to the non-ARI slot/function. The downstream port will convert it back in
* hardware. If ARI is not enabled slot and func are not modified.
*/
static __inline void
pcib_xlate_ari(device_t pcib, int bus, int *slot, int *func)
{
struct pcib_softc *sc;
int ari_func;
sc = device_get_softc(pcib);
ari_func = *func;
if (sc->flags & PCIB_ENABLE_ARI) {
KASSERT(*slot == 0,
("Non-zero slot number with ARI enabled!"));
*slot = PCIE_ARI_SLOT(ari_func);
*func = PCIE_ARI_FUNC(ari_func);
}
}
static void
pcib_enable_ari(struct pcib_softc *sc, uint32_t pcie_pos)
{
uint32_t ctl2;
ctl2 = pci_read_config(sc->dev, pcie_pos + PCIER_DEVICE_CTL2, 4);
ctl2 |= PCIEM_CTL2_ARI;
pci_write_config(sc->dev, pcie_pos + PCIER_DEVICE_CTL2, ctl2, 4);
sc->flags |= PCIB_ENABLE_ARI;
}
/*
* PCIB interface.
*/
int
pcib_maxslots(device_t dev)
{
return(PCI_SLOTMAX);
return (PCI_SLOTMAX);
}
static int
pcib_ari_maxslots(device_t dev)
{
struct pcib_softc *sc;
sc = device_get_softc(dev);
if (sc->flags & PCIB_ENABLE_ARI)
return (PCIE_ARI_SLOTMAX);
else
return (PCI_SLOTMAX);
}
static int
pcib_ari_maxfuncs(device_t dev)
{
struct pcib_softc *sc;
sc = device_get_softc(dev);
if (sc->flags & PCIB_ENABLE_ARI)
return (PCIE_ARI_FUNCMAX);
else
return (PCI_FUNCMAX);
}
/*
* Since we are a child of a PCI bus, its parent must support the pcib interface.
*/
uint32_t
static uint32_t
pcib_read_config(device_t dev, u_int b, u_int s, u_int f, u_int reg, int width)
{
return(PCIB_READ_CONFIG(device_get_parent(device_get_parent(dev)), b, s, f, reg, width));
pcib_xlate_ari(dev, b, &s, &f);
return(PCIB_READ_CONFIG(device_get_parent(device_get_parent(dev)), b, s,
f, reg, width));
}
void
static void
pcib_write_config(device_t dev, u_int b, u_int s, u_int f, u_int reg, uint32_t val, int width)
{
PCIB_WRITE_CONFIG(device_get_parent(device_get_parent(dev)), b, s, f, reg, val, width);
pcib_xlate_ari(dev, b, &s, &f);
PCIB_WRITE_CONFIG(device_get_parent(device_get_parent(dev)), b, s, f,
reg, val, width);
}
/*
@ -1936,3 +2013,82 @@ pcib_power_for_sleep(device_t pcib, device_t dev, int *pstate)
return (PCIB_POWER_FOR_SLEEP(bus, dev, pstate));
}
static uint16_t
pcib_ari_get_rid(device_t pcib, device_t dev)
{
struct pcib_softc *sc;
uint8_t bus, slot, func;
sc = device_get_softc(pcib);
if (sc->flags & PCIB_ENABLE_ARI) {
bus = pci_get_bus(dev);
func = pci_get_function(dev);
return (PCI_ARI_RID(bus, func));
} else {
bus = pci_get_bus(dev);
slot = pci_get_slot(dev);
func = pci_get_function(dev);
return (PCI_RID(bus, slot, func));
}
}
/*
* Check that the downstream port (pcib) and the endpoint device (dev) both
* support ARI. If so, enable it and return 0, otherwise return an error.
*/
static int
pcib_try_enable_ari(device_t pcib, device_t dev)
{
struct pcib_softc *sc;
int error;
uint32_t cap2;
int ari_cap_off;
uint32_t ari_ver;
uint32_t pcie_pos;
sc = device_get_softc(pcib);
/*
* ARI is controlled in a register in the PCIe capability structure.
* If the downstream port does not have the PCIe capability structure
* then it does not support ARI.
*/
error = pci_find_cap(pcib, PCIY_EXPRESS, &pcie_pos);
if (error != 0)
return (ENODEV);
/* Check that the PCIe port advertises ARI support. */
cap2 = pci_read_config(pcib, pcie_pos + PCIER_DEVICE_CAP2, 4);
if (!(cap2 & PCIEM_CAP2_ARI))
return (ENODEV);
/*
* Check that the endpoint device advertises ARI support via the ARI
* extended capability structure.
*/
error = pci_find_extcap(dev, PCIZ_ARI, &ari_cap_off);
if (error != 0)
return (ENODEV);
/*
* Finally, check that the endpoint device supports the same version
* of ARI that we do.
*/
ari_ver = pci_read_config(dev, ari_cap_off, 4);
if (PCI_EXTCAP_VER(ari_ver) != PCIB_SUPPORTED_ARI_VER) {
if (bootverbose)
device_printf(pcib,
"Unsupported version of ARI (%d) detected\n",
PCI_EXTCAP_VER(ari_ver));
return (ENXIO);
}
pcib_enable_ari(sc, pcie_pos);
return (0);
}

View File

@ -27,7 +27,9 @@
#
#include <sys/bus.h>
#include <sys/rman.h>
#include <dev/pci/pcivar.h>
#include <dev/pci/pcib_private.h>
INTERFACE pcib;
@ -46,6 +48,14 @@ METHOD int maxslots {
device_t dev;
};
#
#
# Return the number of functions on the attached PCI bus.
#
METHOD int maxfuncs {
device_t dev;
} DEFAULT pcib_maxfuncs;
#
# Read configuration space on the PCI bus. The bus, slot and func
# arguments determine the device which is being read and the reg
@ -161,5 +171,14 @@ METHOD int power_for_sleep {
METHOD uint16_t get_rid {
device_t pcib;
device_t dev;
} DEFAULT pcib_get_rid;
#
# Enable Alternative RID Interpretation if both the downstream port (pcib)
# and the endpoint device (dev) both support it.
#
METHOD int try_enable_ari {
device_t pcib;
device_t dev;
};

View File

@ -105,6 +105,7 @@ struct pcib_softc
#define PCIB_SUBTRACTIVE 0x1
#define PCIB_DISABLE_MSI 0x2
#define PCIB_DISABLE_MSIX 0x4
#define PCIB_ENABLE_ARI 0x8
uint16_t command; /* command register */
u_int domain; /* domain number */
u_int pribus; /* primary bus number */
@ -126,6 +127,8 @@ struct pcib_softc
uint8_t seclat; /* secondary bus latency timer */
};
#define PCIB_SUPPORTED_ARI_VER 1
typedef uint32_t pci_read_config_fn(int b, int s, int f, int reg, int width);
int host_pcib_get_busno(pci_read_config_fn read_config, int bus,
@ -159,8 +162,7 @@ int pcib_release_resource(device_t dev, device_t child, int type, int rid,
struct resource *r);
#endif
int pcib_maxslots(device_t dev);
uint32_t pcib_read_config(device_t dev, u_int b, u_int s, u_int f, u_int reg, int width);
void pcib_write_config(device_t dev, u_int b, u_int s, u_int f, u_int reg, uint32_t val, int width);
int pcib_maxfuncs(device_t dev);
int pcib_route_interrupt(device_t pcib, device_t dev, int pin);
int pcib_alloc_msi(device_t pcib, device_t dev, int count, int maxcount, int *irqs);
int pcib_release_msi(device_t pcib, device_t dev, int count, int *irqs);
@ -169,5 +171,4 @@ int pcib_release_msix(device_t pcib, device_t dev, int irq);
int pcib_map_msi(device_t pcib, device_t dev, int irq, uint64_t *addr, uint32_t *data);
uint16_t pcib_get_rid(device_t pcib, device_t dev);
#endif

View File

@ -48,6 +48,12 @@ __FBSDID("$FreeBSD$");
#include "pcib_if.h"
int
pcib_maxfuncs(device_t dev)
{
return (PCI_FUNCMAX);
}
uint16_t
pcib_get_rid(device_t pcib, device_t dev)
{

View File

@ -48,6 +48,9 @@
#define PCIE_REGMAX 4095 /* highest supported config register addr. */
#define PCI_MAXHDRTYPE 2
#define PCIE_ARI_SLOTMAX 0
#define PCIE_ARI_FUNCMAX 255
#define PCI_RID_BUS_SHIFT 8
#define PCI_RID_SLOT_SHIFT 3
#define PCI_RID_FUNC_SHIFT 0
@ -65,6 +68,9 @@
#define PCI_RID2SLOT(rid) (((rid) >> PCI_RID_SLOT_SHIFT) & PCI_SLOTMAX)
#define PCI_RID2FUNC(rid) (((rid) >> PCI_RID_FUNC_SHIFT) & PCI_FUNCMAX)
#define PCIE_ARI_SLOT(func) (((func) >> PCI_RID_SLOT_SHIFT) & PCI_SLOTMAX)
#define PCIE_ARI_FUNC(func) (((func) >> PCI_RID_FUNC_SHIFT) & PCI_FUNCMAX)
/* PCI config header registers for all devices */
#define PCIR_DEVVENDOR 0x00
@ -791,6 +797,7 @@
#define PCIEM_ROOT_STA_PME_STATUS 0x00010000
#define PCIEM_ROOT_STA_PME_PEND 0x00020000
#define PCIER_DEVICE_CAP2 0x24
#define PCIEM_CAP2_ARI 0x20
#define PCIER_DEVICE_CTL2 0x28
#define PCIEM_CTL2_COMP_TIMEOUT_VAL 0x000f
#define PCIEM_CTL2_COMP_TIMEOUT_DIS 0x0010
@ -911,3 +918,4 @@
/* Serial Number definitions */
#define PCIR_SERIAL_LOW 0x04
#define PCIR_SERIAL_HIGH 0x08