uadk: fork has issue about uadk HOT 5 CLOSED

linaro commented on July 3, 2024

uadk: fork has issue

from uadk.

Comments (5)

gaozhangfei commented on July 3, 2024

update in https://github.com/Linaro/uadk/tree/nginx-fork

zip and sec has similar issue:

zip: fork in the middle of test

bad status(ctx_st=0x0, s=0x41, t=3)

[59987.624846] {3}[Hardware Error]: Hardware error from APEI Generic Hardware Error Source: 0
[59987.633076] {3}[Hardware Error]: event severity: recoverable
[59987.633080] {3}[Hardware Error]:  Error 0, type: recoverable
[59987.633082] {3}[Hardware Error]:   section_type: PCIe error
[59987.633083] {3}[Hardware Error]:   version: 4.0
[59987.633084] {3}[Hardware Error]:   command: 0x0006, status: 0x0010
[59987.633086] {3}[Hardware Error]:   device_id: 0000:75:00.0
[59987.633087] {3}[Hardware Error]:   slot: 0
[59987.633089] {3}[Hardware Error]:   secondary_bus: 0x00
[59987.633090] {3}[Hardware Error]:   vendor_id: 0x19e5, device_id: 0xa250
[59987.633092] {3}[Hardware Error]:   class_code: 120000

sec:
hpre: fork issue in the middle of test

$ test_hisi_hpre
wd_request_ctx ctx->fd=3
wd_request_ctx ctx->fd=4
<< test finish!
parent pid=30643
child pid=30646
wd_request_ctx ctx->fd=5
wd_request_ctx ctx->fd=6
wd_release_ctx ctx->fd=3
wd_release_ctx ctx->fd=4
HPRE do rsa fail!done=0x2, etype=0x4
Proc-30646, T-30647:hpre rsa-gen 0th status=0 fail!
HPRE do rsa fail!done=0x2, etype=0x4
Proc-30646, T-30648:hpre rsa-gen 0th status=0 fail!
<< test finish!

from uadk.

gaozhangfei commented on July 3, 2024

The issue is forked mm duplicate the old mm, as well as pasid.
dup_mm: memcpy(mm, oldmm, sizeof(*mm));

So iommu_sva_alloc_pasid returns the old pasid, instead of ioasid_alloc a new pasid.
if (mm->pasid)
ioasid_get(mm->pasid);

https://lore.kernel.org/linux-iommu/[email protected]/
can solve the issue.

from uadk.