Re: [PATCH] x86/pci: Add a break condition when enabling BAR

[Bandan Das]

Christian KÃnig <christian.koenig@xxxxxxx> writes:

> Hi Bandas,
>
> thanks for the patch, but this is a known issue with a fix already on
> the way into the next -rc.

Oh great! Thank you, have a pointer to the patch so that I can test ?

> Regards,
> Christian.
>
> Am 07.12.2017 um 09:00 schrieb Bandan Das:
>> On an old flaky system with AMD Opteron 6320, boot hangs
>> with the following trace since commit fa564ad9:
>>
>> [   28.181012] Hardware name: HP ProLiant DL385p Gen8, BIOS A28 09/03/2014
>> [   28.184022] RIP: 0010:lock_acquire+0xd5/0x1e0
>> [   28.185010] RSP: 0018:ffffb7ad818c39a8 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff11
>> [   28.187010] RAX: ffffa074fb39b140 RBX: 0000000000000246 RCX: 0000000000000000
>> [   28.189014] RDX: ffffffffb20a55a9 RSI: 0000000000040009 RDI: 0000000000000246
>> [   28.191012] RBP: 0000000000000000 R08: 0000000000000006 R09: 0000000000000000
>> [   28.193020] R10: 0000000000000001 R11: 00000000dac664b5 R12: 0000000000000000
>> [   28.196013] R13: 0000000000000000 R14: 0000000000000001 R15: 0000000000000000
>> [   28.197011] FS:  0000000000000000(0000) GS:ffffa074fbd00000(0000) knlGS:0000000000000000
>> [   28.201014] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> [   28.201014] CR2: 0000000000000000 CR3: 00000003b6e10000 CR4: 00000000000406e0
>> [   28.205008] Call Trace:
>> [   28.205013]  ? request_resource_conflict+0x19/0x40
>> [   28.207013]  _raw_write_lock+0x2e/0x40
>> [   28.209008]  ? request_resource_conflict+0x19/0x40
>> [   28.209010]  request_resource_conflict+0x19/0x40
>> [   28.212013]  pci_amd_enable_64bit_bar+0x103/0x1a0
>> [   28.213025]  pci_fixup_device+0xd4/0x210
>> [   28.213025]  pci_setup_device+0x193/0x570
>> [   28.215010]  ? get_device+0x13/0x20
>> [   28.217008]  pci_scan_single_device+0x98/0xd0
>> [   28.217011]  pci_scan_slot+0x90/0x130
>> [   28.219010]  pci_scanild_bus_extend+0x3a/0x270
>> [   28.321008]  acpi_pci_root_create+0x1a9/0x210
>> [   28.321014]  ? pci_acpi_scan_root+0x135/0x1b0
>> [   28.324013]  pci_acpi_scan_root+0x15f/0x1b0
>> [   28.325008]  acpi_pci_root_add+0x283/0x560
>> [   28.325014]  ? acpi_match_device_ids+0xc/0x20
>> [   28.327013]  acpi_bus_attach+0xf9/0x1c0
>> [   28.329008]  acpi_bus_attach+0x82/0x1c0
>> [   28.329044]  acpi_bus_attach+0x82/0x1c0
>> [   28.331010]  acpi_bus_scan+0x47/0xa0
>> [   28.333008]  acpi_scan_init+0x12d/0x28d
>> [   28.333013]  ? bus_register+0x208/0x280
>> [   28.333013]  acpi_init+0x30f/0x36f
>> [   28.335010]  ? acpi_sleep_proc_init+0x24/0x24
>> [   28.337013]  do_one_initcall+0x4d/0x19c
>> [   28.337013]  ? do_early_param+0x29/0x86
>> [   28.340013]  kernel_init_freeable+0x209/0x2a4
>> [   28.341008]  ? set_debug_rodata+0x11/0x11
>> [   28.341011]  ? rest_init+0xc0/0xc0
>> [   28.343013]  kernel_init+0xa/0x104
>> [   28.345008]  ret_from_fork+0x24/0x30
>> [   28.345010] Code: 24 08 49 c1 e9 09 49 83 f1 01 41 83 e1 01 e8 73
>> e4 ff ff 65 48 8b 04 25 c0 d4 00 00 48 89 df c7 80 fc 0c 00 00 00 00
>> 00 00 57 9d <0f> 1f 44 00 00 48 83 c4 30 5b 5d 41 5c 41 5d 41 5e 41 5f
>> c3 65
>>
>> Since request_resource() will unconditionally return a conflict for invalid
>> regions, there will be no way to break out of the loop when enabling 64bit BAR.
>> Add checks and exit the loop in these cases without attempting to enable
>> BAR.
>>
>> Signed-off-by: Bandan Das <bsd@xxxxxxxxxx>
>> ---
>>   arch/x86/pci/fixup.c | 7 ++++++-
>>   1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/pci/fixup.c b/arch/x86/pci/fixup.c
>> index 1e996df..8933a1b 100644
>> --- a/arch/x86/pci/fixup.c
>> +++ b/arch/x86/pci/fixup.c
>> @@ -696,8 +696,13 @@ static void pci_amd_enable_64bit_bar(struct pci_dev *dev)
>>   	res->end = 0xfd00000000ull - 1;
>>     	/* Just grab the free area behind system memory for this */
>> -	while ((conflict = request_resource_conflict(&iomem_resource, res)))
>> +	while ((conflict = request_resource_conflict(&iomem_resource, res))) {
>> +		if ((res->start > res->end) ||
>> +		    (res->start < iomem_resource.start) ||
>> +		    (res->end > iomem_resource.end))
>> +			break;
>>   		res->start = conflict->end + 1;
>> +	}
>>     	dev_info(&dev->dev, "adding root bus resource %pR\n", res);
>>