Re: [PATCH] cxl/mbox: Bound the output payload allocation to mailbox payload size

[Davidlohr Bueso <dave@stgolabs.net>]

On Thu, 11 Jun 2026, Richard Cheng wrote:

>CXL_MEM_SEND_COMMAND bounds the user's in.size to the mailbox payload
>size but leaves out.size unbounded, then cxl_mbox_cmd_ctor() calls
>kvzalloc(out.size). A large out.size drives a huge allocation, even
>above INT_MAX it WARNS and taints, on kernel with panic_on_warn=1, it
>will panic.
>The transport __cxl_pci_mbox_send_cmd() already clamps the response copy
>to min(out.size, payload_size, device len), so the bound buffer is never
>written beyond payload_size. Clamp the allocation to payload_size too,
>matching the RAW path.
>
>With the following reproducer[1] , we'll get error logs [2].
>[1]:
>"""
>        #include <fcntl.h>
>        #include <stdint.h>
>        #include <stdio.h>
>        #include <sys/ioctl.h>
>
>        #define CXL_MEM_SEND_COMMAND        _IOWR(0xCE, 2, struct cxl_send_comma
>        #define CXL_MEM_COMMAND_ID_IDENTIFY 1
>
>        struct cxl_send_command {
>                uint32_t id, flags;
>                union { struct { uint16_t opcode, rsvd; } raw; uint32_t rsvd; };
>                uint32_t retval;
>                struct { uint32_t size, rsvd; uint64_t payload; } in;
>                struct { uint32_t size, rsvd; uint64_t payload; } out;
>        };
>
>        int main(void)
>        {
>                static unsigned char buf[512];
>                struct cxl_send_command c = {
>                        .id          = CXL_MEM_COMMAND_ID_IDENTIFY, /* any enabl
>                        .out.size    = 0x80000000,                  /* > INT_MAX
>                        .out.payload = (uint64_t)(uintptr_t)buf,
>                };
>                int fd = open("/dev/cxl/mem0", O_RDWR);
>
>                return ioctl(fd, CXL_MEM_SEND_COMMAND, &c);
>        }

Nit: Maybe not needed in the changelog.

>"""
>[2]:
>  [ 3675.127839] ------------[ cut here ]------------
>  [ 3675.127841] WARNING: mm/slub.c:6841 at __kvmalloc_node_noprof+0x534/0x818,
>  CPU#131: cxl_repro_outsi/4668
>  [ 3675.127853] Modules linked in: nft_masq nft_ct nft_reject_ipv4
>  nf_reject_ipv4 nft_reject act_csum cls_u32 sch_htb nft_chain_nat nf_nat
>  nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 nf_tables bridge stp llc qrtr
>  cfg80211 binfmt_misc nls_iso8859_1 acpi_power_meter nvidia_cspmu acpi_ipmi
>  ipmi_ssif coresight_trbe ipmi_devintf sbsa_gwdt dax_hmem arm_smmuv3_pmu
>  coresight arm_cspmu_module arm_spe_pmu ast nvidia_t410_cmem_latency_pmu
>  nvidia_t410_c2c_pmu ipmi_msghandler cppc_cpufreq mlx5_ib macsec ib_uverbs
>  mlx5_fwctl mlx5_dpll sch_fq_codel dm_multipath nvme_fabrics efi_pstore
>  nfnetlink dmi_sysfs ip_tables x_tables autofs4 ib_core btrfs libblake2b raid10
>  raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor
>  raid6_pq raid1 raid0 linear ghash_ce gf128mul sm4_ce_gcm mlx5_core nvme
>  sm4_ce_ccm nvme_core mlxfw sm4_ce tls nvme_keyring igb sm4_ce_cipher sm4
>  arm_smccc_trng i2c_algo_bit nvme_auth psample i2c_tegra aes_neon_bs aes_ce_blk
>  [ 3675.127894] CPU: 131 UID: 0 PID: 4668 Comm: cxl_repro_outsi Tainted: G
>    W           7.1.0-rc7-cxltest #1 PREEMPT(full)
>  [ 3675.127897] Tainted: [W]=WARN
>  [ 3675.127898] Hardware name:  , BIOS buildbrain-gcid-sbios-45820373-12 Fri
>  Jun  5 07:54:44 AM UTC 2026
>  [ 3675.127899] pstate: 23400009 (nzCv daif +PAN -UAO +TCO +DIT -SSBS BTYPE=--)
>  [ 3675.127900] pc : __kvmalloc_node_noprof+0x534/0x818
>  [ 3675.127902] lr : __kvmalloc_node_noprof+0x520/0x818
>  [ 3675.127903] sp : ffff800102c2fb90
>  [ 3675.127903] x29: ffff800102c2fbc0 x28: ffff0001911d5000 x27:
>  d8eaa73777d13b74
>  [ 3675.127905] x26: 0000000000000001 x25: ffffa73777d13b74 x24:
>  0000000000000000
>  [ 3675.127907] x23: 00000000ffffffff x22: 0000000000000dc0 x21:
>  00000000000029c0
>  [ 3675.127908] x20: 0000000000000000 x19: 0000000080000000 x18:
>  ffff800125340040
>  [ 3675.127910] x17: 0000000000000000 x16: 0000000000000000 x15:
>  0000ffffd627bed8
>  [ 3675.127911] x14: 0000000000000000 x13: 0000000000000000 x12:
>  0000000000000000
>  [ 3675.127913] x11: 0000000000000000 x10: 0000000000000000 x9 :
>  0000000000000000
>  [ 3675.127914] x8 : 0000000000000000 x7 : 0000000000000000 x6 :
>  0000000000000000
>  [ 3675.127916] x5 : 0000000000000000 x4 : 0000000000000000 x3 :
>  0000000000000000
>  [ 3675.127917] x2 : 0000000000000000 x1 : 0000000000000000 x0 :
>  000000007fffffff
>  [ 3675.127919] Call trace:
>  [ 3675.127919]  __kvmalloc_node_noprof+0x534/0x818 (P)
>  [ 3675.127921]  cxl_send_cmd+0x514/0x7e0
>  [ 3675.127926]  cxl_memdev_ioctl+0x7c/0xe0
>  [ 3675.127928]  __arm64_sys_ioctl+0x4a4/0xbc8
>  [ 3675.127931]  invoke_syscall.constprop.0+0xac/0x100
>  [ 3675.127934]  do_el0_svc+0x4c/0x100
>  [ 3675.127935]  el0_svc+0x50/0x2b0
>  [ 3675.127938]  el0t_64_sync_handler+0xc0/0x108
>  [ 3675.127940]  el0t_64_sync+0x1b8/0x1c0
>  [ 3675.127942] ---[ end trace 0000000000000000 ]---

My cxl syzkaller also ran into this issue and I had a similar patch
I never sent out.

Reviewed-by: Davidlohr Bueso <dave@stgolabs.net>

>
>Fixes: 4faf31b43468 ("cxl/mbox: Move mailbox and other non-PCI specific infrastructure to the core")
>Reviewed-by: Kai-Heng Feng <kaihengf@nvidia.com>
>Reviewed-by: Koba Ko <kobak@nvidia.com>
>Signed-off-by: Richard Cheng <icheng@nvidia.com>
>---