LinuxLists.cc - [mkp-scsi:for-next 67/132] drivers/scsi/elx/libefc/efc_els.c:73 efc_els_io_alloc

2022-01-10 07:03:04

Subject: [mkp-scsi:for-next 67/132] drivers/scsi/elx/libefc/efc_els.c:73 efc_els_io_alloc_size() warn: sleeping in atomic context

tree:
git.kernel.org/pub/scm/linux/kernel/git/mkp/scsi.git">https://git.kernel.org/pub/scm/linux/kernel/git/mkp/scsi.git for-next
d9337a2469cac51cca6932
c658e02676821ca834b56c [67/132] scsi: efct: Don't pass GFP_DMA to dma_alloc_coherent()
(https://download.01.org/0day-ci/archive/20220108/[email protected]/config)
9.3.0-22) 9.3.0
kindly add following tag as appropriate
test robot <[email protected]>
Carpenter <[email protected]>
efc_els_io_alloc_size() warn: sleeping in atomic context
ibefc/efc_els.c
2021-06-01 40 struct efc_els_io_req *
2021-06-01 41 efc_els_io_alloc_size(struct efc_node *node, u32 reqlen, u32 rsplen)
2021-06-01 42 {
2021-06-01 43 struct efc *efc;
2021-06-01 44 struct efc_els_io_req *els;
2021-06-01 45 unsigned long flags = 0;
2021-06-01 46
2021-06-01 47 efc = node->efc;
2021-06-01 48
2021-06-01 49 spin_lock_irqsave(&node->els_ios_lock, flags);
^^^^^^^^^^^^^^^^^
2021-06-01 50
2021-06-01 51 if (!node->els_io_enabled) {
2021-06-01 52 efc_log_err(efc, "els io alloc disabled\n");
2021-06-01 53 spin_unlock_irqrestore(&node->els_ios_lock, flags);
2021-06-01 54 return NULL;
2021-06-01 55 }
2021-06-01 56
2021-06-01 57 els = mempool_alloc(efc->els_io_pool, GFP_ATOMIC);
2021-06-01 58 if (!els) {
2021-06-01 59 atomic_add_return(1, &efc->els_io_alloc_failed_count);
2021-06-01 60 spin_unlock_irqrestore(&node->els_ios_lock, flags);
2021-06-01 61 return NULL;
2021-06-01 62 }
2021-06-01 63
2021-06-01 64 /* initialize refcount */
2021-06-01 65 kref_init(&els->ref);
2021-06-01 66 els->release = _efc_els_io_free;
2021-06-01 67
2021-06-01 68 /* populate generic io fields */
2021-06-01 69 els->node = node;
2021-06-01 70
2021-06-01 71 /* now allocate DMA for request and response */
2021-06-01 72 els->io.req.size = reqlen;
2021-06-01 @73 els->io.req.virt = dma_alloc_coherent(&efc->pci->dev, els->io.req.size,
Hellwig 2021-12-14 74 &els->io.req.phys, GFP_KERNEL);
^^^^^^^^^^
2021-06-01 75 if (!els->io.req.virt) {
2021-06-01 76 mempool_free(els, efc->els_io_pool);
2021-06-01 77 spin_unlock_irqrestore(&node->els_ios_lock, flags);
2021-06-01 78 return NULL;
2021-06-01 79 }
2021-06-01 80
2021-06-01 81 els->io.rsp.size = rsplen;
2021-06-01 82 els->io.rsp.virt = dma_alloc_coherent(&efc->pci->dev, els->io.rsp.size,
Hellwig 2021-12-14 83 &els->io.rsp.phys, GFP_KERNEL);
^^^^^^^^^^
2021-06-01 84 if (!els->io.rsp.virt) {
2021-06-01 85 dma_free_coherent(&efc->pci->dev, els->io.req.size,
2021-06-01 86 els->io.req.virt, els->io.req.phys);
2021-06-01 87 mempool_free(els, efc->els_io_pool);
2021-06-01 88 els = NULL;
2021-06-01 89 }
2021-06-01 90
2021-06-01 91 if (els) {
2021-06-01 92 /* initialize fields */
2021-06-01 93 els->els_retries_remaining = EFC_FC_ELS_DEFAULT_RETRIES;
2021-06-01 94
2021-06-01 95 /* add els structure to ELS IO list */
2021-06-01 96 INIT_LIST_HEAD(&els->list_entry);
2021-06-01 97 list_add_tail(&els->list_entry, &node->els_ios_list);
2021-06-01 98 }
2021-06-01 99
2021-06-01 100 spin_unlock_irqrestore(&node->els_ios_lock, flags);
2021-06-01 101 return els;
2021-06-01 102 }
Service, Intel Corporation
.org/hyperkitty/list/kbuild-all@lists.01.org">https://lists.01.org/hyperkitty/list/[email protected]

2022-01-10 07:11:10

by Christoph Hellwig

[permalink] [raw]

Subject: Re: [mkp-scsi:for-next 67/132] drivers/scsi/elx/libefc/efc_els.c:73 efc_els_io_alloc_size() warn: sleeping in atomic context

So while this warning is correct, it is not new. GFP_DMA can sleep
just like GFP_KERNEL. Someone was already trying to give this a spin
on the linux-scsi list, but we really need maintainer helper here.

2022-01-10 16:39:42

by James Smart

[permalink] [raw]

Subject: Re: [mkp-scsi:for-next 67/132] drivers/scsi/elx/libefc/efc_els.c:73 efc_els_io_alloc_size() warn: sleeping in atomic context

On 1/9/2022 11:11 PM, Christoph Hellwig wrote:
> So while this warning is correct, it is not new. GFP_DMA can sleep
> just like GFP_KERNEL. Someone was already trying to give this a spin
> on the linux-scsi list, but we really need maintainer helper here.

Yep - I'll go look at it.

-- james