: hotplug/ia64: SN Hotplug Driver - SN Hotplug Driver code
Quoted 17 lines · Prarit Bhargava
This patch is the SGI hotplug driver and additional changes required for the driver. These modifications include changes to the SN
io_init.ccode for memory management, the inclusion of new SAL calls to enable and disable PCI slots, and a hotplug-style driver.Signed-off-by Prarit Bhargava <[email protected]>Index:
drivers/pci/hotplug/sgi_hotplug.c
===================================================================
---/dev/null(tree:94b3c225a19fb0e688da58fa7c9239db0a743e9c)
+++d42322aa13214a5d099019ff0406dfb328960b98/drivers/pci/hotplug/sgi_hotplug.c(mode:100644 sha1:323041fd41dc19c20ce88e4ae370272cbe09ea97)(quoted diff) +3
-0+static int sn_hp_slot_private_alloc(struct hotplug_slot *bss_hotplug_slot, + struct pci_bus *pci_bus, int device) +{
Review
Quoted 10 lines
(quoted diff) +10
-0+ struct pcibus_info *pcibus_info; + struct slot *slot; + + pcibus_info = SN_PCIBUS_BUSSOFT_INFO(pci_bus); + + bss_hotplug_slot->private = kcalloc(1, sizeof(struct slot), + GFP_KERNEL); + if (!bss_hotplug_slot->private) + return -ENOMEM; + slot = (struct slot *)bss_hotplug_slot->private;
Please remove this cast. Since private is a void* there is never a cast needed to or from this.
Review
Quoted 1 line
(quoted diff) +1
-0+ bss_hotplug_slot->name = kmalloc(33, GFP_KERNEL);
I would feel better if you would use a #define here like the other drivers do. They normally name is SLOT_NAME_SIZE or something similar.
Review
Quoted 8 lines
(quoted diff) +8
-0+ sprintf(bss_hotplug_slot->name, "module_%c%c%c%c%.2d_b_%d_s_%d", + '0'+RACK_GET_CLASS(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_GROUP(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_NUM(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + MODULE_GET_BTCHAR(pcibus_info->pbi_moduleid), + MODULE_GET_BPOS(pcibus_info->pbi_moduleid), + ((int)pcibus_info->pbi_buscommon.bs_persist_busnum) & 0xf, + device + 1);
*eek* Sorry, but that looks really ugly. Wouldn't it be enough do name it like the device that would be in slot, something like DOMAIN_BUS_SLOT and maybe one extra number?
Quoted 1 line
+static struct
hotplug_slot*sn_hp_destroy(void)
Please remove the space before the function name
Quoted 1 line
+static u8
sn_power_status_get(struct hotplug_slot *bss_hotplug_slot)
Why not do this few instructions in get_power_status() directly? Or at least mark it inline, you only use it once and it does not really consume much stack space.
Review
Quoted 2 lines
(quoted diff) +2
-0+{ + struct slot *slot = (struct slot *)bss_hotplug_slot->private;
Remove cast.
Review
Quoted 2 lines
(quoted diff) +2
-0+static void sn_slot_mark_enable(struct hotplug_slot *bss_hotplug_slot, + int device_num)
Add inline or inline it by hand.
Review
Quoted 2 lines
(quoted diff) +2
-0+{ + struct slot *slot = (struct slot *)bss_hotplug_slot->private;
Cast.
Review
Quoted 4 lines
(quoted diff) +4
-0+static void sn_slot_mark_disable(struct hotplug_slot *bss_hotplug_slot, + int device_num) +{ + struct slot *slot = (struct slot *)bss_hotplug_slot->private;
The same.
Review
Quoted 4 lines
(quoted diff) +4
-0+static int sn_slot_enable(struct hotplug_slot *bss_hotplug_slot, + int device_num) +{ + struct slot *slot = (struct slot *)bss_hotplug_slot->private;
Cast.
Review
Quoted 4 lines
(quoted diff) +4
-0+ if (rc == PCI_SLOT_ALREADY_UP) { + dev_dbg(slot->pci_bus->self, "is already active\n"); + return -EPERM; + }
IIRC most other drivers handle this case as success. Greg, your opinion?
Review
Quoted 11 lines
(quoted diff) +11
-0+ if (rc) { + dev_dbg(slot->pci_bus->self, + "insert failed with error %d sub-error %d\n", + rc, resp.resp_sub_errno); + return -EIO; + } + + sn_slot_mark_enable(bss_hotplug_slot, device_num); + + return 0; +}
Replacing the last function call with 4 lines of instructions won't hurt I think. IMHO sn_slot_mark_enable() should be killed. Maybe you add a comment here to tell what's going on if you want to make it clear.
Review
Quoted 4 lines
(quoted diff) +4
-0+static int sn_slot_disable(struct hotplug_slot *bss_hotplug_slot, + int device_num, int action) +{ + struct slot *slot = (struct slot *)bss_hotplug_slot->private;
Cast.
Review
Quoted 1 line
(quoted diff) +1
-0+ if (action == PCI_REQ_SLOT_ELIGIBLE && rc == PCI_SLOT_ALREADY_DOWN) {
I would feel better if you add extra braces around the tests, but I'm probably just a bit paranoid about this things. Greg?
Review
Quoted 3 lines
(quoted diff) +3
-0+ dev_dbg(slot->pci_bus->self, "Slot %s already inactive\n"); + return -ENODEV; + }
Again this might better be a success.
Review
Quoted 7 lines
(quoted diff) +7
-0+ if (action == PCI_REQ_SLOT_DISABLE && rc) { + dev_dbg(slot->pci_bus->self,"remove failed rc = %d\n", rc); + return rc; + } + + return rc; +}
One return should be enough for everyone.
Review
Quoted 3 lines
(quoted diff) +3
-0+static int enable_slot(struct hotplug_slot *bss_hotplug_slot) +{ + struct slot *slot = (struct slot *)bss_hotplug_slot->private;
Cast.
Review
Quoted 1 line
(quoted diff) +1
-0+ num_funcs = pci_scan_slot(slot->pci_bus, PCI_DEVFN(slot->device_num+1,
Add spaces before and after '+'. I don't feel good with this "+1" at all, this is some kind of strange.
Review
Quoted 1 line
(quoted diff) +1
-0+ PCI_FUNC(0)));
Greg, what do you think. Isn't just 0 better here?
Please wrap the line one level higher and keep the second argument as one, so add the break behind "slot->pci_bus,".
Review
Quoted 3 lines
(quoted diff) +3
-0+static int disable_slot(struct hotplug_slot *bss_hotplug_slot) +{ + struct slot *slot = (struct slot *)bss_hotplug_slot->private;
Cast.
Review
Quoted 7 lines
+static int
get_power_status(struct hotplug_slot *bss_hotplug_slot, u8 *value)+{(quoted diff) +5
-0+ down(&sn_hotplug_sem); + *value = sn_power_status_get(bss_hotplug_slot); + up(&sn_hotplug_sem); + return 0; +}
sn_power_status_get() should be easily be inlinable to here. This would allow to make the up()/down() protect even less instructions.
Quoted 7 lines
(quoted diff) +7
-0+static int sn_hotplug_slot_register(struct pci_bus *pci_bus) +{ + int device; + struct hotplug_slot *bss_hotplug_slot; + int rc = 0; + + /*
Review
Quoted 10 lines
(quoted diff) +10
-0+ * Currently only four devices are supported, + * in the future there maybe more -- up to 32. + */ + + for (device = 0; device < SN_MAX_HP_SLOTS ; device++) { + if (sn_pci_slot_valid(pci_bus, device) != 1) + continue; + + bss_hotplug_slot = kcalloc(1,sizeof(struct hotplug_slot), + GFP_KERNEL);
Use "sizeof(*bss_hotplug_slot)" and add a space after the ','.
Review
Quoted 3 lines
(quoted diff) +3
-0+ bss_hotplug_slot->info = + kcalloc(1,sizeof(struct hotplug_slot_info), + GFP_KERNEL);
Same here.
Review
Quoted 9 lines
(quoted diff) +8
-0+static int sn_pci_hotplug_init(void) +{ + if (!rc) + registered = 1; + else { + registered = 0; + break; + }
Please use {} for the if also to make it look saner.
Eike
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
Eike,
Thanks for the comments -- I'm going through them now.
Quoted 16 lines · Rolf Eike Beer
Review
Quoted 9 lines
(quoted diff) +8
-0+ sprintf(bss_hotplug_slot->name, "module_%c%c%c%c%.2d_b_%d_s_%d", + '0'+RACK_GET_CLASS(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_GROUP(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_NUM(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + MODULE_GET_BTCHAR(pcibus_info->pbi_moduleid), + MODULE_GET_BPOS(pcibus_info->pbi_moduleid), + ((int)pcibus_info->pbi_buscommon.bs_persist_busnum) & 0xf, + device + 1);*eek* Sorry, but that looks really ugly. Wouldn't it be enough do name it like the device that would be in slot, something like
DOMAIN_BUS_SLOTand maybe one extra number?
Unfortunately, yes, the longer name is prefered in this case. The rack class ID, etc., are need to identify locations on the Altix platform.
Quoted 4 lines
Review
Quoted 1 line
(quoted diff) +1
-0+ if (action = PCI_REQ_SLOT_ELIGIBLE && rc = PCI_SLOT_ALREADY_DOWN) {I would feel better if you add extra braces around the tests, but I'm probably just a bit paranoid about this things. Greg?
I was asked to remove the braces in an earlier version. I too would prefer to have them but am going with current comments on the code.
P.
Quoted 16 lines · Prarit Bhargava
Quoted 14 lines · Rolf Eike Beer
Review
Quoted 9 lines
(quoted diff) +8
-0+ sprintf(bss_hotplug_slot->name, "module_%c%c%c%c%.2d_b_%d_s_%d", + '0'+RACK_GET_CLASS(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_GROUP(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_NUM(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + MODULE_GET_BTCHAR(pcibus_info->pbi_moduleid), + MODULE_GET_BPOS(pcibus_info->pbi_moduleid), + ((int)pcibus_info->pbi_buscommon.bs_persist_busnum) & 0xf, + device + 1);*eek* Sorry, but that looks really ugly. Wouldn't it be enough do name it like the device that would be in slot, something like
DOMAIN_BUS_SLOTand maybe one extra number?
Quoted 3 lines
Unfortunately, yes, the longer name is prefered in this case. The rack class
ID, etc., are need to identify locations on the Altix platform.
No, use additional sysfs files if you need to export this kind of information to the user somehow. I think IBM already does this in their drivers in a sane manner (so you might want to copy them, as they have already modified the userspace code to handle those sysfs files...)
Quoted 7 lines
Quoted 4 lines
Review
Quoted 1 line
(quoted diff) +1
-0+ if (action = PCI_REQ_SLOT_ELIGIBLE && rc = PCI_SLOT_ALREADY_DOWN) {I would feel better if you add extra braces around the tests, but I'm probably just a bit paranoid about this things. Greg?
I was asked to remove the braces in an earlier version. I too would prefer to have them but am going with current comments on the code.
Who asked you to remove the extra braces? It was not me. If you like them, please add them, I feel more comfortable with them too.
thanks,
greg k-h
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 23 lines · Greg KH
Quoted 16 lines · Prarit Bhargava
Quoted 5 lines · Rolf Eike Beer
Quoted 5 lines
(quoted diff) +5
-0+ sprintf(bss_hotplug_slot->name, "module_%c%c%c%c%.2d_b_%d_s_%d", + '0'+RACK_GET_CLASS(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_GROUP(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_NUM(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + MODULE_GET_BTCHAR(pcibus_info->pbi_moduleid),Quoted 8 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+ MODULE_GET_BPOS(pcibus_info->pbi_moduleid), + ((int)pcibus_info->pbi_buscommon.bs_persist_busnum) & 0xf, + device + 1);*eek* Sorry, but that looks really ugly. Wouldn't it be enough do name it like the device that would be in slot, something like
DOMAIN_BUS_SLOTand maybe one extra number?Quoted 3 lines
Unfortunately, yes, the longer name is prefered in this case. The rack class
ID, etc., are need to identify locations on the Altix platform.
Quoted 6 lines
No, use additional sysfs files if you need to export this kind of information to the user somehow. I think IBM already does this in their drivers in a sane manner (so you might want to copy them, as they have already modified the userspace code to handle those sysfs files...)
Okay -- I'll go with Eike's suggestion above. D_B_S + 1 number.
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
Review
Quoted 1 line
(quoted diff) +1
-0+ if (action = PCI_REQ_SLOT_ELIGIBLE && rc = PCI_SLOT_ALREADY_DOWN) {I would feel better if you add extra braces around the tests, but I'm probably just a bit paranoid about this things. Greg?
Quoted 7 lines
Quoted 2 lines
I was asked to remove the braces in an earlier version. I too would prefer to have them but am going with current comments on the code.
Who asked you to remove the extra braces? It was not me. If you like them, please add them, I feel more comfortable with them too.
Hrmm ... I can't remember. But someone did definately tell me that they were not required.
P.
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 11 lines · Rolf Eike Beer
Review
Quoted 7 lines
(quoted diff) +4
-0+ if (rc = PCI_SLOT_ALREADY_UP) { + dev_dbg(slot->pci_bus->self, "is already active\n"); + return -EPERM; + }IIRC most other drivers handle this case as success. Greg, your opinion?
Other drivers handle this by returning 1. I changed the return value to 1.
Quoted 9 lines
Review
Quoted 5 lines
(quoted diff) +3
-0+ dev_dbg(slot->pci_bus->self, "Slot %s already inactive\n"); + return -ENODEV; + }Again this might better be a success.
Other drivers handle this by returning 1. I changed the return value to 1.
Quoted 6 lines
Review
Quoted 1 line
(quoted diff) +1
-0+ num_funcs = pci_scan_slot(slot->pci_bus, PCI_DEVFN(slot->device_num+1,Add spaces before and after '+'. I don't feel good with this "+1" at all, this is some kind of strange.
Aside from being used in the above calculation, the slot->device_num is also a bitmask. The 0th bit corresponds to the first device that is active/inactive on a slot.
In the calculation above, the slot we're scanning is PCI_DEVFN(1,0) -- the first device in the slot.
I could flip things around and use slot->device_num to a one-based calculation, but that leads to more "device_num - 1" statements than "device_num" statements in the code.
P.
Quoted 20 lines · Prarit Bhargava
Quoted 19 lines · Greg KH
Quoted 17 lines · Prarit Bhargava
Quoted 4 lines · Rolf Eike Beer
Quoted 4 lines
(quoted diff) +4
-0+ sprintf(bss_hotplug_slot->name, "module_%c%c%c%c%.2d_b_%d_s_%d", + '0'+RACK_GET_CLASS(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_GROUP(MODULE_GET_RACK(pcibus_info->pbi_moduleid)), + '0'+RACK_GET_NUM(MODULE_GET_RACK(pcibus_info->pbi_moduleid)),Quoted 8 lines
Review
Quoted 4 lines
(quoted diff) +4
-0+ MODULE_GET_BTCHAR(pcibus_info->pbi_moduleid), + MODULE_GET_BPOS(pcibus_info->pbi_moduleid), + ((int)pcibus_info->pbi_buscommon.bs_persist_busnum) & 0xf, + device + 1);*eek* Sorry, but that looks really ugly. Wouldn't it be enough do name it like the device that would be in slot, something like
DOMAIN_BUS_SLOTand maybe one extra number?Unfortunately, yes, the longer name is prefered in this case. The rack class
ID, etc., are need to identify locations on the Altix platform.
Quoted 6 lines
Quoted 4 lines
No, use additional sysfs files if you need to export this kind of information to the user somehow. I think IBM already does this in their drivers in a sane manner (so you might want to copy them, as they have already modified the userspace code to handle those sysfs files...)
Okay -- I'll go with Eike's suggestion above.
D_B_S+ 1 number.
D_B_S alone should be a unique identifier, so please don't use anything more if you don't need it to make the slot name unique.
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
Review
Quoted 1 line
(quoted diff) +1
-0+ if (action = PCI_REQ_SLOT_ELIGIBLE && rc = PCI_SLOT_ALREADY_DOWN) {I would feel better if you add extra braces around the tests, but I'm probably just a bit paranoid about this things. Greg?
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
I was asked to remove the braces in an earlier version. I too would prefer to have them but am going with current comments on the code.
Who asked you to remove the extra braces? It was not me. If you like them, please add them, I feel more comfortable with them too.
Hrmm ... I can't remember. But someone did definately tell me that they were not required.
They are not required (at least not for the compiler). But brain v0.5 can parse it much better if they are in there ;)
Eike
Quoted 11 lines · Prarit Bhargava
Quoted 4 lines · Rolf Eike Beer
Review
Quoted 1 line
(quoted diff) +1
-0+ num_funcs = pci_scan_slot(slot->pci_bus, PCI_DEVFN(slot->device_num+1,Add spaces before and after '+'. I don't feel good with this "+1" at all, this is some kind of strange.
Aside from being used in the above calculation, the
slot->device_numis also a bitmask. The 0th bit corresponds to the first device that is active/inactive on a slot.
Quoted 6 lines
In the calculation above, the slot we're scanning is
PCI_DEVFN(1,0)-- the first device in the slot.I could flip things around and use
slot->device_numto a one-based calculation, but that leads to more "device_num- 1" statements than "device_num" statements in the code.
So "+1" is the better way. Maybe you should write a big comment at the first "device_num+1" what magic is going on there. It's likely that someone else will come later and will not understand what's going on here, too.
Eike
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 4 lines · Rolf Eike Beer
Quoted 2 lines · Prarit Bhargava
Greg KH wrote:
Quoted 9 lines
Quoted 4 lines
Quoted 2 lines · Prarit Bhargava
Okay -- I'll go with Eike's suggestion above.
D_B_S+ 1 number.
D_B_Salone should be a unique identifier, so please don't use anything more if you don't need it to make the slot name unique.
On Altix, since the system can be cabled in a number of ways a pointer to the rack ID is definately useful and helps to uniquely identify the slot. I'm going to use a 2 digit rack id, followed by the DDDD:BB:SS value.
P.
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 10 lines · Prarit Bhargava
Quoted 4 lines
D_B_Salone should be a unique identifier, so please don't use anything more if you don't need it to make the slot name unique.On Altix, since the system can be cabled in a number of ways a pointer to the rack ID is definately useful and helps to uniquely identify the slot. I'm going to use a 2 digit rack id, followed by the DDDD:BB:SS value.
Err .. uh .. unless there are any strenous objections of course ;
Quoted 2 lines
P.
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 13 lines · Prarit Bhargava
Quoted 6 lines · Prarit Bhargava
Quoted 3 lines
D_B_Salone should be a unique identifier, so please don't use anything more if you don't need it to make the slot name unique.On Altix, since the system can be cabled in a number of ways a pointer to the rack ID is definately useful and helps to uniquely identify the slot. I'm going to use a 2 digit rack id, followed by the DDDD:BB:SS value.
Err .. uh .. unless there are any strenous objections of course ;
Quoted 2 lines
P.
New patch after comments from Eike and Greg.
P.
This patch is the SGI hotplug driver and additional changes required for the driver. These modifications include changes to the SN io_init.c code for memory management, the inclusion of new SAL calls to enable and disable PCI slots, and a hotplug-style driver.
---
commit 9c0467fa45dcc532938e1dd9afa59c5db82e3701 tree ae5cb3f0a1424b31a67af5e0bf481943c6e8988d parent 82fdc1f1df71b98ce76cd6a520c49c25a2dee34f author Prarit Bhargava <[email protected]> 1810650612 -0400 committer Prarit Bhargava <[email protected]> 1810650612 -0400
arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:0e591f85f73457775645ea36688fb0de3c1d5064) renamed from arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:ae6628cd41f8cc2f94643fce2c4fe21bdd85d2d9) +91 -48
@@ -23,6 +23,14 @@ nasid_t master_nasid = INVALID_NASID; /* Partition Master */ +static struct list_head sn_sysdata_list; + +/* sysdata list struct */ +struct sysdata_el { + struct list_head entry; + void *sysdata; +}; + struct slab_info { struct hubdev_info hubdev; };
@@ -137,23 +145,6 @@ } /* - * sn_alloc_pci_sysdata() - This routine allocates a pci controller - * which is expected as the pci_dev and pci_bus sysdata by the Linux - * PCI infrastructure. - */ -static inline struct pci_controller *sn_alloc_pci_sysdata(void) -{ - struct pci_controller *pci_sysdata; - - pci_sysdata = kmalloc(sizeof(*pci_sysdata), GFP_KERNEL); - if (!pci_sysdata) - BUG(); - - memset(pci_sysdata, 0, sizeof(*pci_sysdata)); - return pci_sysdata; -} - -/* * sn_fixup_ionodes() - This routine initializes the HUB data strcuture for * each node in the system. */
@@ -220,6 +211,15 @@ } +void sn_pci_unfixup_slot(struct pci_dev *dev) +{ + struct pci_dev *host_pci_dev = SN_PCIDEV_INFO(dev)->host_pci_dev; + + sn_irq_unfixup(dev); + pci_dev_put(host_pci_dev); + pci_dev_put(dev); +} + /* * sn_pci_fixup_slot() - This routine sets up a slot's resources * consistent with the Linux PCI abstraction layer. Resources acquired
@@ -238,6 +238,7 @@ unsigned long size; unsigned int bus_no, devfn; + pci_dev_get(dev); /* for the sysdata pointer */ dev->sysdata = kmalloc(sizeof(struct pcidev_info), GFP_KERNEL); if (SN_PCIDEV_INFO(dev) <= 0) BUG(); /* Cannot afford to run out of memory */
@@ -276,7 +277,8 @@ dev->resource[idx].parent = &iomem_resource; } - /* Using the PROMs values for the PCI host bus, get the Linux + /* + * Using the PROMs values for the PCI host bus, get the Linux * PCI host_pci_dev struct and set up host bus linkages */
@@ -292,11 +294,10 @@ bs = SN_PCIBUS_BUSSOFT(dev->bus); SN_PCIDEV_INFO(dev)->pdi_pcibus_info = bs; - if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) { + if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) SN_PCIDEV_BUSPROVIDER(dev) = sn_pci_provider[bs->bs_asic_type]; - } else { + else SN_PCIDEV_BUSPROVIDER(dev) = &sn_pci_default_provider; - } /* Only set up IRQ stuff if this device has a host bus context */ if (bs && sn_irq_info->irq_irq) {
@@ -313,55 +314,57 @@ * sn_pci_controller_fixup() - This routine sets up a bus's resources * consistent with the Linux PCI abstraction layer. */ -static void sn_pci_controller_fixup(int segment, int busnum) +void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus) { int status = 0; int nasid, cnode; - struct pci_bus *bus; struct pci_controller *controller; struct pcibus_bussoft *prom_bussoft_ptr; struct hubdev_info *hubdev_info; void *provider_soft; struct sn_pcibus_provider *provider; - status = - sal_get_pcibus_info((u64) segment, (u64) busnum, - (u64) ia64_tpa(&prom_bussoft_ptr)); - if (status > 0) { - return; /* bus # does not exist */ - } - + status = sal_get_pcibus_info((u64) segment, (u64) busnum, + (u64) ia64_tpa(&prom_bussoft_ptr)); + if (status > 0) + return; /*bus # does not exist */ prom_bussoft_ptr = __va(prom_bussoft_ptr); - controller = sn_alloc_pci_sysdata(); - /* controller non-zero is BUG'd in sn_alloc_pci_sysdata */ - bus = pci_scan_bus(busnum, &pci_root_ops, controller); + controller = kcalloc(1,sizeof(struct pci_controller), GFP_KERNEL); + if (!controller) + BUG(); + if (bus == NULL) { - return; /* error, or bus already scanned */ + bus = pci_scan_bus(busnum, &pci_root_ops, controller); + if (bus == NULL) + return; /* error, or bus already scanned */ + bus->sysdata = NULL; } + if (bus->sysdata) + goto error_return; /* sysdata already alloc'd */ + /* * Per-provider fixup. Copies the contents from prom to local * area and links SN_PCIBUS_BUSSOFT(). */ - if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) { + if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) return; /* unsupported asic type */ - } + + if (prom_bussoft_ptr->bs_asic_type == PCIIO_ASIC_TYPE_PPB) + goto error_return; /* no further fixup necessary */ provider = sn_pci_provider[prom_bussoft_ptr->bs_asic_type]; - if (provider == NULL) { + if (provider == NULL) return; /* no provider registerd for this asic */ - } provider_soft = NULL; - if (provider->bus_fixup) { + if (provider->bus_fixup) provider_soft = (*provider->bus_fixup) (prom_bussoft_ptr); - } - if (provider_soft == NULL) { + if (provider_soft == NULL) return; /* fixup failed or not applicable */ - } /* * Generic bus fixup goes here. Don't reference prom_bussoft_ptr
@@ -370,12 +373,47 @@ bus->sysdata = controller; PCI_CONTROLLER(bus)->platform_data = provider_soft; - nasid = NASID_GET(SN_PCIBUS_BUSSOFT(bus)->bs_base); cnode = nasid_to_cnodeid(nasid); hubdev_info = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); SN_PCIBUS_BUSSOFT(bus)->bs_xwidget_info = &(hubdev_info->hdi_xwidget_info[SN_PCIBUS_BUSSOFT(bus)->bs_xid]); + + return; + +error_return: + + kfree(controller); + return; +} + +void sn_bus_store_sysdata(struct pci_dev *dev) +{ + struct sysdata_el *element; + + element = kcalloc(1, sizeof(struct sysdata_el), GFP_KERNEL); + if (!element) { + dev_dbg(dev, "%s: out of memory!\n", __FUNCTION__); + return; + } + element->sysdata = dev->sysdata; + list_add(&element->entry, &sn_sysdata_list); +} + +void sn_bus_free_sysdata(void) +{ + struct sysdata_el *element; + struct list_head *list; + +sn_sysdata_free_start: + list_for_each(list, &sn_sysdata_list) { + element = list_entry(list, struct sysdata_el, entry); + list_del(&element->entry); + kfree(element->sysdata); + kfree(element); + goto sn_sysdata_free_start; + } + return; } /*
@@ -413,15 +451,16 @@ ia64_max_iommu_merge_mask = ~PAGE_MASK; sn_fixup_ionodes(); sn_irq_lh_init(); + INIT_LIST_HEAD(&sn_sysdata_list); sn_init_cpei_timer(); #ifdef CONFIG_PROC_FS register_sn_procfs(); #endif - for (i = 0; i < PCI_BUSES_TO_SCAN; i++) { - sn_pci_controller_fixup(0, i); - } + /* busses are not known yet ... */ + for (i = 0; i < PCI_BUSES_TO_SCAN; i++) + sn_pci_controller_fixup(0, i, NULL); /* * Generic Linux PCI Layer has created the pci_bus and pci_dev
@@ -430,9 +469,8 @@ */ while ((pci_dev = - pci_find_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) { + pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) sn_pci_fixup_slot(pci_dev); - } sn_ioif_inited = 1; /* sn I/O infrastructure now initialized */
@@ -474,3 +512,8 @@ } subsys_initcall(sn_pci_init); +EXPORT_SYMBOL_GPL(sn_pci_fixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_unfixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_controller_fixup); +EXPORT_SYMBOL_GPL(sn_bus_store_sysdata); +EXPORT_SYMBOL_GPL(sn_bus_free_sysdata);
arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ec5e7a01c5b55f328e0dff8a45b012bc1024e84e) renamed from arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ee4ac68d745f2c91e39d10e2f82cf624cf655edd) +3 -3
@@ -284,7 +284,6 @@ int cpu = nasid_slice_to_cpuid(nasid, slice); pci_dev_get(pci_dev); - sn_irq_info->irq_cpuid = cpu; sn_irq_info->irq_pciioinfo = SN_PCIDEV_INFO(pci_dev);
@@ -305,15 +304,16 @@ return; sn_irq_info = SN_PCIDEV_INFO(pci_dev)->pdi_sn_irq_info; - if (!sn_irq_info || !sn_irq_info->irq_irq) + if (!sn_irq_info || !sn_irq_info->irq_irq) { + kfree(sn_irq_info); return; + } unregister_intr_pda(sn_irq_info); spin_lock(&sn_irq_info_lock); list_del_rcu(&sn_irq_info->list); spin_unlock(&sn_irq_info_lock); call_rcu(&sn_irq_info->rcu, sn_irq_info_free); - pci_dev_put(pci_dev); }
arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9813da56d3113c41fb1b413cfa4c7963cda316cb) renamed from arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9bc4de4a3ec0778aa4724e75457bd253a6b1e45e) +37 -0
@@ -18,6 +18,40 @@ #include "xtalk/xwidgetdev.h" #include "xtalk/hubdev.h" +int +sal_pcibr_slot_enable(struct pcibus_info *soft, int device, void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_ENABLE, (u64) busnum, + (u64) device, (u64) resp, 0, 0, 0, 0); + + return (int)ret_stuff.v0; +} + +int +sal_pcibr_slot_disable(struct pcibus_info *soft, int device, int action, + void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_DISABLE, + (u64) busnum, (u64) device, (u64) action, + (u64) resp, 0, 0, 0); + + return (int)ret_stuff.v0; +} + static int sal_pcibr_error_interrupt(struct pcibus_info *soft) { struct ia64_sal_retval ret_stuff;
@@ -187,3 +221,6 @@ return 0; } + +EXPORT_SYMBOL_GPL(sal_pcibr_slot_enable); +EXPORT_SYMBOL_GPL(sal_pcibr_slot_disable);
drivers/pci/hotplug/Kconfig (mode:100644 sha1:9c4a39ee89b57b5065f575b86b2798ed31dc2fc6) renamed from drivers/pci/hotplug/Kconfig (mode:100644 sha1:1a4d4ca2a4dc386d3f63189ecc9cf30f806b947e) +3 -2
@@ -187,9 +187,10 @@ config HOTPLUG_PCI_SGI tristate "SGI PCI Hotplug Support" - depends on HOTPLUG_PCI && IA64_SGI_SN2 + depends on HOTPLUG_PCI && (IA64_SGI_SN2 || IA64_GENERIC) help - Say Y here if you have an SGI IA64 Altix system. + Say Y here if you want to use the SGI Altix Hotplug + Driver for PCI devices. When in doubt, say N.
drivers/pci/hotplug/Makefile (mode:100644 sha1:24ed904514c5043095dfe484877dd38cdfa0f3b6) renamed from drivers/pci/hotplug/Makefile (mode:100644 sha1:93c120ddbd395f5ddeddcc3cc5d52bd31a47dcc4) +1 -0
@@ -14,6 +14,7 @@ obj-$(CONFIG_HOTPLUG_PCI_SHPC) += shpchp.o obj-$(CONFIG_HOTPLUG_PCI_RPA) += rpaphp.o obj-$(CONFIG_HOTPLUG_PCI_RPA_DLPAR) += rpadlpar_io.o +obj-$(CONFIG_HOTPLUG_PCI_SGI) += sgi_hotplug.o pci_hotplug-objs := pci_hotplug_core.o
drivers/pci/hotplug/sgi_hotplug.c (mode:100644 sha1:87035eae97bd8c2b867efd0e2f1504dd9d023282) renamed from dev/null (tree:f8014bc6a8330d22e2dbb0cac34d87e8a5352278) +579 -0
579 lines not on this page · raw message
include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:c1bf22505ff339ea7a5255ec9a8a60031397e9a8) renamed from include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:0abf539cb0414d752fcf14d9318bf1fa00278d32) +4 -0
@@ -151,4 +151,8 @@ extern int pcibr_ate_alloc(struct pcibus_info *, int); extern void pcibr_ate_free(struct pcibus_info *, int); extern void ate_write(struct pcibus_info *, int, int, uint64_t); +extern int sal_pcibr_slot_enable(struct pcibus_info *soft, int device, + void *resp); +extern int sal_pcibr_slot_disable(struct pcibus_info *soft, int device, + int action, void *resp); #endif
include/asm-ia64/sn/pcidev.h (mode:100644 sha1:49711d00ad04f77d90467daf33bfd13d549e2109) renamed from include/asm-ia64/sn/pcidev.h (mode:100644 sha1:9610fcc635450e57bf9be693fb8fa5e22eb60be1) +6 -0
@@ -23,6 +23,8 @@ #define SN_PCIBUS_BUSSOFT(pci_bus) \ ((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) +#define SN_PCIBUS_BUSSOFT_INFO(pci_bus) \ + (struct pcibus_info *)((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) /* * Given a struct pci_dev, return the sn pcibus_bussoft struct. Note * that this is not equivalent to SN_PCIBUS_BUSSOFT(pci_dev->bus) due
@@ -56,6 +58,10 @@ extern void sn_irq_fixup(struct pci_dev *pci_dev, struct sn_irq_info *sn_irq_info); extern void sn_irq_unfixup(struct pci_dev *pci_dev); +extern void sn_pci_controller_fixup(int segment, int busnum, + struct pci_bus *bus); +extern void sn_bus_store_sysdata(struct pci_dev *dev); +extern void sn_bus_free_sysdata(void); extern void sn_pci_fixup_slot(struct pci_dev *dev); extern void sn_pci_unfixup_slot(struct pci_dev *dev); extern void sn_irq_lh_init(void);
Quoted 3 lines · Prarit Bhargava
On Altix, since the system can be cabled in a number of ways a pointer to the rack ID is definately useful and helps to uniquely identify the slot. I'm going to use a 2 digit rack id, followed by the DDDD:BB:SS value.
Again, what's wrong with the sysfs file for the rack id? Don't clutter up the slot name if you don't have to.
DDDD:BB:SS should be unique on it's own, right?
thanks,
greg k-h
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 7 lines · Greg KH
Quoted 3 lines · Prarit Bhargava
On Altix, since the system can be cabled in a number of ways a pointer to the rack ID is definately useful and helps to uniquely identify the slot. I'm going to use a 2 digit rack id, followed by the DDDD:BB:SS value.
Quoted 9 lines
Again, what's wrong with the sysfs file for the rack id? Don't clutter up the slot name if you don't have to.
DDDD:BB:SS should be unique on it's own, right?
thanks,
greg k-h
Okay -- will do.
P.
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 4 lines · Greg KH
On Wed, May 18, 2005 at 09:57:13AM -0400, Prarit Bhargava wrote:
DDDD:BB:SS should be unique on it's own, right?
Greg,
I'm looking at the latest 2.6.12-rc4 sources and the IBM php code but don't see any references to additional sysfs files. Some of the other drivers do implement extra sysfs files. Can you give me a pointer to (as you said) a sane implementation?
Thanks,
P.
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 9 lines · Prarit Bhargava
Quoted 7 lines · Prarit Bhargava
Quoted 5 lines · Prarit Bhargava
Quoted 4 lines
D_B_Salone should be a unique identifier, so please don't use anything more if you don't need it to make the slot name unique.
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
On Altix, since the system can be cabled in a number of ways a pointer to the rack ID is definately useful and helps to uniquely identify the slot. I'm going to use a 2 digit rack id, followed by the DDDD:BB:SS value.
NAK this. I'm changing the name of the sysfs file.
Thanks,
P.
Quoted 10 lines · Prarit Bhargava
Quoted 4 lines · Greg KH
On Wed, May 18, 2005 at 09:57:13AM -0400, Prarit Bhargava wrote:
DDDD:BB:SS should be unique on it's own, right?
Greg,
I'm looking at the latest 2.6.12-rc4 sources and the IBM php code but don't see any references to additional sysfs files.
IBM has 2 drivers in the tree, you were probably looking at the other one :)
Look at the rpaphp_slot.c:rpaphp_sysfs_add_attr_location() function.
thanks,
greg k-h
Quoted 5 lines · Prarit Bhargava
Quoted 4 lines · Prarit Bhargava
Quoted 3 lines · Prarit Bhargava
Quoted 2 lines
D_B_Salone should be a unique identifier, so please don't use anything more if you don't need it to make the slot name unique.
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
On Altix, since the system can be cabled in a number of ways a pointer to the rack ID is definately useful and helps to uniquely identify the slot. I'm going to use a 2 digit rack id, followed by the DDDD:BB:SS value.
Err .. uh .. unless there are any strenous objections of course ;
New patch after comments from Eike and Greg.
Something I missed in the previous patch:
+static int sn_hp_slot_private_alloc(struct hotplug_slot *bss_hotplug_slot,
+ struct pci_bus *pci_bus, int device)
+{
+ struct pcibus_info *pcibus_info;
+ struct slot *slot;
+
+ pcibus_info = SN_PCIBUS_BUSSOFT_INFO(pci_bus);
+
+ bss_hotplug_slot->private = kcalloc(1, sizeof(struct slot),
+ GFP_KERNEL);
+ if (!bss_hotplug_slot->private)
+ return -ENOMEM;
+ slot = bss_hotplug_slot->private;
+
Better do it this way:
slot = kcalloc(1, sizeof(*slot), GFP_KERNEL);bss_hotplug_slot->private = slot;
This way the size of the memory area is always the correct one (which is the preferred way of operating *g*).
I think it's really time for something like kmalloc0() or whatever to stop this type of kcalloc() abuse.
Eike
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerNew patch after Eike's and Greg's comments regarding the physical location (path) sysfs file, and additional minor cleanups.
This patch is the SGI hotplug driver and additional changes required for the driver. These modifications include changes to the SN io_init.c code for memory management, the inclusion of new SAL calls to enable and disable PCI slots, and a hotplug-style driver.
---
commit 7a9a966fc8d33d4f8201f54306e0d994b33240ec
tree c1a822866428a99c849b2a12fd2607ad38167271
parent b81b1aaaee60af81a593181bf11049fa43a8090d
author root <[email protected]> 1810731575 -0400 committer root <[email protected]> 1810731575 -0400
arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:0e591f85f73457775645ea36688fb0de3c1d5064) renamed from arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:ae6628cd41f8cc2f94643fce2c4fe21bdd85d2d9) +91 -48
@@ -23,6 +23,14 @@ nasid_t master_nasid = INVALID_NASID; /* Partition Master */ +static struct list_head sn_sysdata_list; + +/* sysdata list struct */ +struct sysdata_el { + struct list_head entry; + void *sysdata; +}; + struct slab_info { struct hubdev_info hubdev; };
@@ -137,23 +145,6 @@ } /* - * sn_alloc_pci_sysdata() - This routine allocates a pci controller - * which is expected as the pci_dev and pci_bus sysdata by the Linux - * PCI infrastructure. - */ -static inline struct pci_controller *sn_alloc_pci_sysdata(void) -{ - struct pci_controller *pci_sysdata; - - pci_sysdata = kmalloc(sizeof(*pci_sysdata), GFP_KERNEL); - if (!pci_sysdata) - BUG(); - - memset(pci_sysdata, 0, sizeof(*pci_sysdata)); - return pci_sysdata; -} - -/* * sn_fixup_ionodes() - This routine initializes the HUB data strcuture for * each node in the system. */
@@ -220,6 +211,15 @@ } +void sn_pci_unfixup_slot(struct pci_dev *dev) +{ + struct pci_dev *host_pci_dev = SN_PCIDEV_INFO(dev)->host_pci_dev; + + sn_irq_unfixup(dev); + pci_dev_put(host_pci_dev); + pci_dev_put(dev); +} + /* * sn_pci_fixup_slot() - This routine sets up a slot's resources * consistent with the Linux PCI abstraction layer. Resources acquired
@@ -238,6 +238,7 @@ unsigned long size; unsigned int bus_no, devfn; + pci_dev_get(dev); /* for the sysdata pointer */ dev->sysdata = kmalloc(sizeof(struct pcidev_info), GFP_KERNEL); if (SN_PCIDEV_INFO(dev) <= 0) BUG(); /* Cannot afford to run out of memory */
@@ -276,7 +277,8 @@ dev->resource[idx].parent = &iomem_resource; } - /* Using the PROMs values for the PCI host bus, get the Linux + /* + * Using the PROMs values for the PCI host bus, get the Linux * PCI host_pci_dev struct and set up host bus linkages */
@@ -292,11 +294,10 @@ bs = SN_PCIBUS_BUSSOFT(dev->bus); SN_PCIDEV_INFO(dev)->pdi_pcibus_info = bs; - if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) { + if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) SN_PCIDEV_BUSPROVIDER(dev) = sn_pci_provider[bs->bs_asic_type]; - } else { + else SN_PCIDEV_BUSPROVIDER(dev) = &sn_pci_default_provider; - } /* Only set up IRQ stuff if this device has a host bus context */ if (bs && sn_irq_info->irq_irq) {
@@ -313,55 +314,57 @@ * sn_pci_controller_fixup() - This routine sets up a bus's resources * consistent with the Linux PCI abstraction layer. */ -static void sn_pci_controller_fixup(int segment, int busnum) +void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus) { int status = 0; int nasid, cnode; - struct pci_bus *bus; struct pci_controller *controller; struct pcibus_bussoft *prom_bussoft_ptr; struct hubdev_info *hubdev_info; void *provider_soft; struct sn_pcibus_provider *provider; - status = - sal_get_pcibus_info((u64) segment, (u64) busnum, - (u64) ia64_tpa(&prom_bussoft_ptr)); - if (status > 0) { - return; /* bus # does not exist */ - } - + status = sal_get_pcibus_info((u64) segment, (u64) busnum, + (u64) ia64_tpa(&prom_bussoft_ptr)); + if (status > 0) + return; /*bus # does not exist */ prom_bussoft_ptr = __va(prom_bussoft_ptr); - controller = sn_alloc_pci_sysdata(); - /* controller non-zero is BUG'd in sn_alloc_pci_sysdata */ - bus = pci_scan_bus(busnum, &pci_root_ops, controller); + controller = kcalloc(1,sizeof(struct pci_controller), GFP_KERNEL); + if (!controller) + BUG(); + if (bus == NULL) { - return; /* error, or bus already scanned */ + bus = pci_scan_bus(busnum, &pci_root_ops, controller); + if (bus == NULL) + return; /* error, or bus already scanned */ + bus->sysdata = NULL; } + if (bus->sysdata) + goto error_return; /* sysdata already alloc'd */ + /* * Per-provider fixup. Copies the contents from prom to local * area and links SN_PCIBUS_BUSSOFT(). */ - if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) { + if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) return; /* unsupported asic type */ - } + + if (prom_bussoft_ptr->bs_asic_type == PCIIO_ASIC_TYPE_PPB) + goto error_return; /* no further fixup necessary */ provider = sn_pci_provider[prom_bussoft_ptr->bs_asic_type]; - if (provider == NULL) { + if (provider == NULL) return; /* no provider registerd for this asic */ - } provider_soft = NULL; - if (provider->bus_fixup) { + if (provider->bus_fixup) provider_soft = (*provider->bus_fixup) (prom_bussoft_ptr); - } - if (provider_soft == NULL) { + if (provider_soft == NULL) return; /* fixup failed or not applicable */ - } /* * Generic bus fixup goes here. Don't reference prom_bussoft_ptr
@@ -370,12 +373,47 @@ bus->sysdata = controller; PCI_CONTROLLER(bus)->platform_data = provider_soft; - nasid = NASID_GET(SN_PCIBUS_BUSSOFT(bus)->bs_base); cnode = nasid_to_cnodeid(nasid); hubdev_info = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); SN_PCIBUS_BUSSOFT(bus)->bs_xwidget_info = &(hubdev_info->hdi_xwidget_info[SN_PCIBUS_BUSSOFT(bus)->bs_xid]); + + return; + +error_return: + + kfree(controller); + return; +} + +void sn_bus_store_sysdata(struct pci_dev *dev) +{ + struct sysdata_el *element; + + element = kcalloc(1, sizeof(struct sysdata_el), GFP_KERNEL); + if (!element) { + dev_dbg(dev, "%s: out of memory!\n", __FUNCTION__); + return; + } + element->sysdata = dev->sysdata; + list_add(&element->entry, &sn_sysdata_list); +} + +void sn_bus_free_sysdata(void) +{ + struct sysdata_el *element; + struct list_head *list; + +sn_sysdata_free_start: + list_for_each(list, &sn_sysdata_list) { + element = list_entry(list, struct sysdata_el, entry); + list_del(&element->entry); + kfree(element->sysdata); + kfree(element); + goto sn_sysdata_free_start; + } + return; } /*
@@ -413,15 +451,16 @@ ia64_max_iommu_merge_mask = ~PAGE_MASK; sn_fixup_ionodes(); sn_irq_lh_init(); + INIT_LIST_HEAD(&sn_sysdata_list); sn_init_cpei_timer(); #ifdef CONFIG_PROC_FS register_sn_procfs(); #endif - for (i = 0; i < PCI_BUSES_TO_SCAN; i++) { - sn_pci_controller_fixup(0, i); - } + /* busses are not known yet ... */ + for (i = 0; i < PCI_BUSES_TO_SCAN; i++) + sn_pci_controller_fixup(0, i, NULL); /* * Generic Linux PCI Layer has created the pci_bus and pci_dev
@@ -430,9 +469,8 @@ */ while ((pci_dev = - pci_find_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) { + pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) sn_pci_fixup_slot(pci_dev); - } sn_ioif_inited = 1; /* sn I/O infrastructure now initialized */
@@ -474,3 +512,8 @@ } subsys_initcall(sn_pci_init); +EXPORT_SYMBOL_GPL(sn_pci_fixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_unfixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_controller_fixup); +EXPORT_SYMBOL_GPL(sn_bus_store_sysdata); +EXPORT_SYMBOL_GPL(sn_bus_free_sysdata);
arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ec5e7a01c5b55f328e0dff8a45b012bc1024e84e) renamed from arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ee4ac68d745f2c91e39d10e2f82cf624cf655edd) +3 -3
@@ -284,7 +284,6 @@ int cpu = nasid_slice_to_cpuid(nasid, slice); pci_dev_get(pci_dev); - sn_irq_info->irq_cpuid = cpu; sn_irq_info->irq_pciioinfo = SN_PCIDEV_INFO(pci_dev);
@@ -305,15 +304,16 @@ return; sn_irq_info = SN_PCIDEV_INFO(pci_dev)->pdi_sn_irq_info; - if (!sn_irq_info || !sn_irq_info->irq_irq) + if (!sn_irq_info || !sn_irq_info->irq_irq) { + kfree(sn_irq_info); return; + } unregister_intr_pda(sn_irq_info); spin_lock(&sn_irq_info_lock); list_del_rcu(&sn_irq_info->list); spin_unlock(&sn_irq_info_lock); call_rcu(&sn_irq_info->rcu, sn_irq_info_free); - pci_dev_put(pci_dev); }
arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9813da56d3113c41fb1b413cfa4c7963cda316cb) renamed from arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9bc4de4a3ec0778aa4724e75457bd253a6b1e45e) +37 -0
@@ -18,6 +18,40 @@ #include "xtalk/xwidgetdev.h" #include "xtalk/hubdev.h" +int +sal_pcibr_slot_enable(struct pcibus_info *soft, int device, void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_ENABLE, (u64) busnum, + (u64) device, (u64) resp, 0, 0, 0, 0); + + return (int)ret_stuff.v0; +} + +int +sal_pcibr_slot_disable(struct pcibus_info *soft, int device, int action, + void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_DISABLE, + (u64) busnum, (u64) device, (u64) action, + (u64) resp, 0, 0, 0); + + return (int)ret_stuff.v0; +} + static int sal_pcibr_error_interrupt(struct pcibus_info *soft) { struct ia64_sal_retval ret_stuff;
@@ -187,3 +221,6 @@ return 0; } + +EXPORT_SYMBOL_GPL(sal_pcibr_slot_enable); +EXPORT_SYMBOL_GPL(sal_pcibr_slot_disable);
drivers/pci/hotplug/Kconfig (mode:100644 sha1:9c4a39ee89b57b5065f575b86b2798ed31dc2fc6) renamed from drivers/pci/hotplug/Kconfig (mode:100644 sha1:1a4d4ca2a4dc386d3f63189ecc9cf30f806b947e) +3 -2
@@ -187,9 +187,10 @@ config HOTPLUG_PCI_SGI tristate "SGI PCI Hotplug Support" - depends on HOTPLUG_PCI && IA64_SGI_SN2 + depends on HOTPLUG_PCI && (IA64_SGI_SN2 || IA64_GENERIC) help - Say Y here if you have an SGI IA64 Altix system. + Say Y here if you want to use the SGI Altix Hotplug + Driver for PCI devices. When in doubt, say N.
drivers/pci/hotplug/Makefile (mode:100644 sha1:24ed904514c5043095dfe484877dd38cdfa0f3b6) renamed from drivers/pci/hotplug/Makefile (mode:100644 sha1:93c120ddbd395f5ddeddcc3cc5d52bd31a47dcc4) +1 -0
@@ -14,6 +14,7 @@ obj-$(CONFIG_HOTPLUG_PCI_SHPC) += shpchp.o obj-$(CONFIG_HOTPLUG_PCI_RPA) += rpaphp.o obj-$(CONFIG_HOTPLUG_PCI_RPA_DLPAR) += rpadlpar_io.o +obj-$(CONFIG_HOTPLUG_PCI_SGI) += sgi_hotplug.o pci_hotplug-objs := pci_hotplug_core.o
drivers/pci/hotplug/sgi_hotplug.c (mode:100644 sha1:7d39b45c23bbbb744b32b60781e6b38e228f78c7) renamed from dev/null (tree:f8014bc6a8330d22e2dbb0cac34d87e8a5352278) +615 -0
615 lines not on this page · raw message
include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:c1bf22505ff339ea7a5255ec9a8a60031397e9a8) renamed from include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:0abf539cb0414d752fcf14d9318bf1fa00278d32) +4 -0
@@ -151,4 +151,8 @@ extern int pcibr_ate_alloc(struct pcibus_info *, int); extern void pcibr_ate_free(struct pcibus_info *, int); extern void ate_write(struct pcibus_info *, int, int, uint64_t); +extern int sal_pcibr_slot_enable(struct pcibus_info *soft, int device, + void *resp); +extern int sal_pcibr_slot_disable(struct pcibus_info *soft, int device, + int action, void *resp); #endif
include/asm-ia64/sn/pcidev.h (mode:100644 sha1:49711d00ad04f77d90467daf33bfd13d549e2109) renamed from include/asm-ia64/sn/pcidev.h (mode:100644 sha1:9610fcc635450e57bf9be693fb8fa5e22eb60be1) +6 -0
@@ -23,6 +23,8 @@ #define SN_PCIBUS_BUSSOFT(pci_bus) \ ((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) +#define SN_PCIBUS_BUSSOFT_INFO(pci_bus) \ + (struct pcibus_info *)((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) /* * Given a struct pci_dev, return the sn pcibus_bussoft struct. Note * that this is not equivalent to SN_PCIBUS_BUSSOFT(pci_dev->bus) due
@@ -56,6 +58,10 @@ extern void sn_irq_fixup(struct pci_dev *pci_dev, struct sn_irq_info *sn_irq_info); extern void sn_irq_unfixup(struct pci_dev *pci_dev); +extern void sn_pci_controller_fixup(int segment, int busnum, + struct pci_bus *bus); +extern void sn_bus_store_sysdata(struct pci_dev *dev); +extern void sn_bus_free_sysdata(void); extern void sn_pci_fixup_slot(struct pci_dev *dev); extern void sn_pci_unfixup_slot(struct pci_dev *dev); extern void sn_irq_lh_init(void);
Review
Quoted 4 lines · Prarit Bhargava
(quoted diff) +4
-0+static struct hotplug_slot_attribute sn_slot_attr_path = { + .attr = {.name = "path", .mode = S_IFREG | S_IRUGO}, + .show = sn_read_path_file, +};
No, you did not properly initialize all of the attribute fields, so your module could be unloaded when the sysfs file is open :(
Please use the proper macro for this: __ATTR_RO()
greg k-h
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 8 lines · Greg KH
No, you did not properly initialize all of the attribute fields, so your module could be unloaded when the sysfs file is open :(
Please use the proper macro for this:
__ATTR_RO()greg k-h
New patch using __ATTR_RO.
P.
This patch is the SGI hotplug driver and additional changes required for the driver. These modifications include changes to the SN io_init.c code for memory management, the inclusion of new SAL calls to enable and disable PCI slots, and a hotplug-style driver.
---
commit 84e2ef4f5cd65fca5486a270317fdc2d7c3ab06b tree d9d588b149b4c976384fefc5f523479f035ca748 parent 837895352d3c72c4756fe1789e8a7482f69c6df5 author Prarit Bhargava <[email protected]> 1810814415 -0400 committer Prarit Bhargava <[email protected]> 1810814415 -0400
arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:b20e031391d71f07af8201cc5cbdc2e6120909b3) renamed from arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:ae6628cd41f8cc2f94643fce2c4fe21bdd85d2d9) +92 -50
@@ -23,6 +23,14 @@ nasid_t master_nasid = INVALID_NASID; /* Partition Master */ +static struct list_head sn_sysdata_list; + +/* sysdata list struct */ +struct sysdata_el { + struct list_head entry; + void *sysdata; +}; + struct slab_info { struct hubdev_info hubdev; };
@@ -137,23 +145,6 @@ } /* - * sn_alloc_pci_sysdata() - This routine allocates a pci controller - * which is expected as the pci_dev and pci_bus sysdata by the Linux - * PCI infrastructure. - */ -static inline struct pci_controller *sn_alloc_pci_sysdata(void) -{ - struct pci_controller *pci_sysdata; - - pci_sysdata = kmalloc(sizeof(*pci_sysdata), GFP_KERNEL); - if (!pci_sysdata) - BUG(); - - memset(pci_sysdata, 0, sizeof(*pci_sysdata)); - return pci_sysdata; -} - -/* * sn_fixup_ionodes() - This routine initializes the HUB data strcuture for * each node in the system. */
@@ -220,6 +211,15 @@ } +void sn_pci_unfixup_slot(struct pci_dev *dev) +{ + struct pci_dev *host_pci_dev = SN_PCIDEV_INFO(dev)->host_pci_dev; + + sn_irq_unfixup(dev); + pci_dev_put(host_pci_dev); + pci_dev_put(dev); +} + /* * sn_pci_fixup_slot() - This routine sets up a slot's resources * consistent with the Linux PCI abstraction layer. Resources acquired
@@ -238,10 +238,10 @@ unsigned long size; unsigned int bus_no, devfn; - dev->sysdata = kmalloc(sizeof(struct pcidev_info), GFP_KERNEL); + pci_dev_get(dev); /* for the sysdata pointer */ + dev->sysdata = kcalloc(1, sizeof(struct pcidev_info), GFP_KERNEL); if (SN_PCIDEV_INFO(dev) <= 0) BUG(); /* Cannot afford to run out of memory */ - memset(SN_PCIDEV_INFO(dev), 0, sizeof(struct pcidev_info)); sn_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_KERNEL); if (sn_irq_info <= 0)
@@ -276,7 +276,8 @@ dev->resource[idx].parent = &iomem_resource; } - /* Using the PROMs values for the PCI host bus, get the Linux + /* + * Using the PROMs values for the PCI host bus, get the Linux * PCI host_pci_dev struct and set up host bus linkages */
@@ -292,11 +293,10 @@ bs = SN_PCIBUS_BUSSOFT(dev->bus); SN_PCIDEV_INFO(dev)->pdi_pcibus_info = bs; - if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) { + if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) SN_PCIDEV_BUSPROVIDER(dev) = sn_pci_provider[bs->bs_asic_type]; - } else { + else SN_PCIDEV_BUSPROVIDER(dev) = &sn_pci_default_provider; - } /* Only set up IRQ stuff if this device has a host bus context */ if (bs && sn_irq_info->irq_irq) {
@@ -313,55 +313,57 @@ * sn_pci_controller_fixup() - This routine sets up a bus's resources * consistent with the Linux PCI abstraction layer. */ -static void sn_pci_controller_fixup(int segment, int busnum) +void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus) { int status = 0; int nasid, cnode; - struct pci_bus *bus; struct pci_controller *controller; struct pcibus_bussoft *prom_bussoft_ptr; struct hubdev_info *hubdev_info; void *provider_soft; struct sn_pcibus_provider *provider; - status = - sal_get_pcibus_info((u64) segment, (u64) busnum, - (u64) ia64_tpa(&prom_bussoft_ptr)); - if (status > 0) { - return; /* bus # does not exist */ - } - + status = sal_get_pcibus_info((u64) segment, (u64) busnum, + (u64) ia64_tpa(&prom_bussoft_ptr)); + if (status > 0) + return; /*bus # does not exist */ prom_bussoft_ptr = __va(prom_bussoft_ptr); - controller = sn_alloc_pci_sysdata(); - /* controller non-zero is BUG'd in sn_alloc_pci_sysdata */ - bus = pci_scan_bus(busnum, &pci_root_ops, controller); + controller = kcalloc(1,sizeof(*controller), GFP_KERNEL); + if (!controller) + BUG(); + if (bus == NULL) { - return; /* error, or bus already scanned */ + bus = pci_scan_bus(busnum, &pci_root_ops, controller); + if (bus == NULL) + return; /* error, or bus already scanned */ + bus->sysdata = NULL; } + if (bus->sysdata) + goto error_return; /* sysdata already alloc'd */ + /* * Per-provider fixup. Copies the contents from prom to local * area and links SN_PCIBUS_BUSSOFT(). */ - if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) { + if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) return; /* unsupported asic type */ - } + + if (prom_bussoft_ptr->bs_asic_type == PCIIO_ASIC_TYPE_PPB) + goto error_return; /* no further fixup necessary */ provider = sn_pci_provider[prom_bussoft_ptr->bs_asic_type]; - if (provider == NULL) { + if (provider == NULL) return; /* no provider registerd for this asic */ - } provider_soft = NULL; - if (provider->bus_fixup) { + if (provider->bus_fixup) provider_soft = (*provider->bus_fixup) (prom_bussoft_ptr); - } - if (provider_soft == NULL) { + if (provider_soft == NULL) return; /* fixup failed or not applicable */ - } /* * Generic bus fixup goes here. Don't reference prom_bussoft_ptr
@@ -370,12 +372,47 @@ bus->sysdata = controller; PCI_CONTROLLER(bus)->platform_data = provider_soft; - nasid = NASID_GET(SN_PCIBUS_BUSSOFT(bus)->bs_base); cnode = nasid_to_cnodeid(nasid); hubdev_info = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); SN_PCIBUS_BUSSOFT(bus)->bs_xwidget_info = &(hubdev_info->hdi_xwidget_info[SN_PCIBUS_BUSSOFT(bus)->bs_xid]); + + return; + +error_return: + + kfree(controller); + return; +} + +void sn_bus_store_sysdata(struct pci_dev *dev) +{ + struct sysdata_el *element; + + element = kcalloc(1, sizeof(*element), GFP_KERNEL); + if (!element) { + dev_dbg(dev, "%s: out of memory!\n", __FUNCTION__); + return; + } + element->sysdata = dev->sysdata; + list_add(&element->entry, &sn_sysdata_list); +} + +void sn_bus_free_sysdata(void) +{ + struct sysdata_el *element; + struct list_head *list; + +sn_sysdata_free_start: + list_for_each(list, &sn_sysdata_list) { + element = list_entry(list, struct sysdata_el, entry); + list_del(&element->entry); + kfree(element->sysdata); + kfree(element); + goto sn_sysdata_free_start; + } + return; } /*
@@ -413,15 +450,16 @@ ia64_max_iommu_merge_mask = ~PAGE_MASK; sn_fixup_ionodes(); sn_irq_lh_init(); + INIT_LIST_HEAD(&sn_sysdata_list); sn_init_cpei_timer(); #ifdef CONFIG_PROC_FS register_sn_procfs(); #endif - for (i = 0; i < PCI_BUSES_TO_SCAN; i++) { - sn_pci_controller_fixup(0, i); - } + /* busses are not known yet ... */ + for (i = 0; i < PCI_BUSES_TO_SCAN; i++) + sn_pci_controller_fixup(0, i, NULL); /* * Generic Linux PCI Layer has created the pci_bus and pci_dev
@@ -430,9 +468,8 @@ */ while ((pci_dev = - pci_find_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) { + pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) sn_pci_fixup_slot(pci_dev); - } sn_ioif_inited = 1; /* sn I/O infrastructure now initialized */
@@ -474,3 +511,8 @@ } subsys_initcall(sn_pci_init); +EXPORT_SYMBOL_GPL(sn_pci_fixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_unfixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_controller_fixup); +EXPORT_SYMBOL_GPL(sn_bus_store_sysdata); +EXPORT_SYMBOL_GPL(sn_bus_free_sysdata);
arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ec5e7a01c5b55f328e0dff8a45b012bc1024e84e) renamed from arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ee4ac68d745f2c91e39d10e2f82cf624cf655edd) +3 -3
@@ -284,7 +284,6 @@ int cpu = nasid_slice_to_cpuid(nasid, slice); pci_dev_get(pci_dev); - sn_irq_info->irq_cpuid = cpu; sn_irq_info->irq_pciioinfo = SN_PCIDEV_INFO(pci_dev);
@@ -305,15 +304,16 @@ return; sn_irq_info = SN_PCIDEV_INFO(pci_dev)->pdi_sn_irq_info; - if (!sn_irq_info || !sn_irq_info->irq_irq) + if (!sn_irq_info || !sn_irq_info->irq_irq) { + kfree(sn_irq_info); return; + } unregister_intr_pda(sn_irq_info); spin_lock(&sn_irq_info_lock); list_del_rcu(&sn_irq_info->list); spin_unlock(&sn_irq_info_lock); call_rcu(&sn_irq_info->rcu, sn_irq_info_free); - pci_dev_put(pci_dev); }
arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9813da56d3113c41fb1b413cfa4c7963cda316cb) renamed from arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9bc4de4a3ec0778aa4724e75457bd253a6b1e45e) +37 -0
@@ -18,6 +18,40 @@ #include "xtalk/xwidgetdev.h" #include "xtalk/hubdev.h" +int +sal_pcibr_slot_enable(struct pcibus_info *soft, int device, void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_ENABLE, (u64) busnum, + (u64) device, (u64) resp, 0, 0, 0, 0); + + return (int)ret_stuff.v0; +} + +int +sal_pcibr_slot_disable(struct pcibus_info *soft, int device, int action, + void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_DISABLE, + (u64) busnum, (u64) device, (u64) action, + (u64) resp, 0, 0, 0); + + return (int)ret_stuff.v0; +} + static int sal_pcibr_error_interrupt(struct pcibus_info *soft) { struct ia64_sal_retval ret_stuff;
@@ -187,3 +221,6 @@ return 0; } + +EXPORT_SYMBOL_GPL(sal_pcibr_slot_enable); +EXPORT_SYMBOL_GPL(sal_pcibr_slot_disable);
drivers/pci/hotplug/Kconfig (mode:100644 sha1:9c4a39ee89b57b5065f575b86b2798ed31dc2fc6) renamed from drivers/pci/hotplug/Kconfig (mode:100644 sha1:1a4d4ca2a4dc386d3f63189ecc9cf30f806b947e) +3 -2
@@ -187,9 +187,10 @@ config HOTPLUG_PCI_SGI tristate "SGI PCI Hotplug Support" - depends on HOTPLUG_PCI && IA64_SGI_SN2 + depends on HOTPLUG_PCI && (IA64_SGI_SN2 || IA64_GENERIC) help - Say Y here if you have an SGI IA64 Altix system. + Say Y here if you want to use the SGI Altix Hotplug + Driver for PCI devices. When in doubt, say N.
drivers/pci/hotplug/Makefile (mode:100644 sha1:24ed904514c5043095dfe484877dd38cdfa0f3b6) renamed from drivers/pci/hotplug/Makefile (mode:100644 sha1:93c120ddbd395f5ddeddcc3cc5d52bd31a47dcc4) +1 -0
@@ -14,6 +14,7 @@ obj-$(CONFIG_HOTPLUG_PCI_SHPC) += shpchp.o obj-$(CONFIG_HOTPLUG_PCI_RPA) += rpaphp.o obj-$(CONFIG_HOTPLUG_PCI_RPA_DLPAR) += rpadlpar_io.o +obj-$(CONFIG_HOTPLUG_PCI_SGI) += sgi_hotplug.o pci_hotplug-objs := pci_hotplug_core.o
drivers/pci/hotplug/sgi_hotplug.c (mode:100644 sha1:5fa992b99396b4ece081b649d9f7f4af1857ece1) renamed from dev/null (tree:f8014bc6a8330d22e2dbb0cac34d87e8a5352278) +635 -0
635 lines not on this page · raw message
include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:c1bf22505ff339ea7a5255ec9a8a60031397e9a8) renamed from include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:0abf539cb0414d752fcf14d9318bf1fa00278d32) +4 -0
@@ -151,4 +151,8 @@ extern int pcibr_ate_alloc(struct pcibus_info *, int); extern void pcibr_ate_free(struct pcibus_info *, int); extern void ate_write(struct pcibus_info *, int, int, uint64_t); +extern int sal_pcibr_slot_enable(struct pcibus_info *soft, int device, + void *resp); +extern int sal_pcibr_slot_disable(struct pcibus_info *soft, int device, + int action, void *resp); #endif
include/asm-ia64/sn/pcidev.h (mode:100644 sha1:49711d00ad04f77d90467daf33bfd13d549e2109) renamed from include/asm-ia64/sn/pcidev.h (mode:100644 sha1:9610fcc635450e57bf9be693fb8fa5e22eb60be1) +6 -0
@@ -23,6 +23,8 @@ #define SN_PCIBUS_BUSSOFT(pci_bus) \ ((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) +#define SN_PCIBUS_BUSSOFT_INFO(pci_bus) \ + (struct pcibus_info *)((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) /* * Given a struct pci_dev, return the sn pcibus_bussoft struct. Note * that this is not equivalent to SN_PCIBUS_BUSSOFT(pci_dev->bus) due
@@ -56,6 +58,10 @@ extern void sn_irq_fixup(struct pci_dev *pci_dev, struct sn_irq_info *sn_irq_info); extern void sn_irq_unfixup(struct pci_dev *pci_dev); +extern void sn_pci_controller_fixup(int segment, int busnum, + struct pci_bus *bus); +extern void sn_bus_store_sysdata(struct pci_dev *dev); +extern void sn_bus_free_sysdata(void); extern void sn_pci_fixup_slot(struct pci_dev *dev); extern void sn_pci_unfixup_slot(struct pci_dev *dev); extern void sn_irq_lh_init(void);
Review
Quoted 4 lines · Prarit Bhargava
(quoted diff) +4
-0+static struct hotplug_slot_attribute sn_slot_attrs[] = { + __ATTR_RO(path), + __ATTR_NULL +};
Why are you having an array with only 1 attribute? Just define the one, and live with it. That way you can get rid of all of your odd loops.
And if you do want to stick with an array, use the proper sysfs call to register and unregister all of them at once, don't roll your own code to do this.
greg k-h
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 10 lines · Greg KH
Review
Quoted 4 lines · Prarit Bhargava
(quoted diff) +4
-0+static struct hotplug_slot_attribute sn_slot_attrs[] = { + __ATTR_RO(path), + __ATTR_NULL +};Why are you having an array with only 1 attribute? Just define the one, and live with it. That way you can get rid of all of your odd loops.
Sorry -- all the examples I found used __ATTR_RO in an array. I thought
that was the way you wanted the attributes defined. I would prefer to
have a single attribute as well.
I'll modify and retest ...
Thanks,
P.
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 5 lines · Prarit Bhargava
Quoted 3 lines · Greg KH
Why are you having an array with only 1 attribute? Just define the one, and live with it. That way you can get rid of all of your odd loops.
New patch using a single attribute for the path sysfs file, instead of an array which had one element.
P.
This patch is the SGI hotplug driver and additional changes required for the driver. These modifications include changes to the SN io_init.c code for memory management, the inclusion of new SAL calls to enable and disable PCI slots, and a hotplug-style driver.
---
commit 7dc135b31e990bafe5a211de95def141f88bc9fa tree f236fe54c824d0648484f6abf41c199dd4a28c1c parent fbcad9333f12f809ad3ad303b08c89d8225cf6b1 author Prarit Bhargava <[email protected]> 1810857593 -0400 committer Prarit Bhargava <[email protected]> 1810857593 -0400
arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:b20e031391d71f07af8201cc5cbdc2e6120909b3) renamed from arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:ae6628cd41f8cc2f94643fce2c4fe21bdd85d2d9) +92 -50
@@ -23,6 +23,14 @@ nasid_t master_nasid = INVALID_NASID; /* Partition Master */ +static struct list_head sn_sysdata_list; + +/* sysdata list struct */ +struct sysdata_el { + struct list_head entry; + void *sysdata; +}; + struct slab_info { struct hubdev_info hubdev; };
@@ -137,23 +145,6 @@ } /* - * sn_alloc_pci_sysdata() - This routine allocates a pci controller - * which is expected as the pci_dev and pci_bus sysdata by the Linux - * PCI infrastructure. - */ -static inline struct pci_controller *sn_alloc_pci_sysdata(void) -{ - struct pci_controller *pci_sysdata; - - pci_sysdata = kmalloc(sizeof(*pci_sysdata), GFP_KERNEL); - if (!pci_sysdata) - BUG(); - - memset(pci_sysdata, 0, sizeof(*pci_sysdata)); - return pci_sysdata; -} - -/* * sn_fixup_ionodes() - This routine initializes the HUB data strcuture for * each node in the system. */
@@ -220,6 +211,15 @@ } +void sn_pci_unfixup_slot(struct pci_dev *dev) +{ + struct pci_dev *host_pci_dev = SN_PCIDEV_INFO(dev)->host_pci_dev; + + sn_irq_unfixup(dev); + pci_dev_put(host_pci_dev); + pci_dev_put(dev); +} + /* * sn_pci_fixup_slot() - This routine sets up a slot's resources * consistent with the Linux PCI abstraction layer. Resources acquired
@@ -238,10 +238,10 @@ unsigned long size; unsigned int bus_no, devfn; - dev->sysdata = kmalloc(sizeof(struct pcidev_info), GFP_KERNEL); + pci_dev_get(dev); /* for the sysdata pointer */ + dev->sysdata = kcalloc(1, sizeof(struct pcidev_info), GFP_KERNEL); if (SN_PCIDEV_INFO(dev) <= 0) BUG(); /* Cannot afford to run out of memory */ - memset(SN_PCIDEV_INFO(dev), 0, sizeof(struct pcidev_info)); sn_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_KERNEL); if (sn_irq_info <= 0)
@@ -276,7 +276,8 @@ dev->resource[idx].parent = &iomem_resource; } - /* Using the PROMs values for the PCI host bus, get the Linux + /* + * Using the PROMs values for the PCI host bus, get the Linux * PCI host_pci_dev struct and set up host bus linkages */
@@ -292,11 +293,10 @@ bs = SN_PCIBUS_BUSSOFT(dev->bus); SN_PCIDEV_INFO(dev)->pdi_pcibus_info = bs; - if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) { + if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) SN_PCIDEV_BUSPROVIDER(dev) = sn_pci_provider[bs->bs_asic_type]; - } else { + else SN_PCIDEV_BUSPROVIDER(dev) = &sn_pci_default_provider; - } /* Only set up IRQ stuff if this device has a host bus context */ if (bs && sn_irq_info->irq_irq) {
@@ -313,55 +313,57 @@ * sn_pci_controller_fixup() - This routine sets up a bus's resources * consistent with the Linux PCI abstraction layer. */ -static void sn_pci_controller_fixup(int segment, int busnum) +void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus) { int status = 0; int nasid, cnode; - struct pci_bus *bus; struct pci_controller *controller; struct pcibus_bussoft *prom_bussoft_ptr; struct hubdev_info *hubdev_info; void *provider_soft; struct sn_pcibus_provider *provider; - status = - sal_get_pcibus_info((u64) segment, (u64) busnum, - (u64) ia64_tpa(&prom_bussoft_ptr)); - if (status > 0) { - return; /* bus # does not exist */ - } - + status = sal_get_pcibus_info((u64) segment, (u64) busnum, + (u64) ia64_tpa(&prom_bussoft_ptr)); + if (status > 0) + return; /*bus # does not exist */ prom_bussoft_ptr = __va(prom_bussoft_ptr); - controller = sn_alloc_pci_sysdata(); - /* controller non-zero is BUG'd in sn_alloc_pci_sysdata */ - bus = pci_scan_bus(busnum, &pci_root_ops, controller); + controller = kcalloc(1,sizeof(*controller), GFP_KERNEL); + if (!controller) + BUG(); + if (bus == NULL) { - return; /* error, or bus already scanned */ + bus = pci_scan_bus(busnum, &pci_root_ops, controller); + if (bus == NULL) + return; /* error, or bus already scanned */ + bus->sysdata = NULL; } + if (bus->sysdata) + goto error_return; /* sysdata already alloc'd */ + /* * Per-provider fixup. Copies the contents from prom to local * area and links SN_PCIBUS_BUSSOFT(). */ - if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) { + if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) return; /* unsupported asic type */ - } + + if (prom_bussoft_ptr->bs_asic_type == PCIIO_ASIC_TYPE_PPB) + goto error_return; /* no further fixup necessary */ provider = sn_pci_provider[prom_bussoft_ptr->bs_asic_type]; - if (provider == NULL) { + if (provider == NULL) return; /* no provider registerd for this asic */ - } provider_soft = NULL; - if (provider->bus_fixup) { + if (provider->bus_fixup) provider_soft = (*provider->bus_fixup) (prom_bussoft_ptr); - } - if (provider_soft == NULL) { + if (provider_soft == NULL) return; /* fixup failed or not applicable */ - } /* * Generic bus fixup goes here. Don't reference prom_bussoft_ptr
@@ -370,12 +372,47 @@ bus->sysdata = controller; PCI_CONTROLLER(bus)->platform_data = provider_soft; - nasid = NASID_GET(SN_PCIBUS_BUSSOFT(bus)->bs_base); cnode = nasid_to_cnodeid(nasid); hubdev_info = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); SN_PCIBUS_BUSSOFT(bus)->bs_xwidget_info = &(hubdev_info->hdi_xwidget_info[SN_PCIBUS_BUSSOFT(bus)->bs_xid]); + + return; + +error_return: + + kfree(controller); + return; +} + +void sn_bus_store_sysdata(struct pci_dev *dev) +{ + struct sysdata_el *element; + + element = kcalloc(1, sizeof(*element), GFP_KERNEL); + if (!element) { + dev_dbg(dev, "%s: out of memory!\n", __FUNCTION__); + return; + } + element->sysdata = dev->sysdata; + list_add(&element->entry, &sn_sysdata_list); +} + +void sn_bus_free_sysdata(void) +{ + struct sysdata_el *element; + struct list_head *list; + +sn_sysdata_free_start: + list_for_each(list, &sn_sysdata_list) { + element = list_entry(list, struct sysdata_el, entry); + list_del(&element->entry); + kfree(element->sysdata); + kfree(element); + goto sn_sysdata_free_start; + } + return; } /*
@@ -413,15 +450,16 @@ ia64_max_iommu_merge_mask = ~PAGE_MASK; sn_fixup_ionodes(); sn_irq_lh_init(); + INIT_LIST_HEAD(&sn_sysdata_list); sn_init_cpei_timer(); #ifdef CONFIG_PROC_FS register_sn_procfs(); #endif - for (i = 0; i < PCI_BUSES_TO_SCAN; i++) { - sn_pci_controller_fixup(0, i); - } + /* busses are not known yet ... */ + for (i = 0; i < PCI_BUSES_TO_SCAN; i++) + sn_pci_controller_fixup(0, i, NULL); /* * Generic Linux PCI Layer has created the pci_bus and pci_dev
@@ -430,9 +468,8 @@ */ while ((pci_dev = - pci_find_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) { + pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) sn_pci_fixup_slot(pci_dev); - } sn_ioif_inited = 1; /* sn I/O infrastructure now initialized */
@@ -474,3 +511,8 @@ } subsys_initcall(sn_pci_init); +EXPORT_SYMBOL_GPL(sn_pci_fixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_unfixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_controller_fixup); +EXPORT_SYMBOL_GPL(sn_bus_store_sysdata); +EXPORT_SYMBOL_GPL(sn_bus_free_sysdata);
arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ec5e7a01c5b55f328e0dff8a45b012bc1024e84e) renamed from arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ee4ac68d745f2c91e39d10e2f82cf624cf655edd) +3 -3
@@ -284,7 +284,6 @@ int cpu = nasid_slice_to_cpuid(nasid, slice); pci_dev_get(pci_dev); - sn_irq_info->irq_cpuid = cpu; sn_irq_info->irq_pciioinfo = SN_PCIDEV_INFO(pci_dev);
@@ -305,15 +304,16 @@ return; sn_irq_info = SN_PCIDEV_INFO(pci_dev)->pdi_sn_irq_info; - if (!sn_irq_info || !sn_irq_info->irq_irq) + if (!sn_irq_info || !sn_irq_info->irq_irq) { + kfree(sn_irq_info); return; + } unregister_intr_pda(sn_irq_info); spin_lock(&sn_irq_info_lock); list_del_rcu(&sn_irq_info->list); spin_unlock(&sn_irq_info_lock); call_rcu(&sn_irq_info->rcu, sn_irq_info_free); - pci_dev_put(pci_dev); }
arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9813da56d3113c41fb1b413cfa4c7963cda316cb) renamed from arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9bc4de4a3ec0778aa4724e75457bd253a6b1e45e) +37 -0
@@ -18,6 +18,40 @@ #include "xtalk/xwidgetdev.h" #include "xtalk/hubdev.h" +int +sal_pcibr_slot_enable(struct pcibus_info *soft, int device, void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_ENABLE, (u64) busnum, + (u64) device, (u64) resp, 0, 0, 0, 0); + + return (int)ret_stuff.v0; +} + +int +sal_pcibr_slot_disable(struct pcibus_info *soft, int device, int action, + void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_DISABLE, + (u64) busnum, (u64) device, (u64) action, + (u64) resp, 0, 0, 0); + + return (int)ret_stuff.v0; +} + static int sal_pcibr_error_interrupt(struct pcibus_info *soft) { struct ia64_sal_retval ret_stuff;
@@ -187,3 +221,6 @@ return 0; } + +EXPORT_SYMBOL_GPL(sal_pcibr_slot_enable); +EXPORT_SYMBOL_GPL(sal_pcibr_slot_disable);
drivers/pci/hotplug/Kconfig (mode:100644 sha1:9c4a39ee89b57b5065f575b86b2798ed31dc2fc6) renamed from drivers/pci/hotplug/Kconfig (mode:100644 sha1:1a4d4ca2a4dc386d3f63189ecc9cf30f806b947e) +3 -2
@@ -187,9 +187,10 @@ config HOTPLUG_PCI_SGI tristate "SGI PCI Hotplug Support" - depends on HOTPLUG_PCI && IA64_SGI_SN2 + depends on HOTPLUG_PCI && (IA64_SGI_SN2 || IA64_GENERIC) help - Say Y here if you have an SGI IA64 Altix system. + Say Y here if you want to use the SGI Altix Hotplug + Driver for PCI devices. When in doubt, say N.
drivers/pci/hotplug/Makefile (mode:100644 sha1:24ed904514c5043095dfe484877dd38cdfa0f3b6) renamed from drivers/pci/hotplug/Makefile (mode:100644 sha1:93c120ddbd395f5ddeddcc3cc5d52bd31a47dcc4) +1 -0
@@ -14,6 +14,7 @@ obj-$(CONFIG_HOTPLUG_PCI_SHPC) += shpchp.o obj-$(CONFIG_HOTPLUG_PCI_RPA) += rpaphp.o obj-$(CONFIG_HOTPLUG_PCI_RPA_DLPAR) += rpadlpar_io.o +obj-$(CONFIG_HOTPLUG_PCI_SGI) += sgi_hotplug.o pci_hotplug-objs := pci_hotplug_core.o
drivers/pci/hotplug/sgi_hotplug.c (mode:100644 sha1:cbfa52c4d6b93e34d3b51890b7293934f4e1e198) renamed from dev/null (tree:f8014bc6a8330d22e2dbb0cac34d87e8a5352278) +606 -0
606 lines not on this page · raw message
include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:c1bf22505ff339ea7a5255ec9a8a60031397e9a8) renamed from include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:0abf539cb0414d752fcf14d9318bf1fa00278d32) +4 -0
@@ -151,4 +151,8 @@ extern int pcibr_ate_alloc(struct pcibus_info *, int); extern void pcibr_ate_free(struct pcibus_info *, int); extern void ate_write(struct pcibus_info *, int, int, uint64_t); +extern int sal_pcibr_slot_enable(struct pcibus_info *soft, int device, + void *resp); +extern int sal_pcibr_slot_disable(struct pcibus_info *soft, int device, + int action, void *resp); #endif
include/asm-ia64/sn/pcidev.h (mode:100644 sha1:49711d00ad04f77d90467daf33bfd13d549e2109) renamed from include/asm-ia64/sn/pcidev.h (mode:100644 sha1:9610fcc635450e57bf9be693fb8fa5e22eb60be1) +6 -0
@@ -23,6 +23,8 @@ #define SN_PCIBUS_BUSSOFT(pci_bus) \ ((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) +#define SN_PCIBUS_BUSSOFT_INFO(pci_bus) \ + (struct pcibus_info *)((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) /* * Given a struct pci_dev, return the sn pcibus_bussoft struct. Note * that this is not equivalent to SN_PCIBUS_BUSSOFT(pci_dev->bus) due
@@ -56,6 +58,10 @@ extern void sn_irq_fixup(struct pci_dev *pci_dev, struct sn_irq_info *sn_irq_info); extern void sn_irq_unfixup(struct pci_dev *pci_dev); +extern void sn_pci_controller_fixup(int segment, int busnum, + struct pci_bus *bus); +extern void sn_bus_store_sysdata(struct pci_dev *dev); +extern void sn_bus_free_sysdata(void); extern void sn_pci_fixup_slot(struct pci_dev *dev); extern void sn_pci_unfixup_slot(struct pci_dev *dev); extern void sn_irq_lh_init(void);
Review
Quoted 2 lines · Prarit Bhargava
(quoted diff) +2
-0+ list_for_each(list, &sn_sysdata_list) { + element = list_entry(list, struct sysdata_el, entry);
All instances of list_for_each() and then a list_entry() call can be replaced with list_for_each_entry().
thanks,
greg k-h
Re: [Pcihpd-discuss] [PATCH 5/6]: hotplug/ia64: SN Hotplug Driver
In reply to Rolf Eike BeerQuoted 8 lines · Greg KH
Review
Quoted 2 lines · Prarit Bhargava
(quoted diff) +2
-0+ list_for_each(list, &sn_sysdata_list) { + element = list_entry(list, struct sysdata_el, entry);All instances of
list_for_each()and then alist_entry()call can be replaced withlist_for_each_entry().
New patch with list_for_each_entry.
P.
This patch is the SGI hotplug driver and additional changes required for the driver. These modifications include changes to the SN io_init.c code for memory management, the inclusion of new SAL calls to enable and disable PCI slots, and a hotplug-style driver.
---
commit 7ec91b987e364af6998f9f26a744d0a343e05ca1 tree 7f8b42b9f9d9e31ebd6d26e86f93a8bc22b19848 parent c95c10da25e11764596c4517c02fea3b9df6753c author Prarit Bhargava <[email protected]> 1810947453 -0400 committer Prarit Bhargava <[email protected]> 1810947453 -0400
arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:f8a789c15007b627ec5c3b38fb662a74ad252c7e) renamed from arch/ia64/sn/kernel/io_init.c (mode:100644 sha1:ae6628cd41f8cc2f94643fce2c4fe21bdd85d2d9) +90 -50
@@ -23,6 +23,14 @@ nasid_t master_nasid = INVALID_NASID; /* Partition Master */ +static struct list_head sn_sysdata_list; + +/* sysdata list struct */ +struct sysdata_el { + struct list_head entry; + void *sysdata; +}; + struct slab_info { struct hubdev_info hubdev; };
@@ -137,23 +145,6 @@ } /* - * sn_alloc_pci_sysdata() - This routine allocates a pci controller - * which is expected as the pci_dev and pci_bus sysdata by the Linux - * PCI infrastructure. - */ -static inline struct pci_controller *sn_alloc_pci_sysdata(void) -{ - struct pci_controller *pci_sysdata; - - pci_sysdata = kmalloc(sizeof(*pci_sysdata), GFP_KERNEL); - if (!pci_sysdata) - BUG(); - - memset(pci_sysdata, 0, sizeof(*pci_sysdata)); - return pci_sysdata; -} - -/* * sn_fixup_ionodes() - This routine initializes the HUB data strcuture for * each node in the system. */
@@ -220,6 +211,15 @@ } +void sn_pci_unfixup_slot(struct pci_dev *dev) +{ + struct pci_dev *host_pci_dev = SN_PCIDEV_INFO(dev)->host_pci_dev; + + sn_irq_unfixup(dev); + pci_dev_put(host_pci_dev); + pci_dev_put(dev); +} + /* * sn_pci_fixup_slot() - This routine sets up a slot's resources * consistent with the Linux PCI abstraction layer. Resources acquired
@@ -238,10 +238,10 @@ unsigned long size; unsigned int bus_no, devfn; - dev->sysdata = kmalloc(sizeof(struct pcidev_info), GFP_KERNEL); + pci_dev_get(dev); /* for the sysdata pointer */ + dev->sysdata = kcalloc(1, sizeof(struct pcidev_info), GFP_KERNEL); if (SN_PCIDEV_INFO(dev) <= 0) BUG(); /* Cannot afford to run out of memory */ - memset(SN_PCIDEV_INFO(dev), 0, sizeof(struct pcidev_info)); sn_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_KERNEL); if (sn_irq_info <= 0)
@@ -276,7 +276,8 @@ dev->resource[idx].parent = &iomem_resource; } - /* Using the PROMs values for the PCI host bus, get the Linux + /* + * Using the PROMs values for the PCI host bus, get the Linux * PCI host_pci_dev struct and set up host bus linkages */
@@ -292,11 +293,10 @@ bs = SN_PCIBUS_BUSSOFT(dev->bus); SN_PCIDEV_INFO(dev)->pdi_pcibus_info = bs; - if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) { + if (bs && bs->bs_asic_type < PCIIO_ASIC_MAX_TYPES) SN_PCIDEV_BUSPROVIDER(dev) = sn_pci_provider[bs->bs_asic_type]; - } else { + else SN_PCIDEV_BUSPROVIDER(dev) = &sn_pci_default_provider; - } /* Only set up IRQ stuff if this device has a host bus context */ if (bs && sn_irq_info->irq_irq) {
@@ -313,55 +313,57 @@ * sn_pci_controller_fixup() - This routine sets up a bus's resources * consistent with the Linux PCI abstraction layer. */ -static void sn_pci_controller_fixup(int segment, int busnum) +void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus) { int status = 0; int nasid, cnode; - struct pci_bus *bus; struct pci_controller *controller; struct pcibus_bussoft *prom_bussoft_ptr; struct hubdev_info *hubdev_info; void *provider_soft; struct sn_pcibus_provider *provider; - status = - sal_get_pcibus_info((u64) segment, (u64) busnum, - (u64) ia64_tpa(&prom_bussoft_ptr)); - if (status > 0) { - return; /* bus # does not exist */ - } - + status = sal_get_pcibus_info((u64) segment, (u64) busnum, + (u64) ia64_tpa(&prom_bussoft_ptr)); + if (status > 0) + return; /*bus # does not exist */ prom_bussoft_ptr = __va(prom_bussoft_ptr); - controller = sn_alloc_pci_sysdata(); - /* controller non-zero is BUG'd in sn_alloc_pci_sysdata */ - bus = pci_scan_bus(busnum, &pci_root_ops, controller); + controller = kcalloc(1,sizeof(*controller), GFP_KERNEL); + if (!controller) + BUG(); + if (bus == NULL) { - return; /* error, or bus already scanned */ + bus = pci_scan_bus(busnum, &pci_root_ops, controller); + if (bus == NULL) + return; /* error, or bus already scanned */ + bus->sysdata = NULL; } + if (bus->sysdata) + goto error_return; /* sysdata already alloc'd */ + /* * Per-provider fixup. Copies the contents from prom to local * area and links SN_PCIBUS_BUSSOFT(). */ - if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) { + if (prom_bussoft_ptr->bs_asic_type >= PCIIO_ASIC_MAX_TYPES) return; /* unsupported asic type */ - } + + if (prom_bussoft_ptr->bs_asic_type == PCIIO_ASIC_TYPE_PPB) + goto error_return; /* no further fixup necessary */ provider = sn_pci_provider[prom_bussoft_ptr->bs_asic_type]; - if (provider == NULL) { + if (provider == NULL) return; /* no provider registerd for this asic */ - } provider_soft = NULL; - if (provider->bus_fixup) { + if (provider->bus_fixup) provider_soft = (*provider->bus_fixup) (prom_bussoft_ptr); - } - if (provider_soft == NULL) { + if (provider_soft == NULL) return; /* fixup failed or not applicable */ - } /* * Generic bus fixup goes here. Don't reference prom_bussoft_ptr
@@ -370,12 +372,45 @@ bus->sysdata = controller; PCI_CONTROLLER(bus)->platform_data = provider_soft; - nasid = NASID_GET(SN_PCIBUS_BUSSOFT(bus)->bs_base); cnode = nasid_to_cnodeid(nasid); hubdev_info = (struct hubdev_info *)(NODEPDA(cnode)->pdinfo); SN_PCIBUS_BUSSOFT(bus)->bs_xwidget_info = &(hubdev_info->hdi_xwidget_info[SN_PCIBUS_BUSSOFT(bus)->bs_xid]); + + return; + +error_return: + + kfree(controller); + return; +} + +void sn_bus_store_sysdata(struct pci_dev *dev) +{ + struct sysdata_el *element; + + element = kcalloc(1, sizeof(*element), GFP_KERNEL); + if (!element) { + dev_dbg(dev, "%s: out of memory!\n", __FUNCTION__); + return; + } + element->sysdata = dev->sysdata; + list_add(&element->entry, &sn_sysdata_list); +} + +void sn_bus_free_sysdata(void) +{ + struct sysdata_el *element; + +sn_sysdata_free_start: + list_for_each_entry(element, &sn_sysdata_list, entry) { + list_del(&element->entry); + kfree(element->sysdata); + kfree(element); + goto sn_sysdata_free_start; + } + return; } /*
@@ -413,15 +448,16 @@ ia64_max_iommu_merge_mask = ~PAGE_MASK; sn_fixup_ionodes(); sn_irq_lh_init(); + INIT_LIST_HEAD(&sn_sysdata_list); sn_init_cpei_timer(); #ifdef CONFIG_PROC_FS register_sn_procfs(); #endif - for (i = 0; i < PCI_BUSES_TO_SCAN; i++) { - sn_pci_controller_fixup(0, i); - } + /* busses are not known yet ... */ + for (i = 0; i < PCI_BUSES_TO_SCAN; i++) + sn_pci_controller_fixup(0, i, NULL); /* * Generic Linux PCI Layer has created the pci_bus and pci_dev
@@ -430,9 +466,8 @@ */ while ((pci_dev = - pci_find_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) { + pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pci_dev)) != NULL) sn_pci_fixup_slot(pci_dev); - } sn_ioif_inited = 1; /* sn I/O infrastructure now initialized */
@@ -474,3 +509,8 @@ } subsys_initcall(sn_pci_init); +EXPORT_SYMBOL_GPL(sn_pci_fixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_unfixup_slot); +EXPORT_SYMBOL_GPL(sn_pci_controller_fixup); +EXPORT_SYMBOL_GPL(sn_bus_store_sysdata); +EXPORT_SYMBOL_GPL(sn_bus_free_sysdata);
arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ec5e7a01c5b55f328e0dff8a45b012bc1024e84e) renamed from arch/ia64/sn/kernel/irq.c (mode:100644 sha1:ee4ac68d745f2c91e39d10e2f82cf624cf655edd) +3 -3
@@ -284,7 +284,6 @@ int cpu = nasid_slice_to_cpuid(nasid, slice); pci_dev_get(pci_dev); - sn_irq_info->irq_cpuid = cpu; sn_irq_info->irq_pciioinfo = SN_PCIDEV_INFO(pci_dev);
@@ -305,15 +304,16 @@ return; sn_irq_info = SN_PCIDEV_INFO(pci_dev)->pdi_sn_irq_info; - if (!sn_irq_info || !sn_irq_info->irq_irq) + if (!sn_irq_info || !sn_irq_info->irq_irq) { + kfree(sn_irq_info); return; + } unregister_intr_pda(sn_irq_info); spin_lock(&sn_irq_info_lock); list_del_rcu(&sn_irq_info->list); spin_unlock(&sn_irq_info_lock); call_rcu(&sn_irq_info->rcu, sn_irq_info_free); - pci_dev_put(pci_dev); }
arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9813da56d3113c41fb1b413cfa4c7963cda316cb) renamed from arch/ia64/sn/pci/pcibr/pcibr_provider.c (mode:100644 sha1:9bc4de4a3ec0778aa4724e75457bd253a6b1e45e) +37 -0
@@ -18,6 +18,40 @@ #include "xtalk/xwidgetdev.h" #include "xtalk/hubdev.h" +int +sal_pcibr_slot_enable(struct pcibus_info *soft, int device, void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_ENABLE, (u64) busnum, + (u64) device, (u64) resp, 0, 0, 0, 0); + + return (int)ret_stuff.v0; +} + +int +sal_pcibr_slot_disable(struct pcibus_info *soft, int device, int action, + void *resp) +{ + struct ia64_sal_retval ret_stuff; + uint64_t busnum; + + ret_stuff.status = 0; + ret_stuff.v0 = 0; + + busnum = soft->pbi_buscommon.bs_persist_busnum; + SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_SLOT_DISABLE, + (u64) busnum, (u64) device, (u64) action, + (u64) resp, 0, 0, 0); + + return (int)ret_stuff.v0; +} + static int sal_pcibr_error_interrupt(struct pcibus_info *soft) { struct ia64_sal_retval ret_stuff;
@@ -187,3 +221,6 @@ return 0; } + +EXPORT_SYMBOL_GPL(sal_pcibr_slot_enable); +EXPORT_SYMBOL_GPL(sal_pcibr_slot_disable);
drivers/pci/hotplug/Kconfig (mode:100644 sha1:9c4a39ee89b57b5065f575b86b2798ed31dc2fc6) renamed from drivers/pci/hotplug/Kconfig (mode:100644 sha1:1a4d4ca2a4dc386d3f63189ecc9cf30f806b947e) +3 -2
@@ -187,9 +187,10 @@ config HOTPLUG_PCI_SGI tristate "SGI PCI Hotplug Support" - depends on HOTPLUG_PCI && IA64_SGI_SN2 + depends on HOTPLUG_PCI && (IA64_SGI_SN2 || IA64_GENERIC) help - Say Y here if you have an SGI IA64 Altix system. + Say Y here if you want to use the SGI Altix Hotplug + Driver for PCI devices. When in doubt, say N.
drivers/pci/hotplug/Makefile (mode:100644 sha1:24ed904514c5043095dfe484877dd38cdfa0f3b6) renamed from drivers/pci/hotplug/Makefile (mode:100644 sha1:93c120ddbd395f5ddeddcc3cc5d52bd31a47dcc4) +1 -0
@@ -14,6 +14,7 @@ obj-$(CONFIG_HOTPLUG_PCI_SHPC) += shpchp.o obj-$(CONFIG_HOTPLUG_PCI_RPA) += rpaphp.o obj-$(CONFIG_HOTPLUG_PCI_RPA_DLPAR) += rpadlpar_io.o +obj-$(CONFIG_HOTPLUG_PCI_SGI) += sgi_hotplug.o pci_hotplug-objs := pci_hotplug_core.o
drivers/pci/hotplug/sgi_hotplug.c (mode:100644 sha1:5ab8f0fa11bd462c6c97e7a1c1c4a47263698dd7) renamed from dev/null (tree:f8014bc6a8330d22e2dbb0cac34d87e8a5352278) +598 -0
598 lines not on this page · raw message
include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:c1bf22505ff339ea7a5255ec9a8a60031397e9a8) renamed from include/asm-ia64/sn/pcibr_provider.h (mode:100644 sha1:0abf539cb0414d752fcf14d9318bf1fa00278d32) +4 -0
@@ -151,4 +151,8 @@ extern int pcibr_ate_alloc(struct pcibus_info *, int); extern void pcibr_ate_free(struct pcibus_info *, int); extern void ate_write(struct pcibus_info *, int, int, uint64_t); +extern int sal_pcibr_slot_enable(struct pcibus_info *soft, int device, + void *resp); +extern int sal_pcibr_slot_disable(struct pcibus_info *soft, int device, + int action, void *resp); #endif
include/asm-ia64/sn/pcidev.h (mode:100644 sha1:49711d00ad04f77d90467daf33bfd13d549e2109) renamed from include/asm-ia64/sn/pcidev.h (mode:100644 sha1:9610fcc635450e57bf9be693fb8fa5e22eb60be1) +6 -0
@@ -23,6 +23,8 @@ #define SN_PCIBUS_BUSSOFT(pci_bus) \ ((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) +#define SN_PCIBUS_BUSSOFT_INFO(pci_bus) \ + (struct pcibus_info *)((struct pcibus_bussoft *)(PCI_CONTROLLER((pci_bus))->platform_data)) /* * Given a struct pci_dev, return the sn pcibus_bussoft struct. Note * that this is not equivalent to SN_PCIBUS_BUSSOFT(pci_dev->bus) due
@@ -56,6 +58,10 @@ extern void sn_irq_fixup(struct pci_dev *pci_dev, struct sn_irq_info *sn_irq_info); extern void sn_irq_unfixup(struct pci_dev *pci_dev); +extern void sn_pci_controller_fixup(int segment, int busnum, + struct pci_bus *bus); +extern void sn_bus_store_sysdata(struct pci_dev *dev); +extern void sn_bus_free_sysdata(void); extern void sn_pci_fixup_slot(struct pci_dev *dev); extern void sn_pci_unfixup_slot(struct pci_dev *dev); extern void sn_irq_lh_init(void);